THE TRANSFORMATIONAL-GENERATIVE PARADIGM

AMSTERDAM STUDIES IN THE THEORY AND

HISTORY OF LINGUISTIC SCIENCE

E. F. K. KOERNER, General Editor

Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board

Henning Andersen (Albany, N.Y.); Raimo Anttila (Los Angeles) Tomaz V. Gamkrelidze (Tiflis); Klaus J. Kohler (Kiel)

J. Peter Maher (Hamburg);Ernst Pulgram (Ann Arbor, Mich.) E. Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Honolulu)

Volume 1

E. F. K. Koerner, ed.

The Transformational-Generative Paradigm and Modern Linguistic Theory

THE

TRANSFORMATIONAL-GENERATIVE PARADIGM

AND

MODERN LINGUISTIC THEORY

edited by

E. F. K. KOERNER

with the assistance of

JOHN ODMARK and J. HOWARD SHAW

AMSTERDAM / JOHN BENJAMINS B.V.

1975

© Copyright 1975 - John Benjamins B.V. ISBN 90 272 0901 4/90 272 0902 2

No part of this book may be reproduced in any form, by print, photoprint, microfilm or any other means, without written permission from the publisher.

FOREWORD

This first volume of the "Current Trends in Linguistic Theory" (CILT) series reflects the fact that the possibilities in theory con­struction allow for a much wider spectrum than students of linguistics have perhaps been led to believe. It consists of articles by scholars of differing generations and widely varying academic persuasions: some have received their initiation to the trade within the framework of transformational-generative grammar, some in one or the other struc­turalist mould, yet others in the philology and linguistics of part­icular languages and language families. They all share, however, some doubts concerning characteristic attitudes and procedures of present-day 'mainstream linguistics'. All want, not a uniformity of ideolog­ical stance, but a union of individualists working towards the advance­ment of theory and empirical accountability.

In Spring 1973 J. Peter Maher and I first conceived the idea of a volume of the present nature; we regret that its publication has taken so long. Unfortunately, Professor Maher's introduction, originally in­tended to appear in this volume, could not be included for technical reasons. Readers interested in seeing this introductory article are kindly referred to a forthcoming issue of Historiographia Linguistica.

I would like to express my gratitude to the following people -in addition to the contributors to this volume - for their assistance and patience: Mrs. Clara Rothmeier for typing the manuscript; Messrs. Howard Shaw and John Odmark for their advice on editorial matters and proofreading; Dr. Anne Betten for helping me with the index, and, last

VI FOREWORD

but not least, Mr. J. L. Benjamins for his continued interest and finan­cial support.

A Portuguese translation of a number of articles included in the present volume under the editorship of Professor Marcelo Dascal of the University of Campinas, Brazil, is in preparation.

Amsterdam, 5 March 1975 E. F. K. K.

Technical Note: Because of the absence of a sign for the schwa

in phonetic transcription in the italics element at hand, the following sign has been used to designate the e in citation forms: 3. Please note also that, on several occasions, the di­acritic for length has been placed somewhat too far above the vowel in question, e.g., a, e, o, etc.

C O N T E N T S

Preface v

I. SYNTAX AND SEMANTICS

Dwight Bol inger: Meaning and Form: Some fallacies of asemantic grammar 3

Adam Makkai: Stratificational Solutions to Unbridgeable Gaps in Transformational-Generative Grammar 37

Fred . Peng: Non-Uniqueness in the Treatment of the Separabil­ity of Semantics and Syntax in Compound Expressions 87

II. PHONOLOGY AND MORPHOLOGY

Hsin-I Hsieh: How Generative is Phonology? (On listing phonolog­ical surface forms in the lexicon) '109

Michael Kenstowicz: Rule Application in Pre-Generative American Phonology 145

Leonhard Lipka: Prolegomena to "Prolegomena to a Theory of Word-Formation":, A reply to Morris Halle 175

Royal Skousen: On the Nature of Morphophonemic Alternation . . .185 Danny D. Steinberg and Robert K. Krohn: The Psychological Valid­

ity of Chomsky and Halle's Vowel Shift Rule 233

III. LINGUISTIC THEORY AND THE PHILOSOPHY OF LANGUAGE

Raimo Anttila: Generalization, Abduction, Evolution, and Lan­guage 263

Bruce L. Derwing and Peter R. Harris: What is a Generative Gram­mar? 297

Edward R. Maxwell: On the Inadequacy of the Tree as a Formal Con­cept in Linguistic Analyses 315

VIII TABLE OF CONTENTS

Wal burga von Raffi er Engel: Language Acquisition and Common Sense 321 Uhlan V. Slagle: On the Nature of Language and Mind 329

IV. EPISTEMOLOGY AND HISTORY OF LINGUISTICS

Lyle Campbell: Epistemologiaal dilemmas and the Transformational-Generative Paradigm 351

Dell Hymes: Pre-War Prague School and Post-War American Anthro­pological Linguistics 359

Esa Itkonen: Transformational Grammar and the Philosophy of Sci­ence 381

Biographical Notes 447 Index of Names 457

I. S Y N T A X A N D S E M A N T I C S

MEANING AND FORM SOME FALLACIES OF ASEMANTIC GRAMMAR*

DWIGHT BOLINGER

Every so often the scientific theorist finds it politic to climb down from the heights and appeal to common sense. Those of us who were parents in the 1930's well remember the theories of child-rearing that prevailed - asepsis was the ideal, the more untouched by human hands an infant was the better, and if some enterprising dealer in plastics could have found a way to wrap a fetus in cellophane it probably would have been done. Psychologists since then have well-nigh totally faced about, and a mother's love and kisses are again respectable as well as natural. Lin­guistics has seen more than one such retreat from artifici­ality. It was common sense that showed the mindlessness of twenty years ago, and that spurred the return to 'natural­ness' a decade later.

This little preamble is by way of saying that the the­sis I am going to sustain is not one that would surprise the man on the street. Tell him that if two ways of saying something differ in their words or their arrangement they will also differ in meaning, and he will show as much sur­prise as if you told him that walking in the rain is condu-

This represents a lecture given 4 April 1973 at the New York Acad­emy of Sciences and again, with modifications, at King's College, Aberdeen, Scotland, 24 April 1973.

4 DWIGHT BOLINGER cive to getting wet. Only a scientist can wrap himself up in enough sophistication to keep dry under these circumstances.

Here is how I mean to define my target. One of the prin­ciples - I could almost say fetishes - of current formal lin­guistics is the notion that underlying whatever communica­tion one human being transmits to another is a deep structure in which every relationship relevant to meaning is set forth. The communication that gets transmitted is subject to all the accidents of transmission and is therefore a distortion of the bedrock structure. Among the possible distortions are actual bifurcations - two different ways of saying the same thing, two different surface structures, mapped on to a sin­gle underlying structure by means of different transforma­tions. For instance, it was claimed for a while that the to of the infinitive and the -ing of the gerund were merely al­ternate complementizers and when you said Be likes to write and he enjoys writing any difference there was existed only between the verbs like and enjoy, not between to write and writing - the two main verbs merely selected different com­plementizers. The difference between to write and writing became part of the automatism of language.

The other side of the picture - two things the same in form but different in meaning rather than the same in mean­ing but different in form - is better known because it ap­peals more to our sense of the unusual. It is the basis of most puns, the funeral home that advertises a lay-away plan, the athletic girl who loves the sun and air (son and heir), and any number of more professional curiosities that have been invented, such as Chomsky's The shooting of the hunters was terrible. Obviously if the accidents that strike the surface can produce two different things stemming from the same deep structure, they can also produce two same things

MEANING AND FORM 5 stemming from different deep structures. I say little about this because there is little to be said. One can hardly ar­gue the fact that many sentences starting with i t , for in­stance, can be taken in more than one way - It's fun to eat differs depending on whether it refers to what is eaten or to the act of eating; and when Perry Mason says in an old movie I wonder if I wasn't expected to find out , you can take him as saying in effect, "Possibly I wasn't expected to" or "I'll bet I was expected to".

It would be hard to quarrel with the doctrine of same­ness and difference as an abstract scientific principle. The idea that things can be the same but different or different but the same is prerequisite to science - only by shutting our eyes to differences can we see that all legumes are a single family or that the gravitation of an apple hitting the earth is the same as that of a moon revolving about a planet. The problem is not in the principle but in the way linguists have sometimes interpreted it. I question whether any botanist would define his field so as to say that the variation among legumes has nothing to do with it; but lin­guists have tended to define linguistics so as to say that variation in surface structures that have the same deep structure is irrelevant to the one thing that matters most in language, namely meaning. They have insisted on absolute identity with any difference defined out of the way.

This attitude has been around a long time. It charac­terized much of the work in phonology until very recently. In dealing with the sound system of a language it is use­ful to think of an underlying system of contrastive units, phonemes or features, where a speaker, in two utterances of the same word, say, necessarily deviates within a certain range of tolerance without his hearer even being aware of

6 DWIGHT BOLINGER it. Similarly one may find an identical system being used by another speaker, but with the physical traits of each sig­nal differing slightly in ways that mark him as an individ­ual or as the speaker of a different dialect, but with each unit still having the same communicative value as before. It is not too far-fetched to claim that cases like these are identical linguistically but different sociologically. The deviations can reasonably be defined out of the field.

What happens when these notions of systemic identity and irrelevant difference are carried up the ladder into morphology and syntax? With morphology it still makes sense to think of the plurality of geese and the plurality of hens as the same entity despite the difference in ways for form­ing the plural. Also, in describing the differences between speakers we would allow that if one say eethev and another says eyethev they are still using the same word; we know the origins of both and we can see the identity of usage. We may learn something of the speakers - their social group or their individual psychology - by observing the differences, but these can be ignored linguistically, in at least some contexts, since they do not affect the content of a commu­nication. They may even be beyond the control of the speak­ers. He does not manipulate them to ring changes on his mes­sage.

Where the mischief begins is in syntax. Differences in the arrangement of words and in the presence or absence of certain elements are assumed not to count. What is supposed to matter is the underlying deep structure, which is capa­ble of producing, through transformations, divergent struc­tures that mean exactly the same thing. The motive for as­suming this is not only the search for simplicity, f or ways of stating rules or laws just once instead of again and

MEANING AND FORM 7

again, but also the yen that our modern linguist has for be­ing a psychologist. If there is such a thing as a universal deep structure, it must reflect something about the human psyche,- and many conjectures have been made about the human infant springing from the womb with his noun phrases and relative clauses all ready to light up as soon as they are plugged into a particular language.

Obviously the idea that even in syntax one could have identity within difference could not have gained currency without some empirical support. The classical case is that of the passive voice. If some differences of meaning are ignored, it is possible to say that John ate the spinach and The spinach was eaten by John are the same. They report the same event in the real world. The same entities are present and they are in the same relationship of actor and patient. But if truth value were the only criterion of iden­tity in syntax we would have to say - as some have recently been trying to say - that John sold the house to Mary and Mary bought the house from John are just as much the same as the active-passive pair, and to seek some way of deriv­ing them f a common base. Linguistic meaning covers a great deal more than reports of events in the real world. It expresses, sometimes in very obvious ways, other times in ways that are hard to ferret out, such things as what is the central part of the message as against the peripher­al part, what our attitudes are toward the person we are speaking to, how we feel about the reliability of our mes­sage, how we situate ourselves in the events we report, and many other things that make our messages not merely a rec­ital of facts but a complex of facts and comments about facts and situations.

If one wants to.believe, as I do, that in syntax there

8 DWIGHT BOLINGER is no such thing as two different surface structures with the same deep structure (that is, with the same meaning), how does one come to grips with the idea? Nobody has counted how many of these imagined cases of identity exist, so if you vanquish one there is always another one waiting for you. The only answer I know is to find the cases that have the greatest inherent plausibility, and on which the strongest claims of identity have been staked, and to take them onone by one. A sampling of such a procedure is what I offer now. But before I start on that line let me say that this is not an indictment of transformational grammar any more than of traditional grammar, for both have made the assumption of identity. Transformationalists were not the ones who first pronounced the equivalence between active and passive or between attributive adjectives and reduced clauses. It is true, however, that transformational grammar may have the most at stake because of the firmly entrenched position that the idea occupies in the system.

I start with what are for me the most difficult cases, where I am almost compelled to agree that there is no dif­ference. Almost, but not quite. I refer to the ones that are 'caused by surface transformations1. I suspect that the rea­son for virtual identity despite apparent difference is that transformations at this level are a linguistic reality and not a theoretical fiction. When a speaker echoes the ques­tion Would you like to have some tea? with Yes, I would like to have some , or Yes, I would like to, or Yes, I would, in­stead of Yes, I would like to have some tea, I believe that he would immediately understand if it were pointed out to him that his answer is an abbreviation - in fact, if one were to ask him Why didn't you say "Yes, I would like to have some tea"? I think he would reply by saying either But

MEANING AND FORM 9

that's what I said or Why should I say all that if I can say just part of it? The transformations involved are de le t ion and pronominalizat ion.

As I sa id , i t i s harder to find semantic d i f fe rences , and where one finds them they appear to be unpredictable from t h e i r inner s t ruc tu re and dependent on some larger d i s ­course r e l a t i o n s h i p , as i f an accidenta l difference in form could only produce an accidenta l difference in meaning. Some­times a f a i r l y long sentence" is open to a se r i e s of t runca­t ions among which any difference in meaning would be next to impossible to f ind. For example, in answer to the ques­t ion Who might have done it better than John might have done it? one can have the following:

Joe might have done i t b e t t e r than John might have done i t . Joe might have done i t b e t t e r than John might have done. Joe might have done i t b e t t e r than John might have. Joe might have done i t b e t t e r than John might. Joe might have done i t b e t t e r than John. Joe might have done i t b e t t e r . Joe might have. Joe might. Joe.«

(The second of these is suspect; I am not sure that done is in the same function as in the original.) Of course there is a FUNCTIONAL difference here that those of us with an ear for intonation can quickly detect: all the answers use the same intonation curve, and when the answer is reduced to the monosyllable Joe it becomes a little awkward; the three-word answer Joe might have is about the lowest one can go in syllabic weight to accommodate the intonation comfort­ably. To continue with the matter of intonation for a mo­ment, one can see a clear functional difference in two re­plies to the invitation Why don't you go shopping with me?

10 DWIGHT BOLINGER

One,

i s p o l i t e and w i s t f u l . The o t h e r ,

want don't

I to go shopping with you.

is unmistakably rude. If the full answer is used on the first intonation,

one is more strongly impelled to finish with but I can't, and the whole reply sounds a bit brusque. I believe that the effect of repeating the full sentence in a case like this is one either of mocking the original speaker or pretending that he is so dense that it is necessary to repeat all words to make him understand. Combining the mockery with the low terminal pitch, which is normal intonationally because with no new information there are no pitch rises, gives a scorn­ful finality just right for repudiation. Of course, if what is repeated does not echo another person's actual words, one has unadulterated finality. Compare the wistful intonation of

Ma wish mine nice I were as as 's.

ry

MEANING AND FORM 11

with the finality of , . . every bit . Ma

have work it . mine . as nice I to so that will be as . ry's is. in which the addition of is is more appropriate. The effect of finality can be seen when complements are retained at the final low pitch: H needs money and he means to have money; You'll accept what I tell you to do and you'll do what I tell you to do.

There are other effects that can be got by repetition besides the extra bulk for intonational purposes and the mocking echo. One is the ancient device of plurality. If I say She bought a ved dress, a green one, and a blue one, I give you a mere list of her purchases. But if I say She bought a red dress, she bought a green dress, and she bought a blue dress, you will infer that she bought excessively. One difference in form here comes from pronominalization, not deletion; but the two are the same in their main effects, namely in shortening and in not repeating the same words.1 Another effect is that of separation. This is found when the element to be removed is not deleted but pronominalized. If I say George came in the room and turned off the lights, ordinarily I would be taken to mean that George performed the actions in sequence, as two parts of a complex plan. But if I say George c in the room and he turned off the lights, it 'is probably either two separate events that are conjoined or two linked events conceptually separated (for example, his coming into the room is reported, but his turn­ing off the lights is complained about). As in so many oth­er places, the to of the infinitive behaves in this same way, much like a pronoun. The phrase the ability to read and write letters is more likely to be interpreted as 'a letter-

12 DWIGHT BOLINGER reading-and-writing ability' than as 'a reading ability plus a letter-writing ability'. But if the to is retained, the ability to read and to write le t ters , the probabilities are reversed - the to helps to split read off from write-letters and sets it up as an intransitive verb. Still anoth­er effect of repetition is admonitory, when used with some­one's name, as in Mary wants to eat my soup hut Mary isn't going to get the chance. I suspect that this is a side ef­fect of the repetition of a personal name as a kind of re­proof.

Except for the admonitory use, the effects of repeti­tion mentioned thus far seem to be more or less systematic. That is, we regularly have the option of repeating some­thing in full to get extra intonational weight, to suggest plurality, or to indicate that a conjunction applies to a whole sentence rather than part of one. But admonitory rep­etition picks up a meaning through casual association - we repeat a person's name as a form of reproof in direct ad­dress, and manage to carry a suggestion of it over into in­direct address. There are other such casual associations in our repeating or not repeating an element of a sentence. At the time that Dwight Eisenhower was suffering from heart attacks, a cynical cartoon of Richard Nixon was published which pictured the two men standing at the foot of a stair­way and Nixon saying to Eisenhower, Race you to the toy of the stairs. The omission of the subject I and the auxilia­ry is common in such invitations. I suspect it comes by way of a blend with the imperative, which might also be used as an invitation in such a context, Race me to the top of the stairs. In any case, Race you to the top of the stairs is unambiguously an invitation; I'll race you to the top of the stairs is not. The regular deletion of the subject in

MEANING AND FORM 13 the imperative gives, by reversal, an admonitory effect similar to the one just mentioned in connection with proper names, when the subject you is included: You do as I say is stronger than Do as I say, and Come heve, you is stronger still than Come heve. (Of course when the you is a vocative it merely contrasts with some other you, for example You sit heve, Jane, and you sit heve, Mavy,) There are similar deletions in questions which have picked up special mean­ings. If I am sampling a food and say to you Like a taste? you are apt to interpret my invitation as less ceremonious and hence more sincere than if I said Would you like a taste?

It is obvious from these examples that deletions and pronominalizations may be specialized in function. An in­stance of particular specialization is the use of answers with deleted main verbs as strong affirmations or denials. In answer to Do you claim that you weve theve on the night of August 22? one may say, naturally, just Yes, or Yes, I claim etc., but the answer I do is the most positive. The KINDS of differences in meaning that one finds with struc­tures that differ only by deletion or pronominalization may not be the same, or may not be as striking, as those involv­ing change of order or change of lexical material. But these contrasts are obviously being exploited and it is not too far-fetched to suppose that even here there is some poten­tial difference in function whenever there is a difference in form.

As I said, transformations at this level I believe are a linguistic reality. Speakers actually PERFORM, in some sense, the operations of deletion and pronominalization. Where transformations have been set up at supposedly deeper levels the claim of identity-despite-difference is easier to disprove. No better example could be found to start with

14 DWIGHT BOLINGER than the passive voice with all its battle scars. Even the most ardent advocates of underlying identity have been obliged to retreat from their front-line position on this one, and concede that there is some kind of difference in emphasis, a way of highlighting certain elements so that ac­tive and passive are not always interchangeable. If one asks the question What did John do? one hardly expects the an­swer *The spinach was eaten by John. What did John do? is talking about John and the answer talks about the spinach. But there is more than this difference between active and passive, as I have found in working with prepositional verbs in English (Bolinger 1973a). Why is it that one can say both The dog walked under the bridge and Generations of lovers have walked under the bridge but The bridge has been walked under by generations of lovers strikes us as at least tol­erable while *The bridge was walked under by the dog seems absurd? Or why is it that the passive of Nobody is to camp beside this lake, This lake is not to be camped beside by anybody! seems acceptable, but the passive of My sister camped beside the lake, *The lake was camped beside by my sister, is peculiar? After giving a set of examples like these to a class of seventy first-year college students, I found that when I had thrown out all the random responses there was a ten per cent consistent agreement on what the students felt was the reason for their willingness to ac­cept a passive sentence with a prepositional verb - it had to represent something actually DONE TO something. What they were saying in effect was that the passive is marked for transitivity. The speaker has to be thinking of a patient that is somehow affected by the action. For generations of lovers to pass beneath a bridge makes it romantic. For a dog to walk under it is just that - you have a spatial re-

MEANING AND FORM 15 lationship between dog and bridge, and nothing more. If a rancher warns that his lake is not to be camped beside by anybody he obviously has in mind the potential damage to the lake. But for someone's sister to camp there merely tells where she is. A little investigating shows that simple verbs are subject to the same restriction. We can say George turned the pages or The pages were turned by George; some­thing happens to the pages in the process. But while we can say George turned the corner we cannot say *The corner was turned by George - the corner is not affected, it is only where George was at the time. On the other hand, if one were speaking of some kind of marathon or race or a game in which a particular corner is thought of as' an objective to be taken, then one might say That corner hasn't been turned yet, I can-say The stranger approached me or I was approached by the stranger because. I am thinking of how his approach may affect me - perhaps he is a panhandler. But if a train approaches me I do not say *J was approached by the train, because all I am talking about is the geometry of two posi­tions. There are also power relationships involved. Though we can say both Private Smith deserted the army and The gen­erals deserted the army, to say that The army was deserted by Private Smith is comical while The army was deserted by all its generals is normal. This shows, I think, that pas-sivization cannot be defined on a particular set of verbs. It demands access to the speaker's intentions, to the mean-. ing of whether or not an effect is produced. The passive depends on a very specific semantic feature, that of tran­sitivity.

Having disposed I hope of one old war horse that had already been put to pasture for other reasons, I turn to another that is a bit more skittish to deal with. So far

16 DWIGHT BOLINGER as I know, there has never been any doubt in any grammari­an's mind about the absolute equivalence between sentences having and sentences omitting the relative word that. My way of wording this prejudices the case somewhat - to say that the word that has been omitted implies that it was there in the first place, and accordingly some semantic trace of it may be left. We know historically that this is not true - sentences using that and sentences not using it have been around ever since English began to be English. The question is, does a sentence such as I noticed you were there mean the same as one such as I noticed that you weve theve? For traditionalist and transformationalist alike, they have been regarded as in free variation. I will not go into the various restrictions on this supposed variation, which are a long story, more than just to say that there do exist environments where a that is required and others where a that is excluded (see Bolinger 1972). The most conspicu­ous instance of required that is one serving as subject in its own clause, in the case of adjective clauses: *They ar­rested the man shot the policeman requires an added that, but They arrested the man the policeman shot, where the that would be an object in its own clause, can do without it. That is the kind of restriction that can be stated nicely in transformational terms. The question is whether the process is merely one of introducing a that transforma­tionally under set grammatical conditions, or one of mean­ing from which the grammatical restrictions flow as corol­laries. One way of helping to decide the question - it is too complex for me to say that it will thereby truly be decided - is to look for minimally distinct pairs, one mem­ber of which contains a that while the other lacks it. The theory I am going on is that the word that is still - in

MEANING AND FORM 17

very subtle ways - the same word that it was when it first began to be used to head subordinate clauses, namely a demon­strative. If we look at situations where speakers are volun­teering information, where no question has been asked and no answer is implied, but what is being said comes out of the blue, it is unnatural for the word that to be used. If I step into a room and want to drop a casual remark about the weather I may say The forecast says it's going to rain. It would be odd for me to say *The forecast says that it is going to rain. But if you ask me What's the weather word for tomorrow? I have a choice; The forecast says that it is go­ing to rain is normal. If we think of that in its fundamen­tal deictic or anaphoric use as a demonstrative, we see that it is appropriate when the clause in question does not re­present a disconnected fact but something tied in with a previous matter to which that can point back, just as it does in That man insulted me, meaning the man referred to before. If I see you at the side of the road struggling with a tire and feel charitable I may go over and say to you, by way of an opener, I thought you might need some help. To say I thought that you might need some help suggests a question already brought up - if you were a huffy sort of fellow and looked up at me as if wondering what business it was of mine, then I might shrug my shoulders and say by way of answer to the implied question, J just thought that you might need some help. Look at it another way. Suppose that clause is used without its subordinating verb - the distinction be­tween old and new crops up again. Take an exchange like

"I didn't know that." - "Know what?" - "That Jack's held down six jobs at the same time."

Try leaving off the that in this case where a that anaphora

18 DWIGHT BOLINGER has already been introduced. On the other hand, suppose no that anaphora is present and the speaker is offering some­thing new:

"I want to tell you something." - "What?" - "Jack's held down six jobs at the same time."

The use of that here is just as odd as its omission in the other case. You will notice that the main verbs I used, know and tell, are both verbs that are perfectly free to take a clause introduced by that,

The other side of the problem of that and its omission is the supposedly suppletive relationship between that and the set comprising who, whose, and which. Just as that can be seen as basically demonstrative, so the other relatives can be seen as basically interrogative, and as lexemes in their own right, whose interrogative origin of course is a historical fact. The contrast between that and which shows up in a minimal pair such as the following:

This letter that came yesterday, that you remember had no stamps on it, was postmarked four weeks ago.

*This letter that came yesterday, that incidentally had no stamps on it, was ...

The normal use of incidentally is to call a hearer's atten­tion to a side topic which is new to the discourse. It is incompatible with anaphoric that, but quite compatible with a word that raises a new 'question1. We could if we wished use which in the first example, to refresh the hearer's memory, bringing the topic up anew; but there is hardly any choice in the second. The same contrast may be seen between that and who, for example in relation to intonation. In the example I want to get word to him as soon as possible about someone else that (who) I knew was available , the most com­patible intonations are the following:

MEANING AND FORM 19

else ... someone

that I knew was available. vail e 1 se ... someone who I knew was a

able.

In the first, availability is not at issue; it has been brought up before. In the second, the hearer is informed of it.

The independent meaning of a that or a which is a tough point to get across because of its very subtlety and the in­frequently with which using one or the other or neither is a matter of life and death. Unfortunately we have tended too often to see the importance of a question of language in terms of the importance of the message. The two are not re­lated.

Another element that has been viewed as a transforma­tionally introduced particle is the pronoun it in a number of different constructions. Again it has been all schools of thought that have dealt thus cavalierly with this little word. Wallace Chafe (1970:101), for example, talking about sentences like It's h o t , It's late, and It's Tuesday, says that it "need not reflect anything at all in the semantic structure". In cases of extraposition, as for instance to err is human, It's human to err, the it has been regarded as a pronominal copy of the displaced infinitive. But when we take an inclusive look at the various manifestations of it (Bolinger 1973), we find not only a great deal of seman­tic similarity but also the possibility of combining usages which are supposedly distinct and independent. Arthur Schwartz (1972:70-71) sensed this whan he said that "the surface it [with infinitives etc.] is not really a pronomi-

20 DWIGHT BOLINGER

nal Substitute for the proposition, but closer to the imper­sonal situation it of It is cold today or It is crowded in heve". I maintain that Schwartz is right and that it is the same word throughout, with a reference system that is very loose and open, drawing its semantic specification from the context. When the caller on the talk show said It seems to me that in the early 'sixties it was more fun, the it was more fun could just as easily have been expressed with things were more fun. We are willing to accept things as a very general and inclusive term, and should do no less with i t . I mentioned combinations of supposedly different uses of it as one way of showing that the same it is involved. Take a sentence like It's too hot to play tennis and utterly out of the question to do anything else. We slide over this with perfect ease and only if we are prejudiced with a know­ledge of grammar are we apt to realize that the first half, It's too hot to play tennis, involves a weather expression, while the second half, [It's] utterly out of the question to do anything e l s e , involves an extraposition. If someone asks you How is it in your room? you may readily answer It's hard to study. Try combining that question with the other arrangement of the infinitive: "How is it in your room?" - "'To study is hard. " The it of It's hard to study has to be more than a pronominal copy of the infinitive -it is the same situational it as in the question, How is IT in your room? Again we can combine them: It's noisy and hard to study. If the it were two different words, this ought to give the impression of a zeugma, like saying *I have to brush my teeth and hair. I believe that i t , like that, is a word with a meaning in its own right, and two such constructions as It is fun playing tennis and Playing tennis is fun do not mean quite the same, even though in

MEANING AND FORM 2 1 many and perhaps most situations we could manage with either one.

I have already mentioned my next example. Up to six or seven years ago it was generally held in transformational circles that the infinitive and the gerund are selected as complements by particular verbs in a kind of blind automatic process that has nothing to do with separate meanings for those two forms. That notion has now been given up, but like other old articles of faith it dies hard, and one indirect manifestation of it is still lurking around. I refer to the idea that there need not be any feature present in the verbs that take infinitive complements that causes them to do so, and similarly with the verbs that take gerunds (Dingwall 1971; cf. Bolinger 1974b). In other words, the association of verb with infinitive or gerund is still arbitrary. Proof of this is supposed to be found in the fact that there are pairs of synonyms one member of which takes infinitive com­plements. An example of such a pair is refuse and spurn. Since there seems to be no relevant difference in meaning between them, the choice of different complements must be arbitrary, so the reasoning goes:

He refused to accept the job. *He spurned to accept the job. He spurned my helping him. ?He refused my helping him.

The problem here is to show that the minimal pair He refused to accept the job and *He spurned to accept the job once more embodies a difference in meaning, only now the differ­ence produces an anomaly in one of the sentences. More pre­cisely, there is something about the meaning of the verb spurn that is incompatible with the meaning of the infini­tive. Suppose we try to get a fix on spurn and refuse by

22 DWIGHT BOLINGER looking at some of the other complements that go with them. We can say He refused the offer, He refused the invitation . . . bid, advice. We cannot say *He refused the idea, He re­fused the solution, He refused the truth - but with spurn these are all right. We can say After having it on trial he refused it, but we cannot say *After owning it for years he refused i t . Again, spurn is all right. There is obviously something about the meaning of refuse that faces somehow in a different direction from that of spurn. I hypothesize that it is a feature (if you like to call such things features) that might be called 'future orientation'. One can refuse an offer, and accordingly refuse a gift, a car, a dog, or even an idea if it is thought of as something offered. But one may not refuse something that one already possesses. The feature of future orientation fits the meaning of the infin­itive, which as a number of people have pointed out is some­thing on the order of 'hypotheticalness'. There are other pairs like refuse and spurn that show this same contrast of orientation. Take remember and recall. They are synonyms in sentences like I remembered my adventure and I recalled my adventure ; but whereas Remember to phone me is normal, *Re­call to phone me is not. Remember, like refuse, embodies that future orientation. It brings things AHEAD OF one's mind, not back of it. If I say At that moment I remembered my wife, remember suggests something to be done. But in At that mo­ment I recalled my wife all we have is a backward look. The companion pair of remember-recall is forget-overlook: He forgot his sister when he went tells us that an action he was supposed to carry out in the future was left undone. He overlooked his sister when he went merely tells us that she failed to get his attention. The picture of language as an automaton in which you punch the button reading refuse

MEANING AND FORM 23 and an infinitive pops into the slot is false to the facts. The infinitive has a meaning and refuse has a compatible meaning. There is nothing more mysterious about the harmony between refuse and the infinitive than there is between to drink and coffee.

Let me give now a case of supposed free variation, that of our two sets of indefinite pronouns (cf. Bolinger 1974c). Someone, anyone, no one, and everyone belong to one set; somebody, anybody, nobody, and everybody belong to the other. Here we have to take on all the authors of the handbooks, including Jespersen, who could see no difference in meaning between them. My including them among my examples is the result of a friendly challenge. A colleague who knew my po­sition in general about difference in form requiring differ­ence in meaning dared me to find a difference here, and I tried to oblige. First I gave a pair of contrasting situa­tions to a group of thirteen graduate students who were told to choose somebody or someone according to which seemed to fit the meaning best. The first situation read like this:

"Who's the present for?" I asked. "Somebody

He gave me an intimate look. very special, very dear to me," he said. Of course it had to be me, but I concealed my blushes.

The second situation read like this: "Who's the present for?" I asked.

somebody" _ Oh, he said, like meaning it was none of someone , my business. "You don't know him. Her. Them."

The vote was unanimous, with someone for the first and some­body for the second. Knowing what we know about pronouns it should not surprise us that meanings having to do with dis­tance, intimacy, and the relationships between the speaker

24 DWIGHT BOLINGER and others should be built into them, and that appears to be what has happened with the indefinites. My hypothesis about one is that it is marked for nearness, in both a spatial and a psychological sense. I gave a more elaborate test later to another group and asked them to comment on their own reac­tions. Several did so, and the gist of the answers conformed to the hypothesis that I had set up. As one worded it, "[-one] intimacy, definiteness, individuality; [-body] distance, in­definite reference, collectivity". Allow me to point out here - as bearing on something to be elaborated on in a mo­ment - the fact that the one of the indefinite compounds someone_, anyone, etc. has unmistakable ties to the word one as an independent indefinite pronoun, as in What can one say?

I mentioned earlier another transformation about which grammarians have had second thoughts, the one that was sup­posed to yield attributive adjectives. The classical form of this derives a noun phrase such as an empty house from the same underlying source as a house that's empty. It has been clear for some time that the relationship between these two structures is not as obvious as it once appeared. In fact, the supposed deep structure actually gives us less information than the surface structure ; you can see this by the behavior of a great many adjectives. Take one such as l o o s e . I may say The dog is loose, meaning that he is not tied up. I can say Where is the dog that's loose? but I am not apt to say *Where is the loose dog? On the other hand I can say A loose dog is apt to be a danger to the neighbor­hood. Or take an adjective such as handy. The tools are handy is ambiguous - it may mean tools that are made in such a way that they are very useful, or it may mean just that the tools happen to be easy to reach. But if we say the handy tools we select just one of these meanings, the one

MEANING AND FORM 25 that refers to how the tools are made, the way they really are. An adjective that is placed before the noun is not just any adjective that can occur after the verb be , but is one that can be used to do more than describe a temporary state -it has to be able to characterize the noun. *Where is the loose dog? is an unlikely sentence because it refers to a temporary state. A loose dog is apt to be a danger to the neighborhood is normal because we are making a generaliza­tion in which it is necessary to characterize certain dogs AS IF they formed a class. We can say the people asleep but not *the asleep people because we are not characterizing them, only telling how they are at the moment. But when the adjective aware began to be used as a synonym of alert , it was able to move before the noun: He's a very aware peron. An adjective that can only refer to a temporary state has to follow the noun: money galore. Even if we play with the deep structure so as to set up more than one source for these ob­vious differences, it still does not follow that an empty house means the same as a house that1 s empty. Sometimes they are interchangeable, but other times they are not, for the simple reason that the explicit predication in one makes a difference in the way the information is presented to the hearer.

I could go on with more examples in detail but it would only be repeating the same story. There is the so-called particle movement transformation, by which a structure such as haul in the lines is supposed to yield haul the lines in. But these do not mean the same, as can be seen by comparing the compatibility of saying They hauled in the lines but really didn't get them in with the contradictoriness of say­ing *They hauled the lines in hut really didn't get them in. One transformationalist skeptic refused to accept this nau-

26 DWIGHT BOLINGER tical pair as evidence of a contrast, but John Beatty re­ports (personal communication) that he has asked sailors about the two sentences and "they all hold that there is a difference, which deals with completive action. They hauled in the lines but didn't get them in is possible, but *They hauled the lines in but didn't get them in is not." A good example of how a theory can get in the way of reality. There is the pseudo-passive, concerning which it has been claimed (cf. Mihailovic 1967) that there is no difference in mean­ing from the active, for example Be accidentally drowned in the river and He was accidentally drowned in the river. Yet even without an agent expressed or implied there is a dif­ference. If we say Re stupidly drowned we view him as an actor in the causal chain, even though he may not have been a willing or even a conscious one - we can add He stupidly drowned; why couldn't he have been more careful? But to say He was stupidly drowned; why couldn't he have been more care­ful? is odd - we are more apt to say why couldn't they have fenced off the safe area so he could have told how far to venture out? The pseudo-passive, like the real passive, puts the responsibility on other shoulders than those of the vic­tim.

To get the other side of the picture I want to turn now to cases where it has been claimed that there IS a dif­ference but there really is none, and the supposed differ­ence turns out to have been created in the linguist's mind through a confusion of competence and performance. It is essential to look at this side because actually it has of­ten been the failure to see a sameness at one level that has led to the failure to see a difference at another.

I shall give just one new example to make this point and then double back ver a couple of previous ones to show

MEANING AND FORM 27

that they illustrate the same thing. My new example is with the imperative using the auxiliary verb do (see Bolinger 1974d). The claim has been made that do is a dependable test for true imperatives (Schreiber 1972), since it is never used except with actions that can really be commanded: Do be careful, Do try harder, but not *Do be glad, *Do own the property. If this is true, then the function of do in such constructions must not be the same as that of the ordinary do that we find in negative and affirmative sentences, where it is normal to say He did own the property or He didn't own the property. In other words, either we have two homonyms, both spelled do, or somehow the single word do has to be tagged twice in the dictionary. But when we look at the sit­uation in which do imperatives are used, we find that do is still the same old word after all. Imagine that someone knocks at your house and you throw open the door and in the very act of opening it you greet him with Do come in! People I have asked about this find such an ungrounded use of do unnatural - they can imagine throwing the door open and im­mediately saying Come in , but not Do come in. Do come in re­quires an interval, however short, of the person's standing there and NOT coming in. The do that is added is the same affirmative do that contradicts other negatives: "He didn't do it." - "Yes he did!" The test of course is whether we can falsify the other claim about do being normal only with actions that can truly be commanded. A counter-example is a sentence such as I don't care whether you are successful or not, but do be happy; that's the most important thing in life. One cannot be happy by an act of will; but do in this sentence is normal because it is built on a prior negation.2

I have said that the mistake in a case like this is due to a confusion of competence with performance. The com-

28 DWIGHT BOLINGER

petence consists in the meaning of the word do that we carry in our heads. It is a constant, relating to the concept of affirmation. The performance resides in the chance associa­tion of do with imperatives on the one hand and declaratives on the other. Since imperatives and declaratives are dif­ferent, when do intersects with them one gets a different impression, and reads that difference into the word do, like concluding that Joe Smith is a different person when you meet him on a dark night. If we look back at two of the ear­lier examples I gave we find the same thing happening. The word it is a lexical constant which shows up in association both with situational expressions such as those having to do with weather and with extrapositions. In order to get a syn­tactical sameness such as the supposed one between To study is hard and It 's hard to study we have to ignore the reality of i t . To achieve a false sameness we have to create a false difference. The other case was the demonstrative that - it is a lexical constant which has been adopted into the scheme of relative clauses while still retaining its demonstrative meaning. It would probably be going too far to claim that all particles which are supposedly just the product of transformations are really words in their own right which belong in the lexicon as much as in the grammar; but at least enough has been said to make one want to take a second look at most of them. Such things as the to of the infinitive, the be of the passive, and the there of there was deserve to be restudied for what they may contain as independent words.

Returning to my main thesis of difference in form nec­essarily correlating with difference in meaning, I would say that here again there has been a confusion of competence and performance, exactly the reverse of the kind that has

MEANING AND FORM 29 been claimed by many transformationalists. Instead of there being an underlying sameness in active and passive with the differences being relegated to style, focus, or what-not, I would say that there is an underlying difference with the samenesses being due to performance variables. If you are asked What happened to the train? and you answer It was wrecked by the engineer3 you could just as well have answered The engineer wrecked it. There is nothing in the performance situation that cannot be satisfied by one answer as well as by the other. But if you are asked Who was responsible? you are going to prefer the passive voice. The fact that a con­trast that we carry in our competence is relevant does not mean that it is relevant all the time. It only means that it is there when we need it. Here I put in my word again for common sense. If a language permits a contrast to sur­vive, it ought to be for a purpose. When we look at what has happened historically to the accidental contrasts that have cropped up, at the avidity with which speakers seize upon them to squeeze in a difference of meaning, come what may, I think we can form a proper appreciation of linguistic econ­omy. It is'not normal for a language to waste its resources.

If what I think is true of the cases I have cited turns out to be true generally and there really are no syntactic differences that are of no consequence semantically, we will have to expect some changes in our ideas about surface struc­ture and deep structure. It will not necessarily affect sur­face identities that are correlated with deep differences. This after all is commonplace in the lexicon, where homonyms are plentiful. But there will no longer be surface differ­ences correlated with absolute identities in deep structure. Instead of claiming that structures ARE the same, we will be looking for the samenesses that they CONTAIN, and on top of

30 DWIGHT BOLINGER

that trying to identify and define the elements that make them different. Many transformations will be affected, and some will have to be abandoned. I cannot see, for example, how the particle movement transformation can possibly hold up, since the difference in meaning stems from a change in the constituents of the sentence. The example I used was They hauled in the lines and They hauled the lines in. In the first member of the pair in is a constituent of the verb. In the second it is at least partially a constituent of lines - the lines were in; that is why it is contradictory to say *They hauled the lines in but really didn,t get them in. Or take the some-any rule, which Robin Lakoff (1969) and William Labov (1972) have done a thorough job of debunking. Any other rule that treats a lexical difference in such an offhand manner will have to go the same route.

While I am suggesting that there will have to be changes in our view of underlying structure or underlying paraphrase, I am not suggesting that we give it up. The things that peo­ple say are too often a kind of shorthand, and if we are to interpret it we have to elaborate on it. But I do not be­lieve that we can always hope to do this in the chiseled manner that deep structure analysis has demanded. For one thing I believe that many surface structures are not trace­able to single underlying structures or neat embeddings of one kind or another, but have to be viewed as syntactic blends. For another I find the relationships at times too subtle or too general to be built on the actual structures that they supposedly reflect. An example of this sort that came my way recently was an odd fact (reported in Schlach­ter 1973:31-34) that Michael Brame noted regarding idioms, which is that they may be broken up if one of the broken halves is retained in a relative clause. For instance, we

MEANING AND FORM 31 do not say *The lip service displeased me , but we can say The lip service that they paid displeased me. The question that must be asked is whether idioms and relative clauses are relevant per se. Are there other ways of fleshing out an idiom besides providing its missing half? Take pay lip ser­vice again:

*Lip service is unsatisfactory. *The lip service is unsatisfactory. The lip service that they paid was unsatisfactory. The lip service that was all they they expected would not

have been enough for me. Lip service alone is not satisfactory. That sort of lip service would never satisfy me.

Not only can the sentence get along without the rest of the idiom, it can get along without the clause. Now take an ex­pression that does not involve an idiom:

*The folly displeased me. The folly that he was guilty of displeased me. Such folly displeased me. Folly always dispeased me.

A relative clause improves things, though there is no col­location, such as pay with lip service, that has any spe­cial claim on f o l l y ; but other devices effect the same im­provement.

Either the situation with idioms and relative clauses is one of many superficially similar situations each of which requires a different analysis, or all such cases are fundamentally the same though in a rough and ready way that does not answer to one coherent syntactic treatment. I would embrace the second alternative and say that the ex­planation lies in a condition normally imposed on discourse, which is that a speaker will not introduce a noun phrase un­less he can assume that its referent is within the grasp of his hearer. One typical place where this breaks down is in

32 DWIGHT BOLINGER

the speech of children when they address adults. A little girl approaciies a stranger and inquires Have you seen Doro­thy? The appropriate rejoinder for an adult is Who is Doro­thy? If someone says I left the room because the stench was overpowering we have no difficulty putting two and two to­gether and inferring that there was a stench in the room. But on hearing *I left the room because the disgust was over­powering we can only ask What disgust? The speaker has vio­lated a canon of discourse by giving us no way to discover a referent for the noun. Had he said the disgust that I felt there would have been no problem. Idioms are perhaps special because in addition to the lack of any referent, the noun phrase may be meaningless if we fail to provide the rest to the idiom somewhere nearby. A relative clause is no way. Prior context is another, as in

*Tabs on everybody is 1984ish. Kissinger can keep tabs on his friends, but tabs on everybody is pretty 1984ish, isn't it?

Or instead of being meaningless the imcomplete idiom may convey the wrong meaning, as happens with take offense by contrast with take umbrage:

'«The umbrage was understandable. *The offense was understandable.† The umbrage (offense) that they took was understandable. Their umbrage was understandable. *Their offense was understandable.†

The last example, if it means anything, refers to people who are aggressors rather than victims.

Just as idioms are a special case, so are degree words used epithetically. My example *The folly displeased me is wrong because folly is like an adjective: it is used only

f Starred in the relevant sense.

MEANING AND FORM 33 to describe a noun, and the hearer is given no clue as to whether there is anything in the situation that is being described as foolish. The folly of it displeased me provides the missing element. (Ordinarily a concrete noun will be taken to imply "There is such an entity in the situation" and cause no trouble: The comedy displeased me.) A more ob­vious example of an epithet is bastard:

*As I walked out the front door the bastard came toward me. As I walked out the front door the bastard who had insulted

me came toward me. Jones was standing out there, and as I walked out the front

door the bastard came toward me.

The first example is unacceptable because there is no clue to anybody who is being called a bastard.

You can see how easy it is, comparatively speaking, to explain a discourse constraint of this kind using ordinary language, and how difficult it would be to put it in pre­cise syntactic terms short of simply listing all the syn­tactic resources whereby a speaker can make known that a noun phrase has a referent and what the referent is . More is inferential than syntactic. Yet the syntacticist will determinedly seize upon one incidental - and as it turns out only occasional - feature of a broad phenomenon such as this to use it in a derivational scheme for relative clauses. At best, this approach shows a failure to see things in the whole before trying to analyze the parts. There are times -to garble somewhat Einar Haugen's (1972:312) allegory of the Procrusteans and the Heracleans - when the best proce­dure is not to tease the data but to wade in on all fours. I can sympathize with those who try to do more, and still be happy to have chosen a topic that exempts me from it, and allows me to end my discourse by proclaiming that God's in his heaven and when we say two things that are different we mean two different things by them.

34 DWIGHT BOLINGER

NOTES

1 In other respects as well. Certain quantifier pronouns are the same as quantifier adjectives, and appear to be the result of deletion, e.g., He has some money reduced to He has some. The syntax of the to of the infinitive, when the lexical infinitive itself is dropped, is quite sim­ilar to that of the personal pronouns, e.g., I hated IT but I had TO.

2 The sentence *Do b happy unless you have a real reason for feeling sad has been proposed as a counter-example to the theory that do with imperative is based on a prior negation. This assumes that unless is negative, but as Michael Geis (1973) demonstrates, unless is positive, as can be seen using some and any:

If you don't have any objection, I'11 wait. *Unless you have any objection, I'll wait. Unless you have some objection, I'll wait.

When if not replaces unless, do b happy becomes acceptable: If you don't have a real reason for feeling sad, then do happy.

Geis deals effectively with another case of false identity. Unless and if not are not the same in meaning.

REFERENCES

Bolinger, Dwight. 1972. That's That. The Hague: Mouton. ____ . 1973. "Ambient It Is Meaningful Too". Journal of Unguis-

tics 9.261-70. . 1974a. "Transitivity and Spatiality: The passive of the

prepositional verbs". Linguistics at the Crossroads ed. by Adam Makkai. The Hague: Mouton, in press.

. 1974b. "A Semantic View of Syntax: Some verbs that govern infinitives". Festschrift for Archibald A. Hill ed. by Edgar C. Po­lome, Werner Winter, and Mohammad A. Jazayery, inpress . The Hague: Mouton.

MEANING AND FORM 35

Bolinger, Dwight. 1974c. "The In-Group: One and its compounds". Current Trends in Stylistics ed. by Paolo Valesio. The Hague: Mouton, in press.

. 1974d. "Do Imperatives". Journal of English Linguistics 8.1-5 (March 1974).

Chafe, Wallace L. 1970. Meaning and the Structure of Language. Chicago: Univ. of Chicago Press.

Dingwall, William Orr. 1971. "On So-called Anaphoric To and the Theory of Anaphora in General". Journal of English Linguistics 5.49-77.

Geis, Michael L. 1973. "If and Unless". Issues in Linguistics: Papers in honor of Henry and Renée Kahane ed. by Braj B. Kachru, Robert B. Lees, et al.., 231-53. Urbana, I11.: Univ. of Illinois Press.

Haugen, Einar. 1972. The Ecology of Language. Stanford, Calif.: Stan­ford Univ. Press.

Labov, William. 1972. "Negative Attraction and Negative Concord in En­glish Grammar". Language 48.773-818.

Lakoff, Robin. 1969. "Some Reasons Why There Can't Be Any Some-Any Rule". Language 45.608-15.

Mihailovic, Ljiljana. 1967. "Passive and Pseudo-passive Verbal Groups in English". English Studies 48.316-26.

Schachter, Paul. 1973. "Focus and Relativization". Language 49.19-46. Schreiber, Peter A. 1972. "Style Disjuncts and the Performative Analy­

sis". Linguistic Inquiry 3.321-47. Schwartz, Arthur. 1972. "Constraints on Movement Transformations".

Journal of Linguistics 8.35-85.

STRATIFICATIONAL SOLUTIONS TO UNBRIDGEABLE

GAPS IN THE TRANSFORMATIONAL-GENERATIVE PARADIGM

TRANSLATION, IDIOMATICITY, AND MULTIPLE CODING

ADAM MAKKAI

0.0. The present paper surveys some areas of recent linguistic theory that have proven the Chomskyan paradigm (and its various notational vari­ants) to be unworkable, if by a 'working grammatical theory' we mean a theory that squarely faces the task of informing us of what people actu­ally do. Hence the main argument of this paper will be that the Chomskyan distinction between 'competence' and 'performance' must be understood in the light of the pragmatics and the ecology of human interaction in a non-imaginary society. My solutions are suggested within the framework of Stratificational-Cognitive Grammar (henceforth SG) and Pragmo-Ecolog-ical Grammar (PEG), my own approach within the family of stratificational grammars. The areas covered here will be (1) translation, (2) deep struc­ture, (3) idioms and, finally, (4) multiple coding in speech. 1.0. It is by now more than intuition that tells us that TG cannot han­dle translation in any systematic, -ad hoc manner. It is a moot point whether transformationalists have dealt with the problem, for the reply could always be made that the problem simply did not occupy their atten­tion. My point here is that is a LOGICAL IMPOSSIBILITY to achieve any sort of adequate translation from L1 to L2 under TG assumption, whether

38 ADAM MAKKAI

one espouses the Chomsky-Katz-Jackendoff 'lexicalist-interpretivist' po­sition, or the McCawley-Lakoff-Ross 'generative semantics' position. It will be shown that, if TG is to work at all, one would have to practice both the semantics-centered versions of TG and the syntax-centered ones which, according to Emmon Bach (1971), has been proven to be impossible. Essentially Bach says that according to advanced mathematical testing by Stanley Peters and R.W. Ritchie the lexicalist-interpretivist and the generative semantics positions are logically incompatible. The question arises whether or not 'mathematical testing' is relevant for evaluating human grammars, and whether competence and performance are indeed two truly separate sides of the coin of human speech. For linguists in the TG and PEG traditions, competence and performance coincide in many areas. The stratificationalist David G. Lockwood, for instance, speaks about 'ideal performance' in his Introduction to Stratiƒicational Linguistics (1972). 1.1. In what follows, I will describe the various TG positions AS IF THEY ATTEMPTED TO DESCRIBE WHAT HUMAN BEINGS REALLY DO. That the TG-oriented reader will cry 'this is not cricket' is to be expected. My re­ply to such a defense is: A LIN G U I S T I C THEORY O U G H T TO BE A B O U T WHAT HUMAN BEINGS DO, AND IF IT IS NOT, IT IS NOT A VIABLE, SERIOUS LINGUISTIC THEORY, U M E R E L Y AN INTELLECTUAL GAME. Intellectual games, as all games, subdivide into (a) harmless, entertaining games, (b) challenging, sporting games, designed to strengthen the mind and the body, and (c) crooked games, designed to get the better of a socially or emotionally inferior victim by an aggressor. (For a theory of games from the emotion­al point of view, see Eric Berne's popular psychological study of 1966, Games people Play). It is my contention that viewed as a game, TG exhib­its all three of these characteristics. In so far as it is harmlessly entertaining and challenging, there is nothing one could object to. Alas, it also exhibits the characteristics of crooked games and as such has caused the profession of linguistics as well as individuals actual harm. Aspects of this will be documented and discussed below.

STRATIFICATIONAL SOLUTIONS 39

1.2. In his forthcoming book, Introduction to Human L i n g u i s t i c s , Victor H. Yngve of the University of Chicago, recalling his conversations with Chomsky at M.I.T. writes:

The freeing of linguistic data from personal bias is far from a triv­ial point, for it is all too easy to be led astray. One of the most serious problems with Chomsky's early (1957) monograph was an inade­quate treatment of the relation between theory and observation in science. In that study we find recurrent discussions of this issue, for example: "One function of this theory is to provide a general method for selecting a grammar for each language, given a corpus of sentences in that language" (p. 11). "Clearly, every grammar will have to meet certain external conditions of adequacy; e.g., the sen­tences generated will have to be acceptable to the native speaker" (pp.49-50).

Yet these were not the strictures actually employed in that monograph: "One way to test the adequacy of a grammar proposed for L is to determine whether or not the sentences that it generates are actually grammatical, i.e., acceptable to a native speaker, etc. We can take certain steps toward providing a behavioral criterion for grammaticalness so that this test of adequacy can be carried out. For the purposes of this discussion, however, suppose that we assume intuitive knowledge of the grammatical sentences of English and ask what sort of grammar will be able to do the job of producing these in some effective and illuminating way. We thus face a familiar task of explication of some intuitive concept, in this case the concept of 'grammatical in English', and more generally, the concept of 'grammatical'" (p.13).

Now the task of explication has its place, but it brings the danger of mistaking the intuitive concept either for solid data or for some self-evident theory that somehow need not be tested against data (emphasis added, A.M.). There is the further danger that the author's "intuitive knowledge of the grammatical sentences of En­glish" may be biased by his theoretical preconceptions in such a way as to bolster his arguments at the expense of scientific truth (em­phasis added, A.M.). Let's see what Chomsky is led to do when what his intuition would include as 'grammatical in English' differs from what is acceptable to a native speaker.

The whole point of that monograph was to argue the merits of the author's transformational approach to granimar, with the author taking a strong point of advocacy. In the course of the arguments, certain simple processes of sentence formation are postulated and accepted intuitively as characteristic of English (p.21). These processes in­volve the embedding of sentences in other sentences, and would gene­rate such strings as "If either the man who said that if either the woman who reported that it is raining, is wrong, or the boy rode his bicycle, then I will be happy, is arriving today, or you are sad, then Bill was right." The further course of the argument hinges on

40 ADAM MAKKAI

whether strings such as these are grammatical English sentences or not. Now it would seem that any realistic criterion of 'external adequacy' would reject such strings as not being acceptable to the native speaker. This is particularly true in light of the fact that Chomsky intends no limit to the recursive processes used, so that much more complex examples would also be produced. BUT THEN HE COULD NOT GO ON TO 'PROVE' THAT 'ENGLISH IS NOT A FINITE STATE LAN­GUAGE' AND HIS ARGUMENT IN FAVOR OF TRANSFORMATIONS WOULD CRUMBLE (special emphasis added, A.M.).

Faced with a discrepancy between observed facts and the pre­dictions of theory, one would expect a revision of theory: "We shall continue to revise our notions of simplicity and character­ization of the form of grammars until the grammars selected by the theory do meet the external conditions" (p.54).

But we find instead an argument for keeping the theory and re­moving the discrepancy by actually accepting such sentences as grammatical: "Note that many of the sentences of the form (... of the one cited above) will be quite strange and unusual" (they can be made less strange by replacing 'if' by 'whenever', 'on the as­sumption that', 'if it is the case that', etc., without changing the substance of our remarks). BUT THEY ARE ALL GRAMMATICAL SEN­TENCES, formed by processes of sentence construction so simple and elementary that even the most rudimentary English grammar would contain them. They can be understood, and we can even state quite simply the conditions under which they can be true. (Yngve forth­coming, pp.16-18, quoted from a privately-circulated manuscript, by permission of the author.)

In the rest of the chapter Yngve recalls how Chomsky was immediately and widely criticized for this contradiction. Having been at M.I.T. at the time, and having engaged Chomsky personally in conversation about these matters, Yngve recalls the birth of the forced distinction between 'competence' and 'performance'. Chomsky's answer was that for a sen­tence to be grammatical in English, it can meet adequacy conditions in the competence which it does not necessarily meet in performance. Be­tween competence and performance, competence is the more important one; it is, in fact, the basis of the description of the language.

In this manner, then, Chomsky may have patched up the most glaring contradiction of his theory by creating (for his purposes) an unbridge­able gap between competence and performance. So far it would have been merely an entertaining or a challenging game, but alas, in its later stages the game also became a crooked one. Whenever a critic of the

STRATIFICATIONAL SOLUTIONS 41

'Chomskyan paradigm' pointed out an irreconcilable contradiction between fact and theory, Chomsky and his disciples intoned: "You are not inter­ested in real linguistics ... You are talking about matters of performance Real linguistics is the description of competence in terms of transfor­mations and symbolic logic". By using this form of rhetoric, the MIT-es­tablishment accomplished several things at once:

(a) They explained away the contradictions found by the critics; (b) they projected an image of progress, mentalism, sophistication,

and revolution in science; (c) they succeeded in intimidating everybody who dared disagree with

them. Since 'performance' always carries the attribute 'mere' whereas 'competence' was equated with true linguistics, interest in data and facts became peripheral; honest, data-oriented lin­guists became 'lowly taxonomical data gatherers' perhaps not worthy of promotion and salary. Incredible and unprecedented in­equality in hiring and publishing (at least in the United States) was the result. The MIT-establishment became rigid and impene­trable; in short, a closed system alien in spirit to democracy and academic freedom.

1.3. I will now raise the question of what is better to do: Play a crooked game while staying honest (as if playing poker with a group that cheats thereby losing one's shirt); get out of the game and change pro­fessions; or, devise a counter-game, whose apparent crookedness is but a means to show" to the onlookers that the opponent is the one, whose game is really crooked. Since I have tried the first solution and found it not to work, and since I will not change professions, I beg the read­er's indulgence in allowing me to devise a counter-game of my own. It is very simply this: I will interpret Chomsky's position at face value, AS IF HE DID NOT HAVE THE CONVENIENT EXCUSE OF RELEGATING MY OBSERVATIONS TO 'PERFORMANCE'; IN SHORT, AS IF THE COMPETENCE-PERFORMANCE DICHOTOMY D I D N O T E X I S T . I trust that this is not a 'crooked game', but merely a challenging or an entertaining one. I predict that it will help unter­stand what makes me object to the Chomskyan paradigm thereby enabling those who believe in it to remove some of its useless ballast and retain only that which is actually useful in it: The study of surface syntax

42 ADAM MAKKAI

and its sadly neglected pedagogical applicability in teaching English to foreigners or the teaching of composition to children.

1.4. According to my face-value reading of Chomsky's Aspects of the Theory of Syntax (1965) the meaning of a sentence (or a set of sentences) is understood by the ideal speaker-hearer only after semantic projection has taken place, where the semantic component is supposed to interpret the syntactically 'creative' component, the deep structure of what has been said or found in writing, It can be argued, of course, that Chomsky did not mean to say this is what people do. My answer, as before, is that (a) he either ought to have meant it (in which case he would have been plainly wrong), or (b) he ought not to have said it (in which ease there would have been no linguistic revolution). Since my declared purpose is to engage the reader in the logic of a counter-game whose purpose is to show the motivation of the opposition, I will turn tables on Chomsky and pretend that he does not have the convenient excuse of invoking the 'but I am not talking performance' injunction. I do not recognize the relevance and the intellectual legitimacy of using the performance-competence dis­tinction in order not to have to make a theory account for what people do while at the same time promoting an irrelevant and secretly computer-ori­ented linguistic theory alien to life, people's needs, encoding and de­coding, the translation process, poetry, psychology, puns, and double coding. Hence I will deliberately pretend that semantic projection is something that happens in the human brain, right after reading or hearing something, as if happening a few split seconds after the received sentence was processed by human perception. Let us imagine that we are confronted with the following stanza by Horace:

(1) Integer v i t a e , seelerisque purus, Non eget mauris iaeulis neque arou,

Nec venenatis gravida sagittis Phusce, Pharetra.

If were a native Roman living in Horace's days, I would, no doubt, have an easier time of 'interpreting' the meandering syntax of this sentence, forced, as it happens to be, by the metrical constraints of the Alcaeic

STRATIFICAIIONAL SOLUTIONS 43

meter. The non-native student of Latin, of course, cannot really grasp this sentence until and unless he has rearranged it in prose form, some­what as follows: (Homo) purus sceleris (et) integer vitae, 'non eget iaoulis mauris (obligatory ablative where the logical case is accusative) neque arcu, nec pharetra gravida venenatis sagittis, (0) Phusce! After the syntactic relationships of modifier and modified, the use of the ab­lative after egeo, -ere 'need' have been 'projected', i.e., looked up in the internalized grammar with the lexemes found ('looked up') in the in­ternalized lexicon, the translator 'reads' the sentence as:"Phuscus, (I am telling you that) (one, a man) with a life of integrity and pure from sins needs no Moorish javelins nor a quiver heavy with poisoned arrows!" The sentence simply remains unintelligible as long as the agreements and governments demanded by Latin syntax are not worked out. A native speaker of Latin probably did it in much shorter time by using intuitive jumps, the contemporary foreign student of Classical literature labors at the sentence with dictionary in hand at a relatively slow pace. This, of course, has been known from antiquity to the present. It is a basic move in my deliberately devised counter game to state that if 'semantic pro­jection rules' in the sense of Katz and Fodor (1963) and Katz and Postal (1964) were to be relevant to human behavior, that is the linguistics of what people do, they would have to be understood and reinterpreted as DECODING PROCEDURES PERFORMED BY THE HUMAN HEARER-TRANSLATOR'S BRAIN. But human speakers, unlike computers, have no decoding algorithms built into their brains, that is algorithms which function in ordered fashion taking one step at a time. The human brain, being infinitely more subtle and complex than any computers, is of course capable of pretending that it is taking one linear step at a time, and struggling students of Latin who look up each word in the dictionary before understanding Horace's stanza quoted above, may approximate the computer in slowness and ineffi­ciency. But even these slow students, by virtue of being human, will even­tually discover that the sentence is in the Alcaeic meter and that it, therefore, exhibits aesthetic beauty. The point I am making here is simply that by asserting that the meaning of sentences is intimately tied to

44 ADAM MAKKAI

their structure, Chomsky has not said anything new. (I repeat that I am interpreting Chomsky throughout this paper at face value, as a deliberate strategy. In so doing I 'refuse to play his game' and invoke the rules of 'my own game'. Whether the reader agrees with my conclusions or not, I will have demonstated that the argument is not so much about essence but about political power in linguistics nowadays, The real question, of course, is really this: Who is allowed to call the shots in what constitutes an acceptable 'game' for the science of linguistics? The answer: He who suc­ceeds in making others believe that he has the right to do so. In short, success justifies, and we are right back in the Andersen's fairy tale 'The Emperor's New Clothes'.)

Since I refuse to play the game of 'semantic projection rules' on a competence basis where it does not matter whether 'semantic interpreta­tion' is psychologically real or not, and whether it precedes, coincides with, or follows the structural parsing of the sentence in real time, I will arbitrarily assume that (despite Chomsky) there is such a thing as 'semantic projection', but I will rename it as the DECODING OF LEXEMIC CONSTITUENCY NETWORKS CHARACTERIZED BY ORDERED 'AND' AND ORDERED AND UN­ORDERED 'OR' NODES. Examples below will make clear what I mean. I fur­thermore claim that this way of looking at 'semantic projection rules' (that is, by reinterpreting the human decoding process stratificationally) brings the theory of Strati ficational Grammar into immediate and real contact with real human beings and their brain processes. In SG, the LEXOTACTICS (read 'surface structure' prior to morphophonemic processing) of a sentence may lead to one or more SEMEMIC TRACES (read: may be ambig­uous) and the hearer-reader must decide whether from the context or from additional phonological clues (Sememic Traces will henceforth be abbre­viated ST. )

1.4. We may enquire what happens when a person has to translate the sen­tence:

(2) visiting relatives can be a nuisance.

If Chomsky's 'semantic projection rules' had any relevance for human lin-

STRATIFICATIONAL SOLUTIONS 45

guistics (to borrow Yngye's term), one would expect that these semantic projection rules would tell us whether the gerundival or the participial meaning of the form -ing was intended depending on what the deep struc­ture of the sentence was. Thus DS1 would read something like (somebody visits relatives) ( ((it)) is capable of being a nuisance); with DS2 be­ing approximately (relatives visit) ( ((they)) are capable of being a nuisance). Thus, assuming again that performance and competence have been hammered apart artificially in order to save the theory, the real human hearer-reader would perform projectioni (the gerundival interpretation), and match it against the immediately preceding linguistic and extralin-guistic context; if it does not fit, he would perform projection2 (the oarticipial interpretation) and accept it if it fits. The question I am asking is this: Is this, in fact, what people do? Don't be hasty and prejudge your answer. I am not necessarily saying that people never do this at all, under any circumstances. As a matter of fact, - as we saw above in the case of the student struggling with Horace's verse under certain pedagogically quite relevant circumstances, such a composition, laborious, slow translation etc., the person engaged in his work might quite possibly try to literally 'interpret' a given sentence once this way, then the other. This old and experimentally proven fact of language pedagogy is, however, not the reason why this analysis has become so popular. Rather it is said that it can claim theoretical advantages of 'descriptive adequacy' over Neo-Bloomfieldian Immediate Constituent analysis; that the 1 analysis cannot account for the two different meanings and that, therefore, 1 analysis is mechanistic, taxonomical and inferior, with TG being mentalistic, able to approximate 'explana­tory adequacy' in addition to 'descriptive adequacy' and that it has these prestigeous advantages because of positing a level of linguistic competence underneath surface structure, known as deep structure. It is an ironic fact that Latinate traditional grammar can deal with the sit­uation perfectly well by using the concepts of 'gerund', 'present ac­tive participle', 'singular', and 'plural'. Traditional grammar has fallen into disrepute because the 60 years of structuralism (1900-1960)

46 ADAM MAKKAI

have proven that exotic non-Indo-European languages can frequently be best described on their own terms, without reference to Latin-based ter­minology. This movement started with Franz Boas in the USA, and culmi­nated in Neo-Bloomfieldian structuralism.

It is becoming increasingly clear that Chomsky was following a de­finite sales strategy resembling Madison Avenue advertising replete with both seduction and intimidation when he invented the notion of deep structure. The ingredients were as follows:

(1) Sentence syntax is the central axis of language. (2) If ambiguity is seen in the meaning of a sentence, there must

be two nonambiguous sentences (emphasis on SENTENCES) that are the two respective meanings of the observable ambiguous sen­tence.

(3) The surface sentence relates to 'performance1 as the underlying sentence-like proposition (i.e., the 'deep structure*) relates to competence' .

(4) Surface structure is to mechanism and taxonomical data gathering, as deep structure is to mentalist theory-orientedness.

(5) Surface structure might exhibit observational adequacy, but on­ly deep structure can exhibit descriptive or explanatory ade­quacy.

(6) If a sentence is judged unacceptable by native speakers while nevertheless being a logically constructed, hence laboriously retrievable, sentence the unacceptability of the sentence is merely a matter of limitations on performance which does not interfere with the grammaticality of the sentence in competence.

(7) Surface structure oriented linguists are reactionaries and lack insight; deep structure oriented transformationalists, on the other hand, are revolutionary, daring, and are blessed by the gift of insight.

This is the simple script of seven steps that has conquered the world of linguistic scholarship during the past fifteen years. It is as simple-minded and as effective as a television advertisement promoting a new kind of moutwash which posits that clear breath means social suc­cess and bad breath means social failure; that smart people who want not to offend their lovers, bosses, fellow-travellers in a crowded Volkswa­gen, will use the mouthwash, while those who are slow, stupid, and slug-

STRATIFICATIONAL SOLUTIONS 47

gish will go on offending. We must now take a look at how SG deals with the ambiguous sentence

(2). Stratificational linguistics recognizes that neutralization is a multi-stratum phenomenon, not limited to phonology alone, as was prac­ticed - by and large - in Praguean linguistics.

Fig. 1 A simplified relational network analysis of the 'surface struc­ture1 of visiting relatives can be a nuisance. The Lexo-Morph-emic constituents 1-8 are ORDERED, that is, we read them off from left to right. The status of the higher constituents I, II, and III, which depends on the SEMEMI TRACES here realized in NEUTRALIZATION, is the source of the ambiguity, if the sen­tence is encountered out of context.

48 ADAM MAKKAI

The 'surface st ructure ' of the sentence, of course, remains ambig­uous in i so la t i on , when we do not know who uttered i t to whom, under what circumstances, and for what purpose. In s t r a t i f i ca t i ona l thought we recognize that the sentence is ei ther the real izat ion of sememic trace (ST)! or of ST2:

plane, relative child, penny, etc.

nom. (nuisance, pleasure, joy, etc.)OR adj. (dangerous, lovely, neat, etc.)

possibility 1st degree

Fig. 2

A simplified relational network description of the sememic trace giving rise to the 'present active participle' interpretation with can realizing the S/Plural/. Depending what content sememes are used, the network will account for visiting relatives can be a nuisance, shining pennies can be a joy, visiting children can be neat, flying planes can be dangerous, etc. Some outcomes will be nonsensical, e.g., visiting pennies can be neat, shining relatives can be a nuisance, flying relatives can be a joy, etc. None of these are, of course, 'ungrammatical'; their illformedness rests elsewhere. See below. The s/possibility/ in conjunction with the s/first degree/ gives rise to can, in conjunction with s/second degree/ to may, with s/third degree/ to might, etc. Thus a morpheme in 'Deep Structure' such as Auxiliary will not do in a sememic trace. Will have been, would have been, should have been, etc., are all different sememic propositions accidentally realized by a set of verbs known as Auxiliaries, but their auxiliary nature is not a sememic, but a lexotactic fact of the English language.

STRATIFICATIONAL SOLUTIONS 49

fly, shine, etc.

plane, relative, child, elephant, penny, etc.

Fig. 3

A simplified relational network description of the sememic trace giving rise to the 'gerundival interpretation' with can realizing the S/Sing./. Whereas ST1 had 5 unordered AND nodes, ST2 has 6; whereas ST1 had the Plural sememe in it, ST2 has the Singular sememe; whereas ST1 indicates that a predication is made of some­thing we attribute to relatives (i.e., the fact that they visit), ST2 shows that a predication is made of a non-specific agent's doing something to some goal (i.e., some one flies planes, or some one visits relatives). This non-specific agent is not the Deep Structure Dummy SOME ONE later to be deleted, it is simply realized as 0 on the lexemic stratum with the action sememe realized as the gerund that can carry an object.

nom. (nuisance, joy, etc.) OR adj. (dangerous, etc.)

possibility 1st degree

50 ADAM MAKKAI

Fig. 4

= unordered 'OR' node indicating that either ST1 or ST2 is to be read as the decoding.

STRATIFICATIONAL SOLUTIONS 51

Figures 2, 3, and 4 show the 'participial', the 'gerundival', and the joint either participial or gerundival interpretation of the sen­tence in a stratificational relational network description. Figure 4, in other words, builds together ST1 and ST2 without repeating, however, what is common in the two, that is, the predication that is made of the visiting velatives1 and the visiting relatives2 namely, can be danger­ous.

How does this influence our picture of the nature of the translation process and of the concept of 'Deep Structure'? Let us take a look at some real cases:

A French immigrant who knows no English arrives in the United States and wants to learn how to greet people. He is taught to say How do you do?. "How odd", he thinks to himself musingly, "this means nothing in French". And he says to himself: *Comment faites-vous ça? Later he thinks: "Isn't there a verb enchant in English?" - and he finds it in the dictionary. Later, moved by curiosity, he tries out the new verb when he is introduced to some one, and says (with a French accent, to be sure) I'm enchanted. Those who know French smile, thinking that he was joking deliberately, those who don't pull a sour face and walk away from him. Later during the course of his enculturation in the English-speaking world, he gives up saying I'm enchanted and just says How do you do? routinely.

A few years later he makes the acquaintance of a Dutchman, and they exchange notes on greetings. The Dutchman informs the French­man, that in Dutch it is perfectly all right to ask Hu maakt u het? in the situation where Englishmen and Americans say How do you do?, except for the fact that the Dutch, if rendered literally, yields *How make you it? The Dutchman and the Frenchman agree, though, that the Dutch and the English greeting are closely related, whereas the French greeting is not related to either the Dutch or the English.

A worker from Southern Germany joins the conversation a few months later, and informs our friends that he says Wie geht's? Gruss-gott, es freut mich sehr, Sie kennenzulernen, and to older ladies, if he wants to be very polite, küss' die Hand, Gnädige Frau. The German, after some prodding from the American, translates the German greeting thus: *I kiss your hand, honored lady. This creates great hilarity in the group. Now a Russian joins the conversation and says that he doesn't inquire about how people do things, how they manage, etc., nor does he stoop to kissing people's hands, in fact, he asks no question at all, but wishes them good health and says which means 'be healthy' but is not any form of 'be' and not any adjective 'healthy' but a verb, something like an imperative issued to some one *Thrive! The audience is getting more amazed all the time.

52 ADAM MAKKAI

Now they are joined by Hungarian who observes that neither questions nor commands are very nice, and that he merely wishes people a good day when he says jónapot kivánok, but after the Ger­man starts teasing him a bit and asks "didn't I hear you Hungarians say in Latin that you are the servant of the other fellow?" he ad­mits that he does say szervusz, and even acknowledges that it comes from a mediaeval Latin form servus humillimus domini respectabilis sum "I am the humblest servant of your distinguished lordship". He then adds that Hungarians picked it up from the Austrians and that it is merely a colloquial greeting form among youngsters, students, or members of a family. He. also recalls that the v of szervusz ( = servus) is frequently subsituted by in rapid speech; he, too, of­ten says szerbusz. The German at that point admits having heard this in Austria and among German students as well.

A Mexican also joins the conversation and observes that he says something very close to what the Frenchman says, encantador Señor, or encantado Señora, and the group draws the conclusion after an Italian chimes in and reinforces the Mexican's position, that Latin males are more polite than anybody else, because they exaggerate their pleasure of having met some one new.

The conversation becomes a bit livelier as they are joined by an Indonesian and Japanese. The Indonesian points out that asking questions that one doesn't really mean is the most neutral and po­lite way of behaving, hence the Indonesian apa kabar? 'what is the news?' is really the best greeting; whether the news is good or bad, the addressee will answer kabar baik. 'the news is good', after which they can get down to business. The Japanese finds Japanese customs too hard to explain and limits himself to expounding on how to say good morning in Japanese, and teaches them how to say ohayo gozaimasu and translates it to the amazed group as *it daineth to be early. "It all depends on who you talk to" - he adds apologetically, "this form, you see, contains an honorific."

Our Frenchman, Dutchman, the American, the German, the Russian, the Hungarian, the Indonesian, and the Japanese decide to form a bridge club, and meet regularly, once a week. After the game of bridge, the talk turns to their respective languages. The American had just said you'll find out tomorrow, and they are trying out the sentence in their respective native languages. The Russian is first, and says , or , and gives the usual ex­planation how the choice depends on whether he is on 'thou' terms with some one, or on 'you' terms. The Hungarian chimes in and notices that he can say the same thing in two words, just like the Russian, and that he, too, uses the present tense of the verb with a perfec-tivizing prefix added, megtudod holnap, or megtudjátok holnap, sim­ilarly depending on whether he is close to the addressee, or relates to him formally. The Frenchman opts for tu le sauras demain, vous le saurez demain, and explains how one must form the future tense of the verb savoir 'to know' and why he added le "to make sure that the person knows that what we're talking about is what we mentioned be­fore". To this the Hungarian adds that his choice of the definite

STRATIFICATIONAL SOLUTIONS 53

conjugation megtudom, megtudod, megtudja, megtudjuk, megtudjátok, megtud,iák, as opposed to the indefinite conjugation megtudok (valamit) megtudsz (vaiamit) megtud (valamit), megtudunk, megtudtok, megtudnak (valamit) was also motivated by the fact that the item of information sought by the person who was answered you111 find out tomorrow is kept in mind by the answerer. The Indonesian says Tuan akan kenal esok nari, and explains it as *Sir going-to know next day.

One evening, after losing a number of hands, the American says easy come, easy go. Once again the group is back at the translation game, and each one volunteers his version of how he would express a lack of concern over losing. The German goes first and says wie ge­wonnen, so zerronnen, and explains it as *as won, so depleted, but adds that the fact that each part, both the winning and the losing, has eight syllables in it, makes the saying highly rhythmical, al­most like a little verse. "Hard to forget" - he adds. The Hungarian thinks for a while and says: "The closest I can come to it is by saying *what he gains at the ferry, he loses at the customs (=amit nyer a réven, elveszti a vámon.) He adds that this doesn't really mean "I don't care about my losses"; rather it implies the futility of gaining or winning. The Dutchman echoes the German quite literally. "How remarkable", - they exclaim. The Indonesian and the Japanese in­sist that there is no equivalent saying, since they feel about gam­bling as a serious matter, and the Russian agrees. They decide to call the Mexican who can't think anything appropriate. First he tries *facil viene, facil va, but laughs it off as nonsense, then adjusts it to a better *facilmente viene, facilmente va, but throws that out too. Suddenly he remembers that when he was in Madrid last year he heard an older person say:

Los Dineros del sacristán Cantando vienen, e cantando se ván,

and renders it for the group as *the dollars (coins) of the sacristan come by singing, and go b singing. When they register bewilderment, he explains: "What this means is that the sacristan makes his living by singing in church, but he spends his money by going to the tavern where he also sings ... I guess this is like your easy come, easy go, no, Señores?"

As the evening wears on, they get hungry and go to an inexpen­sive 24-hour restaurant around the corner. The American orders a couple of hot dogs. This starts a speculation right away as to how they would say hot dog in their respective languages. "*Chien chaud is out" -says the Frenchman, but the Mexican counters, "no, we can say dos perros calientes con todo por favor very easy", but then he admits that this is only so in border towns, Puerto Rico, and other Spanish speaking countries where there are a lot of Americans, and he admits that he never heard it in Spain. This gives the Frenchman the idea that perhaps chien chaud is all right in Canada. He calls a Canadian friend on the phone, and reports the answer to be largely negative with a very few sporadic incidents started more as a joke. The Cana­dians his friend knows have borrowed the English term hot dog and

54 ADAM MAKKAI

are pronouncing it /ot dog/. The German shudders at the thought of biting into a *heisser Hund, and the Hungarian registers similar revulsion at the thought of putting a *forro kutya into his mouth. The Russian says " sounds terrible". The Indonesian seems almost sick as he utters: anjing yang panas. Then he orders a couple of hot dogs in English and eats them with great gusto.

But let us abandon our imaginary friends and come to grips with the prob­lem. The examples cited here are all common ones that occur with high frequency. This, then, is not a case of a specialized game in the cul­ture, such as a drawing didactism accompanied by a mnemonic verse. (For a detailed discussion of how to translate such a rhymed drawing didactism see "The Transformation of a Turkish Pasha Into a Big Fat Dummy", in Makkai and Lockwood 1973.)

As I stated above, translation should not be achievable under TG assumptions. Unless this is obvious by now to the reader, I will now summarize the main arguments concerning this proposition.

If the semantic component of a TG grammar 'interprets* the 'Deep Structure', the most we can hope to achieve is to decode the sentence as it was given in the source language. The problem now arises as to how we shall express the sentence in the target language. Let us look at visiting relatives can be a nuisance in French. A convenient translation of the sentence would be:

3. Les parents qui visitent peuvent être ennuyeux, r 4. Visiter les parents peut être ennuyeux.

I take it for granted that the following facts are obvious: (1) We have not mapped the phonology of the English, sentence onto

two different French sentences in an algorithmic way without regard for meaning. (By 'meaning' I mean here morphology, lex-ology, and semology lumped together.)

(2) Neither have we mapped the morphology of the English onto the morphology of the French. That, incidentally, may be a possi­bility. Let us see what it would yield:

STRATIFICATIONAL SOLUTIONS 55

English: French:

visit visit--ing "-er relatives parents can peut, peuvent be être a un, une nuisance ennui, incommodité

Lining the French up we get: 5. *visiter parents peut être un ennui, 6. *visiter parents peuvent être une incommodité

or any conceivable combination of morphological quasi-equivalences. It is obvious that a syntactically undoctored morpheme look-up, no matter how detailed, cannot render justice to translating the sentence even in one sense (say, the 'gerundiva!') let alone in both. If we transfer the lexotactics ('surface structure') of the English into the French, accom­modating the French habit of using les or des before parents and substi­tuting the adjective ennuyeux (both sing. and pl. here) for the nouns ennui, incommodité, etc., we would probably get (4), that is visiter les parents peut être ennuyeux, aided by the similarity on word order, but not (3), which retranslated into English gives:

7. *The relatives who visit can be bothersome.

The only possibility left would be to attempt mapping the 'Deep Structure' of the English onto the 'Deep Structure' of the French. But that cannot be done either. First of all, it cannot be done, because - according to Chomsky - the English sentence has two deep structures. The dilemma arises: Which one is to be mapped onto the French structure? Forthermore: Are there two corresponding French 'Deep Structures'? If so, what are they? But let us imagine, that the English sentence has a DS1 and a DS2. Let us imagine that these are straightforwardly available and, in accor­dance with Chomsky's claims, practically identical. Then we would have:

56 ADAM MAKKAI

English: French: DS1 (SOME ONE visits relatives : (QUELQU'UN visite les parents

((this act)) is capable of ((cette activité)) est capable being a nuisance) d'etre ennuyeuse)

DS2 (relatives are visiting : (les parents visitent QUELQU'UN SOME ONE, ((they)) are ((ils)) peuvent être ennuyeux) capable of being a nuisance)

Translation, however, STILL CANNOT BE ACHIEVED. After all, the 'Deep Structures' exist in order to be interpreted semantically, and not to be transferred from one language to the next without semantic interpre­tation! But this is the lesser objection. The main objection is that the translator simply does not know whether to transfer DS1 or DS2 un­less and until he interpreted them semantically. Thus he could transfer English DS1 as French DS1, but he could just as well transfer it as French DS2, and that would be an error. Thus the separate availability (and partial similarity) of DS1 and DS2 both in English and in French is by no means sufficient in order to translate from English to French, and vice versa.

There may be one way out of this dilemma, and it is the following: If the person engaged in translation INTERPRETS the meaning of the sen­tence of the source language, he has managed to get hold of its MEANING. He could, then, for the purposes of expressing the same MEANING in the target language, GENERATE THAT SAME SEMANTIC REPRESENTATION based on the knowledge of what the meaning of the original sentence was. But this amounts to having to be BOTH AN INTERPRETIVIST AND A GENERATIVIST AT THE SAME TIME!

If TG - Bach's statement notwithstanding - can indeed tolerate such duplicity, it will have remedied the greatest malady that plagues it at the present time. But the price for such a double-standard will be the admission THAT LANGUAGE IS STRATIFIED. The interpretivist is, of course, the DECODER, and the generativist the ENCODER.

Now, whether one is in the process of ENCODING or in the process of DECODING depends on the pragmo-ecological fact that human speech occurs

STRATIFICATIONAL SOLUTIONS 57

not in a vacuum, but in definable and describable circumstances under­standable as SWITCHABLE ROLES. Both the speaker and the hearer are role players; the roles they play are regulated by SOCIAL INTERACTION PSYCHOL­OGY (SIP). A person speaking (orsinging) out loud without anybody to hear him is either mad, trying to make the time pass quicker in solitary con­finement, or walking through a dark forest after midnight; additionally he may be practicing for a recital, dictating to himself as he types, practicing reading in a foreign language, etc. But TG has not even reachec the insight that a person either encodes or decodes, let alone the even more basic fact, that a human being is a role player under SIP. 1.4. The translation process works essentially as it has been presented in my article "The Transformation of the Turkish Pasha into a Big Fat Dummy" (Makkai 1971, Makkai and Lockwood 1972), and can be schematically represented as follows:

See diagram (Fig. 5) overleaf

THE TRANSLATION PROCESS A = Translation from French to English, where A' is the French

input and A the English result. = Translation from English to French, where B' is the English

input, and the French result. B" is the point in the translation of English into French at which th e translator, having properly decoded (i.e., under­stood) the English text, appropriately chooses the French medium (and not, say, Spanish) for the rendition, and at which general human knowledge must aid the translation pro­cess in conjunction with what is available in French.

A" serves the same purpose when we proceed the other way round, from French to English, without accidentally winding up speak­ing German, for instance.

( J = Greatest likelihood of s u c c e s s = Limited likelihood of success

Second greatest likelihood of s u c c e s s - No chance of success = strongly influenced by typology and the

register, tenor, and mode of the text

58 ADAM MAKKAI GENERAL

HUMA

N COGNITION

THE LA

NGUA

GE S

WITCH

*'

A"

Language A

(English)

English

Semology

English

Lexology

English

Morphology

English

Phonology

Lang

uage

(French)

Fren

ch

Semology

Fren

ch

Lexology

French

Morp

holo

gy

French

Phonology

Fig. 5

(THE TRANSLATION P

ROCE

SS)

For

legend a

nd e

xpla

nati

on,

see

p. 58

(above)

STRATIFICATIONAL SOLUTIONS 59

The translator receives through the transmitting channel (air, marks on paper, etc.) minimal morpheme building signals, P H O N E M E S , or M O R P H O N S ; in writing GRAPHEMES or GRAPHOMORPHONS. (The morphons are comparable to the 'morphophoneme' and the graphomorphon to the written version of the 'morphophoneme'.) These morpheme-building, or rather, morpheme-realizing elements are decoded into morphemes, and words. The morpheme and word sequences are further decoded into clauses and sentences, and the clauses and sentences are decoded as one, in some instances as two, or more sememic traces, of which these clauses and sentences are the realizations. Having reached the sememic stratum, the decoder-translator switches roles, becomes the encoder-translator. Due to his cognitive apparatus he will be able to tell W H I C H LANGUAGE TO CHOOSE in a given situation, hence the box on the figure called the 'language switch'. In the re-encoding process he now chooses the appropriate sememic trace for the target language. The sememic trace of the source language and that of the target language may frequently be similar, even identical, if the clauses involved are of an ordinary science article type, or a political-international newspaper article style and, especially, if the languages involved are TYPOLOGICALLY SIMILAR. Thus between German, English, and French there will be many more similar, even structurally identical sememic traces than between English and Hungarian, English and Eskimo, English and Hopi. Whereas this is not the place for me to enter into a detailed criticism of the TG 'univer­salist' hypothesis, I would like to suggest that it, too, can be met head-on (cf. Birnbaum in press, Makkai 1973 and 1974). Having chosen the appropriate sememic trace, the translator-re-encoder procedes to realize that trace according to the lexotactic patterns of the target language. These, too, may bear greater or lesser similarity with the sentence structures of the source language, but any resemblance is strictly acci­dental and is NOT the result of an underlying identical 'Deep Structure'! Again, typological similarity will have a large part in any similarities. The same holds for the morphological stratum. It has been noticed through­out the past 150 years of linguistic scholarship, that languages differ the most in the arrangement of morphemes, that is, in their morphosyntax.

60 ADAM MAKKAI

This is why traditional typology has been concentrating on the morphol­ogies of the languages compared. (See Makkai 1973, in which this fact is discussed in detail and is graphically illustrated in terms of a 4,500-sided diamond in space.) After the appropriate morphotactic arrangements have been made, the re-encoder realizes the morphological material in terms of morphons and phonemes, or grapho-morphons and graphemes. Addi­tionally, he may realize them as motions of the hand (as in the American Sign Language for the deaf) or as distinctive configurations of bumps on paper (as in Braille).

2.0. What is 'Deep Structure', really? An imperfect approximation of the sememic traces of stratificational grammar, additionally hampered by the inept admixture of surface elements and the paranoid assertion that 'Deep Structure' is synonymous with Language Universals. (See Lamb'sreview of Chomsky's Aspects, Lamb 1967.) For a detailed treatment of sememic traces and their interrelationships with sentences see D. G. Lockwood 's Introduction to Stratificational Linguistics (1972).

To illustrate the untenability of the 'Deep Structure' hypothesis, I will list below a number of sentences all of which, according to the 'Standard Revised Theory' of 1965 would have identical 'Deep Structure' slightly altered by a number of surface transformations:

8. Jim and George hope to save Noam's paradigm by throwing pornography at the public.

9. It is hoped by Jim and George that loam's paradigm would be saved by throwing pornography at the public.

10. By throwing pornography at the public, Jim and George hope to save Noam's paradigm.

11. Noam's paradigm will be saved, Jim and George hope, by their throwing pornography at the public.

12. It is by throwing pornography at the public that Jim and George hope to save Noam's paradigm.

13. It is Noam's paradigm that Jim and George hope to save by throwing pornography at the public.

14. It is by throwing PORNOGRAPHY at the public, that Jim and George hope to save Noam's paradigm.

15. It is by throwing pornography AT THE PUBLIC that Jim and George hope to save Noam's paradigm.

STRATIFICATIONAL SOLUTIONS 61 l6. It is NOAM's paradigm that is hoped to saved by

Jim and George throwing pornography at the public. 17 It is Noam's PARADIGM Jim and George hope to save by

throwing pornography at the public. 18. It is Jim and George who hope to save Noam's paradigm

by throwing pornography at the public. 19. It is AT THE PUBLIC that Jim and George are throwing

pornography whereby they hope to save Noam's paradigm.

See overleaf (p. 62), for the diagram representing the 'deep structure' of the sentence "Jim and George hope to save Noam's paradigm by throwing pornography at the public".

For Fig. 6, see next page.

62 ADAM MAKKAI

Jim

and

George

hope

save

paradigm

Noam

has

para

digm

Ji

m and

Geor

ge

Fig. 6

'Deep Structure' of

Jim

an

d G

eorg

e ho

pe

to

save

No

am's

para

digm

by

th

ro

STRATIFICATIONAL SOLUTIONS 63

Figure 6 is in no way meant as the only possible TG analysis of (8); it will, however, serve our purposes here since it represents (with simpli­fications and omissions) that version of 'Deep Structure' which is com­monly taught in linguistics courses in the USA these days. It can be represented also as:

(20) # ( ((Jim and George)) hope to save paradigm ((Noam has paradigm)) ((Jim and George)) throw pornography at the public) #

TG has claimed ever since its early inception, that transformations are meaning-preserving. TG has also claimed that transformations apply in terms of ORDERED RULES. But the arrangement of 8-19 has nothing about it that forces, say, 14 to follow 13, and not the other way round. The order of these sentences makes no difference at all; we might as well start with 19 and work our way back to 8. This is the first observation that needs to be made. Second, and more important, is the fact that na­tive speakers of English DO NOT MEAN THE SAME when saying 8 through 19. The truth is that EACH OF THESE SENTENCES MEANS SOMETHING ELSE. Let us take a brief look at the main differences in meaning:

8. This is the 'neutral' or unmarked version of the sentence. Jim and George are thematic, paradigm and public carry the predict­able 'new information stress' in accordance with their clause and sentence final positions, respectively.

9. Hope is thematic, Jim and George are shifted down to agents, saved has clause final stress, the rest is as in 8.

10. Throwing pornography is thematic, public has clause final stress, Jim and George are downshifted to agents in the adjoined clause, public carries the sentence final stress signalling new information.

11. Noam1 s paradigm carries the theme, Jim and George are down­shifted as agents in a secondary clause, and public carries the sentence final stress of 'new information1.

12. This sentence implies that it is precisely by throwing porno­graphy at the public (and not by some other means) that Jim and George hope to save Noam's paradigm. I.e., throwing pornography is topicalized sememically; the result is the cleft sentence construction on the lexotactic level.

13. Here it is Noam's paradigm that is topicalized; the implication is that it is not anything else they hope to save with their verbal habits, but precisely Noam's paradigm.

64 ADAM MAKKAI

14. This sentence has contrastive stress, which implies a differ­ence in meaning. Otherwise identical with 12, this sentence implies that precisely by throwing PORNOGRAPHY and not kindness) at the public, do they hope to save Noam's paradigm.

15. This sentence has contrastive stress on AT THE PUBLIC implying that they don't throw pornography at each other.

16. Contrastive stress on NOAM. It isn't BLOOMFIELD's paradigm they hope to save, but precisely Noam's.

17. Contrastive stress on PARADIGM. It isn't Noam's income tax they hope to save by throwing pornography at the public, but pre­cisely his PARADIGM.

18. Cleft sentence construction topicalizing Jim and George and not Edith and Elizabeth.

19. Cleft sentence topicalizing AT THE PUBLIC. The implication is that they do not throw pornography at a few isolated individuals, but AT THE PUBLIC.

Needless to say, these explications of the differences in meaning hold­ing in sentences 8-19 are'greatly oversimplified; nor are they the only possible explications possible. Further complications arise if we realize that each cleft sentence construction creates a different the­matic structure and that within each differently thematicized cleft structure a number of different contrastive stresses are possible. Hence the number of permutations and variations on each basic, neutral (or, to use Halliday's term, 'unmarked') construction is very high indeed. According to Halliday (personal communication) the number of possible varieties (each grammatical) accommodating topicalization/clefting, focus shifting via the passive construction, re-thematicization, and contras-tive stress is most probably 2 2 5, that is, up in the millions. It thus makes obviously no sense to order these events with relation to one another. Neither does the speaker ever think of an underlying sentence first which he then procedees to 'transform' in order to carry just the right desired shade of meaning. How could he? After all, the meaning of a sentence cannot be clear to anybody unless and until the structure of it is there for him to interpret it semantically. Ludicrous as it sounds, but it is true: IF THE TRANSFORMATIONAL-GENERATIVE ASSUMPTION OF CHOMSKY THAT THE MEANING OF A SENTENCE IS DERIVED BY SEMANTIC PROJECTION RULES

STRATIFICATIONAL SOLUTIONS 65

FROM ITS DEEP STRUCTURE WERE TRUE, PEOPLE WOULD BE UNABLE TO TALK. The only way they could talk would be if they said first what they were going to say, and then listened to themselves in order to determine what they said. This is a sad comment on the great 'mental ist revolu­tion' indeed!

The -reader is repeatedly asked to remember that we are playing a deliberately constructed counter-game: We are not accusing Chomsky's model of having stated that this is what people do. What is being in­sisted upon here is that it is a legitimate counter-move to pretend that his model actually implies that, once we remove the artificial se­paration of 'performance' and 'competence'. It may be useful at this point to return to the Yngve quotation cited earlier in this paper.

What I am attempting to do here is to show how absurd the theory would look, if it were to be a picture of how we produce and decode dis­course in natural languages. If this counter-game is temporarily accepted (its obvious limitations notwithstanding), it emerges that the revised-standard theory of TG turns into a perverse caricature of itself if cred­ited with the desideratum of 'explanatory power' regarding the very im­portant question of how humans produce and perceive sentences. In this light I hope that my observation above is now clear. To repeat: If we had to project the deep structure of a sentence to 'read it' before we unterstood it; we would literally not be aware of what we said, before we have said it. But I, for one, usually have a more or less clear ideal in my head before I open my mouth (as I hope, does the reader of this paper). I, therefore, postulate that a speaker's intended message does NOT start with the deep structure of the syntax of his sentence, but rather with a configuration of concepts in his consciousness, known in stratificational linguistics as a Sememic Network or a Sememic Trace. But back to Jim and George hope to save Noam's paradigm by throwing porno­graphy at the public.

The basic task of this section of the paper is to point out that since sentences 8-19 do have demonstrably different meanings, it makes no sense to derive them from the same 'Deep Structure'. The truth of the

66 D MAKKAI

matter is that each of the above sentences has its own sememic trace, each differing from every other in some real and appreciable way. This is not to deny, of course, that they also resemble each other to a con­siderable extent. But just because two brothers resemble each other (to quote Lamb's striking analogy offered during his talk at the University of Washington in Seattle in August 1973) it doesn't follow that they can be explained as deriving from one another. In fact, just as even two closely resembling brothers are the descendant of their parents (and their parents at DIFFERENT STATES OF THEIR LIVES, even if the brothers are twins!), two AGNATE sentences (the term is Gleason's) (or, for that matter ANY NUMBER of agnate sentences) are related to one another as daughter languages are related to one another via the proto-language, and never as thought of by beginning undergraduates taking a course in linguistics who think that English has descended trom Sanskrit.

TG supporters might object at this point insisting that for sen­tences 8-19 to come out as they are, a separate sememic trace had to be formed for each, which is repetitious. It would be simpler, one could suggest, to PERFORM MINOR OPERATIONS ON THE SAME TRACE. (The reader is referred here to Lockwood's introduction, 1972, and especially chapter 5 'Sememic Phenomena'.) The problem with this suggestion is that it pre­supposes that the speaker had an earlier (say, the 'unmarked') version on his mind before he came out, say, with 18 It -is Jim and George who ... etc. This is definitely not the case in ordinary speech. The speaker forms THE RIGHT TRACE IMMEDIATELY based on the contextual evidence avail­able to him - as far as possible. Admittedly, there are exceptions. The most significant exception is when a person 'has something on his mind' but 'doesn't quite know how to say it', and tries a number of different ways. This sort of behavior is most evident during composition; the writer will scratch out several sentences, sometimes even whole para­graphs and pages and start all over again. His intention is to tell THE STORY in the most effective way; he is searching for alternative realiza­tions of the same set of contextually interlinked sememic traces. We must, therefore, not dismiss entirely the possibility of performing minor sur-

STRATIFICATIONAL SOLUTIONS 67

gery on the same trace for better stylistic effect during composition. But such trace-surgery is not TRANSFORMATION in any conceivable sense of the term. For one thing the meaning is thereby appreciably altered; sec­ond, the basic proposition has not been altered totally, but only par­tially. The term ANASEMIOSIS seems appropriate to use for trace-surgery. We must remember that anasemiosis only applies when there has in fact been a previous attempt at forming the right trace, but for some reason or other the speaker (and most often the writer) chooses to edit it.

Every once in a while we encounter different wordings which do NOT carry any appreciable difference in meaning. Consider:

20. It bothers me that she snores. 21. Her snoring bothers me.

If these two mean different things to any one, he is entitled to regard them as the results of different traces. It is arguable that bothers is thematic in 20, whereas snoring is thematic in 21. (I do not mean for 21 to read her SNORING bothers me which implies contrastive stress.) I would like to suggest the term ANALOGOSIS for varieties of this sort. Clearly, 1-19 are all instances of analogosis, but they are instances of SEMEMICALLY CONSEQUENTIAL ANALOGOSIS, whereas 20 and 21 (for those speakers at least to whom theme doesn't matter much) may be viewed as SEMEMICALLY NONCONSEQUENTIAL ANALOGOSIS. Analogosis, then, is also not a case of 'transformation' but is motivated by (a) thematic choices, (b) cognate vocabulary structure, or (c) frequency of patterns. Thus if I say:

22. It irritates me that she breathes so loud.

I have uttered a sentence that is closely related to 20 (the syntactic structures are E N A T E ) , but also one which differs from 20. in VOCABULARY. Is saying the same sentence in different words a matter of transformation' Hardly. Consider:

23. He had to go to the foot-doctor. 2k, He had to go to the podiatrist.

68 ADAM MAKKAI

Several interesting questions arise, none of which, incidentally, has been faced by TG grammarians. Is podiatrist a transformation of foot-doctor, or the other way round?

TG has failed to face up to this problem during the past seventeen years. I would suggest that these are instances of A N A L E X I S . The exis­tence of analexis proves that the vocabulary of the English language, for one, is diachronically and dialectally stratified, and that this diachronic and dialectal stratification can reveal itself in the speech of the same person at a given time as synchronic stratification. If I say

25. The teacher walked around the building, versus 26. The educator circumambulated the edifice.

I talked plain English in 25. and stilted latinate English in 26. This choice may mark me socially as normal and young, or weird, old, and pedantic; it may show the informal nature of the situation in which 25. was uttered, and the utmost rigidity of the situation in which 26. de­veloped. A N A L E X I S and ANALOGOSIS may co-operate in the production of different sentences whose traces may differ markedly, or only minimally:

27. Around the building walked the teacher, versus 28. The edifice was circumambulated by the educator.

I hasten to add here that foot-doctor C O U L D , in fact, be derived from podiatrist, especially the foot from the *pod-(os). So could, addition­ally, niece and nephew from *nepot(ism), governor and gubernational from *kybern(es, -etics, cf. cybernetics) along with father, mother and broth­er from *parti(cide), *matri(cide) and *fratri(cide). What WOUld need to be done would be to reinvent Grimm's Law, Verner's Law, and most of the work carried out on Indo-European in the 19th century. There would still, I think, remain insurmountable troubles, as it would be almost impossi­ble to derive dog from can(ine patrol), bird from aviary, moon frm selen-(ology) r lun(ar), r grass from herb(arium) along with water from aqua(rium). The reason why I mention these semantically linked pairs in modern English is that E T Y M O L O G I C A L P H O N O L O G Y , a contemporary of TG-style N A T U R A L P H O N O L O G Y , has attempted similar derivations. (See Lock-

STRATIFICATIONAL SOLUTIONS 69 wood 1975 in press.) The way things are, here and now, modern English is a complex ecological system in which POLYCHRONOLOGICAL SYMBIOSIS is overtly manifest. I can walk into a camera shop and ask for a three-foot tripod without contradicting myself, although I have repeated the same *IE words for '3' and 'foot', respectively; once as processed by Grimm's Law for the Germanic languages, and once as borrowed into modern English from Greek, where Grimm's Law has not been operative. Thus, from the point of view of modern English three-foot is 'older', if we evaluate age from the point of view of English by itself (this would be the ENDO-ECOLOGICAL V I E W ) , but, of course, tripod is the older form, if we look at the question from the EXO-ECOLOGICAL point of view. Here and now, endo-ecologically speaking, tripod is by far the 'younger' form, recorded in the OED as occurring first in 1611 in the sense 'three-legged vessel', and in the photographic sense first in 1825.

Analexis, in point of fact, is ecologically analogous to the various ways in which the sentence Jim and George hope to save Noam's paradigm by throwing pornography at the public can be re-encoded in a number of agnate structures. Just as the various realizations of the sentence do not derive from one another but from the sememic network, analectical lexemes, as the ones cited above, do not derive from another synchronically. Dia­chronic derivations are sometimes valid, and sometimes not; the language must tolerate symbiosis. The implications of 27. and 28. are obvious to native speakers of English.

To illustrate how meaning (semology) can be related to sentences (syntax, lexotactics) without transformations yet in such a way that AGNATE and ENATE structures are accounted for (indeed 'generativity' is nothing else but a confusion regarding enate and agnate structures), I will present here one possible stratificational analysis of a set of related German sentences.

70 ADAM MAKKAI

Nouns : Verbs;

Katze weinen Pferd lachen Kuh spazieren Ochs gehen Hund scherzen Mann sitzen Frau atmen Panzer rennen Bett etc Buch

etc.

1 = Die Kinder schlafen.

2 = Das Kind schläft.

3 = Schläft das Kind?

4 = Schlafen die Kinder?

STRATIFICATIONAL SOLUTIONS 71

Fig. 7/a A Key to Stratificational Diagramming

Other names: AND = conjunction OR = disjunction

ORDERED UNORDERED =

sequential coincidental

D O W N W A R D (-encoding-)

D O W N W A R D (-encoding-)

ordered AND unordered AND

ordered OR unordered OR

tactic diamond tying structures on one stratum to those on a higher stratum

ordered AND unordered AND

U P W A R D (-decoding-)

ordered OR unordered OR

U P W A R D (-decoding-)

72 ADAM MAKKAI

Figure 7 describes the German sentences; 29. Die Kinder schlafen 30. Das Kind schläft 31. Schläft das Kind? 32. Schlafen die Kinder?

Under nouns (also eligible in the sememic trace) we have Katze, Pferd, Kuh, Ochs, Hund, Mann, Frau, etc., under verbs (also eligible in the trace) we have weinen, lachen, spazieren, gehen, sitzen, rennen, at­men, etc. The unordered 'AND' node on top indicates that the speaker, before committing himself to one structure of the other, has the free­dom to form ANY on the appropriate traces as the pragmatics of the sit­uation demand it, without having to resort to any given sentence struc­ture (real or abstract) from which to 'derive' any one alternative. The first choice (unordered 'OR') allows the speaker to choose between de­clarative, interrogative and conditional; as it happens the German con­ditional (or at least one variety of it) has the same word order as the interrogative, e.g.,

33. Schläft das Kind? (interrogative, "does the child sleep"?) 34. Schläft das Kind, so können wir ins Kino (conditional,

"if the child sleeps, we can go to the movies")

On the right hand side of the trace the speaker can choose between plu­ral or singular, hence the trace will 'generate' either

35. Die Kinder schlafen ("the children are asleep") or 36. Das Kind schläft ("the child is asleep")

Since sleeping is not any sort of agency, the sememe S/Medium/ is next conjoined with the various verbs and nouns that are pragmatically eligi­ble for this family of traces.

The actual sentences are ordered downward 'AND' nodes 1, 2, 3, and 4, on the lexemic stratum. These sentences, in turn, are realized by the lexons definite article plural, definite article sg. neut.y Kind, plural verb ending-n-en, the verbal stem schlaf- the present singular

STRATIFICATIONAL SOLUTIONS 73

ind icat ive, and the plural third indicative. Depending on which one is needed, schlaf will be schlaf + en, or schlaf- + t, which is a morphemic-morphophonemic matter, no longer under the jurisdiction of the lexotac-tics, hence omitted on this diagram. Notice, incidentally, that the 'gen­erativity' of such a TRACE FAMILY is very powerful indeed.

37. Atmen die Männer? ("are the men breathing"?) 38. Die Frau sitzt. ("the woman is sitting".) 39. Die Frauen sitzen. ("the women are sitting".)

are merely some of the possibilities accounted for (or 'generated') by this particular trace family. Given the very large number of nouns and verbs that can be predicated to each other in German in some real or imaginary sense, this trace family generates millions of sentences with­out a single grammatical transformation or any ordered rules. It simply shows what is stored in the human brain and what paths the speaker selects when he encodes his experience.

I will rest the case of sememi traces versus 'Deep Structure' here by saying that 'Deep Structure' was an aborted attempt on the part of Chomsky and followers to explain meaning in syntax, by using the artifi­cial examples of 'ambiguous sentences' out of context. The attempt failed, however, because transformationalists were unable to give up the idea that sentences can only derive from other sentences (or sentence-like abstract pre-sentences) and because of the computer-inspired fixation of rule ordering. Since a digital computer can only allow or disallow electricity to cross a set of wires, its choices are always binary and must be ordered. Even though Chomsky and his followers are no longer ac­tively involved in computer work, without the mechanical translation fad of the late 'fifties and 'sixties financed by the Army, Navy, and other branches of the US government (later viciously attacked by Chomsky turned left wing politician) TG would not have developed into the binary-logic bound mechanical artifact that it is today in all of its forms. 3.0. In this section of the paper will focus my attention on idioms

74 ADAM MAKKAI (cf. Makkai 1972.) That TG has had nothing of value to say on the matter has been admitted in print by transformationalists (Binnick 1974). 3.1. LEXICAL IDIOMS are multi-morpheme or multi-word sequences which correlate with a definite syntactic function (verb, noun, etc.) and whose meaning does not follow from the standard lexical meaning of the parts when occurring in other environments. Thus hot dog is not a dog that is hot, but a 'Frankfurter in a bun'. This simple observation in itself completely defeats TG in one simple shot, since this fact cannot be accounted for by any derivation, transformation, or any other arti­fact of the system. If you start with a DS derivation predicating of a certain N to BE ADJ., you will wind up with a surface construction hot dog, which, however, will be stressed the wrong way. The same goes for redcap, which is not a red cáp. The lihite House, which is not white

hóuse, and bláckbird, which is not black bird. Lees in his Grammar of

English Nominalizations (Lees 1960) openly admitted that the generative method has no way of accounting for such semantically aberrant and ir­regularly stressed forms. The point I must reiterate here is that these idioms do not DERIVE from any underlying and syntactically mechanically produceable form. (That these forms, too, have syntax, is commonplace knowledge; but the internal syntax of idioms and their behavior in sen­tences are two independent matters.) In SG the lexical idiom is a complex 'AND' node that leads 'downward' to its constituent lexons and morphemes which, in other environments, are the realizates of other lexemes and sememes with the idiomatic lexeme having its own separate sememe. SG does not commit the error, in other words, of trying to derive hót dòg from some fictitious (the dog ((WH dog BE hot)) ) while due to the sen­sitivity and flexibility of the relational network system it can accu­rately show what an idiom means, how it is realized, and what is does in the sentence. 3.2. SEMEMIC IDIOMS are sentence or clause-length, institutionalized utterances which are the realizations of more than one sememic trace. In this regard, then, sememic idioms resemble 'ambiguous sentences'. Fig­ure 8 shows the sememic idiom don't count your chickens before they're

STRATIFICATIONAL SOLUTIONS 75

hatched. The CONTEXTUAL ADJUSTABILITY PRINCIPLE (C.A.P.) as a function of cognition signals to the individual whether to encode the sentence in the sense 'do not enumerate your fledgling chicks before they are out of their shells' (ST1) or in the idiomatic sense 'refrain from celebrat­ing prematurely' (ST2). The black unordered 'AND' on the left may or may not be called into play regulated by the unordered 'OR' above it; if it is, the decoder decodes IN B O T H SENSES which amounts to PUNNING ON THE IDIOM. Such a situation can arise if a farmer actually counts yet unhatched chicks and some one warns him by saying the proverb; it would be signalled that he is foolish for celebrating prematurely and that he is enumerating unhatched chicks. Failing the activation of this unordered 'AND' node, the decoder reads the sentence either in the sense of ST1 or in the sense of ST2, as the pragmatics of the situation demand it.

For Fig. 8, see overleaf (p. 76)

76 ADAM MAKKAI

Fig. 8 A r e l a t i o n a l network

description of Don't count your chickens before they're hatched.

to context

to cognition

unordered AND (coļncidence concaten-ation) unordered OR

(coincidence dis­junction)

THE IDIOM SHARED STRUCTURE THE MEANING

STRATIFICATIONAL SOLUTIONS 77

What matters is that there is no legitimate sense in which the one sense can be 'derived' from the other, except historically, TRANSDERIVATION, as used by Newmeyer (1972), is [+ tricky - honest]. Historically, of course, the proverbial idiom 'derives' from one of Aesop's Fables in which a foolish person did count chickens before they hatched, just as The White House is a house that is, incidentally, whi te, and as a black­bird is also a bird that is, incidentally, black. (But notice: The Texas White House is a yellow barn, the flying White House is Air Force No. 1, this baby blackbird is white, it must be an albino. )

SG, as can be readily seen, handles idioms in a much more elegant and efficient way than other theories. It accounts for their literal versus their idiomatic sense, indicates what sememic traces they are the realizations of, and does NOT,unwarrantedly, mix diachronic with syn­chronic considerations. 4.0. What, we may ask, is the COMMON DENOMINATOR of the various failings of TG - if by 'failings of TG' we understand that the theory, having driven an artificial wedge between 'performance' and 'competence', will accept the output of left-to-right rewrite rules processed cyclically even if they are counterintuitive, while not being able to account for much simpler cases where more than one possible sense to a sequence of sounds is available, and vice versa.

The basic inadequacy of TG is that it regards human language through MUTATION RULES, instead of looking at it as a SYSTEM OF RELATIONSHIPS (see Lamb 1975, in -press).

4.1. In this portion of the paper, as a closing argument, I will ad­dress myself to the common human experience of DOUBLE CODING. Double coding occurs when a lexeme, a phrase, or a whole paragraph in spoken discourse has a discernable second (or even a third and fourth) meaning beyond the institutionalized, lexico-gramrnatically retrievable meaning. 4.1.1. The simple lexeme yes has - at least - eight commonly recogniz­able meanings, depending on the intonation and the length.

78 ADAM MAKKAI

Fig. 9 Eight commonly recognized meanings of yes in American English

(See opposite page for detailed analysis)

yes1

unmarked. ' Objective affirmative'

yes2

'Enthusiastic approval'

yes3

'Reluctant con­descendence '

yes4

'Tell me more, I am listening'

yes5

'Joyful sudden re­cognition'

yes6

'I think you're kidding me'

yes7

'I am registering my attention with­out committing myself'

yes8

'What you say makes sense at first hear­ing; let me think it over'

STRATIFICATIONAL SOLUTIONS 79

Eight commonly recognized meanings of yes in American English: (1) Is by far the commonest; hence we regard it the 'unmarked'

form. The intonation falls from 2.8 or 2.9 to 1 and signals objective consent. It is typically heard in exchanges such as Q: Do you have change for a dollar? A: Yes1

(2) Means something like 'great' or 'wow!' The pitch rises sharply and briefly from 3 to 4. Q: Would you like me to take you to Hawaii for Christmas? A: Yes2.

(3) Takes up twice as much time as (1) or (2). The pitch drops from 2.3, or 2.5 to a low, drawn-out 1. Q: Are you going to do the dishes? A: Yes3.

(4) Has the voice rising in pitch from a low 1 to the top of the scale; the rise is drawn out over two measures. It sounds al­most like a question. Typically heard when an unknown sales­man approaches a customer over the telephone and starts making an appealing proposal. Q: Sir, I am telling you about our new investment possibilities on Grand Bahama Island where we take our customers by jet economy all paid by us - do you have a minute to talk? A: Yes4.

(5) The pitch drops suddenly and sharply from 4 to 1 in one measure. An old, almost forgotten friend announces himself over the telephone: Q: This is Jack Mulligan calling, your old room mate from college? A: Yes5.

(6) The voice starts at 4, dips down to 2 in a drawn out manner, and goes back up again to 4 filling up two measures. Q: There is this strange costume party at the Taylors tomorrow and I was asked if you'd care to come along, - you see, it is supposed to be a surprise for Joanna. A: Yes6.

(7) The word yes is repeated three or four times in the time of two measures in a colorless, even-keyed fashion held at level 2, as if the speaker is just making noise to keep his inter­locutor talking but isn't really paying serious attention.

(8) Indicates that the person saying yes has heard this kind of question or argument before; is not totally surprised by it; that he is considering it; that he thinks the interlocutor has a point but not one that could not be better stated or improved. Typically heard from university lecturers who are interrupted by a student.

4.1.2. If we were to take transformationalism seriously, we ought to posit a 'Deep Structure' for 'yes' of which these (as well as many oth­er possibilities) would be 'transformations'. They could be called the 'Enthusiasm Transformation', the 'Reluctance Transformation', the 'Tell

80 ADAM MAKKAI

Me More Transformation', the 'I am So Glad Transformation', the 'Kidd­ing Transformation', the 'Noncommittal Transformation', and the 'I heard that Before Transformation', respectively. The first one may be regarded as the 'Deep Structure of yes'. I have no idea how these transformations would be ordered, or what would trigger them. Depending on where the grammarian stood with regard to phonological theory, he might attempt to generate these intonations as phonetic representations of the 'surface structure' of yes. But I don't think that would really work. For it seems to me that it is in the very nature of these into­nations that the MEANING of the given yes rests; as if phonetics were NOT theend result of a transformational cycle that maps surface struc­tures into systematic phonetics, but - in cases such as these - an in­tegral part of the SEMANTICS of the speech act.

Stratificational grammar can easily handle situations like this by virtue of the UNORDERED AND (the coincidence concatenation) concept in COGNITION. The one line of the node goes directly to the sememe, the lexeme, and the morpheme, realized by the phonemes /y/, /e/, /s/, with the socially institutionalized major meaning 'affirmative' activated. At the same time - and without any order - another line from the same unordered AND node in cognition can go to the sememes s/yes1/, s/yes2/, s/yes3/, s/yes4/, S/yes5/, S/yes6/, s/yes7/, and s/yes8/ ON A N Y ORDER, and at ANY TIME during the same conversation, several times.

Playwrights, actors, and stage directors know this extremely well. Additionally, of course, there will be stage instructions as to what kind of facial expression to put on while pronouncing one of the appro­priate 'yes'-affirmatives. 4.2. The phenomenon of poetry is, as I see it, not explainable without double (or multiple) coding. The simple fact that two words rhyme in a certain environment changes their relative meanings vis-à-vis one an­other. If a child says

The oat

In the hat That chased the vat

STRATIFICATIONAL SOLUTIONS 81

Sat on a mat

And did this And did that

But he looked very sad

one of the meanings of each marked item - in addition to what it means institutionally - is that it RHYMES with each other marked item. In the common tongue-twister

Peter Piper picked a peck of pickled peppers; 't is a peck of pickled peppers Peter Piper picked

perhaps the MAJOR MEANING of the entire tongue twister is the difficulty of pronouncing a series of aspirated initial /p/-s; a fact very much on the speaker's mind when performing the familiar tongue-twister. 4.2.1. Pursuing my previous line of attack, that is, pretending that TG can be taken at face value, let us imagine what kind of transfor­mational rules we would need to make a statement in 'Deep Structure' BECOME a tongue-twister, difficult to pronounce.

We would probably start out with a logical proposition stating that SOMEBODY with a GIVEN NAME and a SURNAME performed in the P A S T an ACTION, the action was CHOOSING and the ITEM chosen was CUCUMBERS that were made SOUR. Then the deep structure would look something like:

X. Y. Past+choose a bundle of sour cucumbers

82 ADAM MAKKAI

Our problem now is this: how does this BECOME a tongue twister? X. Y. must, first of all, undergo the PERSONALIZATION TRANSFORMATION, SO we can call him John Smith. But this isn't enough; John Smith does not 'pop' with heavily aspirated initial /p/-s; so we have to submit John Smith to the POPPING TRANSFORMATION. HOW does that work? The ideal speaker-hearer searches his memory until he finds sounds that 'pop'. /t/ and /k/ are also eligible, in a sense, since initially, they, too, are heavily aspirated. Thus, strictly speaking, there is no reason why the deep structure cannot become something like Tommy Tinker took a tank of tangy tomatoes, where the POPPING TRANSFORMATION (=initially aspi­rated stop) would be present. So we have to specify that /t/ and /k/ must be out; by having a P-POPPING TRANSFORMATION, a T-POPPING TRANSFOR­MATION, and finally, a K-POPPING TRANSFORMATION, unless we want to run the risk of winding up with the wrong tongue twister, such as Tommy Tinker, r, With the /k/-popping, Kelly Galley collected a kilo of chlorinated cucumbers.

Each specific tongue-twister would have to go its own specific POPPING TRANSFORMATION, and even then we are still not in the clear. For Tommy Tinker took a tank and Kelly Calley collected a kilo would result in AD HOC TONGUE-TWISTERS, Whereas Peter Piper WOUld result in a widely recognized, INSTITUTIONALIZED TONGUE TWISTER. Thus the trans­formational rules would have to be rendered context sensitive with reg­ard to INSTITUTIONALIZED OUTCOME versus RANDOM OUTCOME.

The import of this observation cannot be overemphasized. It simply means that GENERATING SOMETHING I S NOT THE SAME AS CREATING IT. Gener­ating an utterance logically implies that we knew IN ADVANCE what the utterance was going to be, and then laboriously accounted for it by re­assembling it. Functionalists and structuralists were, then, more honest, because they never claimed to have done anything else BUT decompose ut­terances that were ready-made; the TG grammarian, on the other hand, prides himself on 'generating1 new sentences when what he, in fact, DOES do is merely re-assemble new sentences from the parts of previously ob­served and analyzed ones. Being a speedier way of operating, this creates

STRATIFICATIONAL SOLUTIONS 83

the mirage of progress and 'mentalism': MAN himself is in control behind the gears. But is he really? 5.0. In this paper I have deliberately chosen those areas of man's lin­guistic behavior which, if TG were to be understood as a theory describ­ing WHAT P E O P L E D O , would show beyond a reasonable doubt THAT T H I S IS NOT HOW MAN BEHAVES. In so doing I may have been prejudiced and unfair to transformational-generative grammar. It could be argued that there is no justification for doing this. But there is. TG, and especially its earlier miltant phase, has accused the 'Neo-Bloomfieldians' of all sorts of omissions and sorts of intellectual neglect of which they were not really guilty. Yet, in doing so, TG achieved its greatest positive contribution to modern linguistics: It has managed to ask a set of ques­tions that was previously asked only very timidly and occasionally, or not at all. TG, in its systematic war against behaviorism accomplished a clearer understanding of what is REAL about human language behavior, and what is IMAGINARY. AS it often happens in history of science, TG became the victim of its own method. It is only fitting and just, there­fore, that we linguists, in the name of progress and fair play, do the same to TG as TG did to structuralism: Even at the cost of drawing de­liberate caricatures of the system. For in such caricatures the objec­tive researcher will see the vestiges of positive accomplishment as well as the swamps and the quicksands whence there is no return.

If the present paper has managed to draw such a caricature of TG while arousing the reader' interest in stratificational linguistics, I have achieved my goal.

POSTWORD The stratificational analysis of visiting relatives can be a nui­

sance presented in this paper (pp. 44 ff.) is not the only possible one. In fact, it is possible to show that already on the lexemic level two tactic analyses can be carried out. It was my intention throughout this paper to keep the stratificational diagramming — a taxing tech­nicality — to a bare minimum and present the ' philosophy' of the matter at hand.

84

REFERENCES

Bach, Emmon. 1971. "Syntax since Aspects". Monograph Series on Lan­guages and Linguistics 24.1-17. Washington, D.C.: Georgetown Univ. Press.

Binnick, Robert I. 1974. Review of Makkai 1972. International Journal of American Linguistics 40:2.155-57.

Birnbaum, Henrik. In press. "How Deep Is Deep Structure?". Proceedings of the Eleventh International Congress of Linguists ed. by Luigi Heilmann, vol.2. Bologna: Il Mulino.

Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. (10th printing, 1972.)

. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.

Katz, Jerrold J., and Jerry A. Fodor. 1963. "The Structure of a Seman­tic Theory". Language 39.170-210:

Lamb, Sydney M. 1967. Review o'f Chomsky 1965. American Anthropologist 69.411-14.

. 1975. "Mutations and Relations". The 1st LACUS FORUM ed. by Adam Makkai. Columbus, S.C.: The Hornbeam Press, in press.

Lockwood, David G. 1972. Introduction to Stratificational Linguistics. New York: Harcourt, Brace & World.

. 1975. "Quasi-Etymological and 'Natural' Phonology as Two Varieties of the Same Mistake". The 1st LACUS FORUM ed. by Adam Makkai. Columbus, S.C: The Hornbeam Press, in press.

Makkai, Adam. 1971. "The Transformation of the Turkish Pasha into a Big Fat Dummy". Working Papers in Linguistics 3:4.267-73. Honolulu: Univ. of Hawaii, Dept. of Linguistics. (Repr., in rev. form, in Makkai & Lockwood 1973:307-1,5.)

. 1972. Idiom Structure in English. The Hague: Mouton. ' . 1973. "A Pragmo-Ecological View of Linguistic Structure and Language Universals". Language Sciences 27.9-22.

. 1974. "Take One on 'Take': Lexo-ecology illustrated". Lan­guage Sciences 31.1-6.

. Forthcoming. "Systems of Simultaneous Awareness: Possible stratificational approaches to formal poetry and music". Paper pre-

85

sented at the First International Conference on Stratificational Linguistics, Kirkland-Seattle: Summer Inst. of Linguistics; Univ. of Washington.

Makkai, Adam, and David G. Lookwood, eds. 1973. Readings in Stratifica­tional Linguistics. University, Ala.: Univ. of Alabama Press.

Makkai, Valerie Becker, ed. 1972. Phonological Theory: Evolution, and, current practice. New York: Holt, Rinehart & Winston.

. In press. "Systematic versus Autonomous Phonemics: A third alternative". Proceedings of the Eleventh International Con­gress of Linguists ed. by Luigi Heilmann, vol.2. Bologna: Il Muli­no.

Newmeyer, Frederick J. 1972. "The Insertion of Idioms". Papers from the Eighth Regional Meeting of the Chicago Linguistic Society, 294-302. Chicago: Chicago Linguistic Society.

Postal, Paul M., and Jerrold J. Katz. 1964. An Integrated Theory of Linguistic Descriptions. Cambridge, Mass.: MIT Press.

Yngve, Victor H. MS. Introduction to Human Linguistics. (A forthcom­ing book, based on lectures delivered at the Univ. of Chicago.)

. 1975. "The Dilemma of Contemporary Linguistics". The 1st LACUS FORUM ed. by Adam Makkai. Columbus, S.C.: The Hornbeam Press, in press.

NON-UNIQUENESS IN THE TREATMENT OF THE SEPARABILITY OF SEMANTICS AND SYNTAX IN COMPOUND EXPRESSIONS*

FRED . PENG

0.0 INTRODUCTION. The separability of semantics and syntax has been the focus of linguistic discussion for many centuries.1 In recent years opinion has accepted one of two opposing views: the two realms are entirely separable, or not separable at all.

One more complication arises when examining these two positions. It is that those who advocate that semantics and syntax are insepara­ble take two differing approaches. I shall refer to them as the Syn­tactic Approach on the one hand and the Semantic Approach on the other hand. The syntactic approach, represented by scholars like Bloomfield and Hockett, insists that semantics and syntax are not separable, be­cause syntax (or, more broadly, grammar) subsumes semantics. The se­mantic approach, advocated recently by McCawley2 and perhaps also by Bach and Chafe, asserts that semantics and syntax are not separable, because semantics subsumes syntax (in its strict sense). In contra­distinction to these two approaches, I shall refer to the approach adopted in the first position mentioned above (i.e., 'entirely separa­ble') as the Conventional Approach whose proponents may be said to be Katz and Fodor (and perhaps Chomsky as well).

The purpose of this paper is not at all an attempt to solve the overall dispute among the varying approaches but to show that no mat-

The substance of this paper was first presented at the Eleventh In­ternational Congress of Linguists in Bologna, Italy, in August 1972. A somewhat shortened version appeared in Language Sciences 29.13-19 (Feb. 1974) under the title "On the Separability of Semantics and Syntax".

88 FRED C. C. PENG

ter which approach is applied to particular linguistic phenomena, say, compound nouns, there exists the problem of non-uniqueness, and that each approach has its own inadequacy.

There are three good reasons for the fact that compound nouns rather than sentences are chosen to illustrate the point. First, com­pound nouns in any language are always in a state of flux, more so than any other kind of linguistic entities; that is, new compounds are created practically every day, while old ones are forgotten. No such rapid change-overs take place among prepositions or verbs. Thus com­pound nouns reveal an important area of linguistic creativity which en­hances the function exercised by the interplay of semantics and syntax. Second, if the meaning of any sentence in a language is not the sum total of the meanings of the individual words in that sentence, the same can be said of a compound noun, except that a compound noun is structurally always less complex than a sentence. In this respect, I am of the opinion that unless the separability of semantics and syntax is succinctly accounted for within the scope of a compound noun, no in­sight will be gained regarding the separability of semantics and syntax in general by plunging straight into the discussion of sentences. Third, as has been suggested by Lehrer (1971:21), one major point over which the Conventional Approach and the Semantic Approach differ is with re­spect to lexical insertion. That is, there is a difference as to whe­ther lexical items are inserted after all the base rules and before all the transformational rules, or after some transformations. In this con­nection, insertion is an interesting process by which the complexity of compound nouns can be tested, since some compound nouns, e.g., postman3, are in fact treated as lexical items and others are not, e.g., paper cup.

1.0 CHARACTERISTICS OF COMPOUND NOUNS. Before I proceed to discuss the three varying approaches briefly indicated above, let me first out­line some latent characteristics of English compounds that have thus far escaped the scrutiny of linguists. Although three-word compounds, such as teacher service materials, and four-word compounds, such as

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 89

anthropology curriculum study project, are not uncommon, I shall res­trict myself in the present study to two-word compounds.

Two-word compounds denote objects or concepts. Underlying such de­notations is the fact that all compounds in existence were created at one time or another by competent speakers of the language (normally, we may assume, by native speakers). For instance, it is a well-known fact that the compound noun iron curtain was created (or coined) by Winston Churchill. Whether or not we can trace the creator of each compound is of course immaterial here. The main point is that a compound noun is usually created by combining two words together which are already known to the creator and the other members of his speech community. If so, then the creator of each comoound -to -be must undergo a period of deci­sion-making during which he is compelled to choose, from among thousands of words, two particular words that are compatible with each other. For the compatibility of the words so chosen there is sometimes no easy or obvious explanation. I believe this is how such compound nouns as pa­per towel, butter knife, iron curtain, and liquid paper were created.

What goes on during that period of decision-making, no matter how brief it is, is an interesting process to which we should give some thought. First, the process must involve the task of identification (or association) in its cognitive sense; that is, the creator sees certain qualities in an object or concept and identifies them with the refer­ents of the two words he is to choose. Second, the process must involve the task of lining up the words chosen in a proper order; that is to say, the creator sees two ways to line up two words but must choose one way over the other in order to name a new object or concept properly.

At first sight, we may presume that all compounds created in this fashion are alike in that they have one and only one particular-rela­tionship that holds between the words of each such compound. The truth is that they are not all alike. Paper towel, butter knife, and iron curtain will suffice to illustrate this point. For one thing, a pa­per towel is indeed made of paper, but a butter knife is not made of butter, nor is the iron curtain made of iron. For another thing, butter knife has the 'word-relationship' that pertains to some kind of function

90 FRED C. C. PENG

relevant to both butter and knife, such as using the latter to cut the former, whereas neither paper towel nor iron curtain expresses any such function. And while iron curtain involves the word-relationship that changes the original quality of each word involved, say, from concrete to abstract, the words in paper towel and butter knife undergo no such changes.

Facts such as these are so common that they have gone unnoticed by the three approaches in question. But I must point out that phenomena of this nature are present in many, if not all, languages. Mandarin is a language full of compound nouns similar to those mentioned above, e. g., pu tai "cloth bag" (comparable to paper bag), ts'ai tao "vegetable knife" (comparable to fruit knife), and t'ie mu "iron curtain". Other languages like Japanese, e.g., zaru soba "a particular kind of noodles served in a drainer-like container", and German, e.g., Haushund "house dog", may also be mentioned in passing.

Given this much about compound nouns, we are now ready to turn to the discussion of the Conventional Approach, the Syntactic Approach, and the Semantic Approach, in that order, to see how well each of these han­dles compound nouns in terms of semantics and/or syntax.

2.0 CONVENTIONAL APPROACH. The basic assumption of this approach is that semantics and syntax are separable. Katz and Fodor's dictum of 1963 that linguistic description minus grammar eqeials semantics may serve as evidence. Within the semantic domain of their approach three apparatuses are needed: (1) a dictionary, which provides a set of readings for each lexical item; (2) a set of projection rules, which amalgamates the read­ings of the constituents of a construction, and (3) a set of selectional restrictions, against which the combinations of readings are checked, so that those which violate the selectional restrictions are discarded. In other words, "if an item consists of two constituents, one which has a set of m readings attached to it ... and one which has a set of n read­ings attached to it, the mn combinations of one reading of each constit­uent are formed, those combinations in which a reading for one constit­uent violates a selectional restriction in the reading for the other

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 91

constituent are discarded, and from each of the remaining combinations a reading for the whole item is constructed in a manner specified by the projection rule" (McCawley 1968:128-29). But the semantic theory so proposed presupposes the existence of phrase structures.

How then does this approach interpret the meaning(s) of an English compound? To demonstrate, let us take bachelov friend as an example (which is alluded to by Katz [1964:523, note 12], but has never been interpreted).

Following Katz and Fodor, the compound bachelor friend is a con­struction which has its place in the tree diagram of the sentence He is my bachelor friend. The first item of the compound, bachelor, must be represented according to the dictionary, as having four distinct sen­ses, namely:

hache lor

(Human) (Male)-

(Not-young) (Never-married)

(Young) [Knight who is serving under the standard of another]

<Ω 2>

[One having the academic degree conferred for completing the first four years of college]

I <Ω>

(Animal) (Male)

(Young) [Fur seal when without a mate during breeding time] ļ

Fig. 1

Similarly, the second item of the compound, f r i e n d , may also be repre­sented in a tree form as having the following distinct senses:

friend4

noun (Human) [A person who knows

and likes another] [A person who favors and supports]

[A person who belongs to the same side or group]

92 FRED C. C. PENG

The amalgamation of bachelor and friend is accomplished by a projection rule, yielding the amalgam consisting of the set of 12 possible derived paths from the combination of the two lexical items provided by syntax. But only nine derived paths actually obtain, as the path that contains the semantic marker (Animal) in bachelor is incompatible with any path in f r i e n d , because of selectional restrictions. This means that the compound, bachelor friend* is nine-ways ambiguous, as predicted by the approach. Katz and Fodor would probably stop interpreting the compound at this point and move on to the next item in the sentence, namely my* for the next amalgamation. But let us consider the compound a little further. We can ask, for instance, whether the specification of nine meanings for bachelor friend has accounted for all of its meanings. The answer is negative. The reason is that the above interpretation has missed certain important meanings. Observe that it has failed to take note of such meanings of the compound as a female friend who has the B.A. degree versus a male friend who has the B.A. degree and a friend who was never married and has the B.A. degree (as against a

friend who has the B.A. degree and is married and a friend who is

single and does not have the B.A. degree).

What is wrong with the Conventional Approach then? The trouble seems to be threefold: (1) there is no provision of the dichotomy of (Male) and (Female) markers for the item friend* (2) for the item bachelor, the semantic marker (Male) under the node (Human) is opposed not by (Female) but implicitly by (Male or Female), and (3) the Conven­tional Approach proposed by Katz and Fodor is inadequate, because it does not allow for the existence of simultaneous distinguishes, e.g., (B.A. and unmarried).

Suppose that the dichotomy of (Male) and (Female) is provided for in the lexical item friend as in the following:

See diagram on facing page!

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 93

friend5

| noun

(Human)

and likes and sup- longs to and likes and sup- longs to another] ports] the same another] ports] the same

side of side of group] group]

Fig. 3

Six distinct senses now exist for friend. Of the 24 possible derived paths between bachelor and friend one-half will be discarded by way of selectional restrictions. But note that although there are now 12 ac­ceptable paths, the meaning a male friend who has the B.A. degree and was never married remains unaccounted for.

The inadequacy of the Conventional Approach becomes more serious when we begin to consider other compounds, e.g., house dog and liquid paper, not to mention paper towel, butter knife and iron curtain, whose word-relationships simply cannot be handled by the Conventional Approach. The reason is straightforward. Note that when discarding the compound spinster insecticide (1964:508) as semantically anomalous, Katz and Fodòr argue that "the path for insecticide does not contain the seman­tic marker (Human) which is necessary to satisfy the selection restric­tion associated with spinster." In line with this argument, then, the compound house dog must also be discarded, because on the basis of the dictionary entries for house and dog the projection rule will predict that house dog is semantically anomalous, since the path for dog does not contain the semantic marker (Inanimate), which is necessary to sat­isfy the selection restriction associated with house. Obviously, this result is absurd, because house dog is a perfectly good compound.

3.0 SYNTACTIC APPROACH. Before we move on to the discussion of the Syntactic Approach and in order not to do Hockett injustice, let me first quote a few passages from him regarding the separability of se-

94 FRED C. C. PENG

mantics and syntax, because he does admit some separability. He consid­ers this question: To what extent is grammar (or 'the rest of grammar' for those for whom the term habitually includes phonology) separable from semantics (cf. Hockett 1968:69)? In reply to this self-imposed question, Hockett says two things: (1) A certain kind of separability is clear when there is a discrepancy, e.g., oats. Like scissors orpants, oats has no matching singular. (2) "But when a difference in the gram­matical behavior of two words parallels a difference in their meanings, or when words that are used in similar ways have similar meanings, I see no reason to assume any separability of grammar and semantics, nor any reason to separate either of these from 'the rest of culture'." (Ibid., p.70).6

Hockett also describes the task of the linguist in the field as being "that of discovering and setting forth all those facts about a language that cannot be inferred from all the other lifeways of the community". He continues, "This would mean the phonological habits, the stock of elementary linguistic forms and their meanings (one cannot in­fer from the rest of the culture that the word for wheat is wheat), the ways in which these forms are combined, permuted, or modified, and the alternations of meaning achieved by such arrangements and rearrangements, and the brute facts of discrepancies such as that of English oats." Ac­cording to Hockett, then, "this divides neatly into unequal parts: (a) the phonological system; and (b) the enormously complex conventions of correspondence between arrangements of phonological material and mean­ings."7 Hockett thus asserts that "This second part is grammar (-and-lexicon) in the classical sense; it is not something different from and in addition to semantics, but, as Bloomfield said in 1914, simply is the semantic system the way the particular language handles the world."

(Hockett 1968:71; italics in the original).8

It seems that a great deal of the above argument involves the mean­ing and application of the terms 'semantics', 'grammar', 'syntax', and even 'morphology'; part of the issue seems to be purely terminological. Hockett assumes that since Bloomfield equates the semantic structure of a language with the morphology and syntax of the language and then with

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 95

its grammatical system, other people must follow suit by using the terms semantics and grammar in the same senses. I disagree. The point is that what Bloomfield (à la Hockett) calls 'semantics or grammar' could very well be split into two portions by others, one being called 'semantics' and the other 'grammar'. This terminological difference may be shown in the following diagram:9

Syntax

L Morphology Semantics or Grammar

Phonology

Hockett

Semantics

Grammar

Phonology

Syntax

Morphologyl \Grammar

Others

Fig. 4

Suppose now that we let Hockett assume that semantics equals gram­mar. His Syntactic Approach then entails classifying all the compound nouns of any language into types. But since it is a regular feature of this approach that no compound noun is an exocentric construction he only has one type of construction at his disposal, namely, endocentric construction. For the latter, Hockett (1958:186) has listed four sub­types :

I. Like 'stone w a l l ' , where the second item is the head II. Like 'operation Coronet', where the first item is the

head III. Like 'as good as that', where the inner constituent is

the head and IV. Like 'd i d not go', where the outer construction is the head.

Of the four subtypes, only the first, i.e., Attribute-Head type, is applicable to the analysis of such compounds as paper towel, butter knife, and iron c u r t a i n . In other words, if we follow Hockett's Syn­tactic Approach, which he regards as "the way the English language han-

96 FRED C. C. PENG

dles the world," most English compounds must invariably be lumped to­gether under one and only one type of endocentric construction, and we cannot do anything else about them. But we have already shown that paper towel, butter knife, and iron curtain are not just similar endocentric constructions; rather, they involve differing meaning-re­lationships. Something must be seriously wrong with the approach if it fails to account for such facts. The best Hockett could hope for, then, would be to further classify the endocentric construction of the Attri­bute-Head subtype. Consequently, paper towel, butter knife and iron curtain would automatically belong to three sub-subtypes. The next prob­lem would be whether paper tiger should be assigned to paper towel or iron curtain or to a separate subtype of its own.

The inadequacy of the Syntactic Approach may be further demonstrated if we consider more carefully what Hockett says about the brute facts of discrepancies. Actually, the brute facts of discrepancies are more widespread in English than Hockett imagines. Take the pairing of state­ment/question in English, for example. Normally, a statement in English has a matching interrogative. But many sentences involving the phrase used to do' not; that is to say, a statement like he used to eat snakes lacks a matching question, the hypothetical one *did he used to eat snakes? being generally regarded as ungrammatical (cf. Peng 1969). (Even if *did he used to eat snakes is acceptable to and actually em­ployed by some, the verb used to lacks its progressive form, past par­ticipial which all the other verbs have in English, assuming of course that the present form is use to or used to with the same pronunciation.) Moreover, Hockett evidently overlooks words like sight and vision when he says that he sees no reason to assume any separability of semantics and grammar. His reasoning is based solely on the premise that a dif­ference in the grammatical behavior of two words must parallel a dif­ference in their meanings and that words which are used in similar ways have similar meanings. But this premise is false, because sight and vision are synonyms, when used in the context of she lost her . . . , that is, she lost her sight and she lost her vision are more or less inter­changeable, but become antonyms, when used in the context of she is a...,

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 97

that is, she is a sight and she is a vision are opposed in meaning, and not at all interchangeable. Note that in either case the environments of sight and vision are identical.

It must follow that the Syntactic Approach has failed to account for something that is not observable in the mere combination of words and their external phonetic forms, something that can only be properly taken care of outside "the ways in which these forms are combined, permuted, or modified, and the alternations of meaning achieved by such arrangements and rearrangements." Obviously, Hockett cannot hope to adequately explain the difference between horse shoes and alligator shoes with regard to the word-relationships between the words in each expression by showing how the three words, horse, alligator and shoes are combined, arranged, rearranged, or permuted.

4.0 SEMANTIC APPROACH. McCawley's position is a little difficult to delineate. For the most part, it is because he has changed from one position to another rather abruptly. Earlier, he believed in the valid­ity of deep structure and vigorously supported the use of semantic se-lectional restrictions: "I will now present an argument that an adequate account of selection must be in terms of semantic selectional restric­tions such as those of Katz and Fodor (1963) and that there is no reason to have the 'syntactic selectional features' of Chomsky (1965) nor the complicated machinery for creating 'complex symbols' which the use of such features entails" (McCawley 1968:133). But more recently he not only rejects the validity of deep structure completely but also abandons the use of selectional restrictions altogether: "The general outline of this argument for rejecting a level of 'deep structure' is, of course, identical to that of Halle's (1959) celebrated argument for rejecting a 'phonemic level'" (McCawley 1970a:172). In the same article (p.167) he also says: "I see no reason for believing that selectional restric­tions have any independent status in linguistics".

This change of thinking is considerable, because earlier he was in line with Katz and Fodor, adhering to the belief that syntax and seman­tics are separable; now he reverses his belief, indicating on the con-

98 FRED C. C. PENG

trary that "there is no natural breaking point between a 'syntactic component' and a 'semantic component' of a grammar such as the level of 'deep structure' was envisioned to be in Chomsky (1965)..." (McCawley 1970a:171-72).10

The current position McCawley takes, as of 1970, may thus be summed up as follows: "The conception of grammar within which I am investigat­ing 'prelexical transformations' such as Predicate raising is a version of transformational grammar in which there is no such level as Chomsky's deep structure', the base component of a grammar generates semantic re­presentations, and the 'dictionary entries' for the various lexical items are in effect transformations which insert those lexical items in place of various complexes of semantic material that may arise through pre­lexical transformations" (McCawley 1970b:52). His semantic representa­tions "are to form the input to a system of transformations that relate meaning to superficial form" ( o p . c i t . , 36-37). It is in this sense that McCawley may be said to be the advocate of semantics subsuming syntax.

Even on this view, however, it is still hard to see how McCawley would handle the kind of compounds with which we have been concerned. This is mainly because he is concerned with semantic representations which more or less follow symbolic logic, 11 and accounts mostly for sentences - in particular, those which have more than one reading -though he does deal with the question of the semantic representations of lexical items (or the question of the relation of lexical items to semantic representations; cf. McCawley 1970b:49). Observe that he em­phasizes that "lexical items may be related in a sufficiently indirect way to semantic representations of sentences in which they appear that they will not directly match portions of those semantic representations" (loc.cit.). As an example of this emphasis, he cites kill; he says "the semantic elements that are involved in a lexical item are separated f rom each other in the semantic representations of sentences involving that lexical item" (ibid.). But the semantic representation of this item en­tails having a clause containing kill. Thus, the semantic representation of something of the form 'x kill y' would be according to McCawley (1970 b:49-50):

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 99

McCawley (1970b:50) then suggests that "English has a system of mecha­nisms for regrouping semantic elements, one of these being a transfor­mation of 'predicate raising', which optionally ajoins a predicate to the predicate of the next higher sentence". At the end of the applica­tion of Predicate raising, figure 5 becomes figure 6 as follows:

When this structure is eventually obtained in which Cause, Become, Not, and Alive are joined together, McCawley (1970b:51) says that "it is that stage of the derivation rather than semantic representation per se at which it is determined what lexical items may be used in a sen­tence with a given semantic representation". What this means, then, is that eyery lexical item must originate from a clause whose semantic representation (containing the semantic elements of the lexical item) must undergo a series of transformations, before that item can be in­serted into a sentence whose semantic representation will in turn under­go a series of transformations before the sentence can be related to superficial forms for phonetic manifestations.

The question we must ask now is how McCawley would relate the se-

Cause

100 FRED C. C. PENG

mantic elements of a compound to the semantic representation of a sen­tence which contains the compound. Since he has already rejected Katz and Fodor's projection rules, which entail selectional restrictions which he has abandoned, he obviously cannot combine (or amalgamate) two lexical items to form an amalgam so that the compound can be inserted as such into a sentence whose semantic representation will then continue to be transformed. The only alternative left for McCawley is that the lexical items involved in a compound are inserted separately. Two choices are conceivable here. First, McCawley could insert each such lexical item individually and directly into the semantic representation of a sentence at a certain stage of the derivation. Second, McCawley could treat a compound noun as if it were a sentence having a semantic representation of its own. Into this pseudo-semantic representation the two lexical items of a compound noun are inserted separately at certain stages of their derivations. On the completion of such lexical inser­tions, the pseudo-semantic representation could then undergo a series of transformations for its own insertion into the semantic representa­tion of a sentence. But note that there is a distinction between the insertion of a verb and that of a noun. The transformations required for the former are 'Predicate raising', as in the case of k i l l , but those required for the latter are called 'Generalized conjunction reduc­tions', both, however, being regarded as Prelexical transformations (cf. McCawley 1970b:52-53). The handling of compound nouns naturally pertains to the latter and the two choices mentioned above belong here. But there is a serious problem in carrying out Generalized conjunction reductions.

The lexical item daughter is said to be a 'transitive noun'; it is broken down into Offspring and Female, the former expressing a binary relation but the latter a one-place predicate (cf. McCawley 1970b:53). In tree representation, daughter becomes

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 101

This tree representation must then be converted into one in which Off­spring and Female are combined into a single constituent as shown in figure 8.

Offspring And Female x

Fig. 8

The problem now is: What are the tree representations of the lexical items bachelor and friend when they occur in the compound bachelor friend?. More precisely, how many lexical items are there which are pro­nounced bachelor and how many lexical items are there which are pro­nounced friend and exactly what tree representation will each such lex­ical item take? Earlier, while agreeing with Katz and Fodor on their four senses of the word bachelor , McCawley (1965:126) favors Weinreich's (1966) conception of 'lexical item' and argues that "there would simply be four lexical items pronounced bachelor rather than a single four-ways ambiguous lexical item". In line with this argument, there would then be three, or possibly six, lexical items pronounced friend. Given the two alternatives suggested above, how would McCawley propose to in­sert these lexical items separately into the semantic representation of a sentence? The only conceivable way out would be to regard a sentence like

(1) I have a bachelor friend

as having nine, or possibly twelve, semantic representations which may be diagrammed as follows: (Ix (bachelor friend (x have y))) where n = 1, ... , or 9 or 12. If (1) is construed in this way, no matter which alternative one chooses the number of prelexical transformations will be enormously increased: However, there is one advantage; that is, one of the tree representations of bachelor may be diagrammed as

102 FRED C. C. PENG

Fig. 9

which can be converted into figure 10 by way of Generalized conjunction reduction as

Fig. 10

When inserted, the meaning a male friend who has the B.A. degree and was

never married of the compound bachelor friend, which the Conventional Approach has failed to account for, can now be taken care of by the Semantic Approach.

On the other hand, the Semantic Approach gives rise to an addi­tional problem which the Conventional Approach does not have. That is, whichever alternative of the two suggested above is chosen, McCawley must decide whether the first or the second item of a compound under­goes generalized conjunction reductions first. Can the decision be ar­bitrary or must there be some kind of ordering principle? With ha.chelor f r i e n d , does it make any difference one way or the other? Does the same decision apply to operation Coronet?. Note that the base component in the Semantic Approach no longer contains a phrase structure grammar. Thus, logically speaking, there should be no information available as to which item of a compound is head and which is its modifier. It seems to me, however, that some kind of ordering is necessary, otherwise no distinc­tion will be made by the rules between house dog and dog house. Observe that a dog house may be a house in which a dog lives but a house dog

(with the primary and tertiary stress pattern) is by no means a dog that lives in a house. Where would McCawley obtain the syntactic information

ON THE SEPARABILITY OF SEMANTICS AND SYNTAX 103

concerning the distinction, when he applies prelexical transformations to the two compounds? Or is the distinction between them unnecessary prior to their insertions into the semantic representations of sentences? If so, I would like to know why, because there is no a priori reason that two distinct compounds have to be differentiated in terms of seman­tic representations. Something must be seriously wrong with the Seman­tic Approach if it cannot account for the distinction between house dog and dog house in obvious and readily available terms of word order, be­cause the Syntactic Approach can account for it quite well by making good use of word order.

5,0 CONCLUSION. I have tried to show how well each of the three ap­proaches handles compound nouns. It looks as if none of them is capable of dealing with compound nouns adequately. Perhaps it is time that we stopped to enquire where we have gone wrong and where we are heading in our theorizing, before we go too far astray. One possibility is to re­examine what language is. Is thinking essentially bound up with lan­guage or not? If thinking can be done independently, language is cer­tainly not genetic or innate. I am inclined to believe that thought and language are two different things, the former being genetic and the lat­ter a cultural product. It is in the light of this distinction that lin­guistic phenomena can be most adequately analyzed.

N O T E S

1 The Stoic anomalists, for example, displayed an important insight into the semantic structure of language, namely, that word meanings do not exist in isolation and may differ according to the collocation in which they are used.

It should be noted, however, that there are variations in approach among generative semanticists.

3 Recently, the expression chairman has become the target for women's lib. As a result, there now exist such expressions as chairwoman, chairperson, and even chairpeople. This makes the address system quite difficult to follow. For instance, one used to say Mr. Chairman!

104 FRED C. C. PENG

or Madam Chairman! Now, do we dare say Mrs. Chairwoman oder Miss Chairwomen or Mrs. Chairperson or Miss Chairperson or Mr. and Mrs. Chairpeople? During the Democratic Convention in Florida in 1972, one delegate addressed the chairman, who happened to be a woman, as Madam Chairwoman three times. This seems to violate the principle of complementation of the gender in the two items. At the International Christian University, a Japanese faculty member was evidently puzzled by such usages and quite innocently addressed the woman chairman at a faculty meeting as Mrs. Chairman, which caused much laughter. Per­haps, before very long, postman will follow suit, yielding postwoman, postperson or even postpeople, by way of analogy.

Nowhere have Katz and Fodor stated that every division within a lexi­cal item has to be binary, and I know of no basis for such a restric­tion. The ternary division after (Human) is perfectly admissible with­in such a tree diagram.

The ternary division after (Male) and (Female) is retained for ob­vious reasons.

We can see now why Hockett (1958:123) defines a morpheme as a minimal meaningful unit, instead of as a meaning-carrier.

We can also see that Hockett rests his argument entirely on his du­ality of patterning.

The passage from Bloomfield 1914 which Hockett has invoked is as follows: "The first task of the linguistic investigator is the anal­ysis of a language into distinctive sounds, their variation, and the like. When he has completed this, he turns to the analysis of the semantic structure - to what we call the morphology and syntax of the language, its grammatical system." (Quoted in Hockett 1968:19).

Even in phonology there are variations; some include morphophonemics within phonology, others do not. This matter, however, falls out­side of the present discussion.

It may be noted that long before McCawley ever realized the futility of deep structure, I had already stated in my review of Chomsky 1965 that there is no such thing as deep structure (see Peng 1969).

11 "I am proposing a system of semantic representation that is along the lines of the notational systems used in symbolic logic" (McCawley 1970b:36), McCawley subsequently departs significantly from symbolic logic.

REFERENCES

Bach, Emmon. 1968. "Nouns and Noun Phrases". In Bach & Harms 1968:19-to 122.

Bach, Emmon, and Robert T. Harms, eds. 1968. Universals in Linguistic Theory. New York: Holt, Rinehart & Winston.

Bloomfield, Leonard. 1914[1915]. "Sentence and Word". Transactions of the American Philological Association 45.65-75. (Repr. in A Leonard Bloomfield Anthology ed. by Charles F. Hockett, 61-69. Bloomington & London: Indiana Univ. Press, 1970.)

Chafe, Wallace L. 1970. Meaning and the Structure of Language. Chicago & London: Univ. of Chicago Press.

Chao, Yuen-Ren. 1934. "The Non-Uniqueness of Phonemic Solutions of Phonetic Systems". Bulletin of the Institute of History and Philo­logy (Academia Sinica) 4:4.363-97. (Repr. in Readings in Linguis­tics I: The development of descriptive linguistics in America 1925-[to 19] 56 ed. by Martin Joos, 4th ed., 38-54. Chicago & London: Univ. of Chicago Press, 1966.)

Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.

Fodor, Jerry A., and Jerrold J. Katz, eds. 1964. The Structure of Lan­guage: Readings in the philosophy of language. Englewood Cliffs, N.J.: Prentice-Hall.

Halle, Morris. 1959. The Sound Pattern of Russian. The Hague: Mouton. (2nd printing, 1971.)

Hockett, Charles F. 1958. A Course in Modern Linguistics. New York: Macmillan. (12th printing, 1967.)

. 1968. The State of the Art. The Hague: Mouton. Katz, Jerrold J. 1964. "Analyticity and Contradition in Natural Lan­

guage". In Fodor & Katz 1964:519-43. Katz, Jerrold J., and Jerry A. Fodor. 1963. "The Structure of a Seman­

tic Theory". Language 39.170-210. (Repr. in Fodor & Katz 1964:479-to 518.)

Lehrer, Adrienne. 1971. "Semantics: An overview". The Linguistic Re­porter 13:4, Supplement 27 (Fall 1971).

106 FRED C. C. PENG

McCawley, James D. 1968. "The Role of Semantics in a Grammar". In Bach & Harms 1968:124-69.

. 1970a. "Where Do Noun Phrases Come From?". Readings in En­glish Transformational Grammar ed. by Roderick A. Jacobs and Peter S. Rosenbaum, 166-83. Waltham, Mass.: Ginn & Co.; Tokyo: Kanto Books Company.

. 1970b. "Semantic Representation". Selected Papers in Gen­erative Semantics [from the Fifth International Seminar in Linguis­tic Theory] ed. by Shin'ichi Harada and Donald L. Smith, 35-53. To­kyo: TEC Company for Language and Educational Research.

Peng, Fred . 1969. Review of Chomsky 1965. Linguistics 49.91-128. Weinreich, Uriel. 1966. "Explorations in Semantic Theory". Current

Trends in Linguistics ed. by Thomas A. Sebeok, vol.3.395-477. The Hague: Mouton. (Sep. ed., with a preface by William Labov, The Hague: Mouton, 1972.)

II. P H O N O L O G Y A N D M O R P H O L O G Y

HOW GENERATIVE IS PHONOLOGY?

HSIN-I HSIEH

0.0 INTRODUCTION

In the current theory of generative phonology, morpheme alternants are not listed in the lexicon.1 These morpheme alternants, which are called surface forms, are not to be derived by rules from their under­lying or base forms which are listed in the lexicon. Thus, the task of the phonologist is to set up an underlying form for each set of morpheme alternants and to 'predict' their contextual occurrences by phonologi­cal rules. Despite the caution shown by earlier generative phonologists in refraining from claiming psychological reality for their model, the current trend among linguists is to ask two major questions, namely, first, are phonological rules psychologically real, and, second, if they are, do surface forms have to be listed in the lexicon, in addition to their underlying forms?

William S-Y. Wang, who has guided and backed me through my long strug­gle as a graduate student, deserves my warmest thanks here. His was the first unequivocal voice of encouragement when I began studying the psychological reality of Taiwanese tone sandhi rules in 1970. - I also want to thank Matthew Chen, who repeatedly urged me to continue my work in this field. Another friend of mine, William Orr Dingwall, also de­serves my sincerest thanks; through his extensive review (Dingwall 1971) my CLS paper (Hsieh 1970) has reached a wider audience than it would have otherwise. - I am grateful to have received comments and suggestions either orally or in writing by many friends and colleagues; in particular, I wish to thank the following linguists: Matthew Chen, John Crothers, W. 0. Dingwall, Chin W. Kim, Robert Krohn, Ariene Mos-kowitz, John and Manjari Ohala, Masayoshi Shibatani, Danny Steinberg, William Wang, and Karl Zimmer. - Oral versions of this paper were pre-

110 HSIN-I HSIEH

Some effort has been made to answer the first question. This effort includes such experiments as those conducted by Berko 1958, Cheng 1968, Ladefoged and Fromkin 1968, Zimmer 1969, Hsieh 1970, Moskowitz (MS), and Manjari and John Ohala (both in 1972). Although these experiments have not always provided indisputable evidence for or against the reality of phonological rules, they certainly help to clarify the nature of the problem.

With the exception of Maher 1969 and Steinberg 1973, few linguists have addressed themselves directly to the second issue. This issue, however, is also of great interest, for by knowing more about the lex­icon, we may gain a better understanding of the question concerning the reality of the rules that are claimed to apply to the lexcical items.

In order to obtain empirical evidence for this highly theoretical issue in phonology, three experiments were conducted. These experiments are designed to study how children and adults acquire morpheme alter­nants involving tone sandhi rules in Taiwanese, a southern variety of Chinese.

1.0 TONE SANDHI IN TAIWANESE

In Taiwanese, lexical tones in the citation forms undergo morpho-phonemic changes called tone sandhi. A set of phonological rules are observed to operate in these changes (detailed discussions of these rules have been made in Cheng 1968, 1973). The domain of the applica­tion of these rules is a 'sandhi phrase', which is extracted from a sentence on syntactic grounds (see Liao 1971, for a detailed descrip­tion). A sandhi phrase is bounded on both sides by a pair of phrase junctures represented by double crosses ( # ) . Morphemes contained in the sandhi phrase are then separated by morpheme junctures represented by single crosses ( + ). Tone sandhi rules apply only to those tones

sented to the Linguistic Society of Hawaii and at the 1972 Annual Meet­ing of the LSA at Atlanta, Georgia. The first draft of this paper was prepared when I was working as an assistant research linguist at the Phonology Laboratory at Berkeley supported in part by an NSF grant.

HOW GENERATIVE IS PHONOLOGY? 1ll

that are in syllables preceding morpheme junctures. Thus, for example, given the sandhi phrase #X+Y+Z#, the rules will apply to X and Y but not to Z.

A set of five rules that affect the five long tones takes two slightly different forms in the two major geographical dialects.2

These two different sets can be summarized as follows: Rules 1) in the '33' dialect 2) in the '21' dialect

a. 55 → 33 b. 35 → 33 53 → 55 d. 21 → 53 e. 33 → 21

a'. 55 → 33 b'. 35 → 21 c'. 53 → 55 d'. 21 → 53 e'. 33 → 21

(Note: The rules are disjunctively ordered.)

Table I: Tone sandhi rules in Taiwanese

As can be seen from Table I, the '33' dialect and the '21' differ in the rule which affects underlying tone 35. While rule b in the '33' di­alect merges tone 35 with tone 55 so that they both appear as surface tone 33, rule b' in the '21' dialect merges tone 35 with tone 33 so that they both appear as surface tone 21.

Assuming that these rules are psychologically real, the direction of the arrows in these rules is justified. This is so because we can predict the merger if the arrows point to the right. An unpredictable split however would occur if the arrows were to point to the left.3

2.0 THREE EXPERIMENTS

2.1 The First Experiment: Design

In our experiment, we used trisyllabic compounds each containing a disyllabic compound modifying a following monosyllabic head-noun. A tri­syllabic compound of this kind has the tonal construction #XY+Z# and the sandhi rules are required to operate on Y, the tone in the second syllable of the constituent disyllabic compound. The trisyllabic com­pound is a productive formation, but the constituent disyllabic compound

112 HSIN-I HSIEH

is not. For example, the disyllabic compound #kin 33 tsi 55# "banana" can be combined with a third syllable head-noun to make such trisyllabic compounds as #kin 33 tsia 55 + diam 21# "banana store", #kin 33 tsia 55 + bi 33# "banana flavor", and so on. The sandhi rules apply without ex­ception to the second syllable of the trisyllabic compound, chang-ing, for example, #kin 33 tsia 55 + bi 33# tp #kin 33 tsia 33 bi 33# "banana flavor", according to rule a. Ideally, there is no limit to the number of such trisyllabic combinations. However, in the case of the disyllabic compounds, either one or both of the elements are bound forms or semi-bound forms and new formulations are generally not permitted.4

For the purpose of our experiment, however, we made up artificial disyllabic compounds. By replacing the first elements of real disyllabic compounds with other actual syllables, we obtain pronounceable but non-occurring disyllabic compounds. For example, by substituting the first element in the actual compound #kin 33 tsia 56# "banana" with an actual syllable tshai 21 (>53) "vegetable", we obtain the artificial compound #*tshai 53 tsia 55#.5

Forty real disyllabic compounds and forty artificial disyllabic compounds created out of these compounds were combined with five head-nouns into trisyllabic compounds. These head-nouns and modifiers are listed in Tables IIa, IIb, and IIc in the Appendix. The subjects were tested for correct forward operation of the tone sandhi rules in com­bining modifiers and head-nouns into compounds, and for correct backward operation of these rules in decomposing the compounds into modifiers and head-nouns. More specifically, four tests were conducted. Test I involves real compounds and the forward operation of the rules; Test II, real com­pounds and backward operation; Test III, artificial compounds and forward operation; and Test IV, artificial compounds and backward operation. 2.2. Subjects and instructions. Three subjects participated in this ex­periment. S1 (subject one) was five years old, S2 seven years old and S3 nine years old,6

The instructions for Test I and Test III, in which forward operation is studied, are identical; the instructions for Test II and Test IV, in

HOW GENERATIVE IS PHONOLOGY? 113

which backward operation is studied, are also identical. In the former two tests, the experimenter gives the child the names of fruits and veg­etables and asks the child to tell him the names of stores, for example, that sell these fruits or vegetables. In the latter two tests, the ex­perimenter gives the child names of stores, for example, that sell fruits or vegetables and asks the child to tell him what fruits or veg­etables are sold in these stores.7

For the sake of convenience in narration, we will henceforth use "forwards", "backwards", "reals" and "fakes" to abbreviate test items that are subject to forward operation, test items that are subject to backward operation, compounds that are real and compounds that are fake, respectively. Accordingly, items used in Test I, Test II, Test III and Test IV are called "forward-reals", "backward-reals", "forward-fakes" and "backward-fakes", respectively.

2.3. Results. These four tests yield a rich body of data that has bearing on several different issues in phonology. To try to analyze all parts of the data here would mean discussing several not necessarily related topics in a single article. We will therefore proceed to our im­mediate concerns after very briefly commenting on the test results.

The results of this experiment show that the subjects do not always succeed in supplying correct answers to the questions. A subject's degree of success in a test varies significantly, depending on whether the "for­wards" or "backwards" are examined and on whether the "reals" or "fakes" are involved. The subject's degree of success in the same test using the same set of compounds also varies according to tones in the second syl­lables, that is, according to different tone sandhi rules. To a lesser extent, his rate also varies according to different head-nouns in the compounds.

The success rates are different among subjects no matter whether they are compared in all four tests or in just a particular test, or even in just a tone category. This difference sometimes corresponds to an age difference.

This is but a very terse summary of the test results. What we would

114 HSIN-I HSIEH

like to discuss in great detail here is, however, the fact that these children did not succeed in supplying correct answers to all stimuli.

Thus, for example, in the "forward-reals" the percentage of correct answers are 36% for SI, 73% for S2 and 76% for S3 when all five tones in the second syllables (to which five different rules apply) as contexts are considered. Although the ratios of success are substantially higher in the backward operation, these ratios range from only 73% to 98%.

None of these rates, particularly those for the forward operation, seem high enough to warrant the claim that these children have learned the phonological rules rather than individual morpheme alternants. Tone sandhi rules that are strictly regular in the neogrammarian sense do not seem to be in the possession of these children. 2.4. Can variable rules account for the results? Of course, one may still argue for the reality of tone sandhi rules in these children by claiming that these rules exist as variable rules. There is no doubt that these rules can be treated as variable rules as they do not apply categorically. It is doubtful, however, that by positing variable rules we will come any closer to a full description of the variations observed. This is so, because in our case as in many other cases, a variable rule only indicates its flexible overall rate of application as determined by the values assigned to the variables. In no way can it provide us with any further information as to whether a particular lexical item subject to a variable rule is (a) never, (b) sometimes, or (c) always affected by the operation of the variable rule. But such information is sometimes necessary as evidenced in the responses of S3 in the "forward-reals".

These responses of S3 can be divided into three classes according to the frequency of rule application in terms of "never", "sometimes", and "always". The rules neyer apply in the first class, sometimes apply in the second class and always in the third class. For example, rule d never applies to item 28a. ku 55 tshai 21 "chives". This rule sometimes applies, sometimes fails to apply to item 27a. eng 53 tshai 21 "water­cress". It always applies to item 25a. pe 21 tshai 21 "Chinese cabbage". Since each disyllabic compound serving as a modifier is matched with five

HOW GENERATIVE IS PHONOLOGY? 115

head-nouns and three trials are made, the consistency or variation in the subject's reactions to a particular disyllabic compound is supported by 15 tokens. Therefore, a consistent success or failure to apply the rule to a particular disyllabic compound cannot be dismissed as merely fortuitous but must be regarded as sufficiently significant.

To capture this significant constancy, the linguist needs a descrip­tive device by which he can successfully ensure the application or non-application of a rule to a particular item subject to the operation of the rule. The variable rule device is obviously insufficient for such a task. For in our case, the variable rule will not be able to provide any information beyond the fact that rule d applies to 40% of the items in base tone 21. Yet, this still tells us nothing about the fact that this rule applies, never to item 28a, only sometimes to item 27a, and always to item 25a.

The most plausible place to store this information would seem to be in the lexicon. There are at least two conceivable ways in which the lexicon can carry this information. In the first way, one will list only the base form of a given item and attach to it a statement indicating whether a particular rule should never, sometimes, or always apply to it. In accordance with this procedure, items 28a, 27a and 25a will be listed in the lexicon in the following way:

28a. "chives". ku 55 tshai 21: rule d never applies.

27a. "watercress". eng 53 tshai 21: rule d sometimes applies.

25a. "Chinese cabbage". pe 21 tshai 21: rule d always applies.

According to the second way, both the base form and the surface form of a given item will be provided in the lexicon. If an item is never sub­jected to a rule, then its surface form will be identical to its base form. If the rule applies without exception, the surface form and the base form will be different. If it applies only sometimes, then the sur­face form has two optional variants, one identical to the base form and the other identical to the expected surface form. To each individual

116 HSIN-I HSIEH

item an instruction will be attached for the appropriate selection of the base form or the sandhi form according to whether the item precedes a phrase juncture or a morpheme juncture.

In this approach, items 28a, 27a, and 25a will be entered in the lexicon as follows:

28a. "chives". ku 55 tshai 21 : base (_#) ku 55 tshai 21: sandhi (_+)

27a. "watercress". eng 53 tshai 21: base (__#)

tshai 21 eng 53 tshai 53: S a n d h l (-+)

25a. "Chinese cabbage". pe 21 tshai 21 : base (_#) pe 21 tshai 53: sandhi (__+)

2.5. Base-forms-only lexicon and sur face-forms-too lexicon. When we further take into consideration the backward operation of rules in the case of S3, however, we become aware that his lexicon must be more com­plex than the types we suggest above. For as we have mentioned, within the same subject, the rate of application of a particular rule often varies significantly according to its direction. Furthermore, in terms of "never apply", "sometimes apply" or "always apply", an item may be described in one way in one direction of the application of a rule with­out necessarily being specifiable in the same way in the other direction. Thus, for example, an item may be always affected by a rule in the for­ward direction and yet never affected by the same rule in the backward direction.

If we use "C" to stand for "always apply", "I" for "never apply" and "I/C" for "sometimes apply", then the following nine different types of lexical items can occur in a lexicon that provides information re­garding the application of a rule in both directions to individual items: Type . I:I ("I" in forward application and "I" in backward application), Type II. I:I/C, Type III. I:C, Type IV. I/C:I, Type V. I/C:I/C, Type VI. I/C:C, Type VII. C:I, Type VIII. C:I/C, and Type IX. C:C.

All but Type III are found in the application of rule a to items

HOW GENERATIVE IS PHONOLOGY? 117

la-10a in Table IIb which carry base tone 55. And slightly fewer types are found in items of other tone categories. For base tone 55, the value of "I" is 55 in forward application and 33 in backward application, and the value of "C" is 33 in forward operation and 55 in backward operation. Although not all eight types have been discovered in a single subject, each subject displays several of the eight types. More specifically, the responses of each of the three subjects to items la-10a can be assigned to appropriate types as shown in Table III:

See overleaf (p. 118) for Table III

118 HSIN-I HSIEH

Subject 1 (aged 5)

Subject 2 (aged 7)

items

forward

: backward

type forward

: backward ty

la. banana

55/33

55

VI

33

55

I

2a. squash

55

33

I 55/33

55

V

3a. lichee

55

33/55

II

55/33

55

V

4a. watermelon

55/33

33/55

V 33

55

I

5a. parsley

55/33

33

IV

33

55

I

6a. pumpkin

55/33

33/55

V 33

55

I

7a. custard apple

55

33

I 55/33

55

V

8a. winter-melon

55/33

33/55

V 33

55

I

9a. honeydew

55/33

33

IV

33

55

I

10a. papaya

55/33

33/55

V 33

55

I

Tabl

e II

I Responses given by S1, S2, and S3 to stimuli carry

tone 55's are assigned to different phonological

HOW GENERATIVE IS PHONOLOGY? 119

Each of the two kinds of lexicon proposed in the previous section can now be revised to accommodate all the nine possible types of lexical items. Both the "base-forms-only" and the "surface-forms-too" kinds of lexicon can be illustrated with the item kin 33 tsia 55 "banana" ap­pearing in forms belonging to types I, II and IX. In the "base-forms-only" lexicon, these types will appear as follows:

Type I. "banana". kin 33 tsia 55#: rule a never applies forward. kin 33 tsia 33+: rule a never applies backward.

Type II. "banana". kin 33 tsia 55#: rule a never applies forward. kin 33 tsia 33+: rule a sometimes applies backward.

Type IX. "banana". kin 33 tsia 55#: rule a always applies forward. kin 33 tsia 33+: rule a always applies backward.

In the "surface-forms-too" lexicon, these types will appear as follows: Type I.

"banana". kin 33 tsia 55: select this form in sandhi position (__+)• kin 33 tsia 33: select this form in base position (_#).

Type II. "banana". kin 33 tsia 55: select this form in sandhi position (__+)• kin 33 i . : select this form in base position ( #). tsia 55 —

Type IX. "banana". kin 33 tsia 33: select this form in sandhi position (__+). kin 33 tsia 55: select this form in base position (__#).

In the base-forms-only lexicon we need to register the base form kin 33 tsia 55 and the sandhi form kin 33 tsia 33 so that the rule can apply in both directions. It would be difficult to argue for the elim­ination of one or both of these two forms on the grounds that in the test the input of a rule is always pronounced by the experimenter to the child, for the stimuli in Tests I and II are words used in every­day speech, and what the experimenter does amounts to reminding the

120 HSIN-I HSIEH

child that a particular word is being "discussed" between him and the experimenter. The experimenter does not in his capacity create a new item for the child to serve as an input to the rule that the child is expected to apply.

If the child used the surface-forms-too lexicon, his first task would be to identify a particular stimulus as, for example, the compound "banana" in his lexicon. If the compound is a type I item, the child will choose between the two alternating forms kin 33 tsio 55 and kin 33 tsia 33, following the instruction for the selection of proper alter­nant according to the context of morpheme or phrase juncture.

One seeming problem we notice about the latter kind of lexicon is that the stimuli are always pronounced in the correct adult forms by the experimenter. But these adult forms may be missing in the child's lexicon as is the case with type I words. However, this is not a real problem. Since a child not being able to pronounce an adult form cor­rectly can often identify the form when uttered by an adult, it is rea­sonable to assume a special faculty in children for perceiving adult forms. Such a faculty probably involves semantic, syntactic, and phon­ological interpretations of words.

Both these two kinds of lexicon seem quite plausible, and the material gathered here does not allow us to argue directly in favor of one or the other of the two hypotheses. It is not clear whether further experiments can be made to study in any conclusive terms the superiority of one or the other of these two contending lexicons. It is far less clear what kind of experiment could be designed for such purpose. How­ever, we may bring in evidence from the study of child language acqui­sition that bears indirectly on this issue. A tentative choice can then be made between these two kinds of lexicon. 2.6. Why do we need surface forms in the lexicon? It has been shown that children spend many years in acquiring adult forms (e.g., C. Chomsky 1966; Clark 1971). It has also been demonstrated that the child's acqui­sition of phonological forms proceeds according to the principle of lex­ical diffusion originally proposed by Wang 1969 (cf. also Hsieh 1972).

HOW GENERATIVE IS PHONOLOGY? 121

According to this principle, the application of a diachronic rule affects, at a particular time, only some individual members of a category rather than the whole category itself. When we consider the matter of child lan­guage acquisition, we find that in the "base-forms-only" approach one has to assume that the child instantaneously acquires adult base forms (serv­ing as inputs to forward rules) and sandhi forms (serving as inputs to backward rules) for all items, but that his rule application ability in individual items develops gradually. In the "surface-forms-too" approach, however, one will have to hypothesize that the child starts with approx­imations of all items and that it is only through several stages of de­velopment that he gradually reaches their adult models.

Therefore, if the child's lexicon is really like what is proposed in the "base-forms-only" hypothesis, it would be reasonable for the child to suddenly make a decision to apply his rules to the fullest de­gree rather than to let this process of acquiring drag on for a long period of time. For all it requires of the child is to send himself, so to speak, an instruction of the following kind: "Regardless of the di­rection of rule application, change all rules from 'never apply' or 'sometimes apply' to 'always apply'." However, as a matter of fact, the learning of adult forms by children continues for many years.

On the other hand, if the child's lexicon resembles the "surface-forms-too" model, the gradualness in the acquisitional process in indi­vidual words is not so hard to understand. For there is no simple and easy way for the child to send himself an instruction of the above kind without referring to adult models. For example, an instruction for the child to convert his infantile alternants of the category of tone 55 to adult forms will have to be something like "change all alternants to tone 33's if they occur in the sandhi position and to tone 55's if they occur in the base position". But this instruction will also mistakenly affect all other tone categories. To avoid this error, the instruction will have to mention in addition that "only those items that are pro­nounced with base tone 55's by the adult will be affected".

Yet, it is quite conceivable that the correspondence rules between adult targets and child approximations which the linguist sees in great

122 HSIN-I HSIEH

transparency is not as clearly comprehended by the child. It is there­fore very likely that the child does not feel confident enough to make a risky, wholesale reshuffle according to an instruction which requires a meticulous adherence to complex and unclear conditions. If this obser­vation is correct it would offer a quite plausible explanation of the lexical gradualness in child language acquisition. From this viewpoint, there is ample reason for us to favor, however tentatively, the "surface-forms-too" lexicon over the "base-forms-only" lexicon.

2.7. The growth of child lexicon. Let us now return to Table III and study it in the light of the "surface-forms-too" lexicon. We find that Type IX items are not found in SI but are found in S2 and S3. By con­trast, type I items are discovered in S1 but not in S2 or S3. While Type IX is identical to the adult type, Type I is an opposite of the adult type. This difference among subjects suggests that it is likely that the present stage of the speech of S2 as well as that of S3 has evolved from a previous stage not unlike the present stage of S1. In other words, in the process of their speech development, these children have tended to abandon their original forms such as those of Type I, eventually to acquire adult forms of Type IX.

Taking Type I and Type IX as the beginning and the end points of speech development in these subjects, we may arrive at several alterna­tive reconstructions of the time order of the nine evolutional types. While we have as yet no objective criteria for choosing from among these alternatives, the sequence that is identical to the numerical progression given previously seems to be a workable hypothesis for our data.8

In the light of this time sequence, the occurrence or absence of these types in different subjects becomes meaningful. We observe that the stages of evolution cover from Type I to Type VI for SI, from Type V to Type IX for S3 and from Type VI to Type IX for S2. As a learner, S1 is less advanced than S3, who in turn is less advanced than S2. This gradation in terms of types is supported by a parallel gradation in terms of precentag.es of rule application. Thus, in the forward operation, the rate of application of rule a increases from 16% for S1, to 83% for S3

HOW GENERATIVE IS PHONOLOGY? 123

and finally to 85% for S2. In the backward operation, the ascending hierarchy with respect to rule a is formed by 32% for S1, 82% for S3 and 98% for S2.

It therefore appears quite reasonable for us to guess, by referring to this time order, that when a child starts to learn Taiwanese tone sandhi alternations, all or at least most of his lexical items will be­long to Type I. As his learning process continues, each individual item will go, step by step, through all nine stages until it becomes a Type IX, that is, an adult type lexical item.

It remains a question whether a lexical item newly added or reintro­duced to the vocabulary of a child will have to start from the "absolute-primitive" type, i.e., Type I, regardless of how far ahead in progress the child may be, or whether it will start from the "relative-primitive" type, that is, the least advanced type in the speaker concerned.

While further study is called for, the l mited data we have had at our disposal seems to suggest that the latter possibility is more likely to be true. For example, item 7a. "custard apple" is a rare or even un­known fruit to the subjects. When the experimenter presented this item to the subjects in Test I, they had a hard time identifying it as a fruit. Thus, it is likely that this item is treated by the subjects as a new or reintroduced word. We can see in Table III that this item belongs to the relative-primitive type of S2, i.e., Type VI. It also belongs to the relative-primitive type of SI, i.e., Type I. It belongs to Type VII of S3 rather than Type V, the relative-primitive type. Considering that S3 was nine years old at the time of the test, the item 7a. "custard apple" was probably not really new to him. If our interpretation of items such as 7a is correct, we may hypothesize that as a new item is admitted to the lexicon, it is treated first as a member of the relative-primitive type.

If we further study Table III (see above), we will find that item la. "banana" is treated as a member of the "relative-latest" type in all three subjects. Thus, it belongs to Type VI in S1, and to Type IX in S3 and S2. There is no doubt that the banana is among the most familiar kinds of fruit for these children. It thus appears to be the case that

124 HSIN-I HSIEH

the advanced status of an item is correlated with the subject's "famil­iarity" with that item. The term "familiarity" here corresponds roughly to the frequency of occurrence of forms. But it also includes such other factors as the speaker's understanding of the semantic and syntactic structure of the forms.

Based on these two observations, it seems reasonable for us to hypothesize that the ten test items form a continuum of familiarity, with the least familar item 7a. "custard apple" (or 5a. "parsley" in the case of S3) occupying one end and the most familiar item la. "ba­nana" occupying the other. More generally, we hypothesize that the scale of "adultness" or "correctness" according to which a child's lexical items are measured essentially corresponds to the scale of the child's familiarity with these items.

2.8. " How does adult lexicon expand? At this point of our inquiry, it may also be asked whether the "adultness" or correctness of an adult lexical form is to some 'extent determined by the adult's familiarity with the form. If the answer is affirmative, it would provide us with some basis for arguing that the adult lexicon is constructed or at least evolves in the same way as the child lexicon.

Although new items are daily introduced into the lexicon of an adult, it is difficult to observe this process on a short-term basis. Nevertheless, experiments can be designed to obtain results that may shed light on the adult's ability and the procedure he uses in acquir­ing new lexical items.

3.1. The second experiment. With this purpose in view, we conducted two additional tests, Tests V and VI, in which we study adults' ability to perform sandhi rules in artificial four-syllable compounds. In these compounds, artificial trisyllabic morphemes serve as modifiers to actual monosyllabic head-nouns in the fourth syllables. The same five head-nouns used in the previous tests are used here again. With the exception of the third elements, syllables in the artificial trisyllabic morphemes are either unreal or real but difficult to identify. However, the third syllables are all common actual words. These artificial trisyllabic mor-

HOW GENERATIVE IS PHONOLOGY? 125

phemes are mixed with disyllabic real fruit names as control items. The instructions are similar to those given to the children in the

previous tests. The experimenter informs the subject that the test is concerned with fruits and their flavors, etc. In Test V, the subject, upon being presented the name of a fruit, is asked by the experimenter to supply the word for the flavor, for example, of the fruit. In Test VI, the subject, having been presented the word for the flavor, for example, of a fruit, is asked to identify the fruit.9

Five adults who are native speakers of Taiwanese complete three trials of the experiment.10

3.2. Results. Upon analyzing the responses supplied by the subjects, we discover several interesting phenomena. First of all, we find that different items governed by the same rule may receive different treat­ments from the same subject. For example, in S4, item 45b which carries surface tone 21 receives an overall "I/C" response. But item 50b, also carrying surface tone 21, always receives an "I" response. (See Table IV in the Appendix for the word list.)

Second, even though two items may receive the same types of re­sponses, the rates of rule application for them may differ greatly. For example, in S4, both 45a (tone 33) and 50a (also tone 33) receive a mixed "I/C" response. But the overall rating of correctness for 45a is only 10%, while that for 50a is as high as 80%.

Third, the rate of rule application in one direction may be signif­icantly different from that in the other direction (cf. Tables Va and Vb).

Fourth, regarding the same test item or same tone category, various subjects may react with various types of answers (in terms of "I", "I/C" and "C") or with different degrees of correctness.

Since individual items regulated by the same rule may be treated in different ways by the same subject, the subject seems to have followed the principle of lexical diffusion rather than observed the neogrammarian rule of absolute regularity or the Labovian rule of variability.

What, then, is the force that propels the lexical diffusion in a

126 HSIN-I HSIEH

subject? Could it be his familiarity, taken in a quite broad sense, with the test items? Before answering this question, some preliminary discus­sion is needed.

The rates of successful application of individual rules for all five subjects, in all four frames, hue, teng, kau and tseng are compared in Table Va (forward application) and Table Vb (backward application).

! Forward Operation

in the frame hue__ in the frame teng_

a b c d e a b c d e

S4 100 100 100 0 10 100 100 100 0 80

S5 100 60 26 0 0 80 90 32 0 0

S6 100 100 6 22 0 90 100 0 30 0

S7 60 50 0 0 0 80 40 0 0 0

S8 100 80 0 0 70 80 80 0 0 60

Table Va Percentages of correct responses given to stimuli in the

frames of hue__ and teng__ in forward operations by S4 through S8.

HOW GENERATIVE IS PHONOLOGY? 127

Backward Operation

in the frame kau__ in the frame tseng

a b c d e a b c d e

S4 100 100 70 0 40 100 100 70 0 40

S5 30 30 0 0 0 30 30 0 0 0

S6 100 90 40 0 0 100 90 10 0 10

S7 60 50 0 10 16 30 20 0 0 20

S8 100 100 20 0 90 100 90 40 0 100

Table Vb Percentages of correct responses given to stimuli in the

frames of kau__ and tseng__ in backward operations by S4 through S8.

128 HSIN-I HSIEH

We see that for all subjects, regardless of the direction of appli­cation, rules a and always apply to a greater degree than, or at least to the same degree as, rules , d and e , with the negligible exception of the backward operation by S8 in the frame tseng_. Thus, there seems to be little doubt that for almost all subjects the applicability of rule a and that of rule b rank as the two highest among the five rules.

Furthermore, for each particular subject, the ranking of rules in each direction is essentially stable. We mean two things by "essentially stable". First, rules a and b always occupy the two top ranks and rules c, d and e always fill up the three lower positions. Second, among the three lower rules, a particular rule consistently occupies the first position. Thus, for example, in S6 the priority sequence of rules in the forward direction is a, b > d > c, e in the frame hue_. These rules remain in the same order in the frame teng__. In spite of the frame, rules a and b persistently occupy the first two positions, and d in­variably fits in the third position, that is, the first among c, d and e.

As regards the hierarchical order of rules c, d and e, it varies according to the subject and the direction. For instance, the above order for S6 will change to that of a, b > e > c, d if the subject shifts from S6 to S8. This order will, however, change to that of a, b > > d, e if the direction shifts from "forward" to "backward".

The consistencies and variations observed in the above may be summarized in the following two statements:

(1) For almost every subject, his rates of applying rules a and b are higher than his rates of applying rules c3 d and e.

(2) The ranking of the rates of application of rules e3 d and e varies according to subjects and directions.

The first statement answers the possible doubt as to whether in our experiment a subject's success or failure in supplying a correct re­sponse to a stimulus is dictated by his whim or due to his ability. The agreement among the subjects with respect to their higher rates of suc­cess in rules a and b contends strongly for a similarity among their abilities in applying different rules rather than a highly improbable

HOW GENERATIVE IS PHONOLOGY? 129

coincidence in their whims. In other words, what we are studying in these subjects is not their whim but their ability to apply the tone sandhi rules.

The second finging, together with the first, indicates that, for almost every subject, his ability to apply rules a and is always stronger than his ability to apply rules c, d and e. His capability of operating with rules c, d and e is furthermore determined by his lin­guistic peculiarity and the direction in which he applies these rules. 3.3. Why do some rules apply more frequently than others? Some may be tempted to argue, on the basis of the foregoing indications, that the subject's normal degree of rule awareness, wich is 100% in the case of actual forms, has considerably diminished when the subject is forced to cope with artificial items. Further, they may argue, the reduction of an individual's degree of awareness follows a set pattern so that under certain generally characterizable conditions, a subject's rates of appli­cation of rules remain in an essentially constant ranking. To state this argument in terms of variable rule, one would perhaps assert that the tone sandhi rules which apply categorically in an actual language situa­tion have systematically been reduced to variable rules in test perfor­mance.

Unfortunately, such an argument does not hold. It is true that some­times our data can be described by variable rules. However, at other times, the variations loom so wide that it is very doubtful that they ought to be considered as being governed by the same rule. In the ex­treme case, we are even forced to write a variable rule for each single item. Thus, for example, in S4, in the forward direction, rule e applies to a degree of as low as 10% in item 45a but to as high as 80% in 50a. In the backward direction, rule e applies to a degree of 40% in item 45b but it applies to a degree of 0% in item 50b. The disparity in the amount of 70% or 40% is so great that it ceases to be meaningful for one to insist on treating s.uch a disparity as mere negligible variation according to one and the same rule. To a lesser degree, disparities of this kind exist elsewhere in the data from all five subjects.

It is now apparent that neither the neogrammarian rule of regular-

130 HSIN-I HSIEH

i ty nor the variable rule can help us explain fully the results in this experiment. For our purpose, we need to hypothesize another kind of human faculty of speech perception and production. 3.4. The power of association. It would seem that the strongest candi­date for this hypothetical linguistic faculty is the "power of associa­tion" or, "analogical power". It is quite plausible that owing to this power a subject responds to a new or unfamiliar word by associating it with one or more already-known words that are similar in some respects to the new or unfamiliar item. He then supplies responses that resemble in some relevant aspects the responses that he would give to the already-known items being associated.

Let us illustrate this power of association with item 44a. mala-hue 21. When a subject is presented this stimulus, he or she probably tries to associate the syllable -hue 21 with one of the several actual items including hue 21 "goods", phue 21 "to match", kue 21 "to pass", etc. that have the diphthong -ue and the tone 21. Suppose that he suc­ceeds in associating the test syllable -hue 21 with the actual form hue 21 "goods", then what he will do next is just to respond as he would to hue 21 "goods". For the choice of the appropriate alternant, i.e., hue 53, he relies on his "surface-forms-too" lexicon. If he happens to have associated -hue 21 with phue 21 "to match", he will have to tell himself, so to speak, that hue 21 is to be treated exactly like phue 21 in the tone. The proper choice would have been phue 53 if phue 21 were involved. Accordingly, his answer is hue 53. If he fails to link -hue 21 with any known syllable, he may be cautious and just repeat what the experimenter has pronounced to him, i.e., -hue 21, without making any change on the tone. It is also possible that he may try harder only to result in a wrong association. Thus, he may associate -hue 21 with the frequently used word hue 55 "flower" that does not have the same tone. He may then respond with the deviant form hue 33. This kind of wrong association helps to explain the otherwise puzzling behavior of S6 (female) who responds with surface tone 33, without any plausible reason, to base tones 53 and 21 in the forward operation.

The experimental results, some of which have puzzled us earlier,

HOW GENERATIVE IS PHONOLOGY? 131

can now be satisfactorily explained in terms of the power of association, We have noticed that, for almost every subject, his rate of applying rule a or rule b is higher than his rate of applying rule c, d or e. Since as we hypothesized, the subject relies on actual words in his lexicon for his association work, the greater the membership of a tone category, the easier a new or artificial member of this category can be associated with old or actual members in this category. As a matter of fact, in Taiwanese there are undeniably many more actual words in tone 55 (undergoing rule a) as well as tone 35 (undergoing rule b) than in tone 53 (undergoing rule c), tone 21 (undergoing rule d) or tone 33 (undergoing rule e ) . According to DOC, a computerized pool of Chinese dialectological data operated by the Phonology Laboratory at the Univer­sity of California, Berkeley, in the dialect of Xiamen (Amory) which is another Min dialect closely related to Taiwanese, the distribution of syllables among the five long tones is as follows: 656 (in tone 55), 623 (in tone 35), 496 (in tone 53), 493 (in tone 21) and 483 (in tone 33). Apparently, due to this fact, the difference in the rates of application between rules a and b on the one hand and rules c, d and e on the other hand remains very stable in all but one subject.

One way to explain the three contending tone categories that form the low-ranking group is by referring to the difference in size among the memberships of these categories. One may hypothesize that these cate­gories differ in size to a degree that is great enough for their ranking to remain stable in each individual subject, and yet not so great as to allow it to stay constant across all subjects. Even though this is not supported by the DOC data cited above, it is entirely possible if we allow the lexicon of an individual to slightly deviate, according to personal pecularity, from the "model lexicon" postulated by the lin­guist.

Another alternative is to hypothesize that, in spite of the similar­ity in size among the three tone categories, a subject, due to his lan­guage background, has different degrees of familiarity with these tone categories.

As regards the variation in rule applicability caused by the dif-

132 HSIN-I HSIEH ference in rule direction, it suffices for us to assume that the sub­ject's power of association for words in their sandhi forms may be different from that for words in their base forms. As for the wide range of variation in the rate of rule application among different test items regulated by the same rule, our explanation is quite simple. Partly because of their phonetic shapes and partly because of their seeming syntactic and semantic make-ups, different test items may have different degrees of "associability" with already-known words, given the lexicon of a particular person taking the test in a particular mood.

3.5. Child language acquisition compared to adult test performance. We are now ready to answer the question that we have posed namely, could it be the subject's "familiarity" with the test items that pro­pels the lexical diffusion in his responses? It must be obvious by now that our answer is affirmative. It appears plausible, on the basis of our experiment, to assume that the associability of individual arti­ficial words in tests for adults is roughly equatable with the child's familiarity with actual forms in language acquisition. As the child be­comes more familiar with certain actual forms, the more "adultly" or correct his pronunciation of these forms will become. Similarly, as the associability of the made-up items in tests increases, the rate of correct responses given to them by a subject will also increase.

If the associability of artificial syllables in tests for adults and the familiarity of children with actual forms in language acquisi­tion are in fact governed by the same principle, that principle is likely to apply also in the acquisition of real new words by adults and in the performance by children in tests. In other words, it is likely that the same principle of familiarity or associability of lexical items governs an individual's language acquisition and test behavior, be the individual an adult or a child.

It will be ideal for our study if we can make the same subjects perform in the function of a child as well as in that of an adult, in both a quasi-acquisition process and in a test.

HOW GENERATIVE IS PHONOLOGY? 133

4.1. The third experiment. In this connection, we are very fortunate to have been able to work with an informant whose special language back­ground in Taiwanese excellently qualifies her for this ideal experiment. The subject, to be called S9, is an undergraduate student. She moved with her mother from Taiwan to the United States at the age of eleven.11 Al­though she is an adult, her mastery of Taiwanese is approximately that of an eleven or twelve-year-old child. In other words, we have in her both an adult mentality and a child language ability. Our next step is to observe her performance in a quasi-acquisition process and in a test situation.

It seems reasonable for us to regard Test I and Test II in which actual words are used as involving quasi-acquisitional processes and Test III and Test IV in which artifical words are used as psycholin-guistic tests per se. So we asked her to take these four tests.12

The four tests were lined up in a rigid sequence of Test I, Test II, Test III, and Test IV. The subject was asked to complete three trials of the sequence. The second trial was a week after the first and the third trial was two days after the second. 4.2. Results. Having analyzed her responses, we discover that, like the other subjects who took these tests, S9 shows a lexical diffusion. To some extent, this lexical diffusion can be also stated as variable rules. In the forward-reals, the variable rule is observed to operate only in items belonging to the category of tone 33; in items of other tone categories, categorial rules operate. In the backward-reals, vari­able rules affect categories of base tone 33 and base tone 53. In the backward-fakes, variable rules affect base tones 21, 53 and 33. And in the forward fakes, the variable rules apply to all five tone categories.

Apparently, different rules operate with different degrees of ef­ficiency. In terms of "categorical application" ("C") and "variable ap­plication" ("V"), these rules can be compared in Table VI displayed on the next page.

134 HSIN-I HSIEH

Forward-fakes

Backward-fakes

Backward-reals

Forward-reals

a_ (for base 55) V

1 (for base 35) V

j[ (for base 21) V V

£ (for base 53) V v V

_d (for base 33) V v V V

Table VI Categorial (= ) and variable (= V) rules applied by S9 in all four test situations. The number of categorial

rules increases from F-F to B-F, to B-R, and, finally, to F-R.

HOW GENERATIVE IS PHONOLOGY? 135

We observe two striking phenomena in this table, First, the degree of successful application of rules measured through the number of cate­gorical rules ascends from forward-fakes to backward-fakes, then to backward-reals and finally to forward-reaIs,13 Interestingly enough, the reals are higher in the hierarchy than the fakes. Second, there is a clear implicational relation in the categorical-or-variable status among different rules. This implicational relation can be formulated as d > > e > a> where X > Y means that "the categorical application of rule X implies the categorical application of rule Y". 4.3. Degrees of detachment from real life situations. From these two observations, plus our knowledge that all five rules examined are cate­gorical in adult speech, we can make the following inference: The status of a rule observed to be categorically applicable in normal adult speech may change in an unusual speech situation such as in a test. Furthermore, the less familiar or the more detached from real life situations the test items are, the more likely the status of a rule in the test will degenerate from that of categorical application to that of variable application. The degeneration in different rules observes an implica­tional constraint so that as the degree of the detachment increases in a test, the categorical status of "weaker rules" (that is, rules that imply) will disappear sooner than the categorical status of "stronger rules" (that is, rules that are implied.). (Although a distinction can be made between unfamiliarity in real words and detachment from real life situation in artificial words, we use the terms "detached", "de­tachment", etc., to refer to test items in general.)

For instance, in the case of S9, the "detachment continuum" is formed by the forward-reals, backward-reals, backward-fakes and for­ward-fakes in an ascending order. In the forward-reals, all but the weakest rule, d, apply categorically. But as the degree of detachment increases in the backward-reals, the second weakest rule, c, becomes variable. As the degree of detachment further increases in the back­ward-fakes, the categorical status of all but the two strongest rules, a and b, disappears. When the degree of detachment again increases in the forward-fakes, none of the rules remain categorical.

136 HSIN-I HSIEH

4.4. The continuum of reality. The mere fact that the categorical ap­plications of rules in all four tests observe the same implicational constraint is sufficient reason for us to suspect that the subject treats these four sets of test material as having four degrees of reality rather than as either real or fake words.

Our conjecture is further supported by the fact that the subject is capable of improving her scores in all four tests regardless of whether real or fake items are used as stimuli.

Forward-fakes

Backward-fakes

Forward-reals

Backward-reals

Average

1 66.25 88.70 85.00 96.30 84.00

2 80.00 92.50 95.20 97.50 91.30

3 86.25 95.00 100.00 100.00 93.70

Average 77.50 92.00 93.40 97.90

Table VII Percentages of rule application by S9 in all four test situations. Three trials are compared. The subject is

observed to have improved in every second trial.

HOW GENERATIVE IS PHONOLOGY? 137

The ratios of success in the four tests are compared in this table. We notice something very interesting here: In all four situations, the subject has improved her ability to supply correct answers as she pro­ceeds from the first trial to the second and further to the third. Such improvement is observed not only when all four situations are considered together but also when each separate situation is examined. Thus, the average score of the subject's answers in all four situations is 84.0% in the first trial, 91.3% in the second trial, and 93.7% in the third trial. When each individual situation is considered alone, such a steady improvement also occurs. No exception is found. In the forward-fakes, for instance, the subject scores 66.25% in the first trial. Her score increases to 80.00% in the second trial, and further to 86.25% in the third trial. No matter how close to or detached from real language en­vironment a test may be, the subject is able to improve her success rate. In other words, she treats the four sets of test materials as four equals rather than as two different groups, one of real words and the other of artificial words.

From these two supporting facts, we infer that the dichotomy be­tween actual and artificial words is more apparent than real. Further, that there is a "continuum of reality" which extends from words that may be called "most real" (such as words appearing in every day conver­sation) on one end to words that may be considered "least real" (such as words used in a most bizzare test) on the other.

Based on this experiment, we claim that an individual (a child or an adult has the ability to absorb or "internalize" new words of any degree of reality by gradually familiarizing them or by associating them with already-known words in his lexicon. 5.0. Conclusion. On the basis of our experiments, we make the follow­ing conclusion regarding the nature of the lexicon:

The expansion of the adult lexicon and the growth of the child lexicon both rely on the same human capability. This ability is mani­fested, on the one hand, as the power to familiarize oneself with new words in actual language acquisition, and, on the other hand, as the

138 HSIN-I HSIEH

power to associate artificial words in tests with real words. With respect to a fixed body of data, an individual may increase

this capability along the time dimension. In general, for a child this would mean to "rectify" the new forms according to their adult models. In the case of an adult, he will have to "regulate" these forms by analogizing them according to already-known words. Of course, the child is not precluded from being able to "regulate", nor the adult from being able to "rectify". During such a process of learning, an individ­ual treats a new word in natural acquisition or in test performance as a somewhat independent item rather than a token of a category. To a certain extent, different items subject to the same rule may share similar stages of development and may thus be described by a variable rule. Yet, often they are so persistently different in their evolu­tionary status that, in addition to treating them as a category, we need to describe them as separate items. For this purpose we need the theory of lexical diffusion (cf. Hsieh 1972).

If child lexicon and adult lexicon grow and expand in the same manner as our experimental results seem to suggest, it is then reason­able for us to infer that the ways in which these two kinds of lexicons are constructed are identical. There is already evidence that the child's lexicon is a "surface-forms-too" kind of lexicon. Accordingly, the adult's lexicon should also be a "surface-forms-too" type. In oth­er words, contrary to current belief that only phonological underlying forms are needed in the lexicon, it is necessary for us to list phon­ological surface forms in the lexicon.

NOTES Note that footnote one has been placed at the hottom of the first two pages of this article, which was originally entitled "On Listing Phon­ological Surface Forms in the Lexicon". 2 There are seven rules that each apply to one of the seven tone cate­gories. In our experiments we are concerned only with the five long tones that occur in the 'long syllables', that is, syllables that do not end in -p, -t or -k. Because of complicated dialectal variations in the two short tones that occur in the 'short syllables', that is, syllables ending in -p, -t or —k, the two short tones are not included in our study. 3 Although some common characteristics are observed among these five rules, and although rule schemata in terms of distinctive features of tone can be formulated (cf. Cheng 1968, 1973), we find it more con­venient in our experiment to treat the change in each tone as being governed by an individual rule rather than by a rule schemata shared by all five tones. 4 This situation can be compared to the type of English compounds as exemplified by blackboard eraser. There, the trisyllabic compound is based on a productive formula but the disyllabic compound is restricted to occur only in such forms as b l a c k b i r d , blackberry or such forms as surfboard, billboard. Other forms such as *blackwall or ^whiteboard are not allowed. 5 This situation is comparable to creating the artificial English com­pound *w h i t e b o a r d on the basis of the real compound b l a c k b o a r d . 6 They were brothers and sister (S2). The parents of these children are native speakers of Taiwanese; they also speak Mandarin Chinese, Japanese and English. These children had lived in the United States for about two years when they took the tests. In varying degrees of fluency, they speak Taiwanese, Mandarin Chinese and English, Taiwanese being their first language. In their daily conversation, the children speak the '33' dialect, although occasionally they show influence from speakers of the '21' dialect. The subjects were tested separately and, whenever possible, without the presence of others. The experimenter (E) is an adult linguist who speaks the '33' dialect as his native language. 7 The child's responses are acknowledged by E and are either marked on a test sheet or recorded by a tape-recorder to be transcribed later. Comments on the correctness of the child's answer are avoided. If a

140 HSIN-I HSIEH

child gives more than one answer to a particular item either because he changes his mind or because E wants him to repeat, then the two or more responses are all accepted. - Attempts were made to keep the four tests in the fixed sequence of Test I, Test II, Test III and Test IV but in vain. It is hard to make the children work according to rigid schedule. The interval between every two trials on the same test ranges from one hour to two weeks. 8 Derek Bickerton, observing that five of the nine possible types occur more frequently than others, has suggested (personal communication) that the evolutional types be ordered according to a constraint whereby al­ternants of morphemes 'produced' by rules applied in the backward direc­tion are at least as advanced as and at most one step more advanced than alternants ' produced' by the same rules in the forward direction. This constraint is supported by the fact that subjects generally perform better in the backward operation than in the forward operation of rules. Accordingly, such a constraint will yield the following time sequence: Type I. I:I, Type II. I/C, Type V. I/C:l/C, Type VI. I/C:C, and Type IX. C:C. Bickerton's solution has the additional merit of being able to predict the exclusion of type III items from the responses of the sub-jects. Yet its defect lies in its failure to accommodate types IV, VII, and VIII which occur despite his constraint. 9 The order in which the test items are presented to the subject is as follows: Each of the first ten items of modifiers, 41a-50a, is combined with head-noun I. Following this, each of the second ten items of modi­fiers, 41b-50b, is extracted from compounds formed with head-noun I. The same process is repeated for head-nouns II, III, IV and V. This completes one trial of the test for a subject. Each subject is asked to make three trials of the test. The interval between each two trials is usually a week but sometimes it is as short as a day or as long as a month. 10 With the exception of S4 who speaks the '21' dialect, all subjects speak the '33' dialect, which is also the dialect the examiner. These subjects are either graduate students or wives of students from Taiwan. 11 Hereafter, she and her mother continued to speak Taiwanese for several years until they finally settled down to a strange way of communication in which she would speak English to her mother and her mother would speak Taiwanese to her. Because of a lack of Taiwanese speakers in her life circle, she normally did not communicate in Taiwanese except with her mother. At an interview prior to the tests, her Taiwanese impressed us as very fluent, though she seemed to be short of literary vocabulary and a socially proper control of the style. 12 Since our experience with SI, S2, and S3 shows that the difference in the head-nouns is not very significant, we used only the head-noun I, tiam 21 "store" in her case. 13 When the overall percentage of correct responses rather than the num­ber of categorical rules is used as the basis of comparison, the rank­ing of the four test situations with respect to their susceptibility to application of rules changes slightly to that of forward-fakes, backward-fakes, forward-reals and backward-reals in an ascending order (cf.Table VII.).

141

REFERENCES

Bailey, Charles-James N. 1973. Variation and Linguistic Theory. Arling­ton, Va.: Center for Applied Linguistics.

Berko, Jean. 1958. "The Child's Learning of English Morphology". Word 14.150-77. (Repr. in Psyoholinguistics: A book of readings ed. by Sol Saporta, 359-75. New York: Holt, Rinehart & Winston, 1961.)

Bickerton, Derek. 1971. "Inherent Variability and Variable Rules". Foundations of Language 7:4.457-92.

Chen, Matthew. 1972. "The Time Dimension: Contribution toward a theory of sound change". Foundations of Language 8.457-98.

Cheng, Chin-Chuan. 1968. "English Stress and Chinese Tones in Chinese Sentences". Vhonetica 18.77-88.

Cheng, Robert. 1968. "Tone Sandhi in Taiwanese". Linguistics 41.19-42. . 1973. "Some Notes on Tone Sandhi in Taiwanese". Linguistics

100.5-25. Chomsky, Carol. 1966. The Acquisition of Syntax in Children from 5 to

10. Cambridge, Mass.: MIT Press. Chomsky, Noam, and Morris Halle. 1968. The Sound Vattern of English.

New York & London: Harper & Row. Clark, Eve V. 1971. "On the Acquisition of the Meaning 'before' and

'after'". Journal of Verbal Learning and Verbal Behavior 10.266-75. Dingwall, William Orr. 1971. "Linguistics as Psychology". A Survey of

Linguistic Science ed. by W. 0. Dingwall, 759-802. College Park, Md.: Linguistics Program, Univ. of Maryland.

Houston, Susan H. 1972. "Contingency Grammar: Introduction to a gener­al theory of competence and performance". Pagers in Linguistics 5:1.10-27.

Hsieh, Hsin-I. 1970. "The Psychological Reality of Tone Sandhi Rules in Taiwanese". Vagers from the Sixth Regional Meeting of the Chica­go Linguistic Society, 489-503. Chicago: Chicago Linguistic Soc.

. 1972. "Lexical Diffusion: Evidence from child language ac­quisition". Glossa 6:1.89-104.

Hyman, Larry H. 1970. "The Role of Borrowing in the Justification of Phonological Grammars". Studies in African Linguistics 1.1-48.

142

Kiparsky, Paul. 1971. "Historical Linguistics". A Survey of Linguistic Science ed. by W. 0.. Dingwall, 576-642. College Park, Md.: Univ. of Maryland, Linguistics Program.

Koutsoudas, Andreas, Gerald Sanders, and Craig Noll. 1974. "On the Ap­plication of Phonological Rules". Language 50:1.1-28.

Krohn, Robert K. 1972. "The Vowel Shift Rule and its Productivity". Language Sciences 20.17-18.

Labov, William. 1969. "Contraction, Deletion, and Inherent Variability of the English Copula". Language 45:3.715-62.

Ladefoged, Peter, and Victoria Fromkin. 1968. "Experiments on Compe­tence and Performance". IEEE Transactions on Audio and Electroacous-tics 16:1.130-36.

Liao, Chiu-chung. 1971. The Syntactic Environments of the Tone Sandhi in Taiwanese, M.A. thesis, Taiwan Normal University.

Maher, J. Peter. 1969. "The Paradox of Creation and Tradition in Gram­mar: Sound pattern of a palimpsest". Language Sciences 7.15-24.

Moskowitz, Arlene I. MS, On the Status of Vowel Shift in English, Ohala, John J. 1973. On the Design of Phonological Experiments, Univ.

of California, Berkeley; Dept. of Linguistics, mimeo. Ohala, Manjari. 1972. "The Abstractness Controversy: Experimental in­

put from Hindi". Paper presented at the LSA Annual Meeting in At­lanta, Georgia, Dec. 1972 [like the original version of J. J. Ohala 1973, above].

Steinberg, Danny D. 1973. "Phonology, Reading, and Chomsky and Halle's Optimal Orthography". Journal of Psycholinguistic Research 2:3.239-to 258..

Vennemann, Theo. 1972. "Phonological Uniqueness in Natural Generative Grammar". Glossa 6:1.105-16.

Wang, William S.-Y. 1967. "Phonological Features of Tone". Interna­tional Journal of American Linguistics 33:2.93-105.

. 1969. "Competing Changes as a Cause of Residue". Language 45:1.9-25.

Zimmer, Karl E. 1969. "Psychological Correlates of Some Turkish Mor­pheme Structure Conditions". Language 45:2.309-21.

143

APPENDIX

Table II: Stimuli used in Tests I, II, III, and IV

a) Monosyllabic morphemes used as head-nouns: I. tiam 21 "store, shop" IV. tsi 53 "seed" II. pi 55 "side" V. bi 33 "flavor, smell" III. tsi 35 "money"

b) Real disyllabic compounds used as modifiers to the head-nouns: la. kin 33 tsiә 55 "banana" 20a. iu 33 a 53 "grapefruit" 2a. tshai 53 kue 55 "squash" 21a. tek 5 sun 53 "bamboo shoot" 3a. nai 21 tsi 55 "lichee" 22a. the 33 a 53 "peach" 4a. si 33 kue 55 "watermelon" 23a. li 55 a 53 "plum" 5a. en 33 sui 55 "parsley" 24a. suāi 33 a 55 "mango" 6a. kirn 33 kue 55 "pumpkin" 25a. pe 21 tshai 21 "Chinese 7a. sek 5 khia 55 "custard ap- cabbage"

ple" 26a. hue 33 tshai 21 "cauliflower" 8a. tang 33 kue 55 "winter-melon" 27a. eng 53 tshai 21 "watercress" 9a. phang 33 kue 55 "honeydew" 28a. ku 55 tshai 21 "chives" 10a. bok 3 kue 55 "papaya" 29a. kua 53 tshai 21 "mustard 11a. phu 33 thә 35 "grape" green" 12a. ong 33 lai 35 "pineapple" 30a. tau 21 tshai 21"bean sprouts" 13a. iũ 33 thә 35 "star-fruit" 31a. len 55 bu 33 "Taiwanese 14a. tshai 53 thau 35 "radish" apple" 15a. leng 33 keng 53 "dragon-eye" 32a. tho 33 tau 33 "peanut" 16a. lai 33 a 53 "pear" 33a. ang 33 khi 33 "red per-17a. le 33 bong 53 "lemon" simmon" 18a. kam 33 a 53 "orange" 34a. ang 33 tau 33 "red pea" 19a. phong 21 53 "apple" 35a. tsui 55 khi 33 "yellow per­

simmon" Note: In the tests, items both 36a. lek 3 tau 33 "green pea"

in b) and c) are arranged 37a. huan 33 be 33 "corn" so that no tvo items of 38a. tshan 33 tau 33 "kidney bean" the same tone appear in 39a. bin 55 tau 33 "bean" a sequence. 40a. ng 33 tau 33 "yellow pea"

144 HSIN-I HSIEH

c) Artificial disyllabic compounds used as modifiers to the head-nouns:

lb. tshai 53 tsiә 55 21b. iû 33 sun 53 2b. lek 3 kue 55 22b. sin 33 a 53 3b. pe 21 tsi 55 23b tang 33 a 53 4b. tho 33 kue 55 24b. tsiu 55 a 53 5b. tho 33 sui 55 25b. tshiu 55 tshai 21 6b. iu 53 kue 55 26b. gu 33 tshai 21 7b. kim 33 khia 55 27b. bi 55 tshai 21 8b. tsa 55 kue 55 28b. ba 5 tshai 21 9b. iü 33 kue 55 29b. tía 55 tshai 21 10b. bi 55 kue 55 30b. hue 55 tshai 21 l1b. iu 53 thә 35 31b. tsui 55 bu 33 12b. gu 33 lai 35 32b. hue 55 tau 33 13b. hue 55 thә 35 33b. pe 21 khi 33 14b. hue 33 thau 35 34b. leng 33 tau 33 15b. ho 55 keng 53 35b. hue 55 khi 33 16b. 55 53 36b. ai 53 tau 33 17b. tsui 55 bong 53 37b. tiong 33 be 33 18b hok 5 a 53 38b. huât 5 tau 33 19b. tshiu 55 53 39b. tsa 55 tau 33 20b. niü 33 a 53 40b. kin 33 tau 33

* * * * *

Table IV: Artificial trisyllabic morphemes used as modifiers in the four-syllable compounds in the forward (a) and backward (b) operations. The tone sequence for these modifiers is 33-33-X. The last syllables, being actual, are annotated.

a) 41a. phala-hue 55 "flower" b) 41b. phala-kau 33 "furrow" 42a. thala-hue 35 "turn" (noun) 42b. thala-kau 33 "monkey" 43a. khala-hue 53 "fire" 43b. khala-kau 55 "dog" 44a. mala-hue 21 "goods" 44b. mala-kau 53 "religion" 45a. nala-hue 33 "meeting" 45b. nala-kau 21 "thick" 46a. phala-teng 55 "lantern" 46b. phala-tseng 33 "bell" 47a. thala-teng 33 "layer" 47b. thala-tseng 33 "emotion" 48a. khala-teng 53 "top" (noun) 48b. khala-tseng 55 "seed" 49a. mala-teng 21 "to hammer" 49b. mala-tseng 53 "to plant" 50a. nala-teng 33 "hard to break" 50b. nala-tseng 21 "quiet"

* * * * *

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY*

MICHAEL KENSTOWICZ

0.0. One of the major differences between generative phonology and earlier approaches to 'process' phonology/morphophonemics is the heavy reliance upon rule ordering in the generative approach and the virtual absence of this device in the latter (Bloomfield's "Menomini Morphophonemics" [1939] being a well known exception). This is all the more remarkable when it is remembered that in many of these earlier descriptions the underlying forms are just as distant from the surface, and the relationships between the rules are just as intricate, as that found in most generative analyses. Two interpretations of this differ­ence between earlier approaches and more recent generative studies are possible: either (1) the earlier phonologists implicitly assumed the device of rule ordering, the lack of any explicit discussion of the matter merely being consonant with the general lack of interest in de­veloping an explicit theory of phonology/morphophonemics; or, alterna­tively (2) they had a different conception (largely implicit) of how rules interact in the conversion of base forms into phonemic/phonetic representations.

In this paper, I will examine the approach to rule application that is implicit in some of the most important descriptions of this pe­riod of American linguistics, i.e., before the advent of 'item and ar-

* An earlier version of this paper was presented at the Conference on Rule Ordering held in Bloomington, Indiana, in April 1973.

146 MICHAEL KENSTOWICZ

rangement' analysis. I will suggest that the conception of rule appli­cation implicit in most of these earlier studies is significantly simi­lar to the position recently sketched in Kisseberth 1973a, according to which (1) the sequencing of the rules in a derivation is governed by a principle of 'minimal opacity'; and (2) most nontransparent rule appli­cations are described in terms of global conditions.

1.0. The first description I shall discuss is Bloomfield's (1887-to 1949) "Menomini Morphophonemics" of 1939, since his approach to rule application is virtually identical to that most commonly employed in generative phonology (e.g., Postal 1968) and stands in sharp contrast to the more prevalent treatment of the question during the period under discussion.

Bloomfield sought, as far as possible, to provide each morpheme with a unique underlying form from which all of its various surface re­alizations could be predicted by a maximally general and economical set of rules:

The process of description leads us to set up each morphological element in a theoretical basic form, and then to state the devia­tions from this basic form which appear when the element is com­bined with other elements. If one starts with the basic forms and applies our statements (§§10 and following) in the order in which we give them, one will arrive finally at the form of the words as they are actually spoken (Bloomfield 1939:105-06).

Thus, for Bloomfield, as for generative phonologists, the underlying representation is converted into the phonetic representation by a se­quential application of rules, where the sequencing of the rules is controlled by the ordering.1 In other words, whether or not a rule applies to a representation is determined not only by whether that re­presentation appears in the requisite context, but also by the relative position of the rule in the ordering. More specifically, any given rule 1 Bloomfield does not use rule ordering in his descriptions of Tagalog

(1917) and Fox (1925), and thus seems to have adopted this device on­ly in the later part of his career. The rule numbers in our discus­sion of Bloomfield are his.

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 147

ri may be applied only after all preceding rules in the ordering have been tried, and once rļ has applied and the next rule rj' is tried, ri may not be tried again.

The overall goal of Bloomfield's "Menornini Morphophonemics" was to achieve a maximally economical and general statement of rules (cf. Bever 1967). It is clear that the particular ordering Bloomfield imposed on his rules achieves this overall economy to a significant degree. How­ever, he used ordering for a number of subpurposes which, as we shall see, would be considered distinct phenomena by linguists approaching the problem of rule interaction from a different point of view.

For example, Bloomfield employed rule ordering in order to permit a rule to apply not only to the underlying representation, but also to a representation that results from the application of another rule (a 'feeding' relationships, according to Kiparsky 1968). For instance, he formulates a Palatalization rule (13) which converts t to and n to s before e, ē, and y. This rule takes underlying pe?t-e "by error" to pe?c-e (ultimately, surface pe?c by other rules, see below). Another rule (10) inserts a 'connective e' in roughly the context +C. If the preceding morpheme ends in a t, it is regularly palatalized to 5: basic pyεt- 'hither' + -m "by speech" → pyzcem (ultimately ic-εw "he calls him hither"). Bloomfield accounts for this situation by ordering the Palatalization rule after Epenthesis, so that the former rule may oper­ate before basic e as well as before epenthetic e. It is obvious that if the Palatalization rule were not permitted to operate before epen­thetic , but was defined instead to operate upon the basic morphopho-nemic representation, then the rule would have to be written in a more complex form so that it would palatalize t not only before e, but also before a morpheme-initial consonant. Allowing Palatalization to apply before basic as well as before epenthetic e's by the device of rule ordering achieved the simplicity of statement that Bloomfield sought.

To cite just one more example of this type, rule (29) lengthens vowels in monosyllabic words: e.g., mw-:ekw "the other eats him" is con­verted to mok by other rules and is then lengthened to Apocope

148 MICHAEL KENSTOWICZ

(24) converts basic āsetε to aset "in return" (cf. asetε-hsem-ew "he lays them to overlap"). Since two-syllable words which lose their final vowel show up with a long vowel, Bloomfield orders the monosyllabic lengthening rule after Apocope; thus, pe?c-e (from basic pε?t-e) becomes pε?c by Apocope, and then pε?c.

Cases in which the outputs of one rule ri,- do not undergo another rule ri, even though the structural description of r-¡ is met, are han­dled by ordering ri before rj (a 'non-feeding' relation, following Kiparsky 1968). For example, recall that the Palatalization rule (13) takes t to before y: Netyanw- becomes neoyanw (ultimately, nicyan "child"). But there are instances of ty sequences on the phonetic sur­face. Rather than treat them as exceptions, and hence complicate the description, Bloomfield sets up a w between the t and the y; this w is then deleted by an independently motivated rule (16) which drops the first of two semi-vowels after a consonant: pehcekonahtyan "sacred bundle, pl.", which is set up as basic pēnt-ēkon-ahtwy-an. The Glide Dropping rule is, of course, ordered after the Palatalization rule so that the latter does not convert t to before yls that come to stand immediately before t as a result of Glide Dropping. And, as far as I have been able to determine, this is the only reason for ordering Glide Drop after Palatalization.

Ordering is also used to deal with cases in which a rule applies in a nonphonetic context (a 'nonbleeding' relation, cf. again Kiparsky 1968). For example, recall the pε/c comes from basic pe?t-e. In the der­ivation of this form, Palatalization (13) must be applied before Apocope (24), for otherwise, the e which conditions the conversion of t to would be lost.

We have seen how Bloomfield employed ordering to achieve a simpli­fication in the statement of his rules. However, there is one case in his "Menomini Morphophonemics" in which the device of ordering has the opposite effect. This becomes evident when rules (12) and (23) are compared. Rule (12) describes what happens to clusters of C+C which, exceptionally, fail to receive an epenthetic e by (10):

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 149

In such forms if the first consonant is other than n it is replaced by h: sēnak-at-k: sēnakāhk-en 'whenever it is difficult1, sēnakah 'whenever it was difficult1; cf. sanākat fit is difficult' ; ..(Bloom-field 1939:109).

In sēnakah, from basic sēnak-at -k , t is replaced by h by (12), and the resultant cluster hk is simplified in final position to h by rule (25), which drops consonants in cluster until only one is left. Rule (23) sim­ply states (p.113): "Clusters of n plus consonant are replaced by h plus consonant." In other words, (23) is the missing part of (12).

Bloomfield does not state why these two rules cannot be combined. In fact, he does not even remark upon their similarity. However, it is possible to reconstruct, at least partially, why he was forced to de­scribe these rules as separate. The rule immediately preceding (23), i. e., rule (22), replaces e by e before n plus consonant. Examples are:

pakam-εnt — 2 2 → pakam-ent —23—> pakam-eht — 2 5 → pakam-eh "if he is struck" εn-εnt — 2 2 → εnent — 2 3 → εneht — 2 5 → εneh "if he is called so"

kε-set-εns-an — 2 2 → kεsetens-an --23→ kesetehsan "thy toes"

Note that in the last two examples, t and n appear before e. Recall that Palatalization (13) converts t and n to 5 and s before e. Thus, Palatali­zation must be ordered before (22). Furthermore, (22) must precede (23), for there are many examples of surface phonetic ehc sequences (e.g., mεhkam "he finds it", apεhsos "deer", etc.). But rule (12) is ordered before Palatalization.2 If (23) were to be combined with (12), this would mean that (22) would have to precede Palatalization. But with such an ordering it would not be possible to block Palatalization in cases such as kesetehsan "thy toes", for (22) would have merged the contrast between ε and e. As we shall see, the more prevalent approach to rule application during this period would be able to overcome the difficulty examples like this present in a rule ordering framework.

I have not been able to determine why (12) must precede (13). If this is unnecessary, then it may be possible to escape from the ordering paradox by imposing the following sequencing of the rules: (13), (22), (12-23).

150 MICHAEL KENSTOWICZ

2.0. I will now turn to a rather different conception of rule ap­plication, one which is much more characteristic of the approach taken to morphophonemics in the United States before the advent of the 'item and arrangement' analyses of the 1940's and 50's.

I shall begin with a brief examination of the approach to rule ap­plication taken by Edward Sapir (1884-1939), the most outstanding prac­titioner of morphophonemics during this period. Sapir's approach is perhaps best exemplified in his descriptions of Takelma (1922) and Southern Paiute (1930). Although written at different points in his ca­reer, these analyses take the same basic approach to phonology. Like Bloomfield, Sapir conceived of morphemes as having (generally) a unique underlying form (the 'primary' or 'organic' form, in Sapir's terminol­ogy), which can be modified by phonological rules, depending upon the contexts in which they appear. The phonological rules have the effect of altering the underlying form to produce a 'non-primary' or 'inorganic' form, which may be phonetically identical with the primary form of an­other word. Unlike Bloomfield, however, Sapir does not evidence any con­ception of an imposed ordering of the rules as they apply to transform the underlying representation into the surface phonetic shape.

But this does not mean that for Sapir the rules provided a 'direct mapping' between the underlying and surface representations, i.e., that all rules are defined to operate on the underlying level ('simultaneous application', in the sense of Postal 1968). On the contrary, there are many examples in the Takelma and Southern Paiute descriptions which show that Sapir considered the underlying representation to be converted into the phonetic representation by a sequential application of rules. How­ever, the sequencing of his rules is not imposed by rule ordering state­ments.

In the Takelma grammar, for example, Sapir formulates a general rule which deaspirates intervocalic consonants if the immediately pre­ceding vowel is unaccented. Such a rule accounts for numerous alterna­tions, such as that exhibited by the 1st per.sg. suffix in words like phelxà-thê "I shall go to war", and phélêxa^de-? "I go to fight". This

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 151

rule interacts with another rule by which h is absorbed into a preced­ing stop, aspirating it if it is unaspirated: i.e., for the velar se­ries, for example, g or kh + h -+ kh. As Sapir (1922:43-44) notes, "Un­der suitable conditions of accent the contraction product kh or fc^may itself become g or gw so that all trace of the original h seems to be lost." One of the examples of the h-loss process that he cites is the form gUen-hekK)agw- "relate", from basic hegw-hagw, a reduplicated root in which the stem vowel is replaced by a. A related form of this root is also cited in which the accent appears on the reduplicated material: gwen-hegwagw-an-i "tell to". Here we see that Sapir clearly assumes a sequential derivation in which the Deaspiration rule applies to the out­put of h-Absorption:

hegw- root hegw-hágw Reduplication hekhw-agw h-A'bsorption hegw-agw Deaspiration

Another more striking example of sequential application is to be found in the Southern Paiute analysis. Under the section entitled "Vo­calic Contraction", Sapir formulates a rule whereby a+i and a+u con­tract into the diphthongs ai and au. He then states:

An au, itself usually contracted from a + u, is sometimes further contracted to a before qw nw, or p the labial vowel being ab­sorbed, as it were, into the following labialized consonant but leaving its quantitative value behind in the lengthening of the preceding a (Sapir 1930:17).

One of the examples Sapir cites to illustrate this process is nauq winqi-"to fight", which surfaces as na-q'winqi. The former representation is said to come from underlying *na-yuq -wi--, with reference to a sec­tion in which a rule deleting intervocalic is formulated. Here, then, is an example of the sequential application of three rules, each ap­plication creating an input to the next rule:

n a - γ u q - w i - n q i u n d e r l y i n g n a - u q ' w i - r i q i Y → é nauq w i - n q i C o n t r a c t i o n na q w i n q i u - -Absorpt ion

152 MICHAEL KENSTOWICZ

An important point to note here is that, in both this example and the preceding one from Takelma, the rules which apply to the outputs created by applications of previous rules are also permitted to apply to the underlying representation as well. Thus, the rule contracting a + to the diphthong au applies to primary a+ as well as to a + clusters created by the loss of y. Similarly, in the Takelma example, the Deaspiration rule applies to basic intervocalic aspirates as well as to those created by the h-Absorption process. Thus, Sapir, like Bloom-field, allows for sequential rule application to deal with 'feeding' re­lationships. However, unlike Bloomfield, for Sapir the sequencing of rules in a derivation is not controlled by rule ordering, for nowhere in the Takelma or Southern Paiute grammars (as well as in any of his other writings, as far as I know) are there to be found statements of the form "rule x must always be applied before rule y". But if the se­quencing of the rules is not controlled by the device of ordering, how then are the correct derivations ensured?

Although, to my knowledge, Sapir never explicitly discussed this question, it is my impression that he assumed (at least for 'feeding' relationships) what might be called a 'free application' principle to the effect that a rule applies every time its structural description is met. My chief reason for believing this is the manner in which Sapir states his rule, each rule is formulated in such a way that it does not make reference to the other. Thus, in the Takelma example where h-Ab-sorption feeds Deaspiration, Sapir's formulation (and discussion) of the latter rule makes no mention of h-Absorption. Similarly, in his formal statement of h-Absorption, nothing is said about the effects of other rules upon the output of this rule:

When standing immediately after a stop, an organic etymologically significant h loses its individuality as such and unites with a preceding media [voiced stop] or aspirated tenuis [voiceless stop] to form an aspirated tenuis (Sapir 1930:43).

It is true that after this formal statement of the rule he remarks: Under suitable conditions of accent (see §23) [where 23 refers to the place in which Deaspiration is formulated] the contraction

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 153

product h or may itself become g or gw so that all trace of the original h seems to be lost (43-44).

However, I do not think that this remark is to be taken as a condition on the application of the rules; rather, it is to be interpreted as an 'aside', explicating the structure of the words in which the application of h-Absorption is obscured by another rule. Similarly, in the Southern Paiute example just discussed, the Contraction rule is formulated sim­ply as a + > au (Sapir 1930:17), with no mention of whether or not the a + sequence is basic, or derived by another rule. Also, the process by which is absorbed into a following labial is formulated in such a way that the derivational source of the diphthongs is not relevant to the application of the rule.

If this interpretation is correct, it means that unlike Bloomfield, who described feeding interactions by rule ordering, Sapir ensures such rule interactions by a general principle of 'free application'. This does not mean, however, that Sapir excluded the possibility of 'non-feeding' or 'anti-feeding' relationships, a position espoused by Kout-soudas, Sanders, and Noll (1971). On the contrary, the recognition of opaque applications was one of the most distinctive traits of Sapir's work. Non-feeding interactions are made consistent with his implicit principle of free application, because in such circumstances Sapir gen­erally formulated his rules in terms of the derivational source of the segment undergoing or conditioning the rule.

This becomes clear when it is remembered that the terms 'organic' and 'primary' versus 'inorganic' and 'secondary' are technical terms for Sapir, referring roughly to the underlying representation versus representations resulting from the application of some phonological rule. For example, in his discussion of the Takelma accents Sapir notes the essential equivalence between a rising accent (marked with a cir­cumflex), which only occurs on long vowels and diphthongs, and a simple 'raised' accent (marked by the grave sign), which appears only on shortened diphthongs and final short vowels, compare Sapir (1930:17): "The rising pitch is for a long vowel or diphthong what the raised

154 MICHAEL KENSTOWICZ

pitch is for a short vowel or a shortened diphthong." In other words, the circumflex and the grave accents are really one and the same accent; the former is limited to long syllables, while the latter occurs on short syllables. Sapir then goes on to note one important point regard­ing the accent in diphthongs. This involves a rule of Epenthesis which inserts a in the context CC:

It is very important to distinguish between the organic diphthongs, in which each element of the diphthong has a distinct radical or etymological value, and secondary diphthongs arising from an i, u, l, , or n with prefixed inorganic a. The secondary diphthongs (ai, au, l, am, an), being etymologically single vowels, are al­ways unitonal in character; they can have the raised, not the ris­ing accent. Contrast the inorganic au of

bilaûkh (<*bilwkh not *bilaukh) he jumped; cf. bïlwâ?8 jumper with the organic au of

gayau he ate it; cf. gayawå?n I ate it Contrast similarly the inorganic an of

k!emankh ( <*k!emrikh, not ! emonieh) he made i t ; cf. femna? s maker

with the organic am of dasmayam he smiled; cf. dasmayama?n I smiled (Sapir 1930:19).

In other words, the equivalence of the circumflex and raised accent is defined on the underlying, organic value of length, not on that value resulting from the insertion of inorganic a. In a rule ordering frame­work, the rule which assigns the surface rising and raised tones would be ordered to apply before the Epenthesis rule, in an opaque fashion. As we have seen, Sapir did not recognize rule ordering conditions. The assumed tone assignment principle is therefore formulated in such a way that it is sensitive to the derivational source of the syllable.

A much more striking instance of the way in which Sapir treated opaque rule applications is to be found in his formulation of the fa­mous Southern Paiute 'Law of Alternating Stresses'. This law is formu­lated in terms of moras: "all odd moras are 'weak' or relatively un­stressed, all even moras are 'strong' or relatively stressed" (Sapir 1930:39). But the moras are defined in terms of the underlying repre-

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 155

sentation: "Every organic short vowel counts for one mora; every long vowel or diphthong for two" (p.38). However, as Sapir frequently points out, there are many phonological rules that alter the underlying sylla­ble structure of the Southern Paiute word; but these have no effect up­on the mora count, and hence no effect upon the placement of accent:

It is very important to note that all inorganic increments and losses have no effect on the mora-construction of the word. Sec­ondary Lengthening of short vowels, pseudo-diphthongization, glide vowels, shortening of long vowels and diphthongs all have no ef­fect (ibid.).

To illustrate this point, Sapir cites a form like qa( )niva(au)nWI (not glossed), which has four and not seven moras, because the underlying, organic form of this word is qanivanWI. The a's of the first and third syllable of qanivarWl are lengthened by a rule lengthening vowels in unaccented syllables (p.19), and the is a 'glide vowel' inserted af­ter a before a labialized spirant or nasal. (Cf. pâ-va-x-i "over the water" in which the first a is organically long and hence takes the stress).

Another process defined in terms of moras is the rule of Vocalic Unvoicing according to which ne\iery weak mora standing before a gemina­ted stop or sibilant loses its voice" (p.39). The opacity of this rule is nicely illustrated by an example like the following, where the vowel of the morpheme -ta- 'with the feet' is secondarily lengthened: nïvwât-a-maya-p-ïya 'went out to test the depth of snow with the feet', for "a primary two-moraed a- would have demanded the incorrect form: *nivwåt-a-mayaApïya" (p.20). That is, if the a- of -ta-- were basically long, the mora structure of the word would be as follows:

nivwat • aamayaap • ïya w s ws w sw s w

We should, therefore, expect the a mora before the p- to devoice. How­ever, since one of the preceding vowels in the word is secondarily lengthened (namely, the vowel of -ta-), the a mora before p- is strong and thus escapes Unvoicing.

156 MICHAEL KENSTOWICZ

To summarize, it seems that Sapir implicitly assumed a theory of rule application having the following features: (1) rules apply sequen­tially to convert the underlying representation into the surface repre­sentation; (2) the sequencing of the rules is controlled by a general principle of 'free application' (as opposed to extrinsic rule ordering statement or conditions); (3) opaque rule applications are provided for by formulating the rules such that they refer to the derivational source ('organic' versus 'inorganic') of a segment.

3.0. A theory of rule application having the above three features is somewhat more apparent in the classic analysis of Tlibatulabal by Sapir's students Morris Swadesh (1909-67) and Carl F. Voegelin (b.1906) published in 1939 (S and V in the following).

S and V begin their paper by drawing a distinction between two kinds of alternations. They observe that the -izy -s, -z alternation in the plural of English nouns is "conditioned only by the phonetic surround­ings of the morphemes", adding:

In the case cited [i.e., the - -s, -z alternation] the alter­nation is regular. It might be more correct to say that the phono­logic basis for the alternation is patent, as we may demonstrate by contrasting with this example another one involving 'irregularities' (P.1).

The example they go on to discuss is the different morphophonemic behav­ior of the English phoneme ƒ. In a limited group of words, morpheme fi­nal ƒ will alternate with v (e.g., l ea f , leaves), while in the majority of cases it does not (e.g., b e l i e f , beliefs). The alternation of ƒ with v is an example of 'non-patent' phonology. S and V then discuss the pos­sibility of dealing with it by setting up two different morphophonemes F and ƒ.

As we shall see, the distinction between 'patent' and 'non-patent' alternations is important in S and V's analysis of Tlibatulabal. It seems to me, however, that this distinction is open to at least two different interpretations. First, 'patent' could mean that the alternation is 'transparent' in the sense of Kiparsky 1971. That is, the phonological rule relating the alternants is phonetically (or for S and V, phonemi-

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 157

cally) true, and the context in which the rule applies is phonetically (phonemically) present (see their remark above "conditioned only by the phonetic surroundings of the morphemes"). All other alternations would then be relatively 'non-patent'. On the other hand, the term 'non-pat­ent' could have a more narrow interpretation, referring only to cases like l e a f , leaves, where it is necessary to make a morphophonemic dis­tinction in nonphonetic terms. It is difficult to determine from a read­ing of their paper which of these interpretations is correct. A crucial case for drawing the line between 'patent' and 'non-patent' would be an example like the Alternating Stress rule in Southern Paiute. The rule is clearly opaque, but evidently it can be formulated without recourse to special capital letter morphophonemes. Would Swadesh and Voegelin have considered this kind of alternation 'patent' (our second interpre­tation) or 'non-patent' (our first interpretation)?

In their analysis of Tübatulabal, Swadesh and Voegelin formulate some dozen morphophonemic rules. I will now briefly discuss some of them, concentrating on S and V's treatment of the interaction between these rules.

The first morphophonemic rule they formulate is one of Apocope: "most morphemes lose their final vowel when they stand in final posi­tion" (p.4). This rule accounts for the vowel-0 alternation evident in the following data:

tәk "to eat" tәka-t "he is eating" in "to do" inә-t "he is going" 51 "to get up" olo-t "he is getting up"

The next rule, called Terminal Unvoicing (TU), devoices stops and af­fricates in initial and in final position. In final position this rule describes alternations like the following:

tawak "to see" tawaga~t "he is seeing" tәkiwit "to eat collect- tәkiwda-t "he is eating

ively" collectively"

Note that this rule critically interacts with Apocope, for the stop comes to stand in final position only as a result of the deletion of

158 MICHAEL KENSTOWICZ

the final vowel. In an ordering framework, Apocope would have to be or­dered before TU. S and V make no ordering statement such as this. In fact, in their discussion of Terminal Unvoicing, no mention is made of the Apocope rule at all. This lack of any reference to Apocope can be explained by the hypothesis that S and V, like Sapir, assumed a 'free application' principle governing the sequencing of the morphophonemic rules as they apply to convert the morphophonemic representation into the phonemic.

Terminal Unvoicing in initial position occurs in imperfectives; the underlying voicing shows up in the corresponding perfectives which are formed by a reduplication of the initial vowel of the stem:

imperf. perf. tlwak ādawek "see" puw ubuw "irrigate" t9k ataki-n "eat" pusk upusk "blow"

Here, too, S and V make no mention of the relationship between TU and the Reduplication rule, which can be explained, I believe, in one of two ways. First, they might have considered Reduplication a morphologi­cal, as opposed to a morphophonemic, rule. If so, the precedence of Re­duplication over TU would follow from a general conception about the nature of grammar. On the other hand, instead of assuming a principle of 'free application', they may have operated with the idea that the sequencing of rules is controlled by a principle of 'minimal opacity' (Kiparsky 1971; Kisseberth 1973a). This would in general favor 'feeding' over 'nonfeeding', and 'bleeding' over 'nonbleeding' interactions. Al­though there are other places in their analysis indicating that S and V implicitly assumed a principle of 'minimal opacity', in this case I think that they would have followed Sapir and considered Reduplication a morphological process, in which case the proper sequencing of the rules would follow from a general conception of the organization of grammar.

S and V then proceed to formulate two rules treating the modifica-

RULE-APPLICATION IN PRE-GENERATIVE AMERICAM PHONOLOGY 159

tion of word-final clusters. In an ordering framework such as Bloom-field's, there would have to be special conditions requiring that each of these rules apply after Apocope. However, in both their formulation and their discussion of these two rules. S and V make no mention of Apocope. One of these rules metathesizes a sequence of h plus sonorant word-finally and could be formulated as follows:

h sonorant — > sonorant h / # mūyh mūhyu-t "celebrate" a?analh anāhlg-t "make it fast"

Note that the application of this rule to muyh (<*mūhyu) and a?analh (<*?1) is possible only by virtue of Apocope. In this case a der­ivation with the proper sequencing of the rules (i.e., Apocope - Meta­thesis) would be guaranteed by a 'free application' principle. However, this principle is not adequate to ensure the proper interaction between Apocope and the second rule modifying final clusters. The latter is a rule that deletes a glottal stop after a liquid or nasal except in fi­nal position. It accounts for the ?-0 alternation in forms like hal?, haldt "sit"; S and V set up *hal?e as the basic form for this stem. In order to convert *hal?e. into hal?, the rules of Apocope and Glottal De­letion must be sequenced in a 'bleeding' fashion, so that the final vow­el is lost, making the ? terminal, and thereby permitting it to escape Deletion. However, such a sequencing of the rules is not consistent with a 'free application' principle. 'Free application' would predict surface hal, because the underlying form *hal?d satisfies both Apocope and Glottal Deletion. Evidently, then, S and V assumed a theory of rule ap­plication which favored 'feeding' over 'nonfeeding' and 'bleeding' over 'nonbleeding' interactions. This interpretation of S and V's implicit conception of rule application would be entirely consistent with our first interpretation of their term 'patent', discussed above. Moreover, this conception is remarkably similar to Kiparsky's 1971 hypothesis that, in the unmarked case, rules are sequenced in a fashion that max­imizes transparency.

S and V then turn to a discussion of the truly opaque aspect of

160 MICHAEL KENSTOWICZ

Tübatulabal morphophonemics, the complex length alternations exhibited in reduplicated and nonreduplicated forms. Although not directly related to the main concerns of this paper, I will briefly review the reasoning they employ in order to arrive at the basis for the length alternations.

The first observation S and V make is that the reduplicated vowel is always long before voiced stops and affricates, and always short be­fore voiceless stops and affricates - the numbering of the data follows that in Swadesh and Voegelin's paper:

( 2) palala ebelala "arrive" ( tâwagî-na-na-la ādawagināna1a "go along causing him to see" (47) kā? in āgīn "cause him to die" ( 7) polonā-n opolonan "beat it for him" (12) togōy?a-n otogōy?an "decoy it for him" ( 9) cicwanaba icîcwanaba "accompany him" (16) kamî^a-n akamï ?an "catch it for him"

They relate this to a general distributional fact of Tlibatulabal: "in those positions where either a voiced or voiceless stop may occur, only a short vowel ever precedes a voiceless stop; either a short or long vow­el may precede a voiced stop" (S and V, p.7). General considerations about the predictability of alternations suggest that there is a rule of neutralization in the language shortening vowels before voiceless noncontinuants. This generalization can then be applied to the redupli­cation vowel: it is basically long, but shows up as short before voice­less stops and affricates.

S and V then move on to consider the length of the reduplication vowel before consonants outside of the noncontinuant class. They observe that these consonants fall into two groups: before one group the redupli­cation vowel is always long, while before the other it appears as short.

(21) halay?i-n āhalay?in "make him wet" (32) yayan lyayan "be timid" (27) wīna-gam īwīnagam "come to give him a present" (22) hūda uhūda "for the sun to be up" (33) yilahī-la iyilahöla "go along happy" (36) wimš*inī-n iwimsfinin "make him to move out of the way"

Since it has already been established that voiceless noncontinuants in­duce a shortening of the previous vowel, they suggest, by a kind of an-

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 161

alogical reasoning familiar to generative grammarians, that the voice­less noncontinuants and the vowel shortening consonants outside of the noncontinuant class share some property in common at a deeper level, on the basis of which the shortening of the preceding vowels can be pre­dicted. However, instead of distinguishing this class by a phonetic feature, as generative phonologists are wont to do, S and V merely group them together into a morphophonemic class of 'shortening conso­nants' (the nonobstruent shortening consonants being set up as a class of capital letter morphophonemes). This class is opposed to the class composed of voiced noncontinuants and the remaining nonobstruents. The latter set is called the 'neutral class', because it permits a preced­ing vowel length contrast to surface phonetically.

S and V then take the logical step of extending this morphophonemic dichotomy beyond initial position. All consonants which permit a preced­ing long vowel on the surface are assigned to the neutral class, while some of the consonants which are always preceded by a short vowel may be analyzed as belonging to the shortening class.

With both the contrast between neutral and shortening consonants, and the rule of Secondary Shortening before shortening consonants at their disposal, S and V are now in a position to tackle the vowel length pattern of the form " " ", i.e., the first, third, and fifth sylla­bles are long, while the even ones are short. Since Reduplication adds an extra syllable, the basic stem will appear in two different shapes on the surface; (1) taWògi-na-na-la, ādawbginanala "go along causing him to see" is such a stem. Words of this nature suggest that there is an al­ternating length rule operative in the language, such that the first, third, and fifth, etc., vowels are lengthened. However, there are many examples of vowels that remain long, regardless of whether they are in an odd or even numbered syllable:

( 4) tōylan ōdōylan "teach him" ( 6) puwan ubuwan "irrigate it for him" (12) togōy?an otogōy?an "decoy it for him"

This suggests that not all instances of surface vowel length are pre-

162 MICHAEL KENSTOWICZ

dictable by the Alternating Length principle. Some vowels must be ana­lyzed as being basically long; they will remain long regardless of whether they are in an odd or even numbered syllable. In addition to this, S and V observe that these basic, nonalternating long vowels pre­vent the Alternating Length rule from lengthening a preceding vowel. Thus, in (2) pәlәla, әbәlәla "arrive", the medial stem vowel is non-alternating and hence, basically long. It prevents the Alternating Length rule (AL) from lengthening the word-initial vowel in the nonre-duplicated form. But in the reduplicated form, the initial vowel of the words is not adjacent to a long vowel, and hence may be lengthened by AL. (In a form like (4) ōdōylān, cf. tōylan "teach him", the length in the reduplicated vowel has not been assigned by AL; instead here the length is a direct result of Reduplication, which makes a copy of the first stem vowel - in this case a long o.) Furthermore, there are ex­amples demonstrating that a basic long vowel inhibits AL from lengthen­ing an immediately following vowel; but these will not be discussed here.

Finally, there are cases in which we would expect a long vowel by AL but do not find it; moreover, in such cases, there is no adjacent nonalternating long vowel to block AL. (3) tewslan "fix it for him", where the initial vowel should be long by AL, is such an example. Re­call, however, that S and V have a rule of Secondary Shortening before the 'shortening consonants'. Words such as tewdlan may, therefore, be explained by setting up the w as a 'shortening consonant'. The under­lying form S and V assign to this stem is *dәwәlә.

Having seen the motivation for the Alternating Length rule and Secondary Shortening (SS), let us now consider the interaction proper­ties of these two rules. Note first that AL is highly opaque. Not only is it opaque by virtue of neutralizing the contrast between long and short vowels in certain contexts; it is also opaque by virtue of SS, which first of all, shortens long vowels lengthened by AL, and second, may shorten long vowels which inhibit AL. In a rule ordering framework, it would be necessary to order AL before SS, for both of these reasons;

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 163

i.e., to permit SS to shorten vowels lengthened by AL and to permit AL to be applied before the contrast between underlying long and short vow­els is neutralized.

But S and V do not operate with the device of ordering; rather, following Sapir, the AL rule is defined in terms of 'light' and 'heavy' vowels, which, in turn, are defined at the underlying level ('light' = morphophonemically short, 'heavy' = morphophonemically long vowels), namely that

the actual length of light vowels can be covered by simple rules: (a) in the syllable adjacent to one containing a heavy vowel, a light vowel is always short; (b) otherwise, a sequence of light vowel syllables alternates in length, the first being long, the next short, and so on (Swadesh and Voegelin 1939:8).

Thus, like Sapir, S and V deal with the opaque application of AL by identifying inputs to the latter in terms of their derivational source, which, in this case, is the morphophonemic representation, as opposed to any representation resulting from the application of a rule.

SS and AL are opaque with respect to a number of other rules al­ready discussed. For instance, the Apocope rule may delete a basic long vowel, which could be crucial to the nonapplication of AL. This happens with a stem such as hal? "to sit", which we have already seen S and V set up as *hal?'d9 the long explaining why the a is short instead of long. Apocopation of the ā, of course., renders the nonapplication of AL opaque. In a rule ordering framework, there would have to be a special statement guaranteeing that an application of AL is tried before an application of Apocope. It seems that S and V need not assume a special statement in order to produce the correct derivation, since AL has ac­cess to derivational history, in this case the underlying representation

SS is opaque with respect to Terminal Unvoicing (TU); this is shown by forms like tāwèk, ādawek "see" (from *dawbgas cf. t ā w e ә g a - t ) , for in adawdk, there is a surface long vowel in position before a voiceless stop. But this voiceless stop descends from an original voiced stop. In a rule ordering framework, we would be dealing with a nonbleeding inter­action requiring the ordering condition that SS must precede TU. No such

164 MICHAEL KENSTOWICZ

statement is to be found in S and V's analysis. Nor, I think, is one really needed, for SS is defined to operate before a morphophonemic class of consonants, i.e., the 'shortening consonants', which seem to be of the same nature as the 'light' and 'heavy' vowels. That is to say, it is a class of segments defined at the underlying level of represen­tation.

To briefly summarize, we have seen that in both Sapir's and Swadesh and Voegelin's work, there is no explicit conception of rule ordering. I have also suggested that the sequencing of rules in their derivations is not controlled by an implicit idea of rule ordering either. Rather, they seem to operate with an implicit conception of rule application having the following features: (1) rules are applied sequentially; (2) the se­quencing of the rules is governed by a principle of minimal opacity; (3) opaque rule applications are formulated in terms of derivational histo­ry.

Perhaps the most convincing evidence supporting this interpreta­tion, in particular (3), is the fact that these writers are able to de­scribe certain situations in a way that is not possible in a rule or­dering framework. One such situation appears in Swadesh and Voegelin's analysis of T'ubatili aba 1 (cf. McCawley 1969), in particular, of certain ?-0 alternations. S and V cite forms like the following, in which a stem-internal glottal stop is missing in the reduplicated form:

(47) kә?i-n āgin " t o c a u s e bira to bite" (48) su?a-n usan "to dry it for him"

S and V set these stems up as *g?i -n and *su?a-n, and formulate a rule which contracts a V?V sequence except in two-syllable unreduplicated bases, where the quality of the contracted vowel is identical with that of the second vowel in the V?V sequence. Thus, g?i-n → gin, and usu?a-n → usan; and the rule is blocked in unreduplicated bases. (The vowel length in these words will be discussed momentarily.)

However, verbs like the following do not contract their V?V se-

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 165

quences:3

(37) udi-na-n u?udinan "untie it for him" (38) agi-na-n a?aginan "cause him to open his mouth for him" (28) we?in āwā?in "pour water"

S and V choose to distinguish these verbs from those like (47) and (48) by assigning the latter underlying short root vowels, and the ones like (37), etc., basic long vowels:

(47) *g9?a (48) *su?a (37) *?üda (38) *?āga (28) *wē?ina

To handle the data presented so far, we may tentatively postulate a rule which contracts a sequence of two short vowels separated by a ? into a long vowel having the quality of the second vowel:

v, ? v 2 — v2

The contraction process can then be used to explain the alterna­tions in verbs like the following:

(31) Y wā-n eywān *ye?9WūR "hold it for him" (44) oli-n olin *?olo "help him up" (45) i?a-n i?an_ *?i?a "give him a drink" (46) owi-n o?owin *?o?owa "mark it"

These forms present a problem for a theory of phonology which attempts to deal with derivational history in terms of rule ordering, however. We have seen how the contraction process applies only to V?V sequences in which the vowels are basically short, not long. Consequently, it will be necessary to order the Contraction rule before AL, so that the contrast between basic and derived long vowels is preserved. The prob­lem here is that the contracted vowel behaves like a two-vowel sequence for the AL rule as far as its effect upon adjacent vowels is concerned. This is clear from a form like olin from underlying *o?oli-n. In order for AL to properly lengthen the , the must be preceded by a short

These data are cited in S and V's phonemic transcription. Hence, the initial glottal stops, which are present phonetically (and, in unre-duplicated bases, morphophonemically), are omitted in the phonemic transcription (cf. McCawley 1969).

166 MICHAEL KENSTOWICZ

vowel at the point where AL applies. However, the Contraction rule yields a long vowel. Consequently, Contraction must apply after AL. But if AL applies first, the contrast between basic and derived long vowels is neutralized, as the following derivations show.

?ūdan ?olin ?u?udan ?o?olin Reduplication

?o?olin AL Contraction

It is not possible to distinguish ?ū?udan from ?ō?ōlin by permitting Contraction to apply to v?V but not V?V sequences because of forms like (28) ewe?in (<*wb?ina). Contraction must not apply to the v?V sequence in 'ewb?ini but must be permitted to apply to the same sequence in ?ō?lin to derive the correct ?ōlin

The Contraction process presents another problem for a theory which attempts to treat derivational history in terms of rule ordering. This has to do with the fact that the quantity of the contracted vowel is dependent upon a prior application of AL. In all of the examples we have seen so far, the contracted vowel has always been long; hence, the tentative formulation of the rule as V1?V2 → V2. However, as S and V note, if both components of the V?v string are short, the contracted vowel is short. This is shown by a form like

(31) yәwān әywāan *yә?9WuR "hold it for him"

The underlying forms here are *yә?әwān and *әә?әān. In the unredupli-cated word, the initial vowel is long because it is lengthened by AL; hence the contracted product is a long vowel. On the other hand, in the derivation of the reduplicated form, the reduplicated vowel is word-initial and gets lengthened by AL yielding әә?әwn. This prevents the first vowel of the V?V sequence from being lengthened by AL. The second vowel of the V?v sequence is not lengthened because it is followed by an underlying long vowel. Thus neither vowel in the V?V string is length' ened, and the resulting contracted vowel is short, . This again leads to an ordering paradox, since the length of the contracted vowel is dependent upon a prior application of AL. But if AL is applied before

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 167

Contraction, the contrast between basic long vowels and derived long vowels is neutralized, preventing Contraction from applying to the prop­er V?V sequences.

We have seen that when the derivational source of a segment is cru­cial to the application or nonapplication of a rule, S and V formulate the rule in terms of derivational history. Consequently, this data pres­ents no problem to them. Their statement of the Contraction rule is as follows:

The rule of contraction is that it takes place between light vow­els separated only by ? except as between two syllables of an unre-duplicated disyllabic stem (47, 48)...The quality of the contract product is that of the second of the two component vowels, as is seen in usan (<*usu?ana), and the quantity is long if one of the components is long, short if both are short (Swadesh and Voegelin 1939:9).

Notice first that their rule identifies the V?V sequences eligible for Contraction by the term 'light' vowel, which, as we have seen, denotes a morphophonemically short vowel. Second, note that this rule predicts the resulting quantity of the contracted vowel in terms of 'long' versus 'short', which is the distinction introduced by the Alternating Length rule. This, then, is a classic example of what has come to be called a 'global rule', i.e., a rule which simultaneously refers to two points in a derivation - in this case the underlying representation and the re­presentation resulting from an application of AL.

4.0. I believe it is fair to say that the conception of rule ap­plication involving the features of sequential application and refer­ence to derivational history is by no means limited to Sapir and his immediate school. Rather, it is characteristic of the work of most lin­guists who have looked at phonology in terms of some underlying repre­sentation linked to a surface representation by a set of rules. I shall close this paper with a few examples from diverse sources which support this claim.

In his sketch of Yuma, Hal pern (1946) discusses a rule which in­serts 'inorganic1 ^ in a variety of contexts. One such context is:

168 MICHAEL KENSTOWICZ

preceding a final consonant and preceded by a consonant which can­not combine with the final consonant to form a final consonant clus­ter: a?av9k (<*a?avk) "he hears", -1 (<*-1) "all of them"; ...(Halpern 1946:253).

In another section dealing with Syncope Halpern notes: A number of suffixes composed of consonant + vowel occur as such when in the final position in the word but have the vocalic ele­ment syncopated when followed by another suffix: ?a»véva "this snake (absolutive)", >?a vévac (<*?aevévac) "this snake (nomina­tive)" (p.256).

Notice that there is no mention of the Epenthesis rule in the discus­sion of the Syncope process. Clearly, Halpern is assuming a sequential rule application whereby underlying ?a-veva-c is syncopated to inter­mediate îa-vëvC) which then has its final cluster broken to yield sur­face ?a-vevbc.

Halpern employs derivational history when dealing with opaque rule applications. In a section entitled "contact between vowels" he formu­lates a rule by which

a is always absorbed by a preceding or following long vowel: n.?1« (<*n*?1(-) "when he says", vi*vák (<*vi*avák) "he is here", wu'kavárek (<*wau'kavárk) "he causes him to like it" (Halpern 1946:255).

In this same section he formulates a rule whereby i* and a* change following w to wa: ?^* (<*? * n*') "her agedness", ?i'wamácәny (<*?i'u'mácny) "my dancing".

He then states: The vocalic element of wa<*fu* is not absorbed by a following vowel: a'waé'mnya (<'*u'') "his pushing away of it", a-waí- (<*a*u* í mn) "his recklessness" (ibid?. ) •

In ather words, the *u —> wa rule gives potential inputs to the a-Ab­sorption process, but they do not undergo it. The imposition of a 'non-feeding' order would be required in a rule ordering framework. Halpern does not employ rule ordering statements or conditions. Instead, the failure of a's from *w to be absorbed is treated as a (global) con­dition on the Absorption process.

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 169

The description of opaque rule applications in these terms (i.e., by what would now be called 'global conditions') is actually quite prevalent in earlier process morphophonemics. To cite just one more example of this type, in Mary and George Foster's 1947 description of Sierra Popoluca, nontransparent rule applications are consistently de­scribed in this fashion. For example, in this lannuaqe there is a gen­eral rule which palatalizes the alveol ars t, c, s, and n to the palato-alveolars f3 c3 3 and ft when (1) preceded or followed by i; (2) pre­ceded by i?; (3) preceded by . This rule critically interacts with an­other whereby a glottal stop metathesizes with a preceding sonorant. Under a section entitled "Limitations to the palatalization rule" the authors state:

When n would be in syllable final position, but because of meta­thesis becomes initial in the following syllable preceding i, it does not palatalize. ?a?ni spa I see (him, her, it) (?an-t.p.pref. + *?i*sT see + -pa inc. suf.) wA?â*pta?nî6 we ave able to live (*WA?a* be able + p inc. suf. + tan- t.p. pref. + '*?it b, live) (M. and G. Foster 1947:11).

In other words, the Palatalization rule is sensitive to the deriva­tional source of n + i sequences. If the n comes to stand before i as a result of Metathesis, it fails to undergo Palatalization. The Palatal­ization rule is also sensitive to the derivational source of the i vow­el:

When the perfective suffix -ne? is followed by any suffix requiring a shift of accent away from the -ne?, the e varies freely with i. Furthermore, neither the n nor a consonant following the i pala­talize following the usual palatalization rules. ?a?ná?mnitá?m wie have looked at it (?an- t.p. pref. + *?a?m look + ne? + ta?m pi. suf.) (op.cit., p. 17).

In a rule ordering framework these limitations of Palatalization would be described by ordering Palatalization before Metathesis and the rais­ing of e to . The Fosters do not employ rule ordering for this pur­pose. Instead, they describe the failure of Palatalization in these contexts in terms of derivational history.

170 MICHAEL KENSTOWICZ

The treatment of opaque rule applications in terms of derivational history is, of course, not limited to descriptions written in the United States. For instance, consider the traditional notion of 'compensatory lengthening'. As this concept is usually interpreted, some segment, typ­ically a consonant, drops out. As a consequence, the syllable is length­ened to compensate for the loss, customarily by a lengthening of the preceding vowel. If this notion is transferred into a process descrip­tion, it implies a sequencing of rules in which first the consonant is lost, and then the syllable is lengthened. But the lengthening process must be sensitive to derivational history, for it would be only sylla­bles that have lost their consonant which would lengthen (see Kisse-berth 1973b, for a more extensive discussion of the import of compensa­tory lengthening).

It is obvious that the spirit of compensatory lengthening cannot be expressed by a theory which attemps to describe derivational history in terms of rule ordering. Another quite parallel example is to be found in the following quote from Col linder's Survey of the Uralie Lan­guages (1957:146):4

When a single consonant between two short vowels disappears in the weak grade and consequently two vowels come to appear in immediate juxtaposition, the narrow vowels i3 u, il are changed into the wider vowels e, o, and ö, e.g. siga 'pig'*- gsg sea; tegu 'deed' ~ teo3 silsi (stem süde-) 'charcoal' ~ söe; 'lugu 'story' ~ loo (the weak forms pronounced monosyllabically, with diphthong or long vowel respectively).

In the above discussion of Estonian, Col linder quite clearly assumes that there is a general rule of Consonant Gradation accounting for the deletion of the consonants in the examples cited, and a separate process shifting narrow vowels into wide vowels, and that this latter process is restricted to vowel clusters resulting from the loss of a consonant through Gradation. That this restriction is required can be illustrated by examples like the following (taken from Harms 1962):

4 Harms 1962 describes this data in terms quite parallel to Collinder.

RULE-APPLICATION IN PRE-GENERATIVE AMERICAN PHONOLOGY 171

nom.sg, gen.sg.

idu eo "sprout" kiu kiu "fiber" magu mao "stomach, taste" sau saua "stick" lagi lae "law" lai laia "wide"

My final example of a global rule is to be found in the descrip­tion of Modern Uigur by the Soviet linguist E. N. Nadzhip (1971). In addition to the normal Turkic process of front-back vowel harmony, Uigur has two additional morphophonemic rules that interact with one another in an interesting fashion in Nadzhip's description. One of these is a reduction process which raises the unaccented low vowels a and ä to when they are not in the initial syllable of the word.

dada "father" dadi-sì "his father" dadå-m "my father" dadi-lár "fathers"

dadi-lir-im "my father and all of his family"

Uigur has another process which Nadzhip calls 'regressive assimilation', describing it as follows:

When an affix with the broad front vowel ä or the narrow non-labi-alised i is joined to monosyllabic words [i.e., stems] with the broad vowel a/ä (...), the accent is shifted to the affix, and the broad vowel of the stem changes into e (Nadzhip 1971:53).

This rule operates in a variety of contexts: bar- "to go ber-iš "going" bar- "to give" ber-is "giving" al- "to take" el-in "to be taken" mal- "cattle" mel-i "his cattle" kät- "to leave" ket-är "he will leave" bar- "to give" ber-är "he will give"

Since e only appears in the initial syllable of a word, the Regressive Assimilation rule can be generalized to operate before any front vowel.

For our purposes, the most interesting feature of Nadzhip's de­scription is his explanation for why Regressive Assimilation is limited to monosyllabic stems. He scates:

172 MICHAEL KENSTOWICZ

There is no regressive assimilation ... when the broad vowel is in the second syllable and the assimilating narrow vowel is in the third syllable, because here the phenomenon of reduction is opera­tive (...), e.g. dada 'father', dadisi 'his father' and not" dadesi (Nadzhip 1971:55; emphasis mine).

In other words, Regressive Assimilation is limited to monosyllabic stems because in medial syllables, the low vowels have already been raised to , i.e., Reduction takes precedence over Assimilation.

Nadzhip then goes on to state: The assimilation does not occur when the narrow vowel which could function as assimilating is itself the result of reduction, i.e. the transition a/ä > i:i has already taken place, e.g. ana 'moth­er', where the narrow vowel of the second syllable is the result of the reduction i < a and therefore cannot assimilate the broad vowel of the first syllable (loc.oit.).

Thus, Nadzhip assumes that Reduction applies before Assimilation, 'bleeding1 the latter rule. But this sequencing of the rules merges the distinction between basic and derived , which is crucially needed in order to determine when Regressive Assimilation applies (compare dadi­si versus mel-i < *mal-i). Nadzhip (loc.cit.), therefore, places a 'global' condition on the Assimilation rule: it does not apply when the i "is itself the result of reduction".

5.0. In conclusion, this brief study of rule interaction in pre­generative phonology suggests several points that are relevant for cur­rent phonological theory. The most obvious, of course, is that rule or­dering is not the only device that can be employed to ensure proper rule interaction. Indeed, since Bloomfield's "Menomini Morphophonemics" is the only major pregenerative description to make use of rule orderinq, one can conclude that, at least on pretheoretical grounds, it is not even the most natural or straightforward approach to the problem. Fi­nally, the recent attention devoted to rule ordering is but another ex­ample of the current critical reevaluation of another aspect of genera­tive grammar, suggesting that an adequate theory of language in general, and phonology in particular, is not nearly as different from the implic­it theory of many earlier linguists as it might have seemed five years ago.

REFERENCES

Bever, Thomas G. 1967. Leonard Bloomfield and the Phonology of the Menomini Language. Ph.D. diss., Cambridge, Mass.: MIT.

Bloomfield, Leonard. 1917. Tagalog Texts with Grammatical Analysis. Urbana, 111.: Univ. of Illinois Press.

. 1925. "Notes on the Fox Language". International Journal of American Linguistics 3.219-32.

. 1939. "Menomiti! Morphophonemics". Travaux du Cercle lin­guistique de Prague 8.105-15. (Repr. in A Leonard Bloomfield An­thology ed. by Charles F. Hockett, 351-62. Bloomington & London: Indiana Univ. Press, 1970.)

Collinder, Björn. 1957. Survey of the Uralic Languages. Stockholm: Almqvist & Wiksell.

Foster, Mary and George. 1947. Sierra. Popoluca Speech. Washington, D.C.: Smithonian Institution; Inst, of Social Anthropology.

Halpern, A. M. 1946. "Yuma". Linguistic Structures of Native America ed. by Harry Hoijer, et al., 249-88. New York: Viking Fund Pubs. in Anthropology.

Harms, Robert T. 1962. Estonian Grammar. Bloomington, Ind.: Research Center for the Language Sciences.

Kiparsky, Paul. 1968. "Linguistic Universals and Language Change". Universals in Linguistic Theory ed. by Emmon Bach and Robert T. Harms, 171-202. New York: Holt, Rinehart & Winston.

. 1971. "Historical Linguistics". A Survey of Linguistic Science ed. by William Orr Dingwall, 577-642, 643-49 (Discussion). College Park, Md.: Univ. of Maryland, Linguistics Program.

Kisseberth, Charles W. 1973a. "Is Rule Ordering Necessary in Phonolo­gy?" Issues in Linguistics : Papers in honor of Henry and Renée Ka-hane ed. by Braj B. Kachru, Robert B. Lees, et al., 418-41. Urba­na, 111.: Univ. of Illinois Press.

. 1973b. "On the Alternation of Vowel Length in Klamath: A global rule". Issues in Phonological Theory ed. by Michael J. Ken-stowicz and Charles W. Kisseberth, 9-26. The Hague: Mouton.

Koutsoudas, Andreas, Gerald A. Sanders, and Craig Noll. 1971. On the

174

Application of Phonological Rules. Bloomington, Ind: Indiana Univ. Linguistics Club, mimeo.

McCawley, James D. 1969» "Length and Voicing in Tiibatulabal". Papers from the Fifth Regional Meeting of the Chicago Linguistic Society, 407-15. Chicago: Univ. of Chicago; Dept. of Linguistics.

Nadzhip, E. N. (alias Nagip, Amir). 1971. Modern Uigur. Transi, into English by D. M. Segal. Moscow: Izd. "Nauka".

Postal, Paul M. 1968. Aspects of Phonological Theory. New York & Lon­don: Harper & Row.

Sapir, Edward. 1922. "The Takelma Language of South Western Oregon". Handbook of American Indian Languages ed. by Franz Boas, vol.2. (= Bureau of American Ethnology; Bulletin 40:2.) Washington, D.C.: Government Printing Office.

. 1930. "The Southern Paiute Language, I: A Shoshonean Lan­guage". Proceedings of the American Academy of Arts and Sciences 65:1.1-296.

Swadesh, Morris, and Charles F. Voegelin. 1939."A Problem in Phonolog­ical Alternation". Language 15.1-10.

PROLEGOMENA TO 'PROLEGOMENA TO A THEORY OF WORD FORMATION' A REPLY TO MORRIS HALLE

LEONHARD LIPKA

1.1. Research within the TG Paradigm has contributed a great deal to a new awareness of problems of methodology in linguistics. Such a statement will be disputed by nobody. It seems, however, that certain methods of traditional scholarship have been largely abandoned by many researchers who work in the TG framework, and have been replaced by oth­er less commendable procedures. It used to be an accepted principle that reading should come before writing and publishing, i.e.,, one had to make sure before claiming to have discovered something, whether oth­ers, working in the same field, had not already come to the same or similar conclusions. Admittedly, this is much more difficult today than it was fifty years ago. But it seems to me that it is even more impor­tant now, at least to attempt to follow this principle, precisely be­cause this is the only way to improve the quality of the terrifying flood of published and semi-published literature in linguistics, and at the same time reduce its quantity.

1.2. The purpose of publication is to prevent duplication of re­search and effort, but also to allow for equally public criticism which ideally should advance the progress of scholarship and increase general knowledge. To achieve this effect criticism does not necessarily have to be sympathetic, but may also cast doubt on very fundamental assump­tions. A case in point is the so-called Chomskyan revolution itself. Strangely enough, however, once a revolution has been successful, the revolutionaries turned establishment themselves rarely accept basic

176 LEONHARD LIPKA

criticism but only admit 'constructive' comments. The following remarks may serve as an illustration. An earlier and shorter version was sub­mitted for publication in Linguistic Inquiry immediately after Halle's (1973) article had appeared in the same journal. Publication was de­clined - as an anonymous referee put it - because: "This appears to be inappropriate as a squib since it is indirectly an attack on the sig­nificance of Halle's work. The tone is not at all constructive."

2.1. After excluding and then reintroducing semantics in its ear­lier stage of development, TG theory later neglected the morphological component (cf. Kastovsky 1971:3), and now seems to be in the process of rediscovering another aspect of language: word formation. Halle's (1973) article Prolegomena to a Theory of Word Formation is symptomatic of this phase. He believes that this field "has been studied only to a very lim­ited extent" and hopes "to attract others into research on this topic" (p.3). One wonders if this invitation is addressed to certain researchers who have already accomplished a considerable amount of basic work in the field. Amongst those names which immediately spring to mind in this con­text one might mention several, beginning with Botha, Brekle, Coseriu, including Dokulil, Erben, Fleischer, Gauger, Gruber, Hansen, Hatcher, Henzen, Kastovsky, Koziol, Lees, Ljung, Malkiel, Marchand, Morciniec, Motsch, Neuhaus, Rohrer, Stein, and finishing with Weinreich and Zimmer. This research has been openly published in book-form or journals and is not confined to mimeographed papers which are only available within a closed circle. The above list can easily be augmented from the exten­sive bibliography in Marchand (1969) and from Stein (1973). Halle only mentions Chapin, Jespersen, and an unpublished paper by Siegel. Appar­ently, he completely ignores the fact that Marchand (amongst others) has developed a comprehensive theory of word-formation, and has applied this theory to a full-scale description of English word-formation. The first edition of this standard work appeared in 1960 and was reviewed in a number of journals. Cf. Brekle-Lipka (1968), Marchand (1969), Lip-ka (1971), Pennanen (1972), Kastovsky (1974).

PROLEGOMENA TO 'PROLEGOMENA TO A THEORY OF WORD FORMATION 177

2.2.1, Halle starts off with the claim that speakers of English know that adjectives such as transformational are "composed of the mor­phemes" trans - form - at - ion - al and that "facts like those" have to be formally represented in a theory of word-formation. The proposed segmentation is by no means a 'fact' but must be based - either implic­itly or explicitly - on a theory, as is the case with any analytical procedure in linguistics, of which segmentation of utterances or words into morphemes (morphs) is one of the most important instances charac­terizing a whole era of linguistics, viz. structuralism. For example anyone only slightly familiar with the methods of structural descriptive linguistics would probably question treating - at - in transformational (or - i - in serendipity which Halle discusses later) as a morpheme oran allomorph1 but would prefer a segment -ation as a linguistic sign. It is true, though, that "structuralism" was not a monolithic block, and that various "structuralists" held different views at different times. This is a point that is often forgotten when "transformationalists" talk about 'taxonomic structuralism'. Of course, it also applies to unspeci­fied general statements about 'TG grammar'. Any improvement on standard work and the great mass of informed opinion is certainly to be welcomed. However, one might expect such developments to be justified against other work in the field. For a sketch of my views on the 'morpheme1 (cf. 3.2.2. ). Halle further suggests that, for example, the entry for write must contain the information that it belongs to the 'non-Latinate' part of the vocabulary. This observation is handled on a higher level of generalization by Marchand's distinction between word-formation on a native and on a foreign basis.

2.2.2. The "idiosyncratic characteristics of individual words" are discussed at length by Halle. This topic is the subject of a whole book (Botha, 1968) on the function of the lexicon in a transformational-

Of course this is not to be confused with -ate as in consultate, pas­sionate, acetate, hyphenate; cf. Marchand (1969:254-59). For -ation see Marchand (1969:259-61).

178 LEONHARD LIPKA

generative grammar. Starting from Chomsky's hypothesis about the lexicon as "the full set of irregularities of the language". Botha treats nomi­nal compounds in Afrikaans in great detail and postulates a phonologi­cal dictionary and a phonological matching rule. The theoretical model proposed by Botha is strongly influenced by Weinreich's thought (cf. Botha 1968:245; Weinreich 1966:445; 1969:59, 74). Halle (4f.) distin­guishes three types of idiosyncrasy in word-formation: a) semantic, b) phonological, and c) restrictions of productivity, and suggests account­ing for them with 'a special filter' through which words have to pass after being generated by word-formation rules. This solution exactly corresponds2 to the postulation of an 'idiom comparison rule' (later 'matching rule') for a) in Weinreich's 1969 model and the 'phonologi­cal matching rule' for b) in Botha (1968). The restrictions under c) -or more precisely all three types of restrictions on rules: semantic, phonological, productivity - can be accounted for in another theoreti­cal framework by Coseriu's concept of 'norm' (cf. Marchand 1969:17,57; Stein 1971; Neuhaus 1971). Although Weinreich does not claim that his theory, published in 1969 but developed and proposed earlier (lectures delivered during the 1966 Linguistic Institute at UCLA), solves all the problems of word derivation, his concepts of a 'simplex dictionary', a 'complex dictionary', an 'idiom list', 'familiarity ratings', and a 'matching rule' seem to be extremely useful and important. They are consistently applied in Lipka (1972; esp.84ff., 128ff.).

2.2.3. Discussing the distinction between "derivational morpho­logy" Halle (6) states: "I know of no reasons why the list of morphemes should not include also the inflectional affixes". At least two reasons might have been found in Motscn (1962): the place of inflectional mor-

2 Cf. Weinreich (1969:74): "The role of the filtering device is to differentiate, among possible words, those that are established from those that are not" [my emphasis, LL]. Cf. the notions 'possible lexical items' and 'gap in the lexicon', both used currently in Gen­erative Semantics.

PROLEGOMENA TO 'PROLEGOMENA TO A THEORY OF WORD FORMATION' 179

phemes in the constituent structure of complex lexical items, and the different degrees of combination potential of lexical and grammatical morphemes. Inflectional morphemes in English and German are usually placed at the end of words, after all derivative morphemes have been added. Combination with the former is much less restricted than with derivational suffixes. Motscn (1962:39) also sets up rules exactly like the "word formation rules" suggested in Halle (16). The relationship between inflexion and word-formation is treated in great detail within the framework of Chomsky-Halle's Sourd Pattern of English in Wurzel (1970:15-104). Halle mentions that word-formation rules will have to include information on selection restrictions. He seems hardly aware of the difficulties of establishing the correct selection restrictions even for very simple everyday words, or of the problem whether 'selec­tion restriction' as such is a justifiable concept in linguistics. See the review of various linguistic judgments on the selection restrictions of eat in Lipka (1972:48-51). The possibility of treating such restric­tions with the notion of 'presupposition' is not mentioned by Halle.

3.1. A theory of word-formation must include an explanation of the fact that complex lexical items differ semantically from the sum of their components. This could be done with the concept of 'lexicalization' which entails the addition of semantic features. Such an approach is sketched in Lipka (1971). The term is not used here in the way it is used now within the framework of Generative Semantics, i.e. for the insertion of lexical items, or the surface realization of a configura­tion of atomic predicates. It is rather meant to indicate that complex lexical items, once they are created from smaller elements and used re­peatedly, can become lexemes in their own right, with a loss of motiva­tion (and perhaps also analysability), and acquire certain specific se­mantic features. Lexicalization is tied up very closely with 'hypo-statization', but the latter process also affects simple lexical items. The lexical item lexicalization itself may serve as an example. As I use it here, I follow the tradition established in Marchand's Catego-

180 LEONHARD LIPKA

ries in 1960. Both this meaning of lexicalization and the one found in Generative Semantics can be said to go back to an underlying sentence 'Something becomes (a) lexical (item)' or probably better from its caus­ative derivative 'Someone causes something to become (a) lexical (item)'. However, in Generative Semantics, the underlying pro-form something re­fers to prelexical elements, or atomic predicates, while in Marchand's and my own one it refers to the morphemes as elements of surface struc­ture which make up a new lexical item that becomes a semantic unit. 'Surface structure' is not used here in the specific technical sense as defined in some transformational-generative model, but referring to any­thing directly observable as opposed to a more abstract 'underlying structure'.

3.2.1. It is no secret that the process of lexical insertion is a mystery far from being solved in the framework of Interpretative or Gen­erative Semantics. Since McCawley's article "Lexical Insertion in a Transformational Grammar without Deep Structure" (1968) -which despite its title does not clarify but only raises the issue -relatively little progress has been made. I suggest that the concept of lexical insertion should be supplemented or replaced by the notion of MORPHEMIC INSERTION. For various reasons it is impossible for me to describe here my views on this problem, or to develop an alternative theory of word-formation. A few hints have been given above. As a sketch, I can add that I largely agree with the conclusions drawn in Kastovsky (1973), and therefore - as in Lipka (1972) - embrace many of the assump­tions of Generative Semantics. If, however, as Kastovsky and I believe, prelexical semantic elements such as CAUSE DO BECOME NEG MILITARY are converted into complex lexical items such as demilitarize , and the pre­lexical element (or atomic predicate) "MILITARY is replaced by the ad­jective military, the feature BECOME NEG by the prefix de-, which is attached to military, and the features CAUSE DO by the suffix -ize" (Kastovsky 1973:290), then it must be morphemes that are inserted, not

PROLEGOMENA TO 'PROLEGOMENA TO A THEORY OF WORD FORMATION' 181

lexical items.3 This, of course, means a return to surface structure, although, not at the expense of neglecting underlying structure (cf. Kastovsky 1971:8f.). As opposed to Chomsky and Halle, one need not re­discover surface structure if one has never given it up.

3.2.2. At this point I should like to sketch briefly my views on the 'morpheme'. I believe that morphemes are the smallest linguistic signs, i.e., meaningful observable segments in which elements of con­tent (e.g., semantic features) are related in an arbitrary way to ele­ments of expression. As opposed to some varieties of structuralism I do not require allomorphs, i.e., phonologically or morphologically conditioned variants of a morpheme, to have identical or even similar phonic shape. Thus, /iz, z, s, 3n/, and Ø are all considered allomorphs of the same plural morpheme in English (cf. Lipka, 1969). In my view 'morphemes' are, therefore, essentially semantic units. This also be­comes evident from my adoption of the concept of 'zero' in linguistics, since 'zero-allomorphs' and 'zero-morphemes' have no phonic expression at all (cf. Kastovsky 1968, esp.31-53). Following Weinreich (1966:432f.), I believe it is useful and descriptively adequate to distinguish between 'major' and 'minor classes of morphemes', which roughly corresponds to the more traditional distinction between 'lexical' and 'grammatical' morphemes. I disagree with Weinreich (1966:433) on the nature of cate­gorial features such as [+Noun, +Adjective] which he believes to be "semantic in the full sense of the word". Both classes of morphemes then, in my view, can be represented as a triplet of features, which could be termed 'phonological', 'categorial' (also including syntactic information), and 'semantic' features. I am fully aware of the fact that this is not sufficient for a complete specification of lexical entries for morphemes in some type of dictionary or lexicon.

Kastovsky1s particular analysis in which de- replaces BECOME NEG may be questioned if one believes that in the inchoatives black/en, redd/ en, Warm/0 (which are homonymous with the corresponding causâtives) the suffix -en and the zero-morpheme represent BECOME.

182 LEONHARD LIPKA

3.3. The distinction between the transformationalist and the lexicalist hypothesis (not to mention Chomsky's conversion from the former to the latter position) is not mentioned once in Halle's article. This is all the more 'surprising, since the reasons why Chomsky adopted the lexicalist position for "derived nominals" (which are never explic­itly defined) in 1968 (first in print as Chomsky, 1970) are exactly the same as those which led Halle to put forward in his Prolegomena: seman­tic and syntactic idiosyncrasy and restrictions on productivity. Chomsky's article had circulated in mimeographed form as Chomsky (1968) but is labelled Chomsky (1972) in Halle's Prolegomena, thus inducing the naive reader to believe this to be a recent paper. While stating that word-formation processes "are typically sporadic and only quasi-productive" (Chomsky 1965:184f.), Chomsky in Aspects still derives r e fusa l , destruc­tion from the respective verbs by a nominalization transformation, be­cause the process is said to be productive. This is a solution which is truly within the generative-transformational spirit, as it accounts both for creativity in language and irregularity in the superficial surface structure. It shows the greatest possible generalization, and, at the same time, assigns secondary importance to surface phenomena. But even for "quasi-productive processes" such as the formation of horrify, ter­rify j telegram, phonograph Chomsky in Aspects arrives at the conclusion: "it is clear that from the point of view of both the semantic and the phonological interpretation it is important to have INTERNAL STRUCTURE [my emphasis, LL] represented in these words" (186). In Remarks on Nom­inali zation, however, Chomsky abandons his earlier approach to "derived nominals". Halle neither mentions this change of position nor the prob­lems for the theory involved.

4.0. An explanation of the phenomena mentioned in 3.1. and 3.2.1. is never seriously attempted in Halle's article. The ambiguity of lexi-c a l i z a t i o n , or rather, the derivation of the two different, but closely related, lexical items by the same general derivative process could never be explained by anything resulting from Halle's Prolegomena.

PROLEGOMENA TO 'PROLEGOMENA TO A THEORY OF WORD FORMATION' 183

Certain extremely productive word-formation processes are not even touched upon in his paper, such as compounding, prefixation, and zero-derivation (cf. Marchand 1969:11-127, 129-208, 359-89; Kastovsky 1968).

5. Two questions must be raised with regard to Halle's article. Firstly, did he take into consideration the large amount of basic re­search which had previously been done on the subject of word-formation? Secondly, has Halle brought up any problems which have not already been treated, or proposed any solution for such problems which have not been offered elsewhere? It seems that the answer to both these questions is no, and for this reason Halle's remarks cannot be regarded as "Pro­legomena to a Theory of Word Formation".

REFERENCES

Adams, Valerie. 1973. An Introduction to Modern English Word-Formation. (= English Language Ser ies , 7.) London: Longman.

Botha, Rudolf P. 1968. The Function of the Lexicon in Transformational Generative Grammar. The Hague: Mouton.

Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.

. 1970. "Remarks on Nominalization". Readings in English Transformational Grammar ed. by Roderick A. Jacobs and Peter S. Ro­senbaum, 184-221. Waltham, Mass.: Ginn & Co. (Repr. in Studies on Semantics in Generative Grammar by Noam Chomsky, 11-61. The Hague: Mouton, 1972.)

Halle. Morris. 1973. "Prolegomena to a Theory of Word Formation". Lin­guistic inquiry 4.3-16.

Kastovsky, Dieter. 1968. Old English Déverbal Substantives Derived by Means of a Zero Morpheme. Dissertation, Univ. of Tübingen. (Print­ed. , Esslingen/Neckar: B. Langer, 1968.)

. 1971. Studies in Morphology : Aspects of English and German Verb Inflection. (= Tübinger Beiträge zur L ingu i s t i k , 18.) Tübingen: G. Narr.

. 1973. "Causatives". Foundations of Language 10.255-315.

184 LEONHARD LIPKA

Kastovsky, Dieter, ed. 1974. Studies in Syntax and Nord-Formation: Se­lected articles of Hans Marchand, Munich: W. Fink.

Lipka, Leonhard. 1969. "Assimilation and Dissimulation as Regulating Factors in English Morphology". Zeitschrift für Anglistik und Amer­ikanistik 17.159-73.

. 1971. "Grammatical Categories, Lexical Items, and Word-Formation". Foundations of Language 7.211-38.

. 1972. Semantic Structure and Word-Formation: Verb-particle constructions in Contemporary English. Munich: W. Fink.

Marchand, Hans. 1969. The Categories and Types of Present-day English Word-Formation: A sychronic-diachronic approach. 2nd rev. & enl. ed. Munich: C. H. Beck. (First ed., 1960.)

McCawley, James D. 1968. "Lexical Insertion in a Transformational Gram­mar without Deep Structure". Papers from the Fourth Regional Meeting of the Chicago Linguistic Society ed. by Bill J. Darden, Charles-James N. Bailey, and Alice Davison, 71-80. Chicago: Univ. of Chica­go; Dept. of Linguistics.

Motsch, Wolfgang. 1962. "Zur Stellung der 'Wortbildung' in einem forma­len Sprachmodell". Studia Grammatica 1.31-50.

Neuhaus, H. Joachim. 1971. Beschränkungen in der Grammatik der Wortab­leitungen im Englischen. Diss., Univ. of Saarbrücken.

Pennanen, Esko. 1972. "Current Views on Word-Formation". Neuphilologi­sche Mitteilungen 73.292-308.

Stein, Gabriele. 1971. 'Primäre und sekundäre Adjektive im Französischen und Englischen. Diss., Univ. of Tübingen. (Printed, Tübingen: Narr, 1971.)

. 1973. English Word-Formation over two Centuries: In honour of Hans Marchand on the occasion of his sixty-fifth birthday. (= Tü­binger Beiträge zur Linguistik^ 34.) Tübingen: G. Narr.

Weinreich, Uriel. 1966. "Explorations in Semantic Theory". Current Trends in Linguistics ed. by Thomas A. Sebeok, vol.3.395-477. The Hague: Mouton. (Sep. ed., with a preface by William Labov, 1972.)

. 1969. "Problems in the Analysis of Idioms". Substance and Structure of Language ed. by Jaan Puhvel, 23-81. Berkeley & Los An­geles: Univ. of California, Press.

Wurzel, Wolfgang Ullrich. 1970. Studien zur deutschen Laut struktur. (= Studia grammatica y 8.) Berlin: Akad.-Verlag.

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION

ROYAL SKOUSEN

Phonetic evolution first obscures analysis, then makes it completely impossible.

Ferdinand de Saussure (1959:155) 1.0. In this paper I will consider the important lin­

guistic problem of accounting for morphophonemic alternation.* The theory of generative phonology has traditionally accoun­ted for morphophonemic alternation by postulating phoneti­cally-statable rules of the form A → / X Y where A and are natural classes of sounds, and X and Y represent strings of segments (Chomsky and Halle 1968:332-40). Genera­tive phonological rules map natural classes of sounds into other natural classes in phonetically-statable environments, which, in some universal phonetic sense, are supposed to ex­plain the mapping that occurs.1 The theory of generative phonology assumes that speakers account for morphophonemic alternation by positing rules that are similar in form and function to low-level phonological rules (Chomsky 1964:65 to 110; Postal 1968). For example, Chomsky (1964 : 71ff.) claims that the rule accounting for the k~s alternation in opaque/opacity is no different phonologically than the alter­nation of t~D in. write/writer or the alternation of vowel length in hit/bid. The k~s alternation is handled by a

A shorter version of this paper was read at a linguistics conference on "The Expanding Domain of Linguistics" held at the University of Texas at Austin, 26 March 1973.

186 ROYAL SKOUSEN phonological rule that changes to s when followed by a high palatal vowel or glide in a following suffix;

The mapping of to s is phonetically-conditioned by the high palatal segment that follows the k. This mapping, it is claimed, is no different from the rules that flap t in-tervocalically or lengthen a vowel when followed by a voiced obstruent. Rules like → s, however, do have surface ex­ceptions. In order to handle such exceptions, generative phonology has developed certain theoretical devices, such as extrinsic rule ordering, exception features, category re­strictions, lexical stratification, and various ad-hoc bound­ary symbols (Chomsky and Halle 1968; chap.8). More recently, some generative phonologists have argued that global rules should be added to the theoretical machinery (Kisseberth 1970; Kenstowicz and Kisseberth 1970; McCawley 1971; Kisse­berth 1972).

In this paper I will discuss the weakest form of this theory, one that would at least capture phonetically-statable regularities that have no surface exceptions. Such situations frequently occur in the historical development of languages. In numerous instances, an historical phonetic change may take place, the result of which is an exceptionless, phone­tically-statable surface regularity in the data. Now the im­portant question is whether new speakers, in learning the language, will account for this exceptionless phonetically-statable regularity by postulating a synchronic rule that reflects, in a direct way, the form of the earlier historical change. In the traditional practice of generative phonology, such a regularity would be captured with no second thought, especially since there would be no surface exceptions. Such

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 187 a theory therefore claims that the language acquisition de­vice, in accounting for morphophonemic alternation, favors phonetically-statable morphophonemic rules. In this paper I will examine this assumption of generative phonology and show that it does not make the correct predictions about his­torical development. Empirical evidence will show, to the contrary, that the overwhelming tendency is for speakers to account for morphophonemic alternation by means of morpho-phonemic-defined surface rules. I shall refer to such rules simply as morphological rules. Tliese morphological rules are not at all phonetically-plausible, nor are they synchroni-cally explainable in phonetic terms, although historically they may have come from phonetically-motivated historical changes. Rather, morphological rules relate well-defined morphological forms to each other, in a direct and positive way.

1.1. A rather clear example of a morphological rule is the rule that accounts for the i~œ ~ alternation found in English verbs like sing, r ing , sink, and drink. The rule for this alternation is clearly not a phonetically-statable rule, since there is no difference in the environment that might explain, in some phonetic sense, the i~œ ~ alternation. His­torically, of course, this alternation is due to ablaut, but synchronically there is no evidence for a phonetically-con­ditioned environment that will motivate the vowel alterna­tion.

A morphological rule that will account for these vowel alternations is as follows: If a one-syllable verb contains an i vowel followed by a sequence of [non-labial nasal] [velar stop], then in the past-tense stem an œ vowel occurs and in the past participle an vowel occurs (for empirical

188 ROYAL SKOUSEN

evidence for t h i s morpho log ica l r u l e , see Skousen 1972:15 to 17) ;

i one-sy l lab le verb stem

non- lab ia l nasal

velar stop æ

past-tense stem

A

' past participle

Note that this morphological rule is restricted by certain phonetically-defined conditions, such as the class of velar stops (preceded by n) and the syllable. In other words, mor­phological rules are defined, in part, by phonetically-sta­table conditions. But the phonemic alternation itself is not phonetically-motivated in any way; it is, in fact, phonemi-cally arbitrary. The alternation could just as well be o~u~a, for instance. There is no phonetic explanation for the map­ping of i to œ , or of i to . Moreover, the alternation it­self is specifically defined so that it occurs only in cer­tain morphological forms. The i~œ ~ alternation occurs on­ly in the verb system and is used to indicate various tenses. This alternation therefore deals directly with the meaning system of the language. Although the alternation is, in part, phonetically-conditioned, it is not phonetically-motivated.

This morphological rule is directional in the sense that the formation of the past tense and the past participle is based on the present-tense form of the verb. This morpho­logical rule is also reversible, in that given the past-tense

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 189

form of the past participle, the speaker can predict the pre­sent-tense form. In other words, the basic lexical form of the verb is recoverable (cf. Ohala 1973 for other properties of morphological rules).

The directionality and reversibility of morphological rules does not, however, guarantee absolute predictability. For example, a verb like bring can take on the i~æ ~ A alter­nation. Forms like hvang and are used by many speakers, especially younger ones, and are readily understood by all speakers. But this fact does not mean that bring must neces­sarily take this alternation. Some younger speakers, for in­stance, may use a past-tense form like bringed for this verb, and English speakers can understand this past-tense form just as readily as brang or brun g. In other words, the morpholo­gical rule accounting for the i-æ ~ alternation is in com­petition with the psychologically-real morphological rule of regular past-tense formation in English. The regular process of past-tense formation can generally apply to any verb in the language, whereas the morphological rule accounting for the i-œ ~ A alternation can only account for one-syllable verbs containing an i vowel followed by a sequence of [non-labial nasal][velar stop]. The verb bring can therefore po­tentially occur in the past tense with either bringed or brang. Both morphological rules are psychologically real, even though they are in competition with one another. Of course, the exceptional past-tense form brought is also in competition with both of these psychologically-real rules, but in this case, brought is not predicted by rule. Instead, speakers must memorize this exceptional past-tense form.

Languages will, of course, contain morphological rules of this sort. But generative phonology has typically avoided

190 ROYAL SKOUSEN postulating morphologically-defined surface rules and, in­stead, has tried to account for morphophonemic alternation by means of phonetically-statable rules. The linguistically-significant question, however, is whether speakers actually prefer phonetically-statable rules when they account for morphophonemic alternation. In this paper I will argue that the acquisition device actually favors morphological rules rather than phonetically-motivated phonological rules in accounting for morphophonemic alternation. I will consider how both theories account for morphophonemic alternation in Finnish and show that only a theory favoring morphological rules can explain the historical development of the language.

2.0. To begin with, let us consider how speakers of Finnish might account for certain alternations between the nominative singular and the corresponding stem form in Finnish. For the vast majority of nouns and adjectives end­ing in a vowel, the stem is identical to the nominative sin­gular, as in the following examples:2

nominative singular stem puhuja 'speaker' puhuja-korkea 'high' korkea-pää 'head' pää-sukka 'sock' sukka-apu 'help' apu-

There are other words, however, which do show alternation. For example, there is a class of nouns and adjectives in Finnish that have nominative singular forms ending in si , but which take stem forms ending in te:3

kuusi 'six' kuute-hirsi 'log' hirte-jälsi 'sapwood' jälte-kansi 'lid' kante-yksi 'one' yhte-

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 191

In all, there are about forty words that belong to this class of nouns and adjectives (Tuomi 1972:186-93). All these words are two syllables in length. Moreover, in each case, the al­ternating si~te segments are preceded by a dental sonorant (n, l, or r), by k, or by a vowel. In the case of a preced­ing k, however, the k of the nominative singular alternates with h in the stem. Thus the stem form of yksi "one" is yhte-rather than ykte-,

There is another class of nouns and adjectives which show a similar alternation. In this class of words, the nom­inative singular ends in i , but the stem ends in e. Unlike the si-te alternation, the consonant preceding the stem-fi­nal vowel in this class of words shows no alternation, as in the following examples:4

kieli "tongue" kiele-kuusi "spruce" kuuse-kaikki "all" kaikke-lahti "bay" lahte-lapsi "child" lapse-

This class of words is much larger than the one having the si~te alternation, there being about two hundred words in this class (Tuomi 1972:151-242). All of the words in the i~e class are also two syllables in length.

One way to account for these alternations is to postu­late three extrinsically-ordered, phonetically-statable rules . The first rule raises a word-final short e to :5

e -* i / //

The second rule changes a sequence of te to si unless the t is preceded by s, h, or another t;6

t → s / i (unless preceded by s, h, or t)

192 ROYAL SKOUSEN

And finally, in order to account for the k~h alternation in words like yksi ~ yhte- 'one', we will need a rule changing to h when followed by t:7

-> h / t

The underlying representation of yksi, according to the typ­ical generative phonological solution, would be /ykte/. In deriving the surface form of the nominative singular, the word-final e would first be raised to , giving ykti. Now the t would be changed to s in the environment of the fol­lowing i , giving yksi. The would not be changed to h in this case since the rule of t →* s had eliminated the t in the immediately preceding stage of the derivation. But the rule → h would apply in the derivation of the stem, where the stem-final e is not word-final and is therefore not raised. Consequently, the t would remain in the stem, there­fore allowing the to be changed to h in the stem:

underlying representation ykte ykte+nä e → i / # ykti t —+ s / i yksi → h / t yhtenä

A word having the alternation ksi-hte therefore has an un­derlying representation which is different from any of its surface forms; the underlying representation ends in kte , which never shows up on the surface. But in all other words showing either the si-te or the i-e alternation, the under­lying representation of the nominative singular is identical to the surface form of the stem. Thus the underlying repre­sentation of hivsi "log" would be /hirte/. In the nominative singular, no suffix is added to the underlying representation Hence, the stem-final of hirte is word-final in the nomina-

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 193

t ive s ingular and is therefore ra ised to . After e has been raised to , a ti sequence has been produced, and the rule t → s app l i e s , giving the f inal surface form for the nomi­nat ive s ingu la r , namely hirsi. On the other hand, when a suf­fix l ike the essive nä is added to the underlying representa­t ion , the stem-final e is not. word-final and thus the e r e ­mains. As a consequence, the t preceding the e in hirtenä will not be changed to s:

underlying r ep re sen t a t i on h i r t e h i r te+nä e → i / # h i r t i — t → s / i h i r s i • → h / _ t

Similarly, the underlying representation of lahti "bay"would be /lahte/. Again, in the nominative singular, the word-fi­nal e would be raised to i. In this case, however, the ti se­quence that is produced will not be changed to si since the t is preceded by h. Again, when a suffix like the essive is added, the underlying e will remain:

underlying representation lahte lahte+na e → i / # lahti — t → s / i → h / t —

2.1. These purported synchronie rules of e-raising and t → s reflect historical changes that occurred in Finnish between 2000 and 2500 years ago (Raun and Saareste 1965:1-2; Posti 1953:53). There was, as we might suspect, an historical change that raised word-final short es. This change was lat­er followed by the palatalization of t to a palatal affricate ts when followed by the high palatal vowel i . The t was not

194 ROYAL SKOUSEN palatalized, however, when it was immediately preceded by a coronal obstruent, such as s, š or t. Later, the palatal affricate ís merged with s (Posti 1953:53-55). In modern Finnish, therefore, the only evidence we can find is for a direct change of t to s. Moreover, the š that, preceded the t has been changed to h (Hakulinen 1961:53). Consequently, in postulating a synchronic rule of the form t → s, the rule is blocked by either s, h, or t. And finally, there was later on an historical change replacing by h when followed by t (Hakulinen 1961:53). Moreover, the extrinsic ordering of these three purported synchronic rules is identical to the chronological ordering of the historical changes (Posti 1953:53; Hakulinen 1961:53):

historical development purported synchronic solution

E > i / _ # e -* i / _ // t > ts / i t —> s / i

(not preceded by s¿ or t ) ( n o t preceded by s, h, or t) : s > s k - ^ h / t s > h

k > h / t

Of course, this similarity to the historical changes cannot necessarily serve as an argument against such a synchronic solution. Speakers may be able to account for morphophonemic alternation by postulating rules that are similar to histor­ical changes; speakers may be able to do internal reconstruc­tion. The question is an empirical one. We cannot apriorily assume what kinds of regularities are linguistically-signif­icant .

2.2. This solution using these phonetically-statable rules is not the only solution that will account for these

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 1 9 5

a l t e rna t ions in the nominal system. Another way is to postu­l a t e ru les which d i r e c t l y r e l a t e ce r t a in morphologically-defined surface forms to other surface forms. For example, the si-te a l t e rna t i on in words l i ke hirsi could be handled d i r e c t l y by means of the following morphological r u l e : If the nominative s ingular of a two-syl lable word ends in si and is preceded by a dental sonorant or a vowel, then te can occur in the stem:8

[dental sonorant si// / t e -

[vowel] /nominative stem singular, 2-syllable

We notice that this morphological rule is, in part, phone­tically-conditioned; that is, the alternating sequences si and te must be preceded by either a dental sonorant or a vow­el . Moreover, the nominative singular must be two syllables long. But the alternation itself is not phonetically-moti­vated. From a psychological point of view, the synchronic alternation of si and te is not explainable in phonetic terms, although diachronically the alternation is, of course, due to phonetically-explainable historical changes. This morpho­logical rule specifically indicates that this alternation di­rectly relates stem forms to nominative singular forms.

Similary, the i-e alternation in words like lahti could be handled by the following morphological rule: If the nomi­native singular of a two-syllable word ends in , then e can occur in the stem:

i# / [consonant] e-<nominative singular, 2-syllable / <stem>

!96 ROYAL SKOUSEN Finally, for the majority of words ending in a vowel in the nominative singular, we may simply state that the stem is identical to the nominative singular. In this morphological rule there is no alternation at all, nor is there any restric tion on the number of syllables.

2.3. These two ways of accounting for the i~e and si­te alternations make different claims about systematic gaps that might occur on the surface in Finnish. For instance, the phonetically-statable rule changing ti to si places re­strictions on the words that can take the i-e alternation. There should be no nouns or adjectives that show an alterna­tion of ti~te unless the t is preceded by an s, h, or t . An underlying representation ending in te must end up as si in the nominative singular if the t is preceded by a vowel or by a dental sonorant, since ti is changed to si in this en­vironment. The rule t → s thus predicts a systematic gap in the class of words showing the i~e alternation. There should be no nominative singulars in the i-e class ending in nti, vti, l t i , or [vowel] ti:

underlying representations

surface forms

| s t i -s te I ht i" -hte t t i - - t t e |

nti~nte rti~rte lti-lte Vti~Vte

On the other hand, the morphological rule accounting for the i-e alternation claims that virtually any allowable consonant

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 197 could occur before the morpheme-final vowel. The only re­striction is that the consonant preceding the morpheme-final vowel does not alternate. Thus the morphological rule for the i-e alternation would allow alternations like [vowel] ti~ [vowel]te , for instance.

In a similar way, the phonetically-statable rule of <?-raising predicts that there will be a systematic gap among those words whose nominative singular forms are identical to stem forms. The e-raising rule predicts that on the surface there should be no non-alternating words ending in e since the e-raising rule will change the underlying word-final e to an i in the nominative singular. Non-alternating words should therefore end in any short vowel except e:

i~i ä~ä a~a ö~ö o~o ~ u~u

e~e

i~i ä~ä a~a ö~ö o~o ~ | u~u

e → i/_# i~e e~e

But the morphological rule stating that the nominative sin­gular and the stem can be identical when the nominative sin­gular ends in a vowel makes no restriction on what kind of vowel the nominative singular ends in. Thus the morphological rule predicts that words ending in e could occur in the class of non-alternating words.

As a result of the historical changes of e-raising and t → s, systematic gaps like these actually occurred in the nominal system of Finnish. Yet the significant question is

198 ROYAL SKOUSEN whether new speakers, when they were confronted with such data in learning the language, were able to account for those systematic gaps. Did they account for the fact that there were no non-alternating words ending in a short e? Was the fact that there were no [vowel]ti~[vowel]te alternations psy­chologically real? Did speakers account for three different surface alternations by postulating that underlyingly there was no alternation at all? In other words, were new speakers able to capture two ordered rules like e-raising and t —> s?

If speakers did capture these phonetically-statable reg­ularities, then the systematic gaps in the surface alterna­tions should have remained. In actuality, they have not. In the modern language we find unalternating words ending in a short e, such as nukke "doll", itse "self", and kolme "three". Recent loans, such as nalle "teddy-bear", are frequently non-alternating. Many names ending in e, such as Mansike "Straw­berry", Baahe (a town), and Aarne (a Christian name), are al­so non-alternating (Penttilä 1963:152). All of these examples simply show that there is no psychologically-real phonologi­cal constraint against non-alternating words ending in e.

All of these examples, however, did not originally ex­ist in Finnish, but are subsequent developments. For example, nukke is believed to have been created from nukka, originally meaning "a piece of cloth" (Toivonen et al. 1958:397); itse is derived from itsek (Itkonen 1965:213-14; Rapola 1966:301); kolme is derived from kolmet, possibly a plural form (Haku-linen 1961:49; Rapóla 1966:312-13). Yet the fact that speak­ers have allowed such examples to enter the language clearly indicates that they are not accounting for the systematic gaps left by the historical change of e-raising. Although at one time there were no surface examples of non-alternating

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 199 words ending in short e, speakers did not account for this regularity, but considered it accidental, thus allowing for exceptions to the historical rule of e-raising to enter the language.

Similarly, there is a word that has entered the i~ al­ternation which violates the systematic gap predicted by the rule t —> s. From the native word neita "virgin", Finnish speakers have created the word neiti "Miss, young girl". In the standard language, this word is non-alternating. In di­alects, however, some speakers have replaced the stem form neiti- by neite-, thus putting it in the i~e class (Toivonen et al. 1958:371). This new alternation is in conflict with the systematic gap left by the historical change of t to s, since there should, be no [vowel ] ti - [vowel ] te alternations in the ~ class.

2.4. Another difference is that the morphological rules specifically mention that only two-syllable words occur with these alternations. On the other hand, the rules of e-raising and t → y s are formulated without reference to the number of syllables in a stem. Consequently, these phonetically-stata­ble rules claim that this restriction on the number of sylla­bles is not linguistically significant, but only accidental. These rules could be used to derive nominal forms that are more than two syllables in length. And not surprisingly, there are forms longer than two syllables that these purported rules could be used on. Consider, for example, the nominalizing suf -fix that ends in s in the singular and te in the stem, as in the word korkeus "height". The stem form is korkeute-. By postulating that the stem form is the underlying representa­tion, we can almost derive the nominative singular by the rules which we already have. The rules of e-raising and t →

200 ROYAL SKOUSEN s give us kovkeusi, In order to get kovkeus, let us postu­late an additional rule that will delete the word-final i in words longer than two syllables. This rule will give the correct surface form of the nominative singular, kovkeus: 9

unterlying representation korkeute korkeute+na e → i/ # korkeuti t → s/ i korkeusi i → 0/ # (in words longer korkeus

than two syllables) This purported synchronic rule of i-deletion reflects the historical development (Hakulinen 1961:39) and also explains why the words showing si~te and i~ alternations are only two syllables in length. But the question is whether speakers view the s~te alternation in kovkeus as underlyingly the same as the surface si-te alternation.

Evidence from analogical change indicates that the two-syllable restriction is »psychologically-real. Consider the word novsi "smelt" borrowed from Swedish.10 The stem form is normally novsi-, like the nominative singular. However, some speakers have postulated a new stem which ends in te rather than si, so that the essive of novsi is novtena rath­er than novsina (Toivonen et al. 1958:393). A similar exam­ple is the word viisi "rickets", originally borrowed form Russian. In Finnish dialects, we still find the stem form viisi-. But also in the dialects, some speakers have replaced the original stem viisi- by the stem form viite- (Itkonen and Joki 1962:789). There are also examples of words taking on the i~e alternation, such as piili "needle", vyyni "grain" saksi "scissors", sipsi "tongs" (Itkonen and Joki 1962:546, and 1969:903, 952, 1036), and, of course, neiti:11

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 201

original alternation analogical alternation norsi~norsi- norsi-norte-riisi-riisi- riisi~riite-piili~piili- piili~piile-ryyni~ryyni- ryyni~ryyne-saksi~saksi- saksi~sakse-sipsi~sipsi- sipsi~sipse-neiti~neiti- neiti~neite-

In all these examples, one rather striking fact is noticed: The only words in which stem forms have been analogically changed to end in e or te are two syllables in length. The phonetically-statable rules say that this is an accident. On the other hand, there are absolutely no examples of words ending in s in the nominative singular that have taken on stem forms ending in te. In no case has the alternation s~te found in korkeus been extended. Luukas "Luke", for instance, has the stem form Luukkaa-, but Luukate- is impossible and never occurs. Similarly, native words like janis "hare" and sotilas "soldier" never fit into this s~te pattern. The pho­netically-statable rules also claim that this lack of exten­sion is accidental. But the morphological rules predict pre­cisely the 'kinds of analogical changes that occur. Stem forms like ¿änite- and sotilate- are totally unacceptable as well as incomprehensible to native speakers of Finnish. The mor­phological rule restricting the t~s alternation in the nomi­nal system to two syllable words directly reflects these lin­guistic intuitions.

2.5. The solution using morphological rules also pre­dicts that these three morphological rules (accounting for the si~te alternation, the i- alternation, and the lack of alternation) are in competition with one another. Thus norsi "smelt" and neiti ."Miss" were originally in the unalternating

202 ROYAL SKOUSEN class of nominals, their stems being norsi- and neiti-. These stem forms have been analogically changed by some speakers to norte- and neite-, But this does not mean that there is no psychologically-real morphological rule relating the unalternating nominal forms, only that it is in competi­tion with the other two morphological rules. Thus we can al­so find words that historically took the i-e alternation that have been shifted into the unalternating class. For example, (silmä)ripsi "(eye)lashn originally had the stem formptpse-. Some speakers have replaced this stem form with ripsi- (Itko-nen and Joki 1962:809). This competition between morphologi­cal rules simply means that speakers cannot absolutely pre­dict, for a given nominative singular form, what the stem form will be. Thus if a two-syllable noun ends in i , there could be three different possible stem endings: either in i, if the nominative singular ends in si. In hearing a word like norsi in the nominative singular, there is no way for a speak­er to know if the stem is norsi-, norse-, or norte- unless he has already heard the stem form. Thus the morphological rules explicitly predict that any of these three forms might occur, but they do not predict which one will actually occur. This competition between morphological rules thus requires speak­ers to learn which alternation a given word takes.

3.0. There are other places in Finnish morphophonemics where the phonetically-statable rule t → s could be used to account for morphophonemic alternation. Consider, for in­stance, the past-tense verb stem. Ordinarily, the past-tense stem ends simply in i. If the verb stem ends in an unround vowel, that vowel is deleted when the past-tense marker i is added, as in the following examples:

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 203

(present) stem p a s t - t e n s e stem purka- ' t o loosen' p u r k i -p ä ä s t ä - ' t o l e t go' p ä ä s t i -luke- ' t o read ' l u k i -to imi - ' t o a c t ' t o imi -

If t h e v e r b s t e m e n d s i n a r o u n d v o w e l , t h e vowe l i s n o t d e ­

l e t e d :

i s t u - ' t o s i t ' i s t u i -sano- ' t o say' sano i -v i i p y - ' t o l i n g e r ' v i i p y i -s ä i l ö - ' t o p re se rve ' s ä i l ö i -

With c e r t a i n v e r b s e n d i n g i n te, ta, and tä, h o w e v e r , t h e p a s t t e n s e e n d s i n si i n s t e a d of ti a s i n t h e f o l l o w i n g e x a m p l e s :

r aken ta - ' t o b u i l d ' r a k e n s i -t u n t e - ' t o f e e l ' t u n s i -ymmärtä- ' t o unders tand ' ymmärsi-v a e l t a - ' t o wander' v a e l s i -t i e t ä - ' t o know' t i e s i -

The t of t h e s t e m m u s t , i n f a c t , be p r e c e d e d by a vowel o r a d e n t a l s o n o r a n t , 1 2 f o r i f t h e t i s p r e c e d e d by s, h, o r t,, t h e n t h e t of t h e p r e s e n t s t e m r e m a i n s :

unohta- ' t o f o r g e t ' u n o h t i -e s t ä - ' t o prevent ' e s t i -

T h i s d i s t r i b u t i o n of t and s i n t h e p a s t t e n s e i s p r e c i s e l y l i k e t h e t~s a l t e r n a t i o n i n t h e n o m i n a t i v e s i n g u l a r . We c a n c l a i m t h a t t h i s s i m i l a r i t y i s n o t an a c c i d e n t , t h a t i t i s p s y c h o l o g i c a l l y r e a l and t h a t s p e a k e r s a c c o u n t f o r i t by u s i n g t h e r u l e t → s t o a c c o u n t f o r t h e s e p a s t - t e n s e fo rms a s w e l l . To do t h i s , h o w e v e r , we m u s t p o s i t a n o t h e r r u l e , one t h a t w i l l d e l e t e an u n r o u n d vowe l when t h e i of t h e p a s t t e n s e i s a d d e d t o t h e v e r b s t e m :

[unround vowel] → Ø / + i

204 ROYAL SKOUSEN

I s h a l l r e f e r to t h i s synchron ic r u l e as d iph thong r e d u c t i o n . Thus rakensi-, the p a s t - t e n s e form of rakenta- " to b u i l d " , i s u n d e r l y i n g l y / r a k e n t a + i / . F i r s t , the unround s t e m - f i n a l a i s d e l e t e d , g iv ing rakenti.Now the r u l e t → s a p p l i e s , g iv ing the c o r r e c t s u r f a c e form, vakensi-:

underlying r ep re sen t a t i on diphthong reduct ion t → s/ i

rakenta+i r aken t i rakensi

In the p a s t - t e n s e stem of unohti- (from the verb stem unohta-"to f o r g e t " ) , we have / u n o h t a + i / as the u n d e r l y i n g r e p r e s e n ­t a t i o n . In t h i s c a s e , as b e f o r e , d iphthong r e d u c t i o n a p p l i e s , but the ti sequence w i l l no t be changed to si because of the p reced ing h:

underlying r ep re sen t a t i on diphthong reduct ion t → s/ i

unohta+i unohti

There i s a c l a s s of ve rbs in F i n n i s h which a l s o t ake p a s t - t e n s e stems ending in si. In t h i s c a s e , however, no t shows up in the s u r f a c e form of the p r e s e n t s tem. Consider examples l i k e the f o l l o w i n g :

present stem s e l v i ä - ' t o c lea r up' s a l a a - 'to h ide ' halua- ' t o want' kohoa- ' t o r i s e ' herää- ' t o awaken'

p a s t - t e n s e stem s e l v i s i -s a l a s i -h a l u s i -kohosi-h e r ä s i -

ímperative stem s e l v i t -s a l a t -h a l u t -kohot-h e r ä t -

Although the t i s not found in the p r e s e n t s tem, i t i s found. for i n s t a n c e , in the i m p e r a t i v e stem. If we p o s t u l a t e t h a t the u n d e r l y i n g r e p r e s e n t a t i o n for t h e s e ve rbs a c t u a l l y con­t a i n s a t , then we can account for the f a c t t h a t the p a s t -tense stem ends in si. For example, the u n d e r l y i n g r e p r e s e n ­t a t i o n of halua- " to want" would be / h a l u t a / . In o rder to

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 2 0 5

form the p r e s e n t s tem, the i n t e r v o c a l i c t would be d e l e t e d . In the i m p e r a t i v e , the s t e m - f i n a l low vowel would be d e l e t e d whenever the i m p e r a t i v e s u f f i x (which beg ins wi th a k) i s added t o t he s tem. When the p a s t - t e n s e marker i i s added t o the u n d e r l y i n g r e p r e s e n t a t i o n , we get / h a l u t a + i / . By us ing the r u l e of d iph thong r e d u c t i o n , we w i l l d e r i v e the form haluti. Now we have a ti s equence , which w i l l be changed to si :

underlying r e p r e s e n t a t i o n ha lu ta ha lu ta+i haluta+kaa diphthong reduct ion h a l u t i t → s / i ha lu s i a-deletion halutkaa £-deletion halua

This particular solution claims that in those cases when the past-tense stem ends in si, the s is psychologi­cally derived from an underlying t. Moreover, the t under­lying the s is part of the stem to which the past-tense marker i is added. The surface s is not part of the past-tense marker. We also observe that on the surface, there are two types of past-tense stems ending in si. The first class has present stems in which a t actually shows up in the sur­face form of the present stem, as in vakenta-. In a verb like halua-, however, no t shows up in the surface form of the present stem, even though a t does show up in a couple other cases. The solution using the phonetically-statable rules claims that this surface difference is not psycholo­gically real. The si in halusi- is essentially no different than the one in ra k e ns i - . Both are derived from underlying sequences of ta + i, thus si does not indicate the past tense, only i does.

3.1. But the historical development shows once more that speakers are not accounting for these si forms in this way. First of all, speakers clearly conceive of the s inpast-

206 ROYAL SKOUSEN tense stems as a sign of the past tense. This can be seen in certain dialects of southwest Finland, where the past-tense stem is frequently replaced by one ending in si rath­er than the historically correct i. For example, in verb stems like istu- "to sit" and sano- "to say", the past stem ended in i historically, giving istui- and sanoi-. In these dialects, however, diphthongs in unstressed positions have lost the off-glide, thus merging the historical past-tense stem with the present stem in verbs like these. In order to offset this semantic collapse, speakers have analogically added a different past-tense ending to verb stems ending in a round vowel, thus preserving the past tense. And, not sur­prisingly, they chose si. These speakers therefore say istusi- and sanosi- (Kettunen 1930:7, 14). This use of si shows that the whole si sequence is psychologically inter­preted as representing the past tense. In other words, si cannot, in every instance, be derived from or related to an underlying stem-internal t.13

3.2. Another seemingly curious fact is that in many dialects the s in forms like rakensi- has been analogically replaced by t, thus giving rakenti- (Rapola 1966:236). On the other Hand, in verbs like halua-, which has the past-tense form halusi-, we never get an analogical form like haluti-. This difference is totally unexplainable if we assume that speakers are accounting for both si sequences as coming from stem-internal t's. Why should s be replaced by t only in forms like rakensi-, but never in halusi-, when underlyingly they both supposedly end in ta+i. The phoneti­cally-statable rules cannot explain this divergency.

But if we look at the surface, the reason is immediately apparent. Verbs like halua- form a special class of their

ON THE NATURE OF MORPHOPHQNEMIC ALTERNATION 2 0 7

own in F inn i sh and have forms t o t a l l y u n l i k e ve rbs having p r e s e n t stems t h a t end in a s h o r t vowel. The p r e s e n t stems of ve rbs l i k e halua- always end in two vowel s , the second of which i s a low vowel . This f i n a l low vowel a c t u a l l y i n d i c a t e s the p r e s e n t t e n s e in t h e s e v e r b s , thus the b a s i c verb form i s halu-. Consequen t ly , in the p a s t - t e n s e s tem, si i s the p a s t -t e n s e s u f f i x . That the b a s i c form of t h i s verb i s s imply halu-. Consequen t ly , in the p a s t - t e n s e s tem, si i s the p a s t -t e n s e s u f f i x . That the b a s i c form of t h i s verb i s simply halu-can be seen by look ing a t the v a r i o u s s u r f a c e stems t h a t t h i s verb h a s :

halu base form halu+a present stem halu+n p a s t - p a r t i c i p l e stem halu+t imperative stem halu+si p a s t - t e n s e stem halu+ta i n f i n i t i v e stem

H i s t o r i c a l l y , t h e r e was only one s u r f a c e stem form, haluta (Rapola 1966:166) . In f a c t , t he ta o r i g i n a l l y se rved as a v e r b a l i z i n g s u f f i x in F i n n i s h , but because of subsequent pho­n e t i c change , speake r s no longer i n t e r p r e t ve rbs l i k e halua-in t h a t way.

On the o t h e r hand, the co r r e spond ing stem forms for rakenta- a r e ve ry much more r e g u l a r . The s u r f a c e mot ivat ion. for r e p l a c i n g si in rakensi- by ti i s ve ry s t r o n g , s i n c e in the o t h e r forms of the v e r b , t c o n s i s t e n t l y shows up:

rakenta base form rakenta present stem rakenta p a s t - p a r t i c i p l e stem rakenta imperat ive stem rakens i p a s t - t e n s e stem rakenta i n f i n i t i v e stem

208 ROYAL SKOUSEN 4,0. There is one other place where we can make use of

the purported rule t → s in order to account for a morpho-phonemic alternation. This occurs in the plural stems of the nouns and adjectives that show the si → te alternation. In these nouns and adjectives, the plural stem is identical to the surface form of the nominative singular. Thus hivsi "log", for instance, has the stem form hirte-, but the plural stem is hivsi-. The underlying representation for the plural marker is i , so in the typical generative phonological analysis the underlying representation of the plural stem would be /hirte +i/. We can produce the plural stem hivsi- by the rules which we already have. The rule of diphthong reduction will delete the stem-final e in the environment of the plural marker i3

giving hivti. Now the t is changed to s, giving the correct surface form of the plural stem:

underlying representation hirte hirte+nä hirte+i+nä e → i I // hirti diphthong reduction hirtinä t —> s / i hirsi hirsinä

Therefore, in nouns and adjectives taking the si~te alternation, the plural stem is, on the surface, identical to the nominative singular. This surface regularity is not, however, directly captured by the solution using phoneti­cally-statable rules. These rules do not consider this sur­face regularity to be linguistically-significant, but only the result of independently .operating rules. In the nomina­tive singular, e-raising and t → s apply; in the plural stem, diphthong reduction and t → s apply. These independently-motivated rules, it turns out, produce the same surface forms in words having the si-te alternation, but this surface fact is only the accidental result of more generally occurring

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 209

phonetically-statable rules and is not linguistically signif­icant.

4.1. Historical reconstruction, however, indicates that speakers did account for this surface regularity - that is, they did relate the plural stem and the nominative singular in words having the si-te alternation. Originally, the his­torical changes similar to the purported synchronic rules of diphthong reduction and t → s affected the plural stems of words like sota "war" and seta "uncle". In the plural stem, when the plural marker i occurred, the stem-final low vowels, which are unround, were deleted, giving soti- and seti- as plural stems. Then the t's in these forms were changed to s's, giving sosi- and sesi- (Hakulinen 1961:41). These forms, however, no longer exist, but instead the plural stems are soti- and seti-, without the historically-correct s's. But in words like hirsi, which shows the si-te alternation, the his­torical plural stems ending in si are retained.

This analogical replacement of s's by t's in words end­ing in a low vowel is very damaging to a theory of phonology that considers phonetically-statable regularities more highly valued than morphologically-defined surface regularities. Pri­or to this analogical replacement, the nominal system was per­fectly regular with respect to the phonetically-statable reg­ularities captured by the rules of e-raising, diphthong re­duction, and t → s. There were no surface exceptions. Such a solution would claim that alternations like sota-sota--sosi-and setä-setä--sesi- are perfectly regular. A theory that op­erates on the assumption that phonetically-statable regular­ities are to be captured, especially when there are no sur­face exceptions to such regularities, must necessarily claim that these alternations are stable and perfectly regular and,

210 ROYAL SKOUSEN more particularly, should not be changed to alternations that directly violate the phonetically-statable regularities.

The phonetically-statable solution predicts that there should not be any change. Thus it cannot explain why there is a change. In some unknown way, the speakers must have im­perfectly learned the rule t → s, restricting it so that it would not apply in the plural stem if the underlying stem ends in a low vowel. Of course, if the underlying stem ends in e, the rule t —> s continues to apply in the plural stem. Thus, we continue to get si forms in the plural stems of words having the si-te alternation. In other words, if we maintain the rule t → s, we must say that the rule has been unexpectedly complicated. In fact, we are forced to use a global rule to account for the change of £ to s in the plural stem. We must now look at the underlying representation, to make sure the lexical representation does not end in a low vowel. Thus in the derivation of plural stems only, the oth­erwise-regular rule t —> s must be blocked if the underlying lexical representation ends in a low vowel. And finally, this change in the rule t → s produces surface exceptions to the original rule, namely sequences of ti on the surface. Prior to the replacement of s by t, there were no exceptions to the rule t → s. From a state of no exceptions, we now have exceptions. This is totally unexplainable if the theory we are operating under claims that exceptionless phonetically-statable regularities are highly favored.

4.2. Consider, on the other hand, a theory which claims that speakers would not posit a synchronic rule t → s in this case, even when there were no surface exceptions to the historical change in the language. Suppose that, instead of accounting for these phonetically-statable regularities,

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 211 speakers accounted for the t~s alternations by means of morphological rules. Such a theory would account for the si-te alternation directly. Since the morphological rule derives the stem from the nominative singular, the nomina­tive singular is the basic lexical form, and consequently the plural stem is easily recognizable since it is identical to the basic lexical form. But the historical plural stems of words like sota and seta are different. The stem forms sota- and setä- would be identical to the nominative singu­lar, but the plural stems would be sosi- and sesi-, which are totally unrecognizable, given, the other forms of these nouns. Moreover, for other nominal forms in which the stem is identical to the nominative singular, the plural stem is regularly formed by simply adding i to the stem, deleting a stem-final unround vowel, and in no way changing the preced­ing consonant, as in the following examples:

nominative singular stem plural stem sivu "side" sivu- sivui-talo "house" talo- taloi-löyly "steam" löyly- löylyi-nenä "nose" nenä- neni-musta "black" musta- musti-

Words like sota and seta have identical nominative singular and stem forms, but their historical plural stem forms sost­anti sesi- were exceptional to the regular rule of plural for­mation. These unrecoverable plural forms were therefore re­placed by soti- and seti- according to the regular morpholo­gically-defined rule of plural-stem formation.

4.3. In some words in which t was changed to s in. the plural stem, the historical s has been retained in the plu­ral stem. But in nearly all these cases, the historically-correct stem form, which originally contained a t, has been

212 ROYAL SKOUSEN replaced by an analogical one containing an s. Originally, words like kytkyt "leash" and kuollut, the past participle of the verb stem kuole- "to die", had the following forms:

nominative singular stem plural stem kytkyt 'leash' kytkyte- kytkysi-kuollut Mead' kuollute- kuollusi-

In the southern Ostrobothnia dialects of Finland, the histo­rical t's in the stem have been replaced by s's, thus:

kytkyt kytkyse- kytkysi-kuollut kuolluse- kuollusi-

This replacement of forms clearly shows that speakers were not accounting for the s in the plural stem by means of a phonetically-statable rule of the form t → s. The s in the plural stem was not underlyingly t . If it had been, there would have been no reason to replace the historical stem forms with ones having s's. Speakers obviously considered the surface s's in the plural stems as underlying and since plural stems are regularly derived from stems by simply de­leting unround vowels and keeping the consonant unchanged, the stem form was assumed to contain an s as well. This ana­logical change shows once again that speakers were not ac­counting for a t-s alternation by means of a synchronic rule t → s.

In the dialects where this analogical replacement did not occur, the historical plural stems have been replaced with regular plural stems, so that in most dialects we have the following type of forms:

kytkyt kytkye- kytkyi-kuollut kuollee- kuollei-

In. no case does a dialect retain the original historical forms of these words (Rapola 1966:237-38; Posti 1953:52). In some

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 213 way the theory must explain why the s in the plural stem was unstable except in the case when the nominative singular al­so ended in si. The original historical forms are regularly preserved in those words having identical nominative singu­lar and plural stem forms. A theory preferring phonetically-statable rules such as t → s cannot explain the stability of plural stems like hirsi- nor the instability of plural stems like sosi- and kytkysi-. A theory using morphological rules, however, readily predicts this instability.

5.0. Over time the purported rule t → s has become much more complicated. When there were no surface exceptions, we had a perfectly regular phonetically-statable rule. After the analogical replacement of sosi- by soti-, the rule had become a complicated global rule. Throughout the subsequent history of the language, this purported rule has also become more and more restricted in its application. For example, since the historical change, there have been a great many words with underlying ti sequences that have entered Finnish. And in no case have such ti sequences been changed by a sup­posed constraint against t i . The ti sequences in loan words from Germanic, Slavic and Swedish sources have remained un­changed. Thus aiti "mother", a Germanic loan, is never changed to äisi, the Slavic borrowing laati- "to draw up" remains the same, and the Swedish loan tikku "stick" is never changed to sikku (Hakulinen 1961:330, 334, 339). The fact is that these words are not exceptional to the purported rule t → s, nor have they ever been exceptional. There has never been any tendency to remove these later exceptions to the historical change of t to s. In addition, there are a large number of words that have been created by Finnish speakers themselves that violate the historical contraint against t i . Included

214 ROYAL SKOUSEN within this group are onomatopoeti.c words, such as tippa "drop" and mutise- "to grumble", and common words such as tati "aunt", neiti "Miss", and koti "home".14 In every case, the historical constraint against certain ti sequences is ig­nored.

5.1. These are not the only exceptions to the histor­ical change of t to s. Subsequent historical changes have also created exceptions. For example, originally there was an adjectival suffix rise which was added to noun stems, as in the examples entense- "former", created from the stem ente- "first", and vetense- "watery", created from the stem vete- "water". Surface exceptions to the historical change of t to s have been created by later historical development in this suffix. The palatal n lost its closure when followed by the fricative palatal s, giving a nasalized high i vowel. Later the nasalized quality of the vowel was lost, palatal­ized á became dental (Posti 1953:26; Hakulinen 1961:52), and ei was reduced to short , giving the forms found in the mod­ern language, entise- and vetise- (Posti 1953:53; Hakulinen 1961:35). Yet these adjective forms are not changed to ensise-or ve s i se - .

Another historical change creating surface exceptions includes gradation. In short closed syllables, geminate t's have been gradated to short t's, yet ti sequences produced by gradation have not been changed to si. For example, in a nominal stem like kuitti "receipt", the ti sequence would not be changed to si because of the preceding t . But in the gen­itive, when the suffix n is added, the geminate t is short­ened, giving kuitin. This surface ti sequence, of course, is never changed to si.

Finally, the long mid vowel in Finnish has been diph­thongized to ie, as in the verb stem tietä- "to know" and in

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 215 the noun tie "road". Historically, these forms come from teetä- and tee (Rapola 1966:355). But these derived ti se­quences are never changed to give sieta- or sie. In other words, the purported synchronic rule t —> s is never extended to ti sequences produced by later historical changes.

The traditional way in generative phonology to account for such surface exceptions is to extrinsically order syn­chronic rules like gradation and diphthongization after the synchronic rule t → s. Of course, this presumes that grada­tion and diphthongization are also captured as phonetically-statable rules.15 A solution of this sort claims that de­spite the depth of the t~s alternation, the underlying regu­larity does not apply to the most underlying sequences of t i , namely those lexical entries such as ä i t i "mother" which have ti sequences in their lexical representations.

5.2. Contrast this situation with an historical change that has taken place in the Häme and Savo dialects of Fin­land. In these dialects, single consonants have been gemi­nated when preceded by a short, stressed vowel and followed by a sequence of two unstressed vocalic segments:

c > c: / [stressed

[vowel

uns t ressed

vowel

uns t ressed

vowel J

c > c: / [stressed

[vowel

uns t ressed

vowel

uns t ressed

vowel J

For example, the illative form kókoon "whole" is changed to kokkoon by this historical change. Typical examples of this historical change are found in many different forms (for ad­ditional examples, see Kettunen 1930:22, 105; Rapola 1966: 25-26) :

p>pp lúpaan > lúppaan "I promise" present tense r>rr heräisi > herräisi "he would awaken" conditional h>hh lihaa > lîhhaa "meat" partitive singular

216 ROYAL SKOUSEN m>mm plmeissä > pimmeissä "dark" inessive plural t>tt káteùteen > káteùtteen "jealousy" illative singular 1>11 kumartèlee > kûmartèllee "he bows" present tense J>JJ ahristàjaa > âhristàjjaa "pursuer" partitive singular v>vv sávea > sâwea "clay" partitive singular n>nn sánoa > sânnoa "to say" infinitive s>ss kîîtollîsuus > kîitollissuus

"thankfulness" nominalization

k>kk pûrottàkaa > pûrottàkkaa "drop" imperative t>tt lîikutètaan > lîikutèttaan impersonal

"it will be moved"

This historical change applied in every place where the con­ditions for gemination were met. Like the historical change of t to s, gemination produced systematic gaps on the sur­face; namely, short consonants occurring in the environment of a preceding short, stressed vowel and a following sequence of two unstressed vowels. But unlike the change of t to s , this systematic gap has remained in the dialects. New speak­ers apparently account for the systematic gap left by the historical change of gemination by postulating a synchronic rule that is identical to the original historical change. Analogical changes do not create exceptions to this system­atic gap. The synchronic rule of gemination eliminates excep­tions in borrowed words that have entered the dialects since the original historical change took place. For example, loan words from Swedish such as sikaari ,"cigar", minuuti "minute'1, and pôliisi "police", which have short consonants in non-geminating dialects, Piave geminated consonants in the geminat­ing dialects of Häme and Savo: sikkaavi, minnuuti, pôlliisi (Rapóla 1966:26). These changes clearly show that Finnish speakers in the geminating dialects have been able to capture the phonetically-statable regularity left by the historical change, that psychologically there really is a phonetic con­straint against ungeminated consonants in certain environ-

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 217 ments.

Moreover, exceptions created by later historical devel­opments are also eliminated. For example, in Tammela, a par­ish of Häme, the third person personal suffix nsä has been changed through historical development. In forms like väkensä "his people", the final ä vowel was lost in this dialect. Clo­sure of the n was subsequently lost, creating a long nasal­ized vowel. Finally, the nasalization was lost, giving vahees. This surface exception to gemination has now been eliminated, giving väkkees (Rapóla 1966:26).

In other words, the synchronic rule of gemination be­haves completely differently than the purported synchronic rule t → s. Gemination extends itself to new forms entering the language, both in borrowed words and in forms created by later historical changes. The rule of gemination remains on the surface, without allowing surface exceptions. It does not become more and more underlying as time goes on. Syn-chronically then, the purported rule t → s and the rule of gemination are totally different. Gemination continues to be­have precisely like the original historical change, by elim­inating all surface exceptions, while the purported synchronic rule t → s never does.

6.0. What we need of course is a theory that will ex­plain why gemination continues to behave like the original historical change, while the supposed rule t → s does not. A theory that claims that new speakers, in accounting for morphophonemic alternation, favor rules that account for phonetically-statable regularities cannot explain why these two historical changes behave so differently. In each case, there were no surface exceptions immediately after the histor­ical change had taken place. Yet in the case of gemination,

218 ROYAL SKOUSEN the phonetic constraint remains; in the other case, the pho­netic regularity is not captured and later surface exceptions are not removed. If the mechanism that accounts for morpho-phonemic alternation in language acquisition favors phoneti­cally-statable surface regularities, then speakers should have been able to maintain the rule t → s as well as the gemination rule.

Instead of making the theory favor phonetically-statable rules, let us consider the possiblity that the theory favours just the opposite. After an historical change involving mor-phophonemic alternation has taken place, the language acqui­sition device will try to account for that morphophonemic alternation by means of morphological rules. We have already seen that such an hypothesis will explain the failure of speakers to account for the historical t~s alternation in Finnish, even when there were no exceptions to that alter­nation. But how will this theory account for the fact that gemination does not appear to be accounted for by morpholo­gical rules, but instead by a phonetically-statable rule? Why is it that speakers, in this case, are able to posit a phonological rule identical to the historical change?

6.1. The answer to this question can be found by con­sidering the consequences of accounting for gemination by morphological rules. What would happen if speakers tried to account for gemination by means of morphological rules? In accounting for gemination in plural nominal stems, for in­stance, the speaker would learn the following morphological rule: If the stem-final syllable is of the form [consonant] [unstressed vowel] and the preceding syllable in the stem is short and stressed, then a geminated consonant will ap­pear in the plural stem, providing the plural stem ends in an unstressed diphthong:

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 219

/ stressed

vowel

unstressed

vowel C: /

unstressed

diphthong

'nominal stem last syllable plural stem

Thus geminated consonants would be found in the following words :

stem plural stem. tapa- "custom" táppoi-lási- "glass" lássei-luku- "chapter" lûkkui-hâllitsîja- "ruler" hállitsijjoi-

But if the final syllable of the stem is preceded by a long, stressed syllable, a geminated consonant will not occur in the plural stem:

áika- "time" aikoi-áiti- "mother" äitei-

Or if the preceding syllable of the stem is short, but un­stressed, a geminated consonant will not be found in the plu­ral stem:

pákanà- "pagan" pákanoi-tekijä- "doer" têkijöi-

And finally, the plural stem itself must end in an unstressed diphthong (that is, a sequence of two unstressed vocalic seg­ments) in order to get a geminated consonant in the plural stem. If the plural stem ends in a single vowel, then gemi­nation will not be found:

sóta- "war" soti-ópettája- "teacher" òpettàji-

In order to determine then when the plural stem would take

220 ROYAL SKOUSEN gemina t i on , the speaker would be forced to s p e c i f y t h a t the p e n u l t i m a t e s y l l a b l e of the stem i s s h o r t and s t r e s s e d , and, in a d d i t i o n , t h a t the p l u r a l stem i t s e l f ends in a sequence of two u n s t r e s s e d vowels .

But the speaker must account for geminated consonan t s in many o t h e r m o r p h o l o g i c a l l y - d e f i n e d forms b e s i d e s j u s t p l u ­r a l s t ems . In each c a s e , however, the speaker must l e a r n t h a t stems wi th s h o r t , s t r e s s e d p e n u l t i m a t e s y l l a b l e s can t ake geminated consonan t s in o t h e r r e l a t e d forms, but only when those r e l a t e d forms end in a sequence of two u n s t r e s s e d vo ­w e l s . Cons ide r , for i n s t a n c e , t he fo l lowing examples from the verb sys tem:

base form present stem 3rd s ingular sano- "to say" sano- sânnoo sáapu- " to a r r i v e " sáapu- sáapuu kuva- " to desc r ibe" kúwaa- kúvvaa lúke- " to read" luke- lukkee pòimi- " to pick" pòimi- poimii

p a s t - t e n s e stem i n f i n i t i v e condi t iona l stem sánnoi- sânnoa sá n n o i s i -sáapui - sáapua s á a p u i s i -kúvas i - kúvata kú vva i s i -l ú k i - lûkkea l ú k i s i -póimi- póimia pó imis i -

We n o t i c e t h a t the ve rbs sâapu and -póimi- never t ake gemi­na ted stems s i n c e the p e n u l t i m a t e s y l l a b l e in t h e i r base form i s not s h o r t and s t r e s s e d . The ve rbs sano- kuva(a)-, and luke- do meet t h i s c o n d i t i o n , so in r e l a t e d forms the consonant w i l l be gemina ted , but only when those r e l a t e d forms end in a sequence of two u n s t r e s s e d vowel s .

6 , 2 , The ne t r e s u l t i s t h a t in a v a r i e t y of d i f f e r e n t m o r p h o l o g i c a l l y - d e f i n e d c a t e g o r i e s , the speaker would s e t up morpho loc i ca l r u l e s , each of which would r e q u i r e gemi-

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 221

nation of a consonant only when that consonant is preceded by a short stressed syllable and followed by a sequence of two unstressed vowels. Rut this environment is the same en­vironment that the historical change took place in. In al­most every type of stem, the speaker is forced to learn the original historical environment for gemination. By assuming that the acquisition device favors morphological rules, we can predict that speakers will recover the historical envi­ronment of gemination. The historical rule that caused the morphophonemic alternation can be recaptured because the process of postulating morphological rules leads directly to the original rule. In other words, the evidence for the phonetically-statable environment is discovered by an acqui­sition procedure that depends upon morphologically-defined surface rules. .

The residues of the historical change of t to s are not like gemination. After the historical change, the t~s alter­nation only showed up in the nominative singular of certain words having stems ending in te; in plural stems and past-tense stems, when the corresponding stem forms ended in te, tä or ta; and in the past tense of some impersonal forms (Posti 1953:49-52; Hakulinen 1961:50-51; Rapola 1966:235-38). A morphological theory of morphophonemic alternation would account for these different alternations by separate morpho­logical rules, and in no way would these morphological rules require the speaker to posit a phonological rule changing t to s in the environment of i. Consequently, the historical phonetic regularity was not recaptured. Thus a theory that favors morphological rules makes the right prediction. It explains why the historical change of t to s was not recap­tured as a synchronic rule, and its also explains why the rule of gemination continues to remain productive, even

2 22 ROYAL SKOUSEN

though i t involves morphophonemic a l t e r n a t i o n .

7.0. In general , most h i s t o r i c a l phonemic changes are not recoverable . This fact has been widely appreciated in t r a d i t i o n a l h i s t o r i c a l l i n g u i s t i c s . 1 6 In accounting for the h i s t o r i c a l development of almost any language, the number of times the h i s t o r i c a l r e g u l a r i t i e s f a i l to remain produc­t ive is ra ther overwhelming. Only r a re ly do we find cases of morphophonemic a l t e r n a t i o n being accounted for in the i r o r ig ina l h i s t o r i c a l form as phone t i ca l l y - s t a t ab l e phonolo­gica l r u l e s . A theory of phonology which assumes tha t pho­n e t i c a l l y - s t a t a b l e r e g u l a r i t i e s are favored cannot predic t the great loss of h i s t o r i c a l ru les tha t ac tua l ly takes p lace . But a theory of morphophonemic a l t e rna t ion favoring morpho­logica l ru les wi l l p red ic t tha t the majority of h i s t o r i c a l changes wi l l not be recaptured.

7 . 1 . Moreover, such a theory can explain the analogical changes that take place in language, while the other is v i r ­t ua l ly incapable of explaining analogy. Analogical change occurs because of psychologica l ly- rea l morphological ru les that already ex i s t p r ior to the analogical change i t s e l f . These fac ts about analogy were observed long ago by Saussure (1959:161-73):

Analogy supposes a model and i t s r egu la r im i t a t i on . An ana logica l form i s a form made on the model of one or more other forms in a c ­cordance with a d e f i n i t e r u l e (161). Analogy i s psychologica l , but t h i s does not suff ice to separa te i t from phonet ic phenomena, for they may a lso be considered psycholo­g i c a l . . . . We must go fu r the r and say tha t analogy i s grammatical. I t suppose awareness and understanding of a r e l a t i o n between forms. Meaning plays no pa r t in phonetic changes, but i t must in tervene in analogy (165). Analogy i s grammatical throughout, but l e t us hasten to add tha t i t s end r e s u l t - c rea t ing - belongs a t f i r s t only to speaking. I t i s the change product of an i so l a t ed s p e a k e r . . . . S t i l l , two things

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 223

must be kept a p a r t ; (J) awareness of the r e l a t i o n t h a t t i e s t o ­gether the product ive forms; and (2) the r e s u l t suggested by the comparison, the form improvised by the speaker to express h i s t h o u g h t . . . . A major pa r t of the ana log ica l phonomenon i s t h e r e ­fore completed before the new form a p p e a r s . . . . I t i s wrong to suppose tha t the product ive process i s a t work only when the new formation ac tua l l y occurs , the elements were a l ready there (165-66).

Thus morphological ru les are morphologically-defined r e l a ­t ions between morphological forms and can be appropria te ly referred to as analogical r u l e s . 1 7 On the other hand, a theory favoring phone t i ca l l y - s t a t ab l e rules offers l i t t l e explanation for analogy, espec ia l ly those cases of analogi­cal change in which regu la r , phone t i ca l l y - s t a t ab l e phonolo­gical ru les would be suddenly transformed into global r u l e s . In f ac t , such global ru les must r e f e r , e i the r d i r e c t l y or i n d i r e c t l y , to morphologically-defined surface r e g u l a r i t i e s , which suggests that the l i n g u i s t i c a l l y - s i g n i f i c a n t r e g u l a r i ­t i e s are exactly those captured by morphological r u l e s .

7 .3 . Analogical change also suggests tha t the l i n ­g u i s t i c a l l y - s i g n i f i c a n t level in phonology is a surface pho­nemic l e v e l . A theory using morphological ru les implies , in f ac t , tha t underlying represen ta t ions are not very deep at a l l . In learning a language, speakers attempt to account for morphophonemic a l t e r n a t i o n at a surface phonemic l eve l . In order to pos tu la te morphological r u l e s , however, a speaker must f i r s t l ea rn , at l e a s t in p a r t , the psycho log ica l ly - rea l , d i s t i n c t i v e sounds in his language since the a l t e rna t ions in morphological ru les are phonemically a r b i t r a r y . Moreover, since morphological ru les refer to semantical ly- and syn­t ac t i ca l l y -de f ined morphological forms, the speaker must f i r s t develop a su f f i c i en t ly complex semantic and syn tac t i c system tha t wi l l require him to learn various morphologically-defined forms. This theory spec i f i ca l l y claims then tha t

2 24 ROYAL SKOUSEN

morphological rules and phonological rules acquired by means of morphologicalization (such as gemination in Finnish) will be learned relatively late in language acquisition. In gen­eral, most morphophonemic alternation will not be accounted for by rules until after the phonological and semantic-syn-tactic systems of the language have been at least partially acquired. Prior to this time, speakers will account for pho-nemically-arbitrary morphophonemic alternation by memorizing individual forms.

7.4. By no means does this imply that all morphopho­nemic alternation is accounted for by morphological rules, or that all phonological rules are acquired by means of mor-phologization. For example, some morphophonemic alternation will be accounted for by universal phonetic constraints, such as the s~z alternation, such as final devoicing in Ger­man, are recoverable as phonetically-statable phonological rules simply because such alternations directly reflect earlier stages that speakers go through in learning a lan­guage. In the case of fin'al devoicing, for instance, there is a stage in language acquisition in which final obstruents are devoiced; and if the linguistic data the child hears directly reflects such a stage in acquisition, then there will be no need to account for such alternations later on by means of either morphological rules or by phonological rules acquired by means of morphologicalization. Instead, inherent universal phonological processes will automatically account for morphophonemic alternations such as these.19

8. A theory of linguistics, like any empirical theory, is only as valuable as it is able to make correct predic­tions about how speakers actually use language. The task of linguistics is two-fold. First, the linguist must propose

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 225 theories that will explicitly predict how speakers account for linguistic data; and second, he must look for empirical evidence that will either support or disconfirm those theo­ries.20 Discovering the linguistically-significant or psy­chologically-real regularities is a difficult task, since there is no a priori method of analyzing linguistic data that does not actually presuppose a linguistic theory. Yet despite this methodological difficulty, a linguistic theo­ry, to be of any value, must make empirically-testable claims about the nature of human language.

In this paper I have considered the fundamental assump­tion of generative phonology - that speakers account for morphophonemic alternation by means of phonetically-statable phonological rules. I must emphasize that the intent of this paper has not been to show that Finnish cannot be described by the methods of generative phonology. In any given synchro­nic stage of Finnish (or of any other language, for that mat­ter) , a generative phonological analysis can always be made to work. Surface exceptions to phonetically-statable regular­ities can'be handled by means of various theoretical devices (such as rule ordering, exception features, derivational his­tory, and so forth). These devices, however, are merely the consequence of the theoretical assumption favoring phoneti­cally-statable regularities. The point is that simply accoun­ting for linguistic data by writing rules and positing addi­tional theoretical machinery (all of which crucially depends on which phonetically-statable regularities are captured by the linguist) is not linguistically-interesting since there is no set of morphological data that cannot, in some way or another, be accounted for by phonetically-statable rules.

Instead, I have argued that the important question is whether such a theory can actually predict, among other

2 26 ROYAL SKOUSEN things, the historical development of languages. A theory favoring phonetically-statable regularities cannot explain, for instance, why the purported phonological rule t —> s in Finnish behaves completely differently than the rule of gemination. Originally, there were no surface exceptions to either of the phonetically-statable regularities captured by these rules. The purported rule t → s, however, is later changed into a global rule, ceases to apply to underlying sequences of ti 9 and is obscured by subsequent phonetic changes that produce derived sequences of ti on the surface. On the other hand, the rule of gemination never becomes a global rule, always applies to new words entering the lan­guage, and eliminates exceptions created by later historical changes.

Of course, the traditional theory of generative phono­logy can be patched up so that it will make the right pre­dictions, but only by incorporating in that theory one that accounts for morphologically-defined surface regularities and which must, in fact, frequently ignore exceptionless, phonetically-statable surface regularities. The evidence therefore is simply in favor of a theory that prefers morpho­logical rules in accounting for morphophonemic alternation. This theory claims that speakers are not really very good at internal reconstruction, and the evidence from historical change is in favor of such a theory.

NOTES 1 This assumption of natural phonological rules is implicit in nearly all generative phonological work. More recently, there has been an ex­plicit attempt to account for the 'naturalness' of generative phonolo­gical rules (cf. Chomsky and Halle 1968, chap.9; Schane 1972). 2 It should be noted here that the term 'stem' refers to an unweakened form. Stems in Finnish may occur in weakened forms, providing the stem ends in a single vocalic segment and is preceded by a stop. In such a case, the stop may be weakened according to the rules of gradation in the language. These rules include the following alternations:

unweakened weakened unweakend weakened pp p 1t 11 tt t rt rr kk p v mp mm t d nt nn n nn

The weakened form of the stem occurs when certain suffixes like the inessive ssa-ssä are added to the stem. In other cases, such as when the essive na~nä is added, the stem is not weakened. Thus, the stem apu- 'help' has the weakened form avu- in the inessive (avussa), but in the essive, it occurs in its unweakened form (apuna). Similarly, the stem sukka- 'sock' has the weakened form suka-. On the other hand, since korkea- 'high' ends in two vocalic segments, kovkea- is not weakened when a weakening suffix is added to the stem (thus korkeassa) . For further details on gradation, cf. Skousen 1972:46-57. 3 The weakened forms of these stems, according to the rules of grada­tion, are kuuåe-, hirre-, j ä l l e - , -, and yhde-. 4 The weakened forms of the stem i- 'all' and lahte- 'bay' are, of course, i- and l a h d e - . The other stem forms (kiele-, kuuse-3,and tapse-) in this set of examples do not have weakened forms simply because i and s are not stops and are therefore not weakened by the rule of gradation.

5 This rule of e-raising can be found, for example, in virtually any generative phonological study of Finnish noun morphology. Cf. McCawley

228 ROYAL SKOUSEN

1963:181, 1964:12; Johnson 1970:103. This rule has been discussed in many different places, e.g., McCawley

1963:181, 1966:1; Wiik 1967:78-80; Johnson 1970:103. More recently, Kiparsky (1973 passim) has devised new theoretical notions in order to constrain this rule and make it work properly. 7 In his earlier work on Finnish, instead of a rule converting into h, McCawley (1963:181) posited a rule changing h into when followed by s. Later, however, he posited a rule of to h when followed by t (cf. McCawley 1964:16). Wiik (1967:46) also has this form of the rule. For the purposes of this paper, I will assume that the alternation

ksi-hte is handled by a separate morphological rule. The status of this alternation is, in any event, marginal. In the standard language and in most dialects, there are only two words that take on this al­ternation, namely yksi "one" and kaksi "two". Historically, no words have by analogy taken on this alternation, whereas several words that historically had this alternation have been analogically changed to take on the i-e alternation. Thus laksi-lahte- "bay" > lahti-Iahte-, haaksi-haahte- "ship"> haàhti-haàhte-3 vaaksi^vaahte- "foam" >vaaht i -vaahte- (Toivonen, Itkonen, and Joki 1958:269; Toivonen 1955:45; Hakulinen 1961 :53). 9 A solution like this is postulated, for example, by McCawley (1964: 25-26) and Wiik (1967:76-77). 10 In the standard language and in many dialects, this word was bor^ rowed as novssi rather than norsi. The form novssi has been unaffected by analogy.

There are also examples of words having the ksi-hte alternation which have taken on the i~e alternation, namely l a h t i , haahti , and vaahti (cf. footnote 8). 12 There is also one verb, lahte- (<*läkte) "to leave", that retains a past tense form ending in ksi. This alternation (lähte—läksi) is therefore parallel to the ksi-hte alternation found in yksi "one" and kaksi "two". Nonetheless, the form läski has in many dialects been replaced by the analogical past-tense form, lähti (Rapola 1966:235-36). 13 Although both sano- and haluna- (actually hallu- in the southwest dialects) now have similar past-tense forms (sanosi and halusi), their other forms remain unchanged. For instance, the infinitive form of sano- is sanno (which involves a gemination of the n und lacks a final t), while the infinitive form of hallu- is halut (which is ungeminated and has a final t). Thus the emergence of forms like sanosi cannot be explained as a reanalysis of sano-; this verb does not merge with verbs like hallu-, but remains distinct from them. (Cf. Kettunen 1930;1—16 for more details on the historical development and the morphological forms of the southwest dialects).

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 229 14 In each of these last three examples, the word-final i is histori­cally a nominal suffix (Hakulinen 1961:112-12; koti , for example, was created from kota "hut", and neiti is from neita. "virgin" (Toivonen et al. 1958:224-371). 15 For empirical evidence that Finnish speakers do not account for gradation by means of phonetically-statable rules, but rather by morphological rules, cf. Skousen 1972:45-57. 16 For example, Saussure (1959:79-100, 153-61, 183-90) clearly under­stood this fact and went to great length to convince other linguists of it. 17 Thus the term 'analogical phonological ruleT as used in Ohala 1973 is quite appropriate. 18 Cf. Harms 1972 for discussion on the status of universal phonetic rules in grammars. 19 Cf. Stampe 1972 for discussion on the acquisition of phonological processes. 20 Cf. the terms 'explanatory adequacy' and 'descriptive adequacy' as defined by Chomsky (1964:28-30).

REFERENCES

Chomsky, Noam. 1964. Current Issues in Linguistic Theory. The Hague: Mouton.

, and Morris Halle. 1968. The Sound Pattern of English. New York & London: Harper & Row.

Hakulinen, Lauri. 1961. Suomen kielen rakenne ja kehitys. 2nd ed. Hel­sinki: Otava.

Harms, Robert T. 1972. Some Non-rules of English. Austin, Tex.: Univ. of Texas at Austin, mimeo.

Itkonen, Erkki, and Aulis J. Joki. 1962-69. Suomen kielen etymologinen sanakirja. Parts III and IV. Helsinki: Suomalais-Ugrilainen Seura.

230 ROYAL SKOUSEN

Itkonen, Terho. 1965. Proto-Finnic Final Consonants. Part I. Helsinki: Suomalais-Ugrilainen Seura.

Johnson, C. Douglas. 1970. Formal Aspects of Phonological Description. (= Project on Linguistic Analysis, 2nd series, No.11.) Berkeley: Univ. of California.

Kenstowicz, Michael, and Charles W. Kisseberth. 1970. "Rule Ordering and the Asymmetry Hypothesis". Papers from the Sixth Regional Meet­ing of the Chicago Linguistic Society, 504-19. Chicago: Chicago Linguistic Society.

Kettunen, Lauri. 1930. Suomen murtet. Part II: Murrealueet. Helsinki: Suomalaisen Kirjallisuuden Seura.

Kiparsky, Paul. 1973. Abs tractness , Opacity, and Global Rules. Bloom­ington, Ind.: Indiana Univ. Linguistics Club, mimeo.

Kisseberth, Charles W. 1970. A Global Rule in Klamath Phonology. Urba­na, 111.: Univ. of Illinois at Urbana, mimeo.

. 1972. "Is Rule Ordering Necessary in Phonology?". Papers in Linguistics in Honor of Henry and Renée Kahane ed. by Braj B. Kachru, Robert B. Lees, et al., 418-41. Urbana, I11.: Univ. of Il­linois Press, 1973.

McCawley, James D. 1963. Finnish Noun Morphology. (= Quarterly Progress Report; Reasearch Laboratory of Electronics, Massachusetts Institute of Technology, 68.) Cambridge, Mass.: MIT.

. 1964. The Morphophonemics of the Finnish Noun. Cambridge, Mass.: MIT, mimeo.

. 1966. Further Revisions of Finnish Rules. Chicago: Univ. of Chicago; Dept. of Linguistics, mimeo.

. 1971. "Global Rules and Bangubangu Tone". Issues in Phono­logical Theory: Proceedings of the Urbana Conference on Phonology (Urbana, Ill., 1971) ed. by Michael Kenstowicz and Charles W. Kis­seberth, 160-68. The Hague: Mouton, 1973.

Ohala, John J. 1973. On the Design of Phonological Experiments. Berke­ley: Univ. of California, mimeo. [- Expanded version of a paper read at the Winter Meeting of the Linguistic Society of America, Atlanta, Georgia, 29 Dec. 1972.]

Penttilä, Aarni. 1963. Suomen kielioppi. Porvoo: W. Söderström. Postal, Paul M. 1968. Aspects of Phonological Theory. New York & Lon­

don: Harper & Row. Posti, Lauri. 1953. "From Pre-Finnic to Late Proto-Finnic". Finnisch-

Ugrische Forschungen 31.1-91.

ON THE NATURE OF MORPHOPHONEMIC ALTERNATION 231

Rapóla, Martti. 1966. Suomen kielen äännehis torian luennot. Helsinki: Suomalaisen Kirjallisuuden Seura.

Raun, Alo, and Andrus Saareste. 1965. Introduction to Estonian Lin­guistics, Wiesbaden: 0. Harrassowitz.

Saussure, Ferdinand de. 1959. Cours in General Linguistics. Ed. by Charles Bally and Albert Sechehaye, transl. by Wade Baskin. New York: Philosophical Library. (2nd ed., 1966.)

Schane, Sanford A. 1972. "Natural Rules in Phonology". Linguistic Change and Generative Theory ed. by Robert P. Stockwell and Ronald K. S. Macaulay, 199-229. Bloomington: Indiana Univ. Press.

Skousen, Royal. 1972. Substantive Evidence for Morphological and Phon­etic Regularities in Phonology. Ph.D. diss., Univ. of Illinois at Urbana. (Printed under the title Substantive Evidence in Phonology^ The Hague: Mouton, 1974.)

Stampe, David. 1972. A Dissertation on Natural Phonology. Ph.D. diss., Univ. of Chicago.

Toivonen, Y. H. 1955. Suomen kielen etymologinen sanakirja. Part I. Helsinki: Suomalais-Ugrilainen Seura.

, Erkki Itkonen, and Aulis J. Joki. 1958. Suomen kielen ety­mologinen sanakirja. Part II. Helsinki: Suomalais-Ugrilainen Seura.

Tuomi, Tuomo. 1972. Suomen kielen käänteissanakirja. Hämeenlinna: Suo­malaisen Kirjallisuuden Seura.

Wiik, Kalevi. 1967. Suomen kielen morfofonemiikkaa. Turku, Finland: Univ. of Turku, mimeo.

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE

DANNY D. STEINBERG and ROBERT K. KROHN

0.0 INTRODUCTION

In recent years the Chomsky and Halle analysis of English phonology has become the leading theory of the English sound system.* Besides hav­ing a profound effect upon linguistics, the theory is now influencing educators concerned with the teaching of reading and spelling (cf. Ward-haugh 1969; Carol Chomsky 1970) and the teaching of English phonology (e.g., Schane 1970). Despite the many compelling aspects of Chomsky and Halle's description of English, however, there is reason to believe that certain important features of their phonological analysis, along with the inferences they make from that analysis regarding the nature of English orthography and the nature of the reading process, may not be valid. The principal aim of the research to be reported here is to asses empiri­cally the validity of particular, crucial aspects of Chomsky and Halle's

* This research was supported by Office of Education contract OEC-O-71-0036 (508), project 1-0527, through the Pacific and Asian Linguistics Institute, University of Hawaii, Donald Topping, Director. Papers based on this article have been presented at the summer meet­ing of the Linguistic Society of America in Ann Arbor, Michigan, July 1973, at the annual convention of the/American Psychological Associa­tion in Montreal, August 1973, and at the XXth International Congress of Psychology in Tokyo, August 1972. - We would like to express our gratitude to Linda Kobayashi and Frederick Jackson for the testing of the subjects and their many helpful suggestions. We would also like to thank them and Elwood Mott for phonetically transcribing the pro­tocols. We are especially indebted to Ms. Kobayashi for her assis­tance in the planning of the experiments.

234 DANNY D. STEINBERG and ROBERT K. KROHN

phonological theory. The investigations of Chomsky and Halle (henceforth C&H) of 1968

into the sound system of English have led them to posit certain highly abstract underlying phonological representations for lexical items, and to posit a set of phonological rules which assign a phonetic represen­tation to these items. Some of the most important rules in the C&H sys­tem are those concerned with the vowel alternation of base and derived forms. Primarily because vowel alternations appear in a number of cases of such related words as divine-dignity, extreme-extremity, and grave-gravity, and because this relationship can be specified with a Vowel Shift Rule (VSR) and certain other rules, C&H claim that speakers of English have internalized a VSR and operate in accordance with it in the production and understanding of lexical items.

Such a rule as the VSR plays an extremely important role in the C&H system of phonology. Since the VSR is regarded by C&H as a general rule, it applies to any underlying phonological representation (UPR) of a lex­ical item having the requisite structural description, unless the item is marked as an exception. Underlying phonological representations are posited in order to accommodate the application of the VSR so that the expected phonetic representation will be generated. An invalid VSR would demand an extensive revision of a great many of the C&H underlying pho­nological forms.

According to the C&H analysis, phonetically different vowels in certain closely related words are derived from a common underlying ab­stract vowel. For example, the second vowels in the related word extreme and extremity are phonetically [I] and [e], respectively. The abstract representation of both of these vowels ist, however, the phoneme /ē/. In the case of extreme, the underlying /ē/ undergoes C&H's Diphthongization (ë → ) and then their Vowel Shift Rule ( → 1). In the case of ex­tremity, the underlying /ē/ undergoes a Taxing rule (ē → e). The pro­cesses for other such pairs of words, e.g., divine-divinity, sane-sanity, are similar. For all of these, c&H posit abstract underlying representations which undergo the same rules that apply to extreme-ex-

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 235

tvemity. Whether English speakers have actually internalized such a rule

as the VSR as C&H claim is questionable, especially since contrary evi­dence has been collected by some investigators. Robinson (1967), in an unpublished Ph.D. dissertation, found that graduate students of English literature produced alternations but that Grade 9 students did not.1 Unfortunately, in that study a group of non-language oriented adult speakers was not tested. More recently, Moskowitz (1972?), in a pilot study, reports that adult subjects (henceforth ss) rarely produce vowel alternation, as does Ohala (1973) in an unpublished paper. That so few experimental studies have been conducted to date concerning such an important hypothesis as the VSR is unfortunate. The present investiga­tion attempts to improve this situation with a thorough and systematic study of the productivity of vowel alternation. Given a meaningful sen­tence context, Ss were required to select one of two suffixes, e.g., -ic or -ity, attach it to a base word, e.g., maze, and then pronounce the novel derived form. If vowel alternation is a valid psychological phenomenon, we would expect Ss to produce a pronunciation of [mæzik] or [mæziti^]. A pronunciation of [me^zik] or [] would raise serious doubts as to the validity and generality of that phenomenon, and also of the VSR, since there would be no alternation to be accounted for. The cases of alternations already in the lexicon would be excep­tions which speakers may or may not deal with according to such a rule as the VSR.

In this research, two experiments were conducted. The first ex­periment presented materials auditorily only, while the second presented orthographic materials as well. Orthographic stimuli were included be­cause given that C&H contend that the orthographical representation of lexical items in English generally represents the underlying phonologi­cal forms of those items, one may well consider the possibility that English orthography may in some way affect ordinary speaker's pronunci-

We are grateful to John Carroll for informing us of this study.

236 DANNY D. STEINBERG and ROBERT K. KROHN

ation of the English vowels in derived forms. The effects of five dif­ferent base vowels [ā], [I], [e], [ōW], and [aw] and five different suffixes - i c , -ical,, - i f y , -ity and -ish are investigated in these ex­periments.

1.0 METHOD

1.1 EXPERIMENT I

Subjects, The Ss were 12 male and 12 female native English speak­ers who were randomly selected from introductory psychology classes at the University of Hawaii. Participation in the experiment fulfilled a course requirement.

Materials and Task. C&H's analysis predicts that certain vowels occurring in the final syllable of a word will change when a derivation of that base word is formed by the addition of certain suffixes. Five of the base vowels which the C&H theory predicts would change were selected for investigation. These critical base vowels and their postulated al­ternations in derived forms are: [ay] - [i] as in divine-divinity, [r^]-[e] as in extreme-extremity, [ēy] - [si], as in sane-sanity, [ow] - [a] as in verbose-verbosity, and [aw] - [] as in pronownce-pronwnciation. (The frequent, and significant, 'error' pronunciation [sic] is omitted.)

The five different suffixes selected for study were - i c , -ical, - i f y , -ity, and -ish. All but the suffix -ish are predicted by C&H to trigger vowel alternation in derived forms. The -ish suffix was included in the materials to see if it also would result in changed derived forms since we do have the example of the Spain-Spanish alternation in English.

There were 26 base form items used in the experiment. Five differ­ent ordinary English words were chosen as experimental items for each of the five different base form vowels thus providing a total of 25 items. One additional special item with ] in the base form, the name Gold­stein, was included at the suggestion of Charles-James N. Bailey (per­sonal communication), The C&H analysis predicts that the vowel in the second syllable would, with the suffix -ian, be realized as [i] in the derived form, Goldsteinian, as in reptile-reptilian.

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 237

The 26 base words and suffixes used in the experiment are shown in Table 1:

TABLE 1 EXPERIMENTAL ITEMS GROUPED BY CRITICAL BASE WORD VOWEL

BASEa SUFFIX CHOICESb BASE SUFFIX

CHOISES

[] - [æ ] [ōwl - fa]

maze (N) -ic *-ity trombone (N) -ic *-ity mundane (A) -ity *-ical overgrown (A) -ity *-ical drape (N) -ify *-ic stone (N) -ify *-ic snake (N) -ical *-ify honeycomb (N) -ical *-ify jade (N) -ish *-ity chrome (N) -ish *-ity

[īy] - [e] [aw] - [] centipede (N) -ic *-ity snout (N) -ic *-ity effete (A) -ity *-ical ground (A) -ity *-ical concrete (N) -ify *-ic house (N) -ify *-ic kerosene (N) -ical *-ify trout (N) -ical -ify Crete (N) -ish *-ity mouse (N) -ish --ity

[ay] - [i] sapphire (N) -ic *-ity snide (A) -ity *-ical termite (N) -ify *-ic tripe (N) -ical *-ify Goldstein (N) -ian *_ity quagmire (N) -ish *-ity

The phonetic symbols indicate the critical vowel of the base word and the C&H predicted vowel in the derived word, respectively. N = Noun, A = Adjective. The asterisk indicates the contextually inappropriate suffix choice.

238 DANNY D. STEINBERG and ROBERT K. KROHN

In this table the two suffix choices that were presented to the 5s with each base word are also shown. It should be noted that only one of the two suffix choices is contextually appropriate, and further, that for each of the five words with the same target base vowel, a different suffix is appropriate to the context provided. In the table, the inap­propriate suffix choice for the provided context is marked with an as­terisk. While only one of the two suffixes yields the appropriate part of speech for the sentence context, nevertheless, the creation of a derived form with either suffix is predicted by the C&H theory to re­sult in the same vowel change (except in the case of -ish).

The task of choosing between two suffixes was presented to 5s so that they might not unduly focus their attention on the pronunciation of the derived form which they were to create. 5s were instructed that the purpose of the research was to gather information concerning suffix preference.

The entire experiment was tape recorded and presented to the 5s wholly auditorily. The 26 items were arranged in a random order for presentation to the 5s. Each base word with its two suffix choices was introduced and presented to the 5s with a brief paragraph-like context. The last sentence in that context had a word deleted. The 5 was re­quired to say that sentence aloud, filling the blank with a derived word that was to be created by adding one of the two suffixes to the base word. The following is what 5s were presented for the item maze:

- The word is maze. A maze is a confusing path. Say maze.

- Ready? [CLICK - a signal to 5 to respond aloud] - One suffix is -ic. Say -i. - Ready? [CLICK] - Another suffix is -ity, Say -ity. - Ready? [CLICK] - [Ss were required to repeat the base word and the suffixes as a

check to determine whether the 5s actually did receive the in­tended stimuli and, further, to determine their pronunciation of the base word.]

- Fill the blank with the word maze plus either ~ic or -ity:

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 239

- The city library used to be a maze of shelves. People had diffi­culty finding their way out once they got in. Then a new librar­ian improved things by arranging the shelves around attractive reading areas. The library was no longer BLANK.

- The word is maze. The suffixes are -ic and -ity. The sentence is: The library was no longer BLANK.

- Ready? [CLICK] - [The S was required to say the whole sentence aloud with the newly

created derived form. ]

Prior to the presentation of any of the 26 experimental items, Ss were presented a sample item and a practice item. On the sample item, which was the base word piano with the suffix choices *-er and -ist, the S listened to someone (on tape) make the response (pianist). On the practice item, which was the base word astronomy with the suffix choices -er and *-ist, the S himself was required to make the response (astrono­mer). The purpose of the sample and practice items is to familiarize the S with the test procedure and the requirements of the task. Pianist ana astronomer were selected as the sample and practice items because neither provides the Ss with a psychological set either to vowel shift or not. However, since vowels are deleted in these items, such items serve to ready Ss for any eventuality.

The recorded experimental text consisted of five main sections: (1) Introductory Instructions, (2) Final Instructions, (3) Sample Item, (4) Practice Item, and (5) Experimental Items.

Procedure. Each S was tested individually and with the same ex­periment tape. The , a graduate student, tested all of the Ss. After being greeted by the E, the Ss were seated at a table on which there was a microphone. The E took a seat at a table nearby, out of the view of the 5, and played the experiment tape which ran about 40 minutes. All of the Ss' responses were recorded on tape. A brief post-experimen­tal interview was conducted to determine if any of the Ss were aware of the true intent of the experiment, the observation of their pronuncia­tion. None of the Ss indicated any such awareness.

Scoring. From the recorded tape of the Ss' responses, two scorers

240 DANNY D. STEINBERG and ROBERT K. KROHN

independently transcribed the Ss' pronunciation of each base word, suf­fix, and derived word. The transcriptions obtained from each scorer were later compared. Any differences were settled by having the scorers re­play, discuss and rescore the disputed items.

1.2. EXPERIMENT II

Subjects. The 5s were 8 male and 8 female native English speakers who were selected on the same basis as Ss in Experiment I. The Ss were placed in one of two groups, 8 to a group, with an equal number of males and females in each. The two groups of Ss are henceforth referred to as the Condition 1 and the Condition 2 5s.

Materials and Task. The materials and task were the same as that of Experiment I except for the addition of two types of supplemental materials, both of which were of an orthographic nature. Thus, Experi­ment IĪ Ss were presented materials visually as well as auditorily.

The Condition 1 Ss received one card on which the base form and the two suffix choices were printed. For the item maze, the following card was presented:

MAZE -IC -ITY

The Condition 2 5s received two cards. The first card was the same card received by the Condition 1 5s. On the second card, however, was printed the two possible derived words. For example, for the item maze, Condition 2 5s received the following two cards:

MA7F -IC MAZIC -ITY MAZITY

The spellings of the derived forms followed this rule: Delete any final silent e of the base word, and then add the suffix. Thus, maze plus - became mazic, and house plus —if became housify.

The recorded experimental text for Experiment II was the same as that for Experiment I except that the Final Instructions were varied to accommodate the supplemental presentation of cards. In those Final In-

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 241

structions, 5s were informed that they would also see cards with words and suffixes printed on them. 5s were given a set of printed cards and were asked to turn over a card whenever a new item was introduced. 5s under Condition 1 were required to turn over one card while those under Condition 2 had to turn over two.

Procedure. The same as Experiment I except that one additional E, also a graduate student, was used to test 5s.

Scoring. The same as Experiment I.

2.0. RESULTS

2.1. EXPERIMENT I

Valid Responses. Since each of the 24 5s made a response (the pronunciation of the created derived word) to each of the 26 experimen­tal items, a total of 624 responses were made in all. Of the 624 re­sponses, 504 responses (24 5s X 21 items) were for the 21 items for which the context appropriate suffix choice was - i c , -ity,.-ify, or -ical while 120 responses (24 5s X 5 items) were for the 5 items for which the context appropriate suffix choice was -ish. Since only one vowel change of any kind was produced out of all of the responses to -ish suffix items (an odd [āe] was the target vowel given by one 5 for quagmirish), the presentation of results will solely be concerned with the non-ish suffix items. Of the 504 responses made to the group of non-ish suffix items, 50 were discarded for various reasons leaving a total of 454 valid responses. The analysis of results, which is pre­sented following the section on discarded responses, will concern only the valid responses.

Discarded Responses. Responses were discarded if the derived word which was produced: 1) had an odd stress (7 cases), e.g., [grawn-diti], 2) had a syllable deleted (11 cases), e.g., [sneYkal], 3) had a syllable added (5 cases), e.g., [^], 4) had a disjuncture, e.g., [hәnikōwm - 1], 5) had a fabricated suffix (4 cases), for example,

242 DANNY D. STEINBERG and ROBERT K. KROHN

[snēYkebl], or 6) if the 5 gave no response (4 cases). A total of 41 responses were discarded according to these criteria. Approximately half of the discards (20) are attributable to three 5s. The other 21 discards were distributed over 10 other 5s.

In addition to these 41 discards, 9 responses of one 5 were dis­carded. All of these responses have a context inappropriate suffix af­fixed to the base word indicating perhaps that the 5 was not concentrat­ing sufficiently on his task. Such a large number of inappropriate suf­fix selections was unusual in the experiment. (None of these discards had a change predicted by the C&H theory.)

Vowel Changes. Only 12 responses (2.6% of all responses) exhi­bited the vowel change that is predicted by the C&H theory. While 34 other vowel changes did occur (7.5%), these were not changes predicted by the C&H theory. A total of 408 responses (89.9%) showed no changes in pronunciation between the critical vowel of the base word and the target vowel of the derived word. These findings are shown in Table 2:

TABLE 2 SUMMARY OF RESPONSES BY EXPERIMENT AND NATURE OF CHANGE

EXPERIMENT ' No. 5s C&H Other None Total I 24 12 34 408 454

% 2.6 7.5 89.9 100.0 II Cond 1 8 7 3 155 165

% 4.2 1.8 93.9 100.0 II Cond 2 8 8 14 133 155

% 5.2 9.0 85.8 100.0 II Cond 1&2 16 15 17 288 320

% 4.7 5.3 90.0 100.0 I & II 40 27 51 696 774

% 3.5 6.6 89.9 100.0

In this table, "nature of change" indicates whether the target vowel in the derived form has changed in accord with the C&H theory [C&H), wheth­er the target vowel has changed but not according to the C&H theory Other), or whether no change has occurred at all None).

C&H Predicted Vowel Changes. The 12 responses which were made in

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 243

accord with the C&H theory were distributed over 11 Ss. Thus, less than half of the Ss produced a derived word with a C&H target vowel change, and only one S provided more than a single instance of that change. The exceptional S produced two C&H changes, both [] - [i] alternations.

It is interesting to note that 10 of the 12 predicted C&H responses occurred when the critical vowel [] appeared in the base word. Ss pro­duced the C&H predicted vowel [i] in the derived word responses for 3 of the 5 different items: sapphire (5 cases), tripe (3 cases), and Goldstein (2 cases). No C&H predicted vowel changes occurred in response to the items snide and termite. The other 2 responses which were predicted C&H changes occurred in the derived form of effete (predicted vowel [e] and snout (predicted vowel []).

Eon-cm Predicted Vowel Changes. The 34 non-C&H target vowel changes (Other) occurred with items having 4 of the 5 different criti­cal base vowels. No changes occurred for items having the critical vow­el [ōw] in the base word.

Over half (18) of the target vowel changes occurred in response to base items having the critical vowel [iy]. The data shows that 12 of the 18 changes for the [] base items appeared in response to one item, effete , and that in all cases the vowel produced in the derived form was [i]. That same target vowel [i] was the only one which appeared in the derived words for the other 6 items with the critical base vowel [], centipede (2 cases), conerete (2 cases) and kerosene (2 cases). A re­latively large number of responses (8) was also given in response to two items with the critical vowel [] in the base word. The items were sapphire (5) and Goldstein (3). Table 3 (upper half) lists the nature and the frequency of all of the non-C&H predicted changes and identi­fies the items to which such responses occurred:

244 DANNY D. STEINBERG and ROBERT K. KROHN

TABLE 3 FREQUENCY OF NON-C&H PREDICTED RESPONSES OTHER) BY CRITICAL BASE VOWEL FOR EXPERIMENTS I AND II.

Experiment I (Total = 34)

ALTERNATION ƒ ITEM Base Derived

ī i 18 effete (12), centipede (2), conerete (2), kerosene (2)

āy æ 3 sapphire (3) ā i 3 Goldstein (3) 1 sapphire

e 1 sapphire

āw a 5 snout (5) i 2 mundane (1), drape (1) ē iy 1 snake

Experiment II Condition 1 (Total = 3) 1 i 3 effete (2) kerosene (1)

Experiment II Condition 2 (Total = 14) iy i 8 effete (5), kerosene 2), conerete (1)

1 tripe

e 1 sapphire

ay æ 1 sapphire

aw a 1 house

i 1 snake

ī i arane

2.2. EXPERIMENT II Valid Responses. Since each of the 16 Ss made a response to each

of the 26 experimental items, a total of 416 responses were made in all. The 8 5s under each of the two conditions provided 208 responses. Of the

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 245

208 responses in each condition, 168 responses (8 Ss X 21 items) were to the 21 items for which -ish was not the context appropriate suffix choice while 40 responses (8 Ss X 5 items) were to the 5 items for which -ish was the context appropriate suffix choice. Since only one vowel change was produced (a C&H predicted vowel change, [i], for the item quagmire) out of all of the responses to -ish suffix items, the presentation of results will solely be concerned with the non-ish suf­fix items.

Of the 168 responses made to the non-ish suffix items of Condi­tion 1, 3 were discarded leaving a total of 165 valid responses. Of the 168 responses of Condition 2, 13 were discarded leaving a total of 155 valid responses. Discards were made according to the same criteria used in Experiment I. The analysis of results will concern only the valid responses.

2.2.1 CONDITION I

Only 7 responses (4.2%) exhibited the vowel change that is pre­dicted by the C&H theory. There were 3 responses (1.8%) which were vow­el changes not predicted by the C&H theory. A total of 155 responses (93.9%) showed no change in pronunciation between the critical vowel of the base word and the target vowel of the derived word.

The 7 responses made in accord with the C&H theory were made by three Ss, mainly in response to base words with the critical vowel [ā] (sapphire, termite, and Goldstein). The 3 Other vowel changes which oc­curred, all occurred in response to base items with the critical vowel [ī]. The 3 responses (1.8% of all responses) were made by 3 different Ss. Two of the responses were to the item effete, one was to kerosene. In all cases it was the [īy] - [i] alternation which occurred.

2.2.2 CONDITION 2

Only 8 responses (5.2%) exhibited the vowel change predicted by the C&H theory. There were 14 responses (9.0%) which were vowel changes not predicted by the C&H theory. A total of 133 responses (85.8%) showed

246 DANNY D. STEINBERG and ROBERT K. KROHN

no change in pronunciation between the critical vowel of the base word and the target vowel of the derived word. A comparison of the distribu­tions of responses for Condition 1 and Condition 2 on the basis of the C&H, Other and None categories shows a statistically significant dif­ference, X2 = 8.63, p < .02. This effect is primarily due to the rela­tive number of Other responses for the Conditions. Table 2 shows the distribution of responses for these Conditions and that for Conditions 1 and 2 combined, for Experiment I, and the distribution for Experiments I and II combined.

The 8 C&H predicted vowel changes occurred in response to two cri­tical base word vowels, [ay] sapphire, termite, snide) and [ow] trom­bone, honeycomb). These responses were distributed over 5 Ss. The 14 Other vowel changes which occurred were distributed over 7 Ss. Table 3 (lower section) lists the frequency and kind of change, and identifies the items to which such responses occurred. Of the 14 changes, 8 oc­curred in response to base items with the critical vowel ī], and 3 to items with vowel [], 2 to [], and 1 to [āw]. 5 of the 8 [īy] responses were made to the item effete, and all three of the [] re­sponses were made to the item sapphire.

2.3. EXPERIMENTS I & II COMBINED

The distribution of responses for both Experiments I and II may be combined to provide an overall assessment of effects; especially since the difference between the distribution of responses for Experiment I and for Experiment II combined (see data in Table 2) is not significant, X2 = 3.56, df = 2.

Vowel Changes. Of the total of 774 responses, 27 (3.5%) are vow­el changes in accord with the C&H theory and 51 (6.6%) are changed not in accord with that theory. There were 696 responses (89.9%) which showed no change whatsoever. Both the number of C&H and Other responses are significantly less than the number of None responses. For the dif­ference between C&H and None-, X2 - 619.03, p < .001, and for Other and" None, X2 = 556.93, p < .001. The trend of no change in the vowels of

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 247

the derived words is clearly the predominant one. A tabulation of the frequency of target vowel changes in the de­

rived words by critical base vowel and nature of change is shown in Table 4:

EXPERIMENTS I & II COMBINED. RESPONSES BY CRITICAL BASE VOWEL, BASE WORD AND

NATURE OF CHANGE, N = 40

Base Suffix cm Other None Total

maze ic 0 0 38 38 mundane ity 0 1 36 37 drape ify 0 2 38 40 snake ical 0 2 32 34

0 5 144 149 % 0. 0 3. 4 96. 6 100.0

centipede ic 0 2 34 36 effete ity 2 19 14 35 concrete ify 0 3 31 34 kerosene ical 0

2

5

29

33 38 ical 0

2

5

29 122 143 % 1. 4 20. 3 78. 3 100.0

trombone ic 3 0 37 40 overgrozjn ity 0 0 37 37 stone ify 0 0 40 40 honeycomb ical 1

4

0

0

34 35 ical 1

4

0

0 148 152 % 2. 6 0. 0 97. 4 100.0

248 DANNY D . STEINBERG and ROBERT K, KROHN

Base Suffix cm Other None Total

snout ic i 5 32 38 ground ity 0 0 36 36 house ify 0 1 39 40 trout ical 0 0 38 38

1 6 145 152 % .7 3.9 95.4 100.0

sapphtre ic 9 7 24 40 snide ity 1 0 31 32 termite ify 3 0 29 32 tripe ical 4 1 31 36 Goldstein ian 3

20

3

1 1

32 38 ian 3

20

3

1 1 147 178 % 11.2 6.2 82.6 100.0

Grand Total 27 51 696 774 Percentage 3.5 6.6 89.9 100.0

The suffix which is listed is the contextually appropriate one for the base item. In reading the table, the results for the base item mundane , for example, indicate that regarding the pronunciation of the target vowel for the derived form (mundanity): 36 of the 37 Ss did not change their pronunciation, 1 S changed in a way not predicted by c&H, and no S changed in accord with the C&H theory.

C&H Predicted Vowel Changes. Of the 27 responses predicted by the C&H theory, 20 were given in response to base items with the vowel [āyj. The frequency of 20 [] responses is significantly greater than that for any of the other base vowels. For the difference between the zero [ey] responses, X2 = 20.00, p < .001, between the 1 [āw] response, X2 = 17.19, p < .001, between the 2 [īy] responses, X2 = 14.73, p < .001, and between the 4 [ōw] responses, X2 = 10,67, p < .01.

Of the 20 [ā] responses, 9 were given to sapphire, 4 to tripe, 3

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 249

each to Goldstein and termite and 1 was given to snide. The largest difference, that between sapphire and snide is significant, X2 = 6.40, p < .02. All other differences are not significant.

Non-C&H Predicted Vowel Changes. Of the 51 Other vowel change responses, 29 were given in response to base items with the vowel [īy]. The 29 [īy] responses is significantly greater than the frequencies for any of the other base vowels. For the difference between the zero [ōw] responses, X2 = 29.00, p < .001, between the 5 [ēy] responses, X2 = 15.11, p < .001, between the 6 [āw] responses, X2 = 8.11, p < .001, and between the 11 [ā] responses, X = 8.11, p < .01. The frequency of 11 [āy] responses, of 6 [āw] responses, and of 5 [ēy] responses is each significantly higher than the frequency of zero [ōw] responses, where X2 = 11.00, p < .001, X2 = 6.00, p < .02, and X2 = 5.00, p < .05, re­spectively. No other difference is significant.

Suffix Differences. A summary of the target vowel changes in de­rived words by context appropriate suffix and critical base vowel for C&H changes and Other changes is shown in Table 5. With regard to the C&H changes, the -ic suffix total is highest with a frequency of 13.

TABLE 5 EXPERIMENTS I & II COMBINED.

C&H AND OTHER RESPONSES BY CRITICAL BASE VOWEL AND SUFFIX. CHOMSKY & HALLE CHANGES

Base Suffix

Vowel ic ity ify ical ian Total ēy -- īy 2 — 2 3 1 — 4 -w a 1 -- 1

Total

9

13

1

3

3

3

4 3 20

5 3 27

250 DANNY D. STEINBERG and ROBERT K. KROHN

OTHER CHANGES

Base Suffix Vowel ic ity ify ical ian Total ēy 0 1 2 2 — 5 īy 2 19 3 5 — 29 -w 0 0 0 0 — 0 -w 5 0 1 0 — 6 ây

Total

7

14

0

20

0

6

1 3 11

8 3 51

While the difference between -io (13) and -ical (5) is not significant, the difference between 13 -io and the 3 -ity and 3 -ify totals is sig-

2 nificant, X = 6.25, p < .02 in both cases. The significant suffix differences apparently are not due to an

effect of the -io suffix alone because most of the -io responses oc­curred mainly in conjunction with one base vowel, [āy]. The frequency of [ay] base item responses is much higher than that of any of the other base item vowels. (The frequency of 17 [āy] base item responses is significantly higher than the zero responses for the base vowel [iy], X2 = 17.00, p < .001, than the 1 response for [iw] and [ēy], X 2= 14.22, p < .001 in both cases, and than the 3 responses for [ow], X = 9.80, p < .01) The significant suffix differences may, therefore, be due to an interaction effect of the -io suffix with the base vowel [ay]. However, because all 9 of the -io responses in the [ay] vowel category were in response to but a single item, sapphire (there was only this one experimental item which both had an [ay] critical base vowel and took an -io suffix), the possibility remains that the ob­served differences are due instead to the effect of some idiosyncratic feature of that particular word.

With regard to thé Other target vowel changes, the -ity and -io suffix items received the highest number of responses. The difference between the frequencies for the -ity (20) and the -io (14) suffix re-

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 251

sponses is not significant, as are the differences between the frequen­cies of the - i c a l (8) and the -ify (6) responses with that of the fre­quency of the -ic responses. However, the differences between the 20 -ity responses and both the 6 -ify and the 8 -ical responses are sig-nificant, X2 = 7.54, p < .01, and X2 = 5.14, p < .05, respectively.

Here, too, as was noted for the C&H predicted changes, the signif­icant suffix differences apparently are not likely due to the effect of certain suffixes alone, for, 19 of 20 -ity responses were made to but one base word with the vowel [iy], effete, and 12 of the 14 -ic responses were to two items sapphire (7) and snout (5). Again, the pos­sibility of an effect due to some idiosyncratic feature of the base word cannot be ruled out.

Sex Differences. No significant difference in the performance of males and females was found to obtain in any aspect of either experi­ment.

3.0. DISCUSSION

Validity of VSR and Allied Rules. The results show that the C&H predicted vowel alternation seldom occurs. The differences between and within experiments show no change in the critical vowel from the base to the derived forms for 90% of the responses. Overall, only 3.5% of the responses affirm the C&H prediction. It is interesting to note that 20 of the 27 C&H predicted vowel change responses were given to base words with the critical vowel [āy]. The item sapphire + ic pro­duced most of these changes. That only the [āy-i] alternation is pro­ductive to any extent indicates the possibility that a vowel specific rule is operating here for some individuals. It is also.worth noting that of the non-C&H predicted vowel changes, most were in response to one item with the critical vowel [] and the -ity suffix, to effete + ity. Again, it is possible that a vowel specific Taxing rule is oper­ating here for some individuals. Such a rule, if it were valid, would be one that operates on the phonetic representation of the base form to provide a corresponding lax vowel in the phonetic representation of

252 DANNY D. STEINBERG and ROBERT K. KROHN

the derived form. Besides clearly indicating that vowel alternation seldom occurs,

the findings show that laxing both in the Trisyllabic environment (base forms taking the -ify3 -ity, and -ical suffixes) and in the suffix-spe­cific environment (base forms taking the -io suffix), also seldom occurs Since C&H's VSR can be a general rule only to the extent that laxing is a general rule (according to C&H, the underlying representation must be laxed in order for the predicted derived form to be realized), further evidence against the generality of the VSR is provided. C&H's claim that the VSR is a psychologically real and general rule is therefore highly dubious.

C&H's contention that the VSR is a psychologically real and general rule is based on the judgment that derived forms with vowels that alter­nate are regular while forms which do not alternate are irregular. (Which of course also assumes morphological derivation to be monosys-temic.) Thus, the derived forms obscene-obscenity, meter-metric, and nation-national are viewed as regular while obese-obesity, soene-soenic, phoneme-phonemic, vocation-vocational, and transformation-transforma­tional are viewed as irregular. Since the findings of this study indi­cate that vowel alternation is largely non-productive, it must be ad­mitted that such rules as the VSR account at best for exceptions, i.e., to the creative pattern, which we state here: there is no (productive) vowel change in derived forms. Thus, just as C&H have regarded verbs such as keep-kept as irregular but those such as seep-seeped as regular so, too, must derived forms with alternating vowels be regarded as ir­regular. Clearly, the productivity of a rule - the creative aspect of language use - is a deciding factor in the judging of regularity.

That the criterion of productivity is assigned so critical a role in the determination of the validity and generality of a rule such as the VSR should not be surprising. Productivity is essential for distin­guishing, as Maher (1971:72) puts it, between generative phonology (the creative generativity of living language) and etymology. Sapir (1921:140) cautions against being "misled by structural features which are mere

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE1S VOWEL SHIFT RULE 253

survivals of an older stage which have no productive life and do not enter into the unconscious patterning", as does Marchand (1969:5) who states: "Productivity of a derivative type therefore cannot be over­looked in a correct description of a linguistic system, and the lin­guist who neglects this particular factor [productivity] will be count­ing 'dead souls' as live people".

Because vowel alternation in derived forms seldom occurred, there is some question as to whether alternations already in the language are to be accounted for by rule at all. It is possible that no rule is in­volved and that simply representations of both the base and the derived forms are what is listed in the lexicon. (See Steinberg 1973 and Hsieh 1972 for details concerning such a proposal.)

However, allowing that speakers may have such rules as the VSR to account for vowel alternations, it must be admitted that these rules account only for exceptions, i.e., exceptions to the creative pattern of no vowel change in derived forms. Since such rules are based on a non-productive exceptional phenomenon, there is little basis for the C&H claim that these rules are 'general'. To complicate the whole of English phonology simply to accommodate some exceptions is hardly jus­tifiable. If rules are to be used in the generation of these exceptional lexical items, it would seem far more reasonable to mark such items in the lexicon to undergo special rules which would generate the alterna­tions. (See Krohn 1972a and 1972b for such a proposal.)

Perhaps a word should be said about some objections which might be raised with respect to our conclusions. If, for example, one wished to argue that the VSR and other allied rules do not operate for the novel derived forms produced in this experiment because such a form, e.g., mazic is not regarded as a meaningful whole word by 5s, then it should be pointed out that this study provides two pieces of evidence to the contrary: (1) Ss generally selected the context appropriate suffix. On­ly 18 context inappropriate choices were made in the entire study. (2) Ss generally assigned normal stress to novel derived forms. They even shifted primary stress from the initial syllable of all multisyllabic

254 DANNY D. STEINBERG and ROBERT K. KROHN

base forms to the pre-suffix syllable in the derived form. Thus, honey­

comb, quagmire, Goldstein, sápphire, kérosene, concrete, and céntipede

which received primary stress on the first syllable, had their stress shifted to the last syllable before the suffix, in their derived forms honeycômbical, quagmirish, Goldsteinian, sapphîric, kerosénical, con-

crêtify, and centipèdic. Such a shift, by the way, is predicted by Hal­le & Keyser's (1970) Main Stress Rule of English. In the very few cases where such a shift did not occur, other errors, most commonly the loss of one or more syllables (e.g., térmify) were also involved; even in these cases the Main Stress Rule appears to be operating. Such evidence strongly indicates that 5s did regard the novel derived forms as mean­ingful whole words.

Another possible objection, one that might be raised by proponents of the C&H analysis is that such an analysis can account for the results of this investigation by taking into account boundary markers.2 It could be said that the alternation or nonalternation of vowels is sim­ply a matter of whether a (non-formative) word boundary (#) appears be­tween the base form and the suffix whenever the Laxing Rule is supposed to operate.3 It might be argued that because novel derived forms such as mundan#ity are not already in the S's lexicon, such forms would not be subject to the (not formalized) C&H rule that changes a # boundary to a + (formative) boundary,4 and that since the Laxing Rule operates

2 We are indebted to Frederick Jackson for his originality in artic­ulating this argument.

3 According to C&H (1968:368) the # boundary is one that is "automat­ically inserted at the beginning and end of every string dominated by a major category, i.e.., by one of the lexical categories 'noun', 'verb', 'adjective1, or by a category such as 'sentence', 'noun phrase', 'verb phrase', which dominates a lexical category".

4 Unless it can be shown that it has independent support, the rule is subject to the criticism of being ad hoc, i.e., of being moti­vated solely by the desire to get the derivation to come out right, or by the need to protect the C&H analysis, from experimental veri­fication of falsification.

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 255

on the base form plus suffix when a + boundary is present and not when a # boundary is present, then if the ë boundary is not removed for novel derived forms, the application of the Laxing Rule would be blocked. In such a case the resulting derived forms would not alternate but would have the same vowel as the base form. Such an outcome would be in accord with the findings of this study and at the same time would preserve the validity of the VSR, Laxing Rule, etc. This solution might be thought to be a viable one until it is realized that the Main Stress Rule would al­so be blocked from shifting the primary stress in derived items by the presence of the # boundary. Since, as was noted previously, stress did indeed shift as predicted in the experiments, it then appears that the # boundary was removed by the Ss. If that is the case, and the Laxing Rule (which is ordered after the Main Stress Rule) could have applied, why didn't alternation occur? Again, it seems necessary to conclude that the set of rules which C&H posit to account for vowel alternation is not a valid one.

Validity of the C&H UPRs. In the C&H analysis, the rules governing the vowel alternation phenomenon provide a major part of the link be­tween UPRs and the corresponding phonetic representations. Without the VSR and allied rules, lexical items with tense vowels in their phonetic representation, vowels such as [īy, ey, āy, ōy, āw, uw, o w ] , cannot be generated from the C&H UPRs. Consequently, the finding that C&H's VSR is virtually nonproductive and thus cannot be a general rule of English renders invalid most of their underlying phonological representations for lexical items. What is required therefore is the postulation of UPRs that are considerably less abstract, i.e. closer to the phonetic level of representation, than they are in the C&H analysis. Proposals per­taining to such representations have been offered by Krohn (1972b) and Steinberg (1973).

UPRs and Dialect Variation. The necessity for a major revision of C&H's UPRs renders less credible their rather extravagent claim (Chomsky & Halle 1968:49) that, "It is a widely confirmed empirical fact that

256 DANNY D. STEINBERG and ROBERT K. KROHN

underlying representations are fairly resistant to historical change, which tends, by and large, to involve late phonetic rules". On the basis of this unfounded claim C. Chomsky (1970:295) asserts that a UPR based orthography would be adequate for "both British and American English, and the vast range of English dialects that exist within each country and around the world". Despite the C&H assertion that it is a "widely confirmed empirical fact" that UPRs are resistant to historical change (one which Kiparsky (1968:187) does not share since he postulates dif­ferent UPRs for two closely related Swiss-German dialects), this study offers reason to believe that this view is erroneous. The invalidation of the VSR as a general rule, with the consequence that underlying forms must be represented at a level closer to the phonetic level, leads one to expect that the UPRs of lexical items will vary considerably from dialect area to dialect area.

Optimality of Current English Orthography. C&H maintain that cur­rent English orthography is nearly optimal (a claim which raises unan­swered questions about pre-Websterian spelling according to the indi­vidual's own lights). For C&H this means that the orthography is "rath­er close" to the UPR (Chomsky & Halle 1968:184 note). According to this view, the alternating vowels of, say, extreme-extremity are not repre­sented by separate symbols in the orthography since their different phonetical realizations are accounted for by general rules. However, since according to the experimental evidence the rules posited by C&H to account for the vowel alternations are not general rules, and since most of C&H's UPRs of lexical items are of dubious validity, the C&H claim about English spelling being optimal is without foundation. No orthography based on C&H's UPRs of lexical items could be optimal (ac­cording to C&H's notion of optimality), for such UPRs do not represent a phonological level that is psychologically real for English speakers (see Steinberg 1973 for further details).

The C&H View of the Speech and Reading Processes. In what is es­sentially an elaboration of the C&H position on language performance, C. Chomsky (1970:291) comments as follows: ".. In the course of ac-

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE'S VOWEL SHIFT RULE 257

quiring his language he [a speaker] has internalized the rules of his phonological system, and as a mature speaker he operates in accordance with them both in speaking and in comprehending the spoken language". She goes on (291-92) to discuss the suitability of current English or­thography for reading: "Consider ... the common items of words such as courage/couvage-ous, or anx i -ous /anx i -e t y , or photograph/photograph-y/

photograph-. Although the phonetic variations are considerable, they are perfectly automatic, and the lexical spellings can ignore them. They will be introduced by the phonetical component. Of course, the conventional orthography ignores them as well. These are good examples of cases where the conventional orthography, by corresponding to lexical spelling rather than phonetic representation, permits immediate direct identification of the lexical item in question, without requiring the readers to abstract away from the phonetic details, and presents the lexical item directly, as it were."

According to this view, the conventional English spelling of lex­ical items, which is close to the C&H UPRs (if it does not directly give them), facilitates the reading process because it permits a reader to recover the meanings of the lexical items rather directly. Conven­tional orthographical representations are thought to provide an input to an internalized underlying representation, thereby obviating the need to use phonological rules in the recovery of meaning. However, given the dubious validity of the C&H VSR and UPRs, Halle and the Chomskys' views on how English spelling facilitates reading seemhighly implausible.

Implications for Teaching Phonology and Reading. Because the learner must first know the phonological rules which relate phonetic representations to UPRs before he can learn a C&H UPR-based orthography, children would ordinarily be halfway through grade school by the time they would be ready to begin to master such an orthography. For, ac­cording to Halle and the Chomskys, " ... full knowledge of the sound system that would correspond to the [UPR] orthography is not yet pos­sessed by the child of six or seven, and may indeed be acquired fairly

258 DANNY D. STEINBERG and ROBERT K. KROHN

late." (Cf. C. Chomsky 1970:301) . Chomsky's solution to this problem of late acquisition of phonological rules is to accelerate the child's normal rate of language acquisition by teaching them more lexical items. With regard to the acquisition of the VSR, she advocates the teaching of a rather sophisticated vocabulary in the early grades. She proposes that "Extending the child's vocabulary to include Latinate forms and polysyllabic derived forms is one of the best ways to provide him with the means of constructing the phonological system of his lan­guage more fully as he matures. He ought to become familiar with word groups such as industry-industrial, major-majority, histor-his torical-

historian, wide-width, sign-signature, etc., and have their relation­ships made explicit for him." (1970:302). However, given the question­able validity of Chomsky and Halle's VSR, other allied rules, and UPRs, it is evident that educators need not concern themselves with the prob­lem of having students acquire such rules and representations. Propos­als of teaching materials and techniques which are based on such as­pects of the C&H phonological analysis of English are clearly not well motivated.

REFERENCES Chomsky, Carol. 1970. "Reading, Writing, and Phonology". Harvard Edu­

cational Review 4.287-309. Chomsky, Noam. 1970. "Phonology and Reading". Basic Studies in Reading

ed. by Harry Levin and Joanna P. Williams, 3-18. New York & London: Harper & Row.

Chomsky, Noam, and Morris Halle. 1968. The Sound Pattern of English. New York & London: Harper & Row.

Halle, Morris, and Samuel Jay Keyser. 1971. English Stress. New York & London: Harper & Row.

Hsieh, Hsin-I. 1972. On Listing Phonological Surface Forms in the Lex­icon. Honolulu: Univ. of Hawaii, Dept. of Asian Languages, mimeo. (Dec. 1972). [Revised version in the present volume.]

THE PSYCHOLOGICAL VALIDITY OF CHOMSKY AND HALLE's VOWEL SHIFT RULE 259

Kiparsky, Paul. 1968. "Linguistic Universals and Linguistic Change". Universals in Linguistic Theory ed. by Emmon Bach and Robert T. Harms, 171-202. New York: Holt, Rinehart & Winston.

Klima, Edward S. 1972. "How Alphabets Might Reflect Language". Language by Ear and by Eye ed. by James F. Kavanagh and Ignatius G. Mattingly, 57-80. Cambridge, Mass.: MIT Press.

Krohn, Robert K. 1972a. "The Vowel Shift Rule and its Productivity". Language Sciences 20.16-18.

. 1972b. "Underlying Vowels in Modern English". Glossa 6:2. 203-24.

Maher, J. Peter. 1971. "Etymology and Generative Phonology in Tradi­tional Lexicon". General Linguistics 11:2.71-98.

Marchand, Hans. 1969. The Categories and Types of Present-Day English Word-Formation: A synchronic-diachronic approach. 2nd rev. and enl. ed. Munich: H. C. Beck.

Moskowitz, Arlene I. 1972? On the Status of the Vowel Shift in English. Unpublished MS.

Ohala, John J. 1973. On the Design of Phonological Experiments. Berke­ley: Univ. of California; Dept. of Linguistics, mimeo.

Robinson, Jane. 1967. The Development of Certain Pronunciation Skills in the Case of Suffixed Words. Diss., Cambridge, Mass.: Harvard Univ.

Sapir, Edward. 1921. Language: An introduction to the study of speech. New York: Harcourt, Brace & World. [Quoted after the 1949 Harvest Books paperback edition.]

Schane, Sanford A. 1970. "Linguistics, Spelling, and Pronunciation". TESOL Quarterly 4:2.137-42.

Steinberg, Danny D. 1973. "Reading, Phonology, and Chomsky & Halle's Optimal Orthography". Journal of Psycholinguistic Research 2:3.239-to 258.

Wardhaugh, Ronald. 1969. Reading: A linguistic perspective. New York: Harcourt, Brace & World.

III. L I N G U I S T I C T H E O R Y A N D P H I L O S O P H Y OF L A N G U A G E

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE

RAIMO ANTTILA

1. THE STARTING POINT. Linguistic literature is dotted with at­tacks against transformational-generative grammar that are often more colorful than correct in their arguments, though even such loose argu­mentation might be a legitimate answer to "clearly ...".* In this pa­per I take up a few aspects of the theory of language and its epistemo­logical foundations that have seriously been neglected by the transfor­mational school.1 The three main aspects to be treated are the human mind, social function, and evolution, the last two expressly denied by Chomsky, the first one inadequately treated by him. Without document­ing the transformational side (to keep the bibliography to size) I want to give emphasis to those currents which have been kept out of the main­stream of recent linguistic debate. In addition to the history of lin­guistics this entails the work of Walburga von Raffler-Engel, J. Peter Maher, Henning Andersen, Esa Itkonen, and others. One objective of this paper is to show how many of those who leave the transformational camp step into a traditional pen without knowing it.2

The mental apparatus essential for language comes out rather clear­ly in current studies of semantic acquisition; thus we can start from

I thank Kay Bikson for background and discussion in rationalism and empiricism, Breyne A. Moskowitz in semantic acquisition, and Michael Shapiro in Peircean philosophy.

1 I combine here the arguments of Anttila 1973a and 1973f in which ful­ler bibliographies are available.

2 I have treated this topic in more detail in Anttila 1973c and 1973d.

264 RAIMO ANTTILA

there and then proceed to other areas. Chomsky has assigned a crucial role to a language acquisition de­

vice (henceforth LAD) for the acquisition of all knowledge. This sounds very good from the point of view of language learning, until we realize that the approach is very arbitrary in many respects: 1) separation of this device from culture (see Raffler-Engel 1970b); 2) separation of intelligence from nature; 3) arbitrary division between input and out­put; 4) the exclusivity of a competence vs. performance dichotomy; 5) the different evaluation accorded to base forms and transformations; 6) the separation of syntax and semantics; and many others. A partic­ularly grave fault of the cybernetic model of this LAD is the abuse known as the 'black-box', i.e., the use of undefined explañantes; this 'model' is thus a diagrammatic petitio principii.

All this is coupled with a strong Cartesian principle that one can trust one's own mind only and not other people's minds, although there was a period in transformational grammar during which Chomsky's 'clear and distinct' Cartesian ideas were laws for others as well. Yet, as Esa Itkonen (1974b) has pointed out, Wittgenstein shows the impossibil­ity of a private language; thus, such a Cartesian/Chomskyan private mind could not be used in society at all. Indeed, we need a true nat­ural rationalism rather than a Chomskyan 'antimental rationalism', be­cause in studying language acquisition we want to penetrate other minds, other minds that do not possess language in its final form.

2. SEMANTIC ACQUISITION. Eve Clark's study of semantic acquisition of 1971 takes perception as the possible source for semantic features. The first semantic features that the child uses are liable to be derived from the encoding of his percepts (A). At a later stage, as he learns more about his language, he will learn which percept-derived features play a linguistic role and which are redundant (I). As Clark points out, nothing is said about what the child perceives. Over-extension is then the phenomenon that tells us to know how the child's semantic field is different from the adult's (D). The initial over-extensions are based on some of the perceived characteristics of the object around the child,

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 265

and thus the features used in the early stages will be related to fea­tures in perceptual development:

Thus, the perceptual features themselves may well belong to the set of universal semantic primitives postulated by Bierwisch. While a great deal of research is obviously needed before we can make such claims conclusive, what data we have clearly point in that direction. (Clark)

The development of perception is crucial here, and Clark can establish a hierarchy of features used: movement, shape, size, sound, taste, tex­ture (but not color) (E).

It is the phenomenon of over-extension that supports Clark's Se­mantic Feature Hypothesis as outlined above. The quotation shows close similarity to the Universal Primitives Hypothesis:

This then leads to the extremely far-reaching, though plausible, hypothesis that all semantic structures might finally be reduced to components representing the dispositions of the cognitive and perceptual structure of the human organism. According to this hy­pothesis semantic features cannot be different from language to language, but are rather part of the general human capacity for language, forming a universal inventory used in particular ways by individual languages (Bierwisch 1970:181-82; quoted also by Clark).

Such semantic components were long in the making, but let us assign the notion to Bierwisch for the sake of the present argument.

Also Slobin's recent article (1971) on developmental psycholin-guistics is an attempt to explore the cognitive prerequisites of lan­guage learning or grammar construction. Although some of his princi­ples, universals, and strategies must be questioned, one can again ac­knowledge his plea for a language - free acquisition model and the pri­macy of cognitive development. This development promises an eventual universal theory of the growth of the mind (Slobin 1971:301). The whole acquisition process no doubt requires a richly structured and active child mind (ibid., 367-70).

3. PERCEPTUAL S Y S T E M S . Even linguists who started out in the straitjacket of transformational grammar are coming more and more to realize that you cannot just push around mere grammatical rules and

266 RAIMO ANTTILA

formalisms and call it 'explanation'. Thus Bever and Langendoen (1971: 451) demonstrate that nothing is really free of behavioral effects:

The main burden of this paper has been to point out that language learning and linguistic evolution are not merely the learning and evolution of grammatical structure, but also of the perceptual and productive systems for speech behavior. The novel structures that the child recognizes as relevant motivation for restructuring his grammar must be sentences he can (at least partially) understand, desire to say, and learn from.

The program is laudable, but this is not a new insight, as implied, but one of the solid traditional themes which is being ignored, in spite of the fact that many have defended such common sense all along (e.g., Raffler-Engel 1970b). According to Bever and Langendoen (1971:434-35) there is an independent perceptual mechanism that relates external' structures. Young children are heavily dependent on "perceptual strat­egies in speech perception, even to the point of overgeneralizing them to sentences to which they should not be applied" (p.437; cf. Clark above). The child can restructure his grammar by a minimal change from the grammar of the preceding stage. Such restructuring is possible only when the previous structure is comprehensible and implies grammatical structures that are close to the already learned structure. The defi­nitions of "minimal" and "close" are "of course" left open (Bever and Langendoen 1971:450). In principle Bever and Langendoen cannot get be­yond the traditional achievements (which they themselves are unaware of):

The question as to what triggers any particular linguistic change seems to us to be wildly premature. However, our claim that lin­guistic evolution is in part a function of the balance between learnability and perceptibility raises the possibility that cer­tain internal cultural developments can themselves motivate a linguistic shift, by changing what the language is used for (op. cit., p.454).

And: Such questions await further empirical and theoretical investiga­tion (456).

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 267

In short, we run into the traditional actuation problem, the total frame of traditional philology; and 'such questions' have indeed been waiting for a long time. I cannot understand why this article is generally hailed as a new theoretical insight, though it deserves credit for tackling a complicated syntactic problem, even if their analysis of this syntax is partially questionable. Their overall framework is solid - and totally traditional (cf. Anttila 1973c).

Another paper that investigates the parallels between language ac­quisition and a historical syntactic problem from English is Naomi Bar­on's of 1972. This is also actual testing rather than speculation; it shows a proper reverence for the past. What she actually was struggling with is the issue of diachronic correspondences vs. actual historical change (cf. Andersen 1972 and 1973): How to get from state A to B. De­scriptive typologies do not tell us how disambiguation takes place (cf. Bever and Langendoen 1971), a typology of processes must cope with it. Baron correctly criticizes Kiparsky's resurrection of the 19th-century notion of imperfect learning by children. This is the constant inter­ference from the adult point of view (as in the work of many others). One factor hampering the study of the two typologies has been the uni­versal assumption that children can only simplify grammars. This is now being rejected. Yet, as Baron (1972:47) noted, "we still lack explicit models of elaboration which would satisfy both students of language ac­quisition and diachronists. I do not yet possess such a model." At this point one can ignore the notion of 'elaboration', because below we will see that it is irrelevant in the explanatory model.

4. THE P E R I P H E R A L L Y OF SYNTAX. What these and similar studies show is that syntax is psycholinguistically a very shallow notion (as already seen by Breal more than two generations ago). Syntactic studies have also used older children as subjects, thus distorting the issue even further. Clark rejects as too syntax-based McNeill's (1968) Gram­matical Relations Hypothesis of semantic acquisition which goes from grammar to vocabulary and Anglin's (1970) Generalization Hypothesis where learning proceeds from concrete to abstract. She finds that wher-

268 RAIMO ANTTILA

ever there is a hierarchy, top features are learned first, in other words, initial features are abstract rather than concrete. This ques­tion of abstractness ties in now with other recurrent themes surveyed above: perception/cognition, general capacity for language, language-free acquisition model, growth of mind, minimal change, closeness, ac­quisition vs. change, and even 'imperfect learning'. These aspects are either the goals sought after or they are the dead ends where the studies must stop, or where they have stopped. The question is: how can one combine goals and dead ends? Is it at all possible?

I think that to a great extent such a task has already been car­ried out in essence already a century ago. Linguists are to be blamed for their own ignorance of the history of scholarship. This has led to rephrasing old achievements as new insights, and some of these old achievements go well beyond such 'current gains'.

5. ABDUCTION AND PERCEPTUAL JUDGEMENT. The solution to many of these problems is simply the observance of Charles S. Peirce's (1839-to 1914) theory of scientific method, above all his theory of abduction. I will briefly review the relevant notions here, based on Knight (1965: 117-23) and Reilly (1970:23-77; see also Anttila 1972:196-203).

All cognition involves inference for Peirce (Knight 1965:116), and all knowledge whatever comes from observation (Reilly 1970:25). The starting point of a reasonable inquiry is the percept and not the sense-impression (Reilly 28). The world of fact or experience makes brutal inroads from without into our ways of thinking (Reilly 29). Experience is a cognitive operation (p.30). There are three modes of reasoning or argument. The order of the logic of the syllogism is where one infers from the rule (major premise) and the case (minor premise) into the result (conclusion). This is deduction. Induction is inference with the order of the procedure reversed: we infer the rule from the case and the result. But the most common type of reasoning is abduction in which the rule and the result are given and we infer the case (Fig.1).

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 269

RULE All the nuts from this bag are cracked

CASE These nuts are from this bag

RESULT These nuts are cracked

Fig.l. The interrelation between deduction (D) , induction (I), and abduction (A) (after Peirce). Major premise = a; minor premise = b; and conclusion =

Abduction is everyday logic par excellence: The surprising fact, C, is observed; but if A were true, would be a matter of course. Hence there is a reason to suspect that A is true (5.189) (Knight 1965:117).

This is a living process of the mind (Rei 1ly 1970:31), extremely falli­ble, but it is man's most important asset. Every item of science came originally from such conjecture, which has only been pruned down by ex­perience. Abduction is an act of insight, coming to us in a flash. It is the first explanatory phase of scientific inquiry, it suggests that something may be; unlike the other modes of argument it introduces a new idea.

The scientific explanation suggested by abduction has two charac­teristics that must be pointed out...: 1) an explanatory hypothe­sis renders the observed facts necessary or highly probable; 2) an explanatory hypothesis deals with facts which are different from the facts to be explained, and are frequently not capable of being directly observable (Reilly 1970:35).

Abduction only suggests that something may be the case. But any under­standing or learning must go through such a suggestion. Abduction stands as the basis for predictions. Deduction infers those predictions, and induction tests them. Deduction proves that something must be (if cer­tain conditions are fulfilled), induction shows that something actually is.

Man's mind is akin to the rest of the cosmos (cf. Reilly 125-28); the role of the instinct in abduction proves the success in guessing right (Reilly 41):

270 RAIMO ANTTILA

Instinct is the tool of science only at the moment when the hypoth­esis to be tested is chosen from among several suggested hypotheses In making such a choice, instinct is a surer instrument than reason (5.445, 6.530) (Reilly 45).

There is a close connection between abduction and perceptual judgment. Abductive inference shades into perceptual judgment without any sharp line of demarcation, in fact the latter are extreme cases of the former. But perceptual judgment is not subject to criticism (cf. Reilly 46-47). Science is built up from perceptual judgments. Perceptual judgment and abductive inference share four important similarities: 1) Both contain elements of generality, 2) both are in some aspect beyond the control of reason, in that neither judgment is the necessary conclusion of an inference. 3) There is a newness or originality in both, and 4) both are interpretative (Reilly 47-50): "... perceptual judgment is an inter­pretation of a perceived object, and though there are several interpre­tations possible, the one actually adopted seems forced upon us" (Reilly 47). Moreover:

The perceptual judgment is interpretative because it is abstract. That is, it represents one or more features of the known object without exhausting the meaning of the object. The knowable aspect which is grasped in the perceptual judgment is only one of several (7.198). Nonetheless the perceptual judgment is true in the sense that "it is impossible to correct it, and in the fact that it only professes to consider one aspect of the percept" (5.568). It is for this reason that the interpretation made in the perceptual judgment is not the only one possible, and the aspect represented in it is really as given as it appears so forcefully to be (Reilly 50-51).

Induction is used for the process of verification, and it also corrects hypotheses. The revised hypothesis is a new abductive inference based on a richer experience (cf. Reilly 63-75), "although abduction is the only process of inference which is genuinely additive of ideas, the inductive process prepares the scientific for making new abductions" (Reilly 76).

The distinction between abduction and induction is crucial in the Peircean system, and it is not generally made in the philosophy of

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 271

science. It is absolutely necessary for the clarity we need, and for the explanation we need. In the question between the respective vir­tues of deduction and induction those who have stressed induction come out better than the deductionists (e.g., among language acquisition scholars: Raffler-Engel 1970a:183, Slobin 1971:334, Ervin 1964:168).

6. ABDUCTION AND ACQUISITION OF KNOWLEDGE. Peirce's theory of abduction provides the most comprehensive frame for studying the prob­lems mentioned above. Of course there remain difficulties, but at least one should utilize what Peirce has given, and then try to go beyond him. Peirce talks about a 'scientist', it is true, but since he is talking about natural perceptual logic we can easily, and legitimately, substi­tute 'child' for 'scientist', because such everyday fallible logic is clearest in children. In this respect children and adults differ only in the amount of experience that weeds out possible abductions. Much of Peirce's program also occurs among other philosophers. The most vehe­ment antirationalist, Hume, had to acknowledge natural relations, empir­ical laws of association. In fact, philosophers and linguists have al­ways stumbled when it has come to define similarity or resemblance. Peirce tackles it head on: since you cannot do away with it, why not make it central. Children provide the clearest evidence, because their experience provides minimal obstacles.

Peirce's theory of abduction gives exactly the language-free ac­quisition model that combines language learning and linguistic change. It gives further support to those linguists who have been arguing in favor of panchronic (or a-, meta-) linguistics (e.g., Raffler-Engel 1953:408-09; Maher 1970:30-31, and myself), for various other reasons. In fact, the theory has been explicitly revived for linguistic and cul­tural learning and linguistic and cultural change by Henning Andersen (1972, 1973). Here is the model that Baron among others is seeking. It is imperative that all linguists get acquainted with it. So far I have heard only two talks on the topics of acquisition and change in which Andersen's work has been referred to - in one sentence each. But I have myself applied it (Anttila 1972), as well as Peirce's theory of signs

272 RAIMO ANTTILA

which enabled me, among other things, to differentiate historical lin­guistics from comparative better than before.

The theory of abduction and perceptual judgment supports abstract-ness as primary. But note that in a way this is a very concrete opera­tion, and instinct, and the division may be arbitrary for the child. In this light Hume's particulars as generals, or Anglin's hypothesis, are not far off, after all. As Wescott (1969:118) has put it: man is an an­imal that specializes in generality. Abstraction is concrete for him. This aspect of abduction deserves serious attention, since it gives the most fruitful synthesis of mind and matter. Because of its importance we will have to return to it below in connection with the evolutionary aspect of language.

Chomsky himself has mentioned approvingly Peirce's attempts to develop a theory of abduction (1972 [1968]:90-96, 171-72). In fact, he stated:

The way in which I have been describing acquisition of knowledge of language calls to mind a very interesting and rather neglected lecture given by Charles Sanders Peirce more than fifty years ago, in which he developed some rather similar notions about acquisi­tion of knowledge in general (Chomsky 1972:90).

Mention of Peirce is perhaps the best part in this book, but Chomsky does not really get down to Peirce's essentials, and goes on to imply that not much was achieved: "Even today, this [the development of a theory of abduction] remains a task for the future" (p.92). Chomsky could have avoided his violations by studying Peirce more closely, and Andersen has shown what can be done. Chomsky is not convinced by Peirce's analogy between human abduction and animal instinct (91). In fact, this is to be expected since Chomsky's position can be charac­terized by the fact that he denies both nature and nurture. But note now that perception/abduction seems to have a strong biological connec­tion in the feature hierarchy established by Clark (E), which seems to follow the evolutionary scale. For me this is a convincing argument of abduction.

In spite of such superficial references to acquisition of knowledge

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 273

Chomsky's position, which has misled a number of linguists, has remained language-centered, though some of his followers would like to see the obvious, at least now and then; compare McNeill's (1968:29) statement:

However, the possibility is worth considering that very little of this is, in fact, a capacity for language. Much of what I have dis­cussed may actually be the linguistic manifestation of a very gen­eral, though still inborn, cognitive capacity - cognition is the process by which the mind gets knowledge. If this is true, then the study of how language is acquired may provide insight into the very basis of mental life. For if the capacity for language is a special case of more general cognitive capacity, it would follow that the latter must have all the universal properties of the former.

And to cite an oppositional view: It would seem that a more plausible candidate for the status of innate ideas would be not the principles of a universal grammar but the principles of a universal logic (Hook 1969:164-65).

7. ANDERSEN'S MODEL OF LEARNING AND CHANGE. Clark's total pro­gram is now seen as a clear manifestation of the Peircean trichotomy abduction-deduction-induction (A-D-I). First the child abduces some­thing, and this is completely covert, not visible or observable. The abduction surfaces only through deduction which infers the predictions suggested by abduction (over-extension). In this way deduction is al­ways an experiment. The inductive test is whether other speakers accept such an output. If they do, the general rule has been verified. The two grammars, the newly inferred one and the one represented by tradition, are the same for all practical needs. But they may of course be totally different in structure. And here lies an important impetus for change (cf. Anttila 1972:197). This is Andersen's theory, and as he has shown, it is the same for all learning: semantic, cultural, phonetic, and so on, as may be shown by means of a diagram:3

Andersen has presented this diagram (even if not exactly in this outer shape) at numerous lectures and discussions. Its basic correct­ness can be inferred also from the fact that it is easy to adopt. At the Bologna Congress in 1972, Elizabeth Traugott presented it as her own, even when Andersen himself was in the audience. One should be more careful to place credit where it belongs, since cases of return to old ideas like this are invariably interpreted as new break­throughs of the transformational-generative paradigm.

274 RAIMO ANTTILA

Learner

Universals

Grammar Grammar

Output 1 Output 2

Fig.2. The crucial link in the language-learning process (after Andersen [1973:778]).

Again, the right observation has been made here and there, even though not followed up, compare McNeill (1968:28).

How do children acquire transformations? Unfortunately, there is no definite answer. However, one view is that the process takes place in the same manner as scientific inference. On the basis of their capacity for language, children formulate hypotheses about regularities observed in parental speech. Each hypothesis is eval­uated against further evidence, such as additional parental speech and parental reactions to a child1s own speech. In pursuing this empirical programme, children may even perform linguistic experi­ments, the equivalent in most respects of the experiments conducted in scientific laboratories.

Credit must go also to Edgar Howard Sturtevant (1875-1952), a neglected giant of American linguistics. Sturtevant presented reinterpretation phenomena ('folk etymology') in exactly the frame I have been pleading for here. When his little son underwent treatment of the ear by irriga­tion with warm water, he made the 'logical' guess (abduction) that ear and irr-igate go together. We know this only through a further deduc­tive derivation. When his nose was treated the same way he called the verb nosigate. Similarly a child who saw four airplanes and learned that is was a formation, connected the first syllable abductively with four. This surfaced when two more planes was called a twomation (Stur­tevant 1947:97-98; cf. Anttila 1972:93). Normally inductive testing

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 275

eliminates such forms, but if they happen to survive the attrition of such experience, analogic change results. Adults need not act different­ly: The women's liberation movement has tried to replace history by herstory and boycott by girlcott.

Many linguists, actually contemporary with Peirce, could be cited for proper understanding of the situation irrespective of the terminol­ogy used. I single out another neglected giant, the Frenchman Michel Breal (1832-1915). He pleaded for a logic of the ordinary people. Only that can explain linguistic change and language acquisition, arguing:

We have only to listen to the conversation of people who are but imperfectly acquainted with a language and to take note of the mistakes which they make, to see that it is by association of this kind that they are usually influenced.

We are now speaking of a rule not yet formulated; a rule at which mankind strives to guess, and which we see children trying to dis­cover. By pre-supposing its existence, the people actually creates it (Breal 1964[1897] : 72).

This is a par t i cu la r l y clear statement about human rule-forming and rule-governed intent ional behavior (cf . Itkonen's contr ibut ion in th is volume) which has obvious implications for evolution in general (see below). Breal (1964[1897]:73) went on to point out: "But the people must not be set problems too d i f f i c u l t for so lu t ion . " In other words, the steps must be 'min imal ' , obey a certain 'closeness', resemblance or s i m i l a r i t y , because: "The memory of people is short ." Moreover:

The intelligence of the masses is here seen on one of its most in­teresting sides: by the simplest means, it wins through the diffi­culties which, in every profession and every art, the material op­poses to the workman (Breal 75).

For those who apply a kind of professional logic to language, e.g., by comparing all the usages of the subjunctive and eliciting the common intersection, Breal (1964:220) can say:

Popular Logic does not proceed in this way. It advances, so to speak, by stages. Starting from a very c-ircumscrib-ed and definite point, it pushes straight ahead, and arrives, in all ignorance, at

276 RAIMO ANTTILA

a stage at which by the nature of things - I mean by the general tenour of the speech - a change is produced.

He has a lot to say about such logic of language, resting entirely on analogy (219-228). In addition, Breal was fully aware of the role played by semantic knowledge in the acquisition of language in general, includ­ing syntax. Current research is just now coming to this, as a modern breakthrough!

Significantly the most recent research into phonological acquisi­tion also tries to develop perceptual models. In this respect I think that the foremost scholar is Björn Lindblom (1972), who also rediscovers the same basic Peircean model as Andersen. Thus both scholars strength­en the findings of the other, though Lindblom's diagram uses a diffe­rent arrangement:

"Babbling" Creative behavior

Perceptual mapping of motor processes

Interaction with environment

Perceptual values. Articulatory interpreta-" tion of favored contrasts

Adult speech

Normalization

Fig. 3. A hypothetical language acquisition device (after Lindblom [1972:73]).

Lindblom further strengthens the arguments for a capacity for language. At the same time it would indeed seem that abduction is an instinct based on human biology. Liljencrants and Lindblom (1972) have further developed a numerical model to reflect the segmenting of perceptual contrast along the human speech apparatus (articulatory space). Here

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 277

we have a biological universal that must be taken into consideration in linguistic explanation. The authors suggest that predictions of phono­logical facts be derived as consequences of the structure of the mecha­nism available for human linguistic communication and the optimization of their use. We will return below to further supporting arguments for optimization and use from the points of view of sociolinguistics and language evolution. All this supports Liljencrants and Lindblom's delib-erately substance-based theory; later on we will see that also variants can have a semiotic function of their own.

8. THE SUBTLETY OF INNATE IDEAS. The above arguments support the criticism Goodman put forward against Chomsky's 'innate ideas' hypothe­sis. Goodman (1967, 1969) points out that first-language learning is actually a second-language acquisition with respect to much of cognition or non-language symbols. At least then first-language learning is a case bil inguai ism with other cognitive systems. He finds Chomsky's argument in favor of innate ideas by no means crude:

It is of exquisite subtlety, like the gossamer golden cloth made for that ancient emperor. But the emperor needs to be told that his wise men, like his tailors, deceive him; that just as the body covered with the miraculous cloth has nothing on it, the mind packed with innate ideas has nothing in it (Goodman 1969:141-42).

Yet we do need the mind and the natural laws of association or the like. As has been mentioned, even the staunchest empiricists admit that. The subtlety (and the arbitrary joints) of Chomsky's hypothesis does indeed remind us of the depth of Taoist wisdom (on the surface of it):

... it is said of Tao: Whoever understands it seems duller, Whoever follows it seems to go backward, Its even path seems crooked. The deep seems shallow, The white seems tarnished, The wholesome seems flawed, The solid seems shaky, The purest seems mixed ... (Tao-te-ching, 41)

278 RAIMO ANTTILA

"And the green is colorless", one feels tempted to add. A clear Taoist theme in transformational grammar has been "go along with the stream" (appeal to authority and fashion). Thus the eastern monistic system agrees with western arbitrary dualism, but this does not support arbi­trary segmentation.

Peirce fills exactly the gap seen from a more abstract level of observation in various individual language acquisition studies. He sup­plies a frame that combines rationalism with empiricism, and which ac­commodates both nature and nurture, and evolution; compare Hook's (1969: 167) criticism of Chomsky with regard to the latter:

I sometimes get the impression, on the basis, of Chomsky's rather offhand dismissal of the theory of evolution, that it is not so much empiricism that he regards intellectually baneful as it is naturalism.

Before going into the wider questions of evolution, and society as a synthesis of the problems of both evolution and language, we have to backtrack a little into the structure of grammar itself, as it reflects the basic design of language.

9. STURTEVANT'S PARADOX AND 'ONE MEANING - ONE FORM' CORRESPONDENCE The interplay between sound change and analogy has been the prime tar­get of phonological investigation for the past decade no matter what terminology has been used (e.g., various rule manipulations that can shift the issue totally into 'sound change'). What has re-emerged is Sturtevant's paradox, and in only a few cases the linguists are aware that they have rediscovered this paradox: sound change is regular and causes irregularity; analogy is irregular and causes regul ari ty (Sturtevant 1947:109). Even if the paradox is not absolute, it is amazingly accurate. Regular sound change can pull paradigms apart, and if analogy occurs, its result tends to be greater regularity in morphology (cf. Anttila 1972:94-97, for documentation).

Why this should be so is understandable from the stratified na­ture of language. The learning of phonology is a prerequisite f or keep­ing the huge number of linguistic signs separate, whereas the shape of

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 279

that phonology is relatively free. As we saw already, the child has to impose a perceptual differentiation on the acoustic scale he hears, and then try out articulatory optimizations based on that (cf. Lindblom 1972; Liljencrants and Lindblom 1972; Andersen 1972; 1973; Anttila 1972: 198). Thus the signs can shift apart, and create ultimately a clash with one of the basic design features of language: In an optimal communica­tion system the same meaning should always be represented by the same form, one of the old traditional insights of Western linguistics. It has gone under many different names. I have been calling it descrip­tively the principle of one meaning - one form (Anttila 1972 passim).4

It constitutes the strongest mental force behind the regularizing ten­dency of analogy. Naturally, none of the proponents of this force have maintained that human language could ever get into such a perfect state in which every meaning would have its own form. That is impossible. The great innovation in language design was the addition of syntagmatic means whereby independent elements can be combined to represent new meanings (concepts, inventions, experiences, etc.) not unlike pit vs. tip in phonology, or tree garden vs. garden tree in syntax. The main difference, and a drastic one, is that phenomena by themselves carry no meaning. Metaphors and metonyms also break the one-to-one relation between form and meaning in natural language.

I have developed a simple notation to describe the tendency toward one meaning - one form. Basically, the one-to-many relations between

4 I do not want to repeat here the history of this concept, suffice it to mention a few of the terms under which it has dwelled: the prin­ciple of optimality (Humboldt), univocability (Vendryes), canon of singularity (Ogden and Richards), the laws of specialization and dif­ferentiation (Brêal), the mental law that shuns purposeless variety (Wheeler), etc. (see Anttila 1972:107-08). Similarly, I prefer to for­go the modern resurrections of this law, as well as Sturtevant's para­dox. Some younger transformationalists are reinventing them, and since many of them believe, this to be a modern breakthrough, they of course keep creating further 'new' names (I will return to this question in another context).

280 RAIMO ANTTILA

meaning and form are of two types. Either one meaning has two forms or one form has two meanings (two here of course covers more than two, but two is used here as the simplest case of many; (cf. Anttila 1972:100). To use a diagram:

Fig. 4. The basic meaning-form relations in language.

Configuration A represents allomorphic alternation, compounds, phrases, and synonymy; configuration represents metaphor, metonymy, polysemy, homophony, and loan translation (cf. Anttila 1972:144). Thus we can write leveling as and split as (in which each variant becomes an independent sign with one-to-one relation between meaning and form). When extension of alternation in a certain subset of vocab­ulary takes place, the net result is still unity for diversity

(e.g., when the original consonant gradation of stops is extended to every consonant in Lapp [cf. Anttila 1972:130]). The yery essence of language,is diversity (either and if it is allowed to prevail it is made some use of in communication. On the other hand, in the right social context there will be shifting toward the 'one meaning - one form' correspondence. Thus the fading of metaphors agrees with this principle or polysemy can split o f f [ o f and off3

through and thorough, etc.). Often both the configurations intersed (bake and ba t c h , for most speakers). If the vari­ation is integrally wound into the culture, simplification does not set in so easily, for example the Christian religious metaphors have gener­ally remained unfaded, though exceptions may be cited.

10. MULTILINGUALISM AS LINGUISTIC VARIATION. Bi1ingualism, or mul­tilingualism, agrees with this general structure of grammar, and they

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 281

provide the clearest passage into sociolinguistics and the function of language use. If we accept the fact that our knowledge of semantic struc­ture is vague and fragmentary, we get the configurations on the lexical side for a meaning like "child" in English as in the following diagram:

child child(ren) kind(ergarten) ped(iatrics) puer(ile) infant(icide) etc.

Fig. 5. Typical complexity in English lexicon.

Here the contexts shift onat least two clearly discernible parameters, grammatical and a scale from homey to technical, not to mention inher­itance vs. borrowing with its own scales. In multilingual ism the con­texts are intricate social webs. In Sauris/Zahre in the Carnian Alps we have the situation sketched below (the branches indicate the rela­tive distance, or familiarity in connotation; cf. the following graphic presentation (after Denison 1969):

German Friulian Italian

Fig. 6. Typical configuration of multilingualism.

282 RAIMO ANTTILA

German is spoken between parents, by grandparents to children, and even by some children to each other, Friulian is spoken in the bar and to ac­quaintances from the surrounding area, and Italian in church, school, and kindergarten, between parents and children, to outsiders, as well as in reading and writing. Note how the circle is closed by the contextu-ally more formal Italian, in that it is allowed also for communication at home between parents and children. Compare the Saurian case with Switzerland in general where German, French, Romansch, and Italian oc­cupy different geographical regions; thus, the country is functionally monolingual in any one community (cf. Kloss 1971:8). The situation is again perfectly parallel to the language-internal facts of morphopho-nemic variation, or other variation with some similarity in outer shape, vs. suppletion. It depends on the speakers where automatic variation ends and suppletion begins in cases like opaque/opacity, s e a t / s i t , heat/ hot, etc. Now, however, through Fishman's (e.g., 1967) efforts diglossia is understood so that the languages have a different context of function, as in the Carnian Alps. Such a functional definition is better or more adequate, and covers then also many dialectal situations (social jargons, slang, etc.). The importance of function comes out also in language planning in that the community must choose one particular language, or if necessary, must assign different roles to different languages. (For the parameters to be manipulated, see Kloss 1967.)

Such questions of function have many implications for communica­tion. They have invaded questions of semantics, e.g. the difference be ween sentences like "I say it to you as an officer" and "... as a pri­vate person". The referent is the same person, but in different contexts these variants take on direct semtotic function. Variants have direct meaning. In philosophy the problem is known as the evening/morning star situation. From anthropology we know that different clothing is used in different contexts, and again reconstructing or positing a unified un­derlying garment is highly questionable. Multilingual ism or diglossia (in the second sense) is of exactly the same type. The variants (the languages) have social, psychological, emotional value, and positing

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 283

an underlying abstract system destroys such messages, because one would be delineating universals of language only. Multi 1 inguai i sm provides evidence against overall dialectal systems, in its suppletion end of the scale. This does not deny the possibility of some degree of over­all systems, but they cannot be as universal as often assumed.

The diagrams presented above (Fig. 5 and 6) have one single point on the semantic side; in other words, the 'referential meaning' was assumed to be the same for all the variants (except for the bake/hatch

-case). 'Social meaning', however, shows that the meaning-form re­lation is one to many both ways; the variants reflect intricate conno­tation networks in the general form of Fig.7 (cf. Giglioli 1972:16).

Fig. 7. Typical branching out of social meaning (intersection = referential meaning).

That variation leads to social meaning in terms of whole languages is no more remarkable than that random pronunciation variation leads to some kind of pronunciation index revealing origin or social attainment of speakers (Anttila 1972:190). In fact, the former source of variation is more understandable than the latter. Sturtevant's paradox shows also that when sound change pulls paradigms apart, variants can more clearly than before represent 'grammatical meaning', the ultimate case being suppletion. Both in paradigms and in multilingual societies variants can remain, if they are tied to a constant function. Since language exists for society as its most important vehicle for communication, the psychological burden of variation is upheld for the society. In other words, sociolinguistic criteria outweigh psycholinguistic ones. Mind

284 RAIMO ANTTILA

and memory have to serve social functions, and if the social context remains stable, variation remains. Thus, for example, the hieroglyphs retained their complex structure, because they served in a uniquely sacred function, and the religious sector remained stable for centu­ries (cf. Giglioli 1972:323, note 10). But when the social configura­tions are uniform and stable enough, bi- or multilingual ism is a bur­den, and a shift to monolingual ism easily occurs. Fishman's studies of language maintenance show that once the different functions of the lan­guages in question are lost, bil inguai ism is a fleeting phenomenon. Of­ten some kind of 'contamination' results, as in England where English gained the upper hand in grammar, and French in almost the entire voca­bulary (if not in the basic everyday lexicon). There are again paral­lels in paradigmatic levelings, e.g., Latin *ieouv/*iecinis gives iecur/iecinovis with the strong case -r throughout (a rather minor reg­ularity, it is true).

The efficiency of the shift / \ > | in languages has considerable economic implications as well. Countries with homogenous language sit­uations are on the whole much better off than multilingual societies. This correlates homogeneity with all kinds of positive factors (Fish-man 1967:24-28; cf. Kloss 1967:7). In other words, a solid state struc­ture is more efficient and economic to operate than an arrangement in a state of flux, through an assembly of tubes.

11. EVOLUTIONARY BIOLOGY. Since language is primarily a system that enables society to face the future with least effort, it is illu­minating to look also into evolutionary biology for parallels. Wescott points out that specialized species are better adapted to past condi­tions and generalized species to future conditions. In other words, as Wescott (1969:107-08) sees it:

Specialized groups are better adapted to actual conditions and generalized groups better adapted to possible conditions.

Specializers, in short, have greater biological potency, while generalizers have greater potentiality.

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 285

There are all kinds of biological rules-of-thumb, parallel to Sturte-vant's paradox and the like, to quote just two already cited by Wescott (1969:108-09):

One such rule in zoology is "Cope's Law" of the survival of the unspecialized, which states, in effect, that, other things being equal, generalized animal groups last longer than specialized ones. Yet, as we have already seen, "other things" are environmental con­ditions, which are (as, be it admitted, they have been throughout Quaternary Period) in a state of relatively rapid flux. When con­ditions are stabler, the "law" reverses itself, and it is the spe­cialized groups that survive more often and for longer periods than the unspecialized.

A more balanced and therefore more reliable zoological rule is "Williston's Law", which states that, as animals evolve, their or­gans tend to become fewer in number but more specialized in func­tion.

For instance, the ancestors of man started out with four legs but re­duced the number of walking limbs to two and specialized the other two for manipulatory purposes (cf. Wescott 1969:109). The implications of these notions to bil inguai ism and analogical leveling are obvious. Spe­cialization, / \ , favors past conditions, and generalization, , carries the future (compare also the economic factors mentioned above). We have seen that the societal conditions have to be right for mul ti 1 inguai ism to thrive; this is true of biology as well, to quote Wescott (1969:106) once more:

Specialization, in short, is a biological tendency which needs to be optimized rather than maximized. But just what constitutes op­timal specialization varies according to conditions.

Man, in fact, is an animal that has "specialized in genevalism" (Wes­cott 1969:118); in our notation:/\> |, where the j-relation can be bent most easily into new situations (of course language is central even here). "Man's evolution [is] unique in showing the dominance of convergence over divergence" (Huxley 1965:11) ( /\ > | ). Huxley is speaking here about Teilhard de Chardin's conceptions as they pertain to his, adding:

286 RAIMO ANTTILA

He usually uses convergence to denote the tendency of mankind, dur­ing its evolution, to superimpose centripetal on centrifugal trends so as to prevent centrifugal differentiation from leading to frag­mentation, and eventually to incorporate the results of differen­tiation in an organised and unified pattern (p.14).

There is a danger that this tendency might destroy the valuable results of diversification and lead to drab uniformity (p.15), and in fact a language of the type of -relations only would be impossible. But very importantly, Teilhard links his ideas with a notion of complexification, the genesis of increasingly elaborate organizations. This involves the universe in all its parts in an enroulement organique sur soi-même or reploiement sur soi-même. For this self-complexification Huxley (1965: 15) suggests the term 'convergent integration'. In other words, we need also the opposite force |>V and A. In language we have seen such a rise of variation and its use for social communication. The optimum balance requirement needs this principle also in diglossia or bil in­guai ism broadly conceived.

These biological notions make one think about vocabulary, espe­cially English vocabulary. The child-roster (cf. Fig.5) displays the strong hybrid character of the English lexicon. It shows specialization of the t y p e / \ , making it possible for English to have an unusually high number of stylistic levels that take on the same function as bi­lingual ism does elsewhere; in other words, English has a built-in diglossia. But this variety exists within one language, which provides the generalization aspect, | , and English has indeed proved to be well adapted for both actual and future conditions.

12. UNIFORMATION IN CULTURE AND LANGUAGE. The closest reference or parallel among sociolinguistic and bil inguai ism studies to the prin­ciple of 'one meaning - one form' [\J > / \ > | ) 1 have found in Joyce Hertzler's article "Social Uniformation and Language" (1957) in which she investigates the reverse factors of differentiation. Uniformation means "a set of related process-es whereby uniformity and similarity of practices in the various social relationships are established in time and space" (Hertzler 1967:172).

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 287

Closely related to uniformation, and in most instances contri­buting to it and functioning as subprocesses, [are] ... stan­dardization, stratification, regimentation, assimilation, ac­culturation, equilization, integration, routinization, unfica-tion, institutionalization, and universalization (172, note 4).

Such social forces underlie unification in language and vice versa. Hertzler (1967:171, note 1), which bears a close resemblance to Wescott1 s 'man as a specializer in general ism'. In short, these notions have a direct contact to Teilhardian ideas in their basic structure.

Generative grammarians have tried to make much of the maxim "chil­dren simplify grammars, adults complicate them". Here then Sturtevant's paradox is supposed to correlate with age (and cognitive development). The maxim, however, is an incorrect generalization. The distinction be­tween simplification and elaboration is a neat requirement of strict tax­onomy. but does not correspond to actual facts, which can be explained through abductions. Further, pidgins are generally quoted as extreme cases of simplification and they are due to adults. And when speakers grow up with it, and create a creole, children elaborate

Formation of pidgins is thus an extreme form of bil inguai ism cor­responding with simplicity, and creolization is a shift to monolin-gualism, creating complexity within the language. This is a reversal of what we have seen before, and tells much about human communication. It displays immediately that pidgins and creoles characterize human lan­guage from a new angle. DeCamp (1971:13) has pointed out that one can speak of creole rather than creoles. At least this supports the general principle of 'one meaning - one form', or Sturtevant's paradox, in that

shows that a natural language cannot operate with mere -rela­tions. In other words, even simplification in pidgins may not be that (cf. DeCamp, loc.cit. ); Hjelmslev may have been right in talking about optimal structure rather than minimal (cf. DeCamp 1971:19). Again we must revert to optimal balance, as mentioned above. Social factors are the crucial catalysts here. Adults striving at limited experience can get to a pidgin. Children have limited experience; thus, they try to keep complexities down. But their development is directed toward wider human communication, and the 'complexities' of a creole are expected. Whinnom

288 RAIMO ANTTILA

(1971:105) has discussed these dichotomies with biological terminology, although he points to the fact that bil inguai ism has no biological equi­valent. Whinnom notes the processes of impoverishment and repair, and how pidgins have temporary advantages over their rivals. In other words, the social context is the supreme filter. In biology we have primary hybridization corresponding to fragmentation, dialect differentiation, or origin and genesis of variants, to put it more generally. Secondary hybridization, interbreeding of distinct species, is the hybridization, and it corresponds to naive foreign language learning and bil inguai ism (cf. Whinnom 1971:91). Now, all language-switching constitutes second­ary hybridization (cf. English vocabulary) and is at the same time as­sociated with bil inguai ism. Tertiary hybridization and relexification occurs in pidgins, in this curious simplification. But a creole must be regarded a primary language; "the hybrid has become a new species" (Whinnom 1971:111). Then decreolization corresponds to the rare biolog­ical phenomenon of despeciation, in which barriers, mainly ecological, are removed. By contrast, decreolization turns out to be a process of primary hybridization (p.112). Thus, the circle of the forces behind Sturtevant's paradox closes and we get eloquent evidence for the opti­mum balance hypothesis.

Wescott (1969:171) has used a scale of behavior that both resem­bles language and depends on it to make inferences about glossogony. He calls this apo-language and divides it into several relatively autono­mous subsystems, among them: 1) song, 2) poetry, 3) writing, 4) mathe­matics, 5) logic, 6) pidgins and creols, and 7) synthetic international languages. This puts at least pidgins in further new light as secondary languages, though not creoles. But creoles can be assimilated more and more by the standard language; hence, we often have a bilinguai ism of a different kind. In this sense logic is another pidgin that helps com­munication in a limited area of experience; it is disambiguation of natural language

13. V A R I A N T S AND COMUNICATI ON. The notation used , etc.) is a very general one and does not cover the details in an exact way

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 289

(e.g., tertiary hybridization). On the other hand, generalization is today the name of the game, and here we see indications of a basic sim­ilarity between grammar itself, religion (e.g., Teilhard), anthropology, biology, and the psycho- and sociolinguistics of bil inguai ism. Most sig­nificantly, all this highlights the semiotic functioning of variants in communication. Thus bil inguai ism makes it quite clear that the search for yery deep unified underlying structures is not necessarily right, though it is one of the basic tenets of modern linguistics. These im­plications support a modern return to some kind of morpholexical alter­nations (Bloomfield), whatever terminology will actually be used. Most explicit in this vein is Shapiro's semiotic theory of markedness that explains morphophonemic alternations:

Generative morphophonemics has been preoccupied with the positing of underlying forms of increasingly higher degrees of abstraction and deriving surface forms by a mechanistic application of ordered rules. This trend has ineluctably led to a major methodological impasse in that neglect of actually occurring morphophonemic alter­nations has rendered even more remote the ultimate goal of lin­guistic research - the explanation of grammar (Shapiro 1972, ab­stract) .

For phonology I have presented similar arguments from a different angle (see Anttila 1973b). Perhaps a reference should also be made to writing and orthography, since they represent a kind of cultural bil inguai ism integrally tied to grammar, in the case of alphabets to some level of phonology. Here also the tradition has favored variants; the morphopho-npmir notation is due to special mass production forces (though still

on their own level). But this notation should not be transplanted into phonology as Chomsky and Halle have done. Remember that Bloomfield defined phonological units as bundles of features; this, like multilin­gual communities, represents a wery central kind of truth. Let a paral­lel from physics convey the narrow-mindedness of the blind search for underlying forms:

MIT scholars in possession of the Jakobsonian distinctive feature notation have denied any theoretical significance to the bundles, giving arguments worthy of Anaxagoras' "snow is not white, because in snow there is only water there is no whiteness". There are no

290 RAIMO ANTTILA

phonemes because there are features and morphophonemes; there is no rain, no snow, no hail, no sleet, because there is precipita­tion and because precipitation is all H2O, plus late rules (Maher 1972).

Of course here again different cultures would assign immediate semiotic function to the variants. In the present case it is easy to see because of the different words assigned to the variants (variants are not even known culturally to be variants; the situation is also true in naive grammar, and this is even more pronounced in cases where a culture seg­ments variants further, e.g., "snow" in Eskimo, or "potato" in Aymara, etc. Such rich vocabulary functions often in a parallel way to bil in­guai ism as we saw above in the English lexicon.

One more reference to grammatical analysis is illustrative. For years transformational grammarians have argued whether transformations change meanings or not. Transformations are after all devices that com­bine surface variants through underlying unified structures. Now the discussion has settled to a position in which all schools seem to accept the fact that transformations do change meaning. We get at least into the intricate delicacies of stylistic variation. And stylistic variation is in no way different in structure from multi lingual ism; there is a considerable overlap even in actual practice (compare again English vo­cabulary). A proper understanding of the semiotics of variants would have saved years of detour for the generative-transformational school of linguists.

14. PSYCHOLOGICAL R E A L I T Y AND GENERALIZATION. We have now seen a wide spectrum of concerns about language and linguistics, ranging from semantics to phonetics, from perception to evolution, from Peirce and Bréal to the 1970's. A psychologically real model of language and its use can only be achieved by paying attention to the theory of abduction, because it combines internal evidence with external, mind with matter, learning with change. It is curious how consistently the transforma­tional-generative paradigm has been able to avoid even any appearance of psychological reality in spite of the rhetoric to the contrary. Of

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 291

course abduction does not solve everything, but the findings presented here should be a basis for further advancement in theory. Another notion that "has been freely used by the transformationalists is 'valid' or 'in-tersting (etc.) generalization'. I think I have demonstrated, by implica­tion, how arbitrarily they use this latter term. The reason seems to be that 'generalization' is a term on the description side corresponding to 'psychological reality' on the innate cognition or universals side. If the latter aspect is inadequately treated, as it is in the transfor­mational-generative model, the former should also be discarded. This is exactly what we have seen happen, though my concern here has been only to expound new parameters for the concept. This is why I have retained rather loosely the terms 'generality', 'generalism', and 'generaliza­tion' that occur in literature. The transformationalists have tried to limit generalization largely to tallying marks on paper, which is total­ly inadequate for linguistic explanation, as should have become clear from the preceding. The main factor in generalization should not be economy of description, but, as I have shown above, optimization and function of use.

Most notable in this connection is that analogy and metaphor represent fundamental generalization. Note how important optimization and function are, since in evolution generaliza­tion and specialization represent configurations that overlap with those of language. Analogy and metaphor stem from per­ceptual judgments, and since transformationalists have denied the human mind, they have had to bar analogy and metaphor as well. The main reason seems to have been the fact that analogy cannot be predicted for certain, and unpredictability is the reason why scientists have not been willing to acknowledge abduction (Knight 1965:118), whereas formal ization post fac­to has been easier. But the essence of abduction and similarity cannot be formalized. Part of the confusion has arisen from the fact that trans­formational ists have used past-oriented models to predict change and the future. What children show, however, is that the unpredicatability of abduction is predictable, although we cannot know in advance the per-

292 RAIMO ANTTILA

ceptual judgment 'forced upon' the cognizer. Children are machines for predicting the future; compare Hook's (1969:163) observation:

"Reason" is not a fixed schematism of mind that controls behavior separate from it, but a pattern of ideas suggested by past, and corrected by present, experience. Its source - to repeat - is his­tory and culture broadly conceived, not a transcendental psychology or ontology.

The prolonged infancy reinforces the role of the community in bringing in the past and the present. The child has no use for history in any form in his early abductions.

This observation brings us directly to the question of a reason­able metaphysics to meet the needs of linguists. The search after it is now on the increase, since more and more linguists realize that trans­formational-generative grammar has also failed in this respect. Such a metaphysics has to start with phenomenology, the 'weak link' in any science, since it deals with both the inner and outer worlds (cf. Knight 1965:71), past and present. It generalizes by "describing and classi­fying the ideas that belong to ordinary experience... without regard to their being valid or invalid or to their psychology" (8.328) (Knight 70) As in perceptual judgment, such generalizations cannot be criticized or evaluated on paper.

In one sense, however, transformational-generative grammar takes a phenomenological detached twist in that it has studied language as a phenomenon unconnected with communication and users. In other words, it has studied an autonomous tool without paying attention to what it is used for, when, and by whom. A tool cannot be autonomous in this way; otherwise one misses the whole point of language, which is intimately connected with the teleology of human activity in general. On this score the philological tradition has been all along much more fruitful, with­out explicit formalization (cf. Anttila 1973e). What philology has also shown is that as in any other historical discipline we need a hermeneu-tic frame of reference, a method of understanding, not observation as in the natural sciences. And now Esa Itkonen has indeed drawn synchronic linguistics tightly into the pale of human intentional, rule-governed

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 293

behavior and shown that it is an unusually pure hermeneutic science without positivistic elements (see Itkonen 1974a+b; also Anttila 1973e). He applies the explicatory method to atheoretical linguistic knowledge (i.e., intuition) which produces theoretical knowledge (i.e., a grammar). As for linguistic behavior> it is clearly intentional, and can best be explained through the practical syllogism, in which hermeneutic and pos­itivistic elements mix inextricably. Note how all this ties in with ab­duction and evolution as discussed above. Abduction is the agent that combines and explains this area where hermeneutics and positivism meet. It corresponds to explication in that we have in both a generalization going down into the detail; and evolution is the counterpart of the teleology of human behavior (partially) explained by the practical syllogism. The correct vs. incorrect of hermeneutics is just replaced by optimality in evolution. Itkonen's and my pleas are compatible also in that both offer rather fundamental arguments for rescuing linguistics as a human science.

REFERENCES

Andersen, Henning. 1972. "Diphthongization". Language 48:1.11-50. . 1973. "Abductive and Deductive Change". Ibid. 49:4.765-93.

Anglin, Jeremy M. 1970. The Growth of Word Meaning. Cambridge, Mass. & London: MIT Press.

Anttila, Raimo. 1972. An Introduction to Historical and Comparative Linguistics. New York: Macmillan.

. 1973a. "Child Language, Abduction, and the Acquisition of Linguistic Theory by Linguists". Proceedings of the International Symposium on First Language Acquisition (Florence3 2-6 Sept. 1972) ed. by Walburga von Raffler-Engel. Ottawa: Univ. of Ottawa Press, in press.

1973b. "Exception as Regularity in Phonology". Proceedings

294 RAIMO ANTTILA

of the Second International Phonology Meeting (Vienna, 4-9 Sept. 1972). Munich: W. Fink, in -press.

Anttila, Raimo. 1973c. "Was There a Generative Historical Linguistics?" Proceedings of the Second International Conference of Nordic and Gen­eral Linguistics (Umeå, 14-19 June 1973 ed. by Karl-Hampus Dahlstedt. Umeå: Skytteanska Samfundet, in press. [Original version was publish­ed in the Preprints of the conference, Univ. of Umeå, 1973, pp. 8-14.]

. 1973d. "Onko strukturalismin aika kielitieteessä jo ohi?". Virittäjä 77:2.109-22. [A revised and shortened German version was published in Historiographia Linguistica 1:2.278-82 (1974) under the title "Ist der Strukturalismus in der Sprachwissenschaft schon pas-seV\]

. 1973e. "Linguistik und Philologie". Linguistik und Nachbar­wissenschaften ed. by Renate Bartsch and Theo Vennemann, 177-91. Kronberg/Taunus: Scriptor.

. 1973f. "Bilingualism and Linguistic-Mental Laws: General background and individual cases". Social Studies in Language Contact ed. by Jacob Ornstein and William F. Mackey. Quebec, P.Q.: Univ. La­val, in press.

Baron, Naomi. 1972. "Language Acquisitions Role in Diachrony: The evo­lution of English periphrastic causatives". Stanford Occasional Pa­pers in Linguistics 2.47-63. Stanford, Calif.: Univ. of Stanford, Nov. 1972.

Bever, Thomas G., and D. Terence Langendoen. 1971. "A Dynamic Model of Evolution in Language". Linguistic Inquiry 2.433-63.

Bréal, Michel. 1964[1900]. Semantics: Studies in the science of meaning. Transi, by Mrs. Henry Cust. New ed., with an introd. by Joshua What-mough. New York: Dover.

Chomsky, Noam. 1972[1968]. Language and Mind. 2nd enl. ed. New York: Harcourt Brace Jovanovich.

Clark, Eve V. 1973. "What's in a Word? On the child's acquisition of semantics in his first language". Cognitive Development and the Ac­quisition of Language ed. by Timothy E. Moore, 65-110. New York & London: Academic Press. [Circulated in MS as early as 1971.]

DeCamp, David. 1971. "Introduction: The study of pidgin and creole lan­guages". In Hymes 1971.13-39.

Denison, Norman. 1969. "Sociolinguistics and Plurilingualism". Proceed­ings of the Tenth International Congress of Linguists vol.1.551-59. Bucharest: Edit, de l'Acad. RSR.

Ervin, Susan M. 1964. "Imitation and Structural Change in Children's Language". New Directions in the Study of Language ed. by Eric H. Lenneberg, 163-89. Cambridge, Mass.: MIT Press.

GENERALIZATION, ABDUCTION, EVOLUTION, AND LANGUAGE 295

Fishman, Joshua A. 1967. "Some Contrasts between Linguistically Homo­geneous and Linguistically Heterogenous Polities". In Lieberson 1967.18-30.

Giglioli, Pier Paolo, ed. 1972. Language and Social Context. Harmonds-worth, Middlesex: Penguin.

Goodman, Nelson. 1967. "The Epistemological Argument". Synthese 17.23-to 28.

. 1969. "The Emperor's New Ideas". Language and Philosophy ed. by Sidney Hook, 138-42. New York: New York Univ. Press.

Hertzler, Joyce 0. 1967. "Social Uniformation and Language". In Lieber­son 1967.170-84.

Hook, Sidney. 1969. "Empirism, Rationalism, and Innate Ideas". Language and Philosophy ed. by S. Hook, 160-67. New York: New York Univ. Press.

Huxley, Julian. 1965. "Introduction". The Phenomenon of Man by Pierre Teilhard de Chardin. New York & London: Harper & Row.

Hymes, Dell, ed.. 1971. Pidginization and Creolization of Languages. Cam­bridge: Cambridge Univ. Press.

Itkonen, Esa. 1974a. Linguistics and Metascience. (= Studia Philosophi-ca Turkensia, 2.) Kokemäki: Risteen Kirjapaino.

. 1974b. "Transformational Grammar and the Philosophy of Sci­ence". (See this volume, Part IV.)

Kloss, Heinz. 1967. "Types of Multilingual Communities: A discussion of ten variables". In Lieberson 1967.7-17.

Knight, Thomas S. 1965. Charles Peirce. New York: Washington Square Press.

Lieberson, Stanley, ed. 1967. Explorations in Sociolinguistics. Bloom­ington, Ind.: Research Center for the Language Sciences. (4th ed., 1973.)

Liljencrants, Johan, and Björn Lindblom. 1972. "Numerical Simulation of Vowel Quality Systems: The role of perceptual contrast". Language 48:4.839-62.

Lindblom, Björn E. F. 1972. "Phonetics and Description of Language". Proceedings of the Seventh International Congress of Phonetic Sci­ences ed. by André Rigault and René Charbonneau, 63-96, 96-97 (Dis­cussion). The Hague: Mouton. [The oral version presented at the Con­gress was entitled "Numerical Models in the Study of Speech Produc­tion and Speech Perception? Some phonological implications".]

Maher, J. Peter. 1970. "A Note on the English Passive, with reference to some general questions of theory". Language Sciences 13.30-31. (Dec. 1970).

296 RAIMO ANTTILA

Maher, J. Peter. 1972. Labrets,, Lenses, Pens and Print: The effect of artifacts on speech and phonological theory. Unpublished MS.

McNeill, David. 1968. "The Creation of Language". Language: Selected readings ed. by R. . Oldfield and J. Marshall, 21-31. Harmonds-worth, Middlesex & Baltimore, Md.: Penguin.

Raffler-Engel, Walburga von. 1953. "Panchronic Linguistics". Zeit­schrift für Phonetik und Allgemeine Sprachwissenschaft 7.408-09.

_. 1970a. "The Concept of Sets in a Bilingual Child". Proceed­ings of the Tenth International Congress of Linguists vol.3.181-84. Bucharest: Edit, de l'Acad. RSR.

. 1970b. "The LAD, our Underlying Unconscious, and More on Felt Sets". Language Sciences 13.15-18 (Dec. 1970).

Reilly, Francis E. 1970. Charles Peirce's Theory of Scientific Method. New York: Fordham Univ. Press.

Shapiro, Michael. 1972. Morphophonemics as Semiotic. Unpublished MS. Slobin, Dan I. 1971. "Developmental Psycholinguistics". A Survey o f Lin­

guistic Science ed. by William Orr Dingwall, 299-400, 401-10 (Discus­sion). College Park, Md.: Univ. of Maryland; Linguistics Program.

Sturtevant, Edgar H. 1947. An Introduction to Linguistic Science. New Haven, Conn.: Yale Univ. Press.

Wescott, Roger W. 1969. The Divine Animal, New York: Funk & Wagnalls. Whinnom, Keith. 1971. "Linguistic Hybridization and the 'Special Case1

of Pidgins and Creoles". In Hymes 1971.91-115.

Addendum :

Bierwisch, Manfred. 1970. "Semantics". New Horizons in Linguistics ed. by John Lyons, 166-84. Harmondsworth, Middlesex & Baltimore, Md.: Penguin Books.

WHAT IS A GENERATIVE GRAMMAR?*

BRUCE L . DERWING a n d PETER R. HARRIS

Genera t ive grammarians have a s s e r t e d t h a t "The goal of the d e s c r i p t i v e s tudy of a language i s the c o n s t r u c t i o n of a grammar" (Chomsky & Ha l l e 1 9 6 8 : 3 ) , where nA grammar of a language p u r p o r t s to be a d e s c r i p t i o n of the i d e a l s p e a k e r -h e a r e r ' s i n t r i n s i c competence" (Chomsky 1 9 6 5 : 4 ) . T h e r e f o r e , in o rde r to a s c e r t a i n what l i n g u i s t i c s i s a l l a b o u t , acco rd ­ing to t h i s v iew, one must i n q u i r e i n t o the n a t u r e of the n o t i o n ' c o m p e t e n c e ' .

This k ind of i n v e s t i g a t i o n might be c h a r a c t e r i z e d as an a t t empt to answer t h i s q u e s t i o n : what deeper ( p s y c h o l o g i c a l or b i o l o g i c a l ) meaning can be a t t a c h e d to the i n t e l l e c t u a l p r o d u c t s of l i n g u i s t i c r e s e a r c h , and what meanings a re ex­c luded? In t e c h n i c a l t e r m s , the problem concerns the i n t e r ­p r e t a t i o n of l i n g u i s t i c t h e o r i e s . The q u e s t i o n posed in the t i t l e of t h i s paper i s thus about the interprétation of genera­tive grammars. Chomsky's w a r r a n t for making deeper c la ims for grammars i s based on p o s i t e d r e l a t i o n s h i p s among a grammar, a l a n g u a g e - a c q u i s i t i o n d e v i c e , and a p r o d u c t i o n - p e r c e p t i o n

This paper i s a compacted vers ion of a two-part paper d i s t r i b u t e d under the same t i t l e a t the 1971 meeting of the Canadian L ingu i s t i c Assoc ia t ion , St . J o h n ' s , Newfoundland. For a more de t a i l ed ana lys i s of the i s sues discussed h e r e , see Harr is (1970) and Derwing (1973), e s p e c i a l l y chap. 8.

298 BRUCE L. DERWING and PETER R. HARRIS

model. In brief, his general claim is this:

The general explanatory theory of language and the specific theory of a particular language that results from application of the gen­eral theory to data each have psychological content, the first as a hypothesis about innate mental structure, the second as a hypoth­esis about the tacit knowledge that emerges with exposure to appro­priate experience. (Chomsky 1966b:12, end of notes).

The entire framework is logically centered on the signifi­cance of the level of descriptive adequacy, which provides justification for both individual grammars and, ultimately, general linguistic theories. The explanatory power of the general theory rests on its ability to provide correct par­ticular theories. As a hypothesis about 'innate mental struc ture' it is corroborated to the extent that it can '(presum­ably in conjunction with some learning theory) account for the acquisition of 'tacit knowledge'. As Chomsky (1957:34) puts it: "The strongest possible proof of the inadequacy of a linguistic theory is to show that it literally cannot apply to some natural language." In terms of his criteria, this means showing that a theory cannot provide a descriptively adequate grammar for some language, that is, a correct de­scription of the competence or tacit knowledge of a native speaker of that language.

By now it should be clear that the requirements placed on a grammar which is to be 'descriptively adequate' are not strong enough to allow for selection from among innumerable alternatives, and thus to make claims about a correct de­scription of competence. Strong generative capacity, agree­ment with native-speaker' intuitions, capturing of so-called 'significant generalizations', and economy of description are not enough to justify any claims about psychological reality. Grammars are underdetermined by the data. Chomsky (1965:21) himself has admitted this possibility:

WHAT IS A GENERATIVE GRAMMAR? 299

Perhaps the day will come when the kinds of data we can now obtain in abundance will be insufficient to resolve deeper questions con­cerning the structure of language.

Peters (1970), Prideaux (1970) and others have recognized that the day has arrived. But it should be emphasized that other kinds of data can be brought to bear only if the con­cept of 'competence', in particular its relation to a model of performance, is clarified. Questions as to which grammar, indeed which kind of grammar, can be said to represent cor­rectly what the native speaker has learned demand explica­tion of the proposed general interpretation for grammars -'competence'.

Three interpretations of the notion 'model of linguistic competence' (or C-model) are implicit in the literature. The first of these is that of an 'idealized performance model' (or P-model), i.e., a model of what the speaker-hearer actu­ally does when he produces or comprehends utterances (ab­stracting away from such extraneous considerations as dis­tractions, shifts of attention, hesitations and the like). Such a model would obviously be of great interest to lin­guists and psychologists alike in that it would seek to ex­plain such fundamental aspects of language use as "the abil­ity of a speaker to understand an arbitrary sentence of his language and to produce an appropriate sentence on a given occasion" (Chomsky 1966b:3). And, indeed, at times Chomsky himself gives the impression that he intends his notion of a generative grammar of C-model to be interpreted as a model of just this sort, i.e., as a set of instructions or rules for language use, or possibly a set of descriptions of suc­cessive performance events. For example (our emphasis):

(1) By a 'generative grammar1 I mean a description of the tacit competence of the speaker-hearer that underlies his actual

300 BRUCE L. DERWING and PETER R. HARRIS

performance in production and perception (understanding) of speech. A generative grammar, ideally, specifies a pairing of phonetic and semantic representations over an infini te range; i t thus constitutes a hypothesis as to how the speak­er-hearer interprets utterances, abstracting away from many factors that interweave with tac i t competence to determine actual performance (Chomsky 1966a:75, note)

(2) The most striking aspect of l inguist ic competence is what we may call the 'creativity of language1 that i s , the speaker's abi l i ty to produce new sentences, sentences that are immedi­ately understood by other speakers although they bear no [sic] physical resemblance to sentences which are ' famil iar ' . The fundamental importance of this creative aspect of normal language use has been recognized since the seventeenth cen­tury at least . . . (1966b:4).

Now in such accoun t s Chomsky seems to be a rgu ing t h a t l i n g u i s t s ought to become concerned wi th e x p l a i n i n g l i n ­g u i s t i c performance or language use ( sen tence p r o d u c t i o n and p e r c e p t i o n ) and we have no q u a r r e l wi th t h a t v iew. Nor wi th the i m p l i c a t i o n t h a t ' compe tence ' r e f e r s (or ought to r e f e r ) to some u n d e r l y i n g performance mechanism. Yet i t i s c l e a r t h a t i t would be q u i t e improper to c h a r a c t e r i z e Chomsky's n o t i o n of ' l i n g u i s t i c competence ' as a model of t h i s s o r t ( h i s remarks above n o t w i t h s t a n d i n g ) , a f a c t wi th Chomsky has been a t g r e a t p a i n s to emphasize and r e -emphas ize on numerous o c c a s i o n s . Even in Syntactic Structures ( i . e . , be­fo re the term ' l i n g u i s t i c competence ' had as ye t been i n ­ven ted) , Chomsky made i t c l e a r t h a t t r a n s f o r m a t i o n a l - g e n e r ­a t i v e grammars of the s o r t he proposed were not t o be i n t e r ­p r e t e d e i t h e r as i d e a l i z e d p r o d u c t i o n or p e r c e p t i o n models , but r a t h e r as models of something e l s e somehow ' n e u t r a l ' between the two. As he pu t i t :

A grammar does not t e l l us how to synthesize a specific ut­terance; i t does not t e l l us how to analyze a part icular given utterance. In fact, these two tasks which the speaker and hearer must perform are essentially the same and are both outside the scope of grammars . . . (Chomsky 1957:48).

WHAT IS A GENERATIVE GRAMMAR? 301 He reiterates this position in no uncertain terms in Aspects, as well (cf. Chomsky 1965:9).

So while Chomsky speaks (and most eloquently) on the matter of the desirability of constructing models on lin­guistic performance which may explain the facts of language use (such as creativity and the like), he has from the very beginning refused to interpret his own work in generative grammar in such terms (except to imply that there must be some necessary connection, as discussed below). The reason for this is quite obvious, we think: it is because genera­tive grammars would certainly fail under such an interpre­tation. In particular, though such grammars do incorporate the requisite feature of recursiveness, which would seem to suggest the possibility of accounting for the creative as­pect of language use, these grammars at the same time lack certain other essential features which any model of lin­guistic performance must certainly possess. One of these, for example, is the feature of selectivity. As Chomsky and Miller (1963:271) themselves point out, the native speaker of a language not only has the ability simply to "produce novel utterances", but also to produce them "on the apro-priate occasion". The normal use of language, in short, is not only 'innovative', but also 'coherent' and 'appropriate to the situation', such that, as Chomsky (1968:11) himself puts it, "We can distinguish normal use of language from the ravings of a maniac or the output of a computer with a random element." Yet generative grammars of the standard sort do not have this requisite property of selectivity. Katz and Postal (1964:166-67) describe the situation which actually does obtain:

The syntactic component, which is the generative source for the whole description, enumerates the infinitive set of sentoids in an order and in a way that must be considered essentially

302 BRUCE L. DERWING and PETER R. HARRIS random from the viewpoint of actual speech production and com­prehension. . . Therefore, there is no provision for describing how speakers equipped with a linguistic description of their language can extract from it just the sentences they wish to produce and just the analyses required to understand the sen­tences produced by others.

Consequently, although a generative grammar may be described as na system of processes and rules that apply in a certain order to relate sound and meaning", it is nevertheless clear, as Chomsky (1967a:399) admits, that "we are not entitled to take this as a description of the successive acts of a per­formance model... - in fact, it would be quite absurd to do so."

There are other essential features which any satisfac­tory performance model must possess, but which generative grammars also lack. To give just one additional illustra­tion, a performance model (assuming that it contains a cen­tral component - 'the linguistic code' - which is common to both language production and perception) must be capable of a bidirectional interpretation, i.e., it must be capable both of translating semantic representations into phonetic ones (on the production or encoding side) and of interpreting pho­netic representations as semantic ones (on the perception or decoding side). This requirement cannot be met by any version of what may properly be called transformational-generative grammar. In particular, no grammatical model which incorporates a phrase-structure component is capable of satisfying such a requirement, due to the fact that the relation defined by the symbol —> (rewrite arrow) in all such grammars is defined, among other things, as asymmetric (Chomsky & Miller 1963:292). Moreover, it is clear that the transformational relation is also to be interpreted in all such grammars as specifying a unidirectional process

WHAT IS A GENERATIVE GRAMMAR? 303 or operation, as well (cf. Chomsky 1961).

In sum, a transformational-generative grammar is in­capable in principle of serving as an idealized model of linguistic performance because it lacks certain requisite properties which any model of this sort must possess. In fact, under a dynamic interpretation of the sort suggested, a generative grammar would seem capable of serving only as a model of random sentence production, being inherently in­capable (because of its unidirectional character) of serv­ing as a model of speech perception of any sort. From this we can conclude that there is no apparent substance to Chomsky's claim that a generative grammar "describes and attempts to account for the ability of a speaker to under­stand an arbitrary sentence of his language and to produce an appropriate sentence on a given occasion" (1966b 3; italics added), nor to the implication that the rules which appear in such grammars "determine how sentences are to be formed, used, and understood" (Chomsky 1965:25).

If generative grammars (or C-models) do not represent attempts to specify what is done in speaking or understand­ing a language, what, then, do they represent? Chomsky's own suggestion is that, while a generative grammar does not purport to represent a performance model per se, it does constitute an essential component of such a P-model. As he puts it in Aspects,

No doubt a reasonable model of language use will incorporate, as a basic component, the generative grammar that expresses the speak­er-hearer's knowledge of his language; but this generative gram­mar does not, in itself, prescribe the character of functioning of a perceptual model or a model of speech production (Chomsky 1965:9).

In an early programmatic paper, Miller and Chomsky (1963) explore the possibility of constructing a model for the lan-

304 BRUCE L. DERWING and PETER R. HARRIS guage user which incorporates a generative grammar as a fun­damental component, and they conclude that any such model must contain

a component M that contains rules for generating a matching signal. Associated with M would be components to analyze and (temporarily) to store the input, ... a heuristic component that could make a good first guess, a component to make the comparison of the input and the internally generated signals, and perhaps others (Miller & Chomsky 1963:465).

The general nature of the link proposed here between the grammar (or C-model) and the remaining components of the more general P-model is clearly the familiar one of analysis by synthesis y in which a whole "group of components perform­ing the functions of storage, preliminary analysis, compar­ison, and control" (Halle & Stevens 1964:610) somehow inter­act to specify how the rules of the grammar are "put to use" in actual linguistic performance. One difficulty with pro­posals of this sort, of course, is that they specify nothing more than an interesting "model and a program for research" as to how a generative grammar might in principle be incor­porated into a workable model of linguistic performance, since almost no substantive work has actually been done (so far as one can ascertain from the published literature) to determine whether or not such a program can, in fact, be carried out. The sort of P-model which seems to be implied by these suggestions can, however, be represented schemati­cally as follows:

X < H »

↓ ↓ LB LI

WHAT IS A GENERATIVE GRAMMAR? 305 In this framework, clearly, the competence model C re­

lates directly only to a relatively small (and unobservable) set of behavioral tokens which we have labeled (after Chom­sky) 'linguistic intuitions' (LI), which may be regarded as interpretations associated with certain properties of the output of the grammar. The relation between the grammar and all other aspects of linguistic behavior (LB), however, as can be seen in our schematic representation, is mediated by at least two component-types of unknown character, dimension and scope. The first of these is the component labeled H in the diagram, which is intended to encompass the various 'heuristics' or 'rules for language use' which are referred to occasionally in the literature (cf. Katz & Postal 1964, and Fodor & Garrett 1966, as well as the two papers just mentioned), while the second is a catch-all component (la­beled X in the diagram) which is meant to stand for any other additional components which any empirically adequate P-model incorporating might require. But regardless of the details (which are, of course, quite impossible even to imagine at this early date, much less specify in any rigorous way), the important point to be made is this: by no stretch of the imagination can there be said to exist any direct or well-specified logical link within the overall P-model between the chief object of current linguistic investigation (the generative grammar or competence model C) and the data of linguistic behavior or language use (with the single ex­ception of that extremely circumscribed sub-class of such behaviors called 'linguistic intuitions'). Only the compo­nent of this performance model is specified in any ex­plicit way, while the remaining components are merely as­sumed to exist and to be somehow operative in relating the grammar to linguistic performance. Thus it is quite impos-

306 BRUCE L. DERWING and PETER R. HARRIS

sible for any logical inferences to be drawn about linguistic behavior from properties of the grammar, and, by the same token, there are no particular conclusions to be drawn from the facts of language use about the appropriate form and structure of grammars. There exists, in short, an inferential gap between the linguist's grammar and the observable facts of linguistic behavior. As a consequence, even abstracting away from such supposedly 'grammatically irrelevant factors1 as memory limitations, distractions, shifts of attention and interest, errors, and the like, it is still clear that we have no empirical basis whatever for thinking of the study of competence "as the study of the potential performance of an idealized speaker-hearer who is unaffected by such ... factors" (Chomsky & Halle 1968:3)1, nor is there any ground for claiming, as Chomsky (1965:9) does, that a generative grammar characterizes "the knowledge of the language that provides the basis for actual use of language by a speaker-hearer". Chomsky himself denies that a generative grammar does this directly, and now we see that in order for him to demonstrate that a generative grammar is capable of doing this even indirectly (i.e., as part of a model of linguistic performance), he is required to specify the details of the operation of the components H and X in his (implied) ideal­ized performance model which show how his grammar might be 'put to use'. Chomsky's competence/performance distinction, in sum, involves far more than mere abstraction away from such 'grammatically irrelevant factors' as the ones men­tioned above, ignoring, as it does, a great deal of addi-1 Cf. Chomsky (1964:52), where he states that "the description of in­trinsic competence provided by the grammar is not to be confused with an account of actual performance ... Nor is it to be confused with an account of potential performance."

WHAT IS A GENERATIVE GRAMMAR? 307

t i o n a l s t r u c t u r e , as w e l l , namely, all of that structure which would be required to demonstrate whether generative grammars play any significant role in language use or not. Chomsky, t h e n , is quite cor­rect in his admission t h a t " t h e q u e s t i o n of how a grammar i s used in p r o d u c t i o n or p e r c e p t i o n of speech i s , of c o u r s e , open" (1967b:125 n o t e ) , but i t should be more g e n e r a l l y r e a l i z e d t h a t in making such an admis s ion , one i s , in e f f e c t , a l lowing t h a t one p o s s i b l e r e s o l u t i o n of the i s s u e might wel l be t h a t t h e r e i s no connec t ion a t a l l , i . e . , t h a t g e n e r a t i v e grammars of the Chomskyan type p l ay a b s o l u t e l y no r o l e in l i n g u i s t i c performance or language use wha t soeve r .

In f a c t , t h e r e i s a l r e a d y a c e r t a i n amount of e x p e r i ­mental ev idence a v a i l a b l e to sugges t t h a t t h i s i n t e r p r e t a ­t i o n i s ve ry l i k e l y to be the c o r r e c t one . To i l l u s t r a t e , a l t hough i t i s c l e a r t h a t t h e r e i s no s t r a i g h t f o r w a r d con­n e c t i o n a t a l l between grammars and the f a c t s of language use which a l lows for any p a r t i c u l a r d e t a i l e d p r o p o s a l s about grammar to be t e s t e d by the s tudy of l i n g u i s t i c b e h a v i o r , i t has n e v e r t h e l e s s been sugges t ed t h a t a few rough-and - r eady measures Of co r respondence might r e a s o n a b l y be d e v i s e d . In p a r t i c u l a r , as e a r l y as 1963 M i l l e r and Chomsky o f f e r ed t he s u g g e s t i o n t h a t

The psychological p l a u s i b i l i t y of a t ransformat ional model of the language user would be s t r e n g t h e n e d . . . , i f i t could be shown tha t our performance on tasks requ i r ing an apprec ia t ion of the s t r u c t u r e of transformed sentences i s some function of the n a t u r e , number and complexity of the grammatical t ransformat ions involved (p.481) .

Fodor and G a r r e t t (1966:141) have r e s t a t e d t h i s p o s i t i o n as fo l lows :

If . . . the grammar i s involved in sentence process ing in anything l i k e the way t ha t ana lys i s -by - syn thes i s models suggest , then we have a r i g h t to expect a very general correspondence between such formal fea tu res of d e r i v a t i o n a l h i s t o r i e s a s , for i n s t ance , length

308 BRUCE L. DERWING and PETER R. HARRIS in rules, and such performance parameters as perceptual complexity, ease of recall, and so on.

Psychologists were quick to seize on this idea, dubbing it (among other things) the "Derivational Theory of Complexity", and a long series of experiments were initiated and carried out by them over a period of several years to assess its gen­eral feasibility. Interestingly enough, except for a few very early and relatively unsophisticated and uncontrolled studies of this sort, almost all serious work which has been carried out along these lines has yielded what can only be charac­terized as definitely negative results. (For details, see the summaries provided by Fodor & Garrett 1966; Bever 1968, and Watt 1970). Watt (1970:144) sums up this whole effort as a series of experiments "designed to show not whether per­formance mirrored competence" (i.e., in all, most or even many respects), but rather merely to test the much weaker hypothesis "whether performative complexity mirrored compe­tence complexity". Thus, even if this weaker hypothesis were upheld, "the stronger version would remain still in doubt; but (of more immediate importance), if it were disconfirmed, then the stronger version would be seriously threatened" (ibid.). By 1970, then, this hypothetical threat had clearly become a very real one; so much so, in fact, that Bever (1968:57) has been disposed to remark that "the failure of linguistic derivational complexity to predict psychological complexity demonstrates that the relation between current grammars and perception is remote if it can be said to be extant at all."

Stated in terms of the overall performance model dia­grammed above, Chomsky's original suggestion that the rela­tionship between the C-model (or component) and the overall P-model must be rather close amounted to a suggestion that

WHAT IS A GENERATIVE GRAMMAR? 309 the contribution of the added components H and X must be re­latively insignificant. But now it is seen that this sugges­tion fails to hold up even when so crude a measure as gene­ral sentence complexity if utilized for purposes of compar­ison between the two domains, i.e., between the properties of the grammar (C-model) and the facts of linguistic behav­ior which this grammar has been said (in some sense) to 'underlie'. We are therefore forced to the conclusion that in any P-model which attempts to incorporate a generative grammar as a fundamental component, the contribution of the unspecified components (provided they can even be specified at all) is going to have to be considerably more substantial than originally assumed. That is, in effect, these added components are going to have to drastically change or radi­cally reshuffle the rules of the generative grammar around if anything like the proper predictions are to follow from the overall model as it relates to the facts of language use. And if this is to be the case, one wonders with Stein­berg (1970:181) what the point might be of positing the C-model in the first place:

If a person is presumed to have one order of competence rules for production and another for understanding, why would he still need another order, that of Chomsky's model of competence? The postula­tion of the existence of an organization of language knowledge such as Chomsky's is theoretically superfluous. A competence model where language knowledge is ordered directly for performance is all that is required.

The option is always open, of course, to attempt to revise the grammar so that predictions about difficulty of processing hold, i.e., so that the Derivational Theory of Complexity is confirmed. In fact, however, the trend seems to be against revision in the face of falsification. Rather, the grammar (and thus the theory on which it is based) is

310 BRUCE L. DERWING and PETER R. HARRIS

h e l d t o be s a c r o s a n c t and t h e interpretation i s r e j e c t e d . C o n s i d e r , f o r e x a m p l e , F o d o r and G a r r e t t ' s ( 1 9 6 6 : 1 5 2 ) c l a i m t h a t

one would bes t i n t e r p r e t negat ive data as showing tha t an accep t ­able theory of the r e l a t i o n between competence and performance models w i l l have to represen t t ha t r e l a t i o n as a b s t r a c t , the de ­gree of a b s t r a c t i o n being propor t iona l to the f a i l u r e of formal fea tu res of de r iva t ions to correspond to performance v a r i a b l e s .

P r e s u m a b l y , i f t h e f a i l u r e o f c o r r e s p o n d e n c e was c o m p l e t e , t h e d e g r e e of a b s t r a c t n e s s wou ld be i n f i n i t e l y l a r g e . The i m p o r t a n t c o n s i d e r a t i o n h e r e i s t h a t t h i s t h i r d p o s i t i o n i n v o l v e s t h e w e a k e s t (and l e a s t f a l s i f i a b l e ) c l a i m s of a l l . And t h i s i n t u r n p o i n t s t o t h e f a c t t h a t t o o much r e l i a n c e i s p l a c e d n o t so much on i n t r o s p e c t i v e e v i d e n c e ( a s i s o f ­t e n c h a r g e d ) , b u t on p u r e l y f o r m a l c o n s i d e r a t i o n s . F o d o r and G a r r e t t c a n n o t i m a g i n e f o r m u l a t i o n s b a s e d on w h a t t h e y c a l l s t r o n g ' i n t e r n a l e v i d e n c e ' s u c c u m b i n g t o " a n y p u r e l y e x p e r i m e n t a l d i s c o n f o r m a t i o n " loc. cit.). T h i s seems t o be a c a s e of ' s a v i n g t h e t h e o r y 1 r a t h e r t h a n ' s a v i n g t h e p h e ­n o m e n a ' . The d a n g e r h e r e , of c o u r s e , i s t h a t one r u n s t h e r i s k of o n e ' s work b e c o m i n g ( t o r e v i v e an o l d d i s m i s s a l p h r a s e ) " o f o n l y m a r g i n a l i n t e r e s t " .

T h r e e b a s i c i n t e r p r e t a t i o n s of t h e c o m p e t e n c e - p e r f o r ­mance r e l a t i o n a r e t h u s i n d i c a t e d :

(a) a competence model is an idealized performance model (b) a competence model is somehow to be incorporated into

a performance model as a component (c) the relation is unspecified, but more abstract (in­

direct) than (b)

Notice first that one could very well give grammars what Lamb (1966) calls a 'dynamic interpretation', which would involve claims about modeling encoding and decoding pro­cesses. The main objections against this seem to be based

WHAT IS A GENERATIVE GRAMMAR? 11

on Chomsky's opinions about the 'logical priority' of com­petence, the lack of certain features in currently proposed grammars, and the lack of interest in psychology which has been traditional in North American linguistics since Bloom-field (and which is still implicit in Chomsky's position, despite what he says). Perhaps the important factor here is that grammars would have to be radically changed in order to achieve any sort of plausibility on this interpretation.

The same kind of factor seems to cause difficulty with the second position mentioned above, as well - competence as a component of a performance device. The crucial point here is that if performance variables cannot serve as po­tential falsifiers, then talk about explanations of 'the facts of language use' must be abandoned. The relevant log­ical connections are the same for both cases, and they are either present or they are not. In other words,

the dictum that performance is to be considered a reflection of competence is to be supplemented by the converse that competence must be viewed as something that can effectively lead to perfor­mance (Schwarcz 1967:51).

The asymmetry which is tolerated at the present time seems to us to be unjustifiable.

Chomsky's more recent writings (e.g., 1970) seem to place him in the abstractionist camp along with Fodor and Garrett. His confusing doctrine of notational variants, his comments on the empirical status of directionality and or­dering in grammars, and his cryptic statement to the effect that competence models and performance models are logically different types of things (and thus presumably not conjoin-able within a single theoretical structure) all seem to point to an admission that competence grammars are extremely remote (if not totally unconnected) from the objects of ex-

312 BRUCE L. DERWING and PETER R. HARRIS

p l a n a t i o n . To p u t i t b l u n t l y , p e r h a p s g e n e r a t i v e g rammars a re semantically a b s t r a c t , t h a t i s incapable of empirical interpretations. I f s o , t h e y a r e i n v u l n e r a b l e , b u t a l s o i r ­r e l e v a n t t o C h o m s k y ' s p r o f e s s e d g o a l s . And i f l i n g u i s t s a r e n o t i n t e r e s t e d i n t h e s e g o a l s - i n e x p l a i n i n g l a n g u a g e u s e and a c q u i s i t i o n - t h e n t h e y m i g h t a s w e l l c l o s e up s h o p and d e c l a r e b a n k r u p t c y .

REFERENCES

Bever, Thomas G. 1968. "A Survey of Some Recent Work in Psychol inguis -t i c s " . Specification and Utilization of a Transformational Grammar ed. Warren J . P l a t h , 1-66. Yorktown Heights , N.Y.: IBM Corporation.

Chomsky, Noam. 1957. Syntactic Structures, The Hague: Mouton. (10th p r i n t i n g , 1972.)

. 1961. "On the Notion of 'Rule in Grammar"1. Structure and Language and its Mathematical Aspects: Proceedings of the Twelfth Symposium in Applied Mathematics ed. by Roman Jakobson, 6-24. Pro­vidence, R . I . : American Mathematical Society.

. 1964. "Current Issues in L ingu i s t i c Theory". In Fodor & Katz 1964'.50-118. (Also s e p a r a t e l y , The Hague: Mouton, 1964.)

. 1965. Aspects of the Theory of Syntax, Cambridge, Mass. : MIT Press .

. 1966a. Cartesian Linguistics: A chapter in the history of rationalist thought. New York & London: Harper & Row.

. 1966b. "Topics in the Theory of Generative Grammar". Cur­rent Trends in Linguistics ed. by Thomas A. Sebeok, vol.3.1-60. The Hague: Mouton. (Also separately, 1966.)

___^ . 1967a. "The Formal Nature of Language". Biological Foun-dations of Language by Eric H. Lenneberg, 397-442. New York: Wiley & Sons.

. 1967b. "Some General Properties of Phonological Rules". Language 43.102-28.

WHAT IS A GENERATIVE GRAMMAR? 313

Chomsky, Noam. 1968. Language and Mind, New York & London: Harper & Row. (2nd enl. ed., 1972.)

. 1970. "Deep Structure, Surface Structure, and Semantic Interpretation". Studies in General and Oriental Linguistics_, pre­sented to Shiro Eattori on the occasion of his sixtieth birthday3 ed. by Roman Jakobson and Shigeo Kawamoto, 52-91. Tokyo: TEC Com­pany. (Also in 1) Noam Chomsky, Studies on Semantics in Generative Grammar, 62-119. The Hague: Mouton, 1972, and 2) Danny D. Stein­berg and Leon A. Jakobovits, eds., Semantics: An interdisciplinary reader in philosophy, l i n g u i s t i c s , anthropology, and psychology, 183-216. London: Cambridge Univ. Press, 1971.)

, and Morris Halle. 1968. The Sound Pattern of English. New York & London: Harper & Row.

Chomsky, Noam, and George A. Miller. 1963. "Introduction to the Formal Analysis of Natural Languages". Handbook of Mathematical Psycholo­gy ed. by R. D. Luce, R. Bush, and E. Galanter, vol.2.269-321. New York: Wiley & Sons.

Derwing, Bruce L. 1973. Transformational Grammar as a Theory of Lan­guage Acquisition. Cambridge: Cambridge Univ. Press.

Fodor, Jerry A., and M. Garrett. 1966. "Some Reflections on Competence and Performance". Psycho Unguis tics Papers: The proceedings of the 1966 Edinburgh Conference, ed. by John Lyons and Roger J. Wales, 135-54. Edinburgh: Edinburgh Univ. Press.

, and Jerrold J. Katz, eds. 1964. The Structure of Language: Readings in the Philosophy of Language. Englewood Cliffs, N.J.: Prentice-Hall.

Halle, Morris, and K. N. Stevens. 1964. "Speech Recognition: A model and a program for research". In Fodor & Katz 1964.604-12.

Harris, Peter R. 1970. On the Interpretation of Generative Grammars. M.Sc. thesis, Edmonton, Alta.: Univ. of Alberta.

Katz, Jerrold J., and Paul M. Postal. 1964. An Integrated Theory of Linguistic Descriptions. Cambridge, Mass.: MIT Press.

Lamb, Sidney M. 1966. Outline of Stratificational Grammar. Washington, D.C.: Georgetown Univ. Press.

Miller, George A., and Noam Chomsky. 1963. "Finitary Models for Lan­guage Users". Handbook of Mathematical Psychology ed. by R. D. Luce, R. Bush, and E. Galanter, vol.2.419-91. New York: Wiley & Sons.

Peters, P. Stanley. 1970. "Why Are There so Many 'Universal' Bases?". Papers in Linguistics 2.27-43.

314 BRUCE L. DERWING and PETER R. HARRIS

Prideaux, Gary D. 1970. "On the Selection Problem". Papers in Linguis­tics 2.238-66.

Schwarcz, R. M. 1967. "Steps toward a Model of Linguistic Performance: A preliminary sketch". Machine Translation 10.39-52.

Steinberg, Danny D. 1970. "Psychological Aspects of Chomsky's Compe­tence-Performance Distinction". Working Papers in Linguistics 2:2. 180-92. Honolulu, Hi.: Univ. of Hawaii; Dept. of Linguistics.

Watt, William 1970. "On Two Hypotheses concerning Psycholinguistics". Cognition and the Development of Language ed. by John R. Hayes, 137-to 220. New York: J. Wiley & Sons.

ON THE INADEQUACY OF THE TREE* AS A FORMAL CONCEPT IN LINGUISTIC ANALYSIS

EDWARD R. MAXWELL

Some years ago, Goodman (1961) suggested that trees as defined in Chomsky (1959) are inadequate to define many re­lationships. Specifically, he stated that the proper graph to represent the phrase

"happy and healthy boys and girls" would be

happy boys girls

healthy Goodman concluded his essay: "If it seems at the present stage of structural linguistics that nothing more will ever be needed than the familiar rudiments of graph-theory, it probably seemed at a comparable stage in the development of physics that nothing more would ever be needed than elemen­tary arithmetic" (Goodman 1961:55).

Gross and Lentin (1970:83) state that labeled brackets and trees with labeled nodes are notationally equivalent, with an algorithm to transfer from one to the other. However, they go on:

By tree I mean a single rooted, non crossing, n-ary branching con­struction commonly used in constituent analysis.

316 EDWARD R. MAXWELL

The c h o i c e of a t r e e t o r e p r e s e n t t h e s t r u c t u r e of a s e n t e n c e i s somewhat a r b i t r a r y and we might c e r t a i n l y have used a more g e n e ­r a l t y p e of g r a p h , f o r example , a g r a p h w i t h c y c l e s . E m p i r i c a l r e a s o n s d i c t a t e d t h e c h o i c e of a t r e e . T r e e s a r e s i m p l e g r a p h s , and a l t h o u g h c e r t a i n phenomena a r e n o t r e p r e s e n t e d ( l i k e r e s t r i c ­t i o n s be tween n o n - c o n t i g u o u s e l e m e n t s ) t h e y a r e q u i t e s a t i s f a c t o r y f o r a f i r s t a p p r o x i m a t i o n : t h e i r s i m p l i c i t y i s n o t bought a t t oo g r e a t a p r i c e .

Gross and L e n t i n ' s l a s t s t a t e m e n t i s , of c o u r s e , d e b a t a b l e . What, in f a c t , i s the f u n c t i o n of the ph rase s t r u c t u r e t r e e and what i n f o r m a t i o n does i t supply about s e n t e n c e s ?

The P . S . t r e e has s e v e r a l f u n c t i o n s : (1) to d e f i n e a h i e r a r c h i c a l r e l a t i o n among c o n s t i t u ­

en t s which r e s u l t in e m p i r i c a l c la ims about c o n s t i t u e n t s t r u c t u r e : e . g . , where does the adverb go— i s i t as impor­t a n t as the s u b j e c t and p r e d i c a t e ?

-Adv

or is it less important?

or are there some adverbs that are more important than oth­ers?

Adv

(2) to define relations between constituents them­selves :

ON THE INADEQUACY OF THE TREE 317 the fact that NPb is lower than and preceded by NPa will re­sult in a morphological change in NPb (i.e. it may pronomi-nalize).

(3) to use the formal constituent structure in the def­inition of transformations:

(a)

(b)

In the (a) tree above, NPb will be subject to the rule of reflexivization while in the (b) tree the NPb will not be subject to the rule because an S intervenes between it and the main sentence.

I don't think there can be any disagreement about the usefulness of P.S. trees within the framework described. To prove that P.S. trees are improper formal devices it must be shown that (a) constituents are not hierarchically or­dered; (b) constituents, like pronouns, are not related to their antecedents ; or (c) that there is no such thing as a transformation.

Concerning the notion of the transformation, there seem to be two general schools within the transformational frame­work: (1) Chomsky's position that t-rules are extremely pow­erful formal devices and should be used rigorously and spar­ingly; (2) Lakoff's, McCawley's and Postal's positions that transformations are the only devices worth discussing and that the best theory is one that employs only t-rules.

318 EDWARD R. MAXWELL The problem with the Chomsky position is that a new

type of rule, the projection rule, must be formulated (as in Jackendoff 1973) to do exactly the same thing as t-rules with exactly the same power. Therefore, neither theory is really any more interesting than the other. And both are interesting as far as they go.

However, if the use of the tree is to be extended be­yond simple phrase structure representation one must pro­vide proof that hierarchical and linear information is es­sential for the representation. For example, generative semantics theory says that since all meanings can be stated in terms of sentences, then trees can be used for semantic representations. The burden of proof is on these theorists to show that semantic information must necessarily have hi­erarchical or linear formal representation to account for semantic relatedness. That is, since trees cannot represent restrictions between non-contiguous elements, and if it is important in semantics to represent such restrictions, then the tree is a poor formal device.

An example of a non-contiguous semantic restriction are the relative relationships for which we use the words "good", "bad" or "short", "tall". That is, the relation between these terms must be represented on some sort of probability scale. For example, we can say that a "good" knife is one that pos­sesses sharpness greater that .7 in the real interval between 0 and 1. The point is that tree representations must be used sparingly.

At this point the science of language needs more work done on the notions of 'notation' and 'notationa! variation'. A devastating criticism often leveled against another system is that it is simply a notational variant of one's own nota­tion. Consider the following notational variations:

ON THE INADEQUACY OF THE TREE 319

(I)

(ID

(III)

All three representations above are, in some sense, no» tational variants. At one level they represent the same in­formation. However, if Y is dependent for its definition on X (like the Det and N are derived from the NP node) then the trees do not represent the same information. But when we are doing feature analysis, either in phonology or in semantics, how do we determine which features are derived from which

320 EDWARD R. MAXWELL

others. And if there is no way to determine, then why not use the Venn diagram model which has no implication of hi­erarchy?

Further study of notational variation should prove very enlightening.

REFERENCES

Chomsky, Noam. 1959. "On Certain Formal Properties of Grammars". In­formation and Control 2:2.137-67.

. 1973. "Conditions on Transformations". A Festschrift for Morris Halle ed. by Stephen R. Anderson and Paul Kiparsky, 232-86. New York: Holt, Rinehart & Winston.

Goodman, Nelson. 1961. "Graphs for Linguists". Structure of Language and its Mathematical Aspects (= Proceedings of the Twelfth Sympo­sium in Applied Linguistics; Providence, R.I.: American Mathematical Society, vol.XII), 51-55

Gross, Maurice, and André Lentin. 1967. Introduction to Formal Grammars. New York: Springer Pub. Co.

Jackendoff, Ray S. 1973. Semantic Interpretation in a Generative Gram­mar. Cambridge, Mass.: MIT Press.

Lakoff, George. 1971. "On Generative Semantics". Semantics: An inter­disciplinary reader in philosophy, l i n g u i s t i c s , and psychology ed. by Danny D. Steinberg and Leon A. Jakobovits, 237-96. New York & London: Cambridge Univ. Press.

*****

LANGUAGE ACQUISITION AND COMMON SENSE

WALBURGA VON RAFFLER-ENGEL

Most funded research of the 1960's followed along the lines of transformational grammar. We might well have gained a little more insight into the complexities of language ac­quisition had these same agencies funded anthropological lin­guistics and sociolinguistics to explore first language ac­quisition, bilingualism, and second language acquisition.

According to Chomsky language was acquired through a unique and specific mechanism and in a manner that was large­ly culture-free. Psycholinguistics was dealt with as if it were clearly separable from sociolinguistics. Chomsky also challenged the previous belief that language develops as a means of communication. The corpus to be examined was iso­lated from its context of speech interaction. These discon­nected utterances were then analysed according to the rules of transformational grammar.

As the main focus was on grammar, and grammar was de­fined in terms of the generative rules of syntax, no effort was made to include any information on the intonation curve or the pattern of the sentence stress. When even such impov­erished data as we had did not fully fit the preconceived theory, the recalcitrant facts were excluded from the final analysis and defined as 'marginal data'.

The corpus was generally obtained in short lab sessions. The emphasis was not on developmental observations but on

32 2 WALBURGA VON RAFFLER-ENGEL

cross-sectional samples. Of all types of longitudinal studies, diaries were the ones most neglected because their compilers supposedly lacked a coherent experimental tech­nique. That the observation of human behavior can lead to significant results only when it is done in the manner of a tightly designed experiment was a tacit assumption underly­ing much of the research in the social sciences of the last decade in the United States. Another common fallacy to which the MIT group of linguists fell prey was the belief that the methodology for analyzing the natural sciences was suitable also for studying man.

Students in the hard sciences were, however, less sin­gle-minded than transformational linguists. They recognized the need for a variety of approaches to complex problems. To quote one example, physicists no longer assume that a solid can have only one constant that relates to the passage of sound waves through it. Variation in experimental data is now clearly explained by the fact that a solid can have many such constants.

In an area of which we know very little - and which in addition is coupled with complex interactions - it might be wise to try a variety of experimental techniques. We know extremely little about what language is and next to nothing about how it develops. To limit our research to one theory, and a comprehensive theory at that, has been one of the ma­jor roadblocks hampering research in child language in re­cent years.

Chomsky created a whole system of linguistic analysis and postulated that his particular theory of syntax corres­ponded to an innate schema, a view which has not been sup­ported by empirical evidence. Chomsky focused entirely on language, and failed both to correlate language with cogni­tion and to look at language as a form of communicative be-

LANGUAGE ACQUISITION AND COMMON-SENSE 3 2 3

havior which, however, has to be correlated with kinesics. I believe that the data support the Piagetian view that cognition precedes language and that language develops in a close tie with other forms of communication. The audit­ory and oral channels work together with the visual mode in the child's attempt to gain understanding and conse­quently to express himself.

One two-year-old boy whom I studied several years ago used both upstairs and downstairs with the meaning of "on the other side of the stairs". When he became aware eventually that the words were not synonymous he acted out the situation and watched his mother's reaction to his saying the one word or the other until he was sure that he understood and used each word correctly. The child manifested an innate desire to learn and was actively involved in that task. The process of language acquisition is one of listening and observing on the receptive side. On the productive side it consists of trial and error in speaking and acting out.

Chomsky has listed what is acquired and, recognizing that there is a certain regularity in the sequence of acqui­sition of a number of linguistic features, he has concluded that these features are innate. He has constructed a hypo­thetical network to connect some of these features and as­serts that the human brain comes equipped with this network in a fashion similar to a computer input.

I am inclined to believe that what is inborn is cogni­tion. Cognition determines the child's approach to learning in general and to language in particular. To find out what syntatic and other linguistic features are acquired at a cer­tain maturational period I would rather look for how they are acquired. It is not the what that explains the how, but the how that explains the what. In conclusion, it seems to me that the way language is used seems to be hereditary. I made

32 4 WALBURGA VON RAFFLER-ENGEL

a statement in 1964 to which I am still holding, that the degree of loquaciousness and other styles of verbalization are transmitted from parent to child. A presentation by Ka­ren Fischer at the International Symposium on First Language Acquisition in Florence, Italy, in 1972 has produced excel­lent documentation on the issue through her research with fraternal and identical twins. Grammar, lexicon, phonology, meaning are not universal, nor are they inborn or hereditary

In my opinion the same holds true for kinesic behavior, but. nobody has yet undertaken conclusive research in that area.

The first attempt to bring together scholars working in kinesics was accomplished by Adam Kendon in a two-day meet­ing on Face-to-Face Interaction at the University of Chicago in 1973. There is now also a Newsletter on Non-Verbal Compo­nents of Communication, Paralanguage, Proxemics, edited by Mary Ritchie Key at the University of California at Irvine. If this field develops as it promises to do we may come much closer to understanding language development. This emerging field of inquiry, so far, seems to shun grandiose theories and is open to solid empirical research. Some research along these lines has already been done but the primary concern remains the establishing of unities of kinetic behavior.

In reference to child development, as I see it, a for­mulation of the basic problems in the acquisition of kin­esics (excluding purely instinctive gestures from that field of study) involves the following, a formulation that paral­lels the basic problems involved in acquisition of verbal language :

1) The use of gestural means for communication is universal and in­nate. It remains to be seen if it is species-specific to man. It is possible that in man the kinetic use of specific parts of the hu­man body (such as the head, the trunk, the hands, etc.) corresponds to the three main divisions of Osgood's semantic differential.

LANGUAGE ACQUISITION AND COMMON-SENSE 325 2) the amount and expanse of kinetic movement may be hereditary ac­cording to race. In most instances the same is reinforced by culture. 3) The specific direction of each kinetic movement is culture-bound and transmitted through learning. The learning process takes place by imitation and through teaching. The proportion of these two means varies by culture and by SEC group.

The above t h r e e p o i n t s can be i l l u s t r a t e d by an example:

Re. (1) All peoples count with the help of t h e i r f i nge r s . Right-handed persons count on t h e i r r i g h t hand and lef t-handed i n d i v i ­duals make use of t h e i r l e f t hand. This d i s t i n c t i o n holds t rue in most cases I have observed, but i t i s complicated by the fact tha t some persons use both hands, touching the ou t s t r e t ched f ingers with the index of the o ther hand. Re. (2) The speed of the movement of the f ingers and the poss ib le involvement of the lower pa r t of the hand in some kind of motion seems to d i f f e r among r a c i a l groups. My observat ions on wr i s t move­ment in counting are not conclus ive . Re. (3) Some cu l tu re s s t a r t counting by s t r e t c h i n g out the index while o thers begin with the thumb. I t i s most confusing to Euro­peans when Americans i nd i ca t e the number "two" by means of the i n ­dex and the middle f inger . Because they expect to see the thumb and the index f inger for the number two, they sometimes i n t e r p r e t t h i s sign as the number t h r e e .

Given t h a t k i n e s i c s and language can be viewed as e s s e n t i a l l y s i m i l a r in what i s i n n a t e and what i s a c q u i r e d in each of t h e s e d i v i s i o n s of human communicative b e h a v i o r , the nex t q u e s t i o n would be how they i n t e r a c t dur ing the developmenta l s t a g e s .

T r a n s f o r m a t i o n a l - g e n e r a t i v e t heo ry has much o v e r s i m p l i ­f i ed the whole i s s u e of communicative development . S o c i o l i n -g u i s t i c s and p s y c h o l i n g u i s t i c s have been too n e a t l y s e p a r a t e d ; and w i t h i n p s y c h o l i n g u i s t i c s the v i s u a l mode has been severed from speech p e r c e p t i o n . In the e x p r e s s i v e domain i t i s i n c r e a s ­i n g l y c l e a r t h a t the o r a l mode i s c l o s e l y t i e d to k i n e s i c s . Small c h i l d r e n say "Thank you" when r e c e i v i n g and when g i v i n g an o b j e c t . For them the meaning of thank you i s a s s o c i a t e d wi th the t r a n s f e r of an o b j e c t .

In the human i n f a n t by the end of the t h i r d month of l i f e

32 6 WALBURGA VON RAFFLER-ENGEL

a profound physiological change is taking place in the vocal organs. The change is approximately completed at nine months of age. At exactly these same dates in the baby's life one also notices definite changes in gestural behavior.

The same developmental parallelism holds true in respect to brain lateralization. At birth it seems that vocalization and kinesics are kinesthetically linked together. It is not conceivable that a baby will cry harder and kick less when he signals hunger or boredom. An exhausted infant may still cry in pain but this is not the purpose of his act of commu­nication.

Only after lateralization sets in are we able to observe vocal and gestural behavior as units to be analysed separate­ly. After the process of lateralization is complete, at about two years of age, children can be seen talking with only a minimum of body motion. I realize that talking of lateraliza­tion is an oversimplification and that neurologists are still far from the truth. Neurolinguists also have not done much research into kinesics as opposed to grasping motions.

In the verbal-kinetic development of children, the sepa­ration of speech preparatory and speech accompanying gestures seem to begin together. This is another area where research is much needed.

In the case of instances which are clearly kinesic it is not always easy to determine if the child considers the gesture as optional or as obligatory, or, when the child sep­arates the two modes, which one he considers the primary one. Is it the smile that goes with the greeting formula, or does "hello" simply accompany that smile? Many children are capa­ble of waving their hand before they can say "bye-bye", but it is hard to tell if this implies a lag in perception or in production.

Pointing to a desired object while vocalizing is common

LANGUAGE ACQUISITION AND COMMON-SENSE 32 7 to all children. Here it is extremely difficult to decide if the pointing gesture is instinctive or kinesic and even if the pointing gesture is instinctive, we do not know if it may not be perceived as kinesic by the child.

I would like to suggest that more research be done on the acquisition of kinesics. It does seem that the use of ges­tures for communication purposes is innate. The actualizations of each kinesic system seem to be culture-bound. The question whether the use of kinesics is species-specific toman remains to be explored.

To enable the researcher to analyze the physical envi­ronment and the societal interaction the use of film or vid­eotape is absolutely essential. Filming allows for a rapid and fairly extensive coverage of the situational setting and of both the verbal and non-verbal interaction of the child with his mother, his siblings, and any of his significant and non-significant others. Later, the researcher can try to iso­late the distinctive variables and investigate their corre­lation. He can observe the ethology of speech, the child's verbal behavior, his vocal non-verbal (paralinguistic) behav­ior, his kinesic movements (e.g., posture, gestures, and fa­cial mimicry), body language, and proxemics.

The study of language acquisition has again to be viewed in the context of the child's total communicative development. The analysis of these data will take a lot longer than writ­ing grammars of children's speech. To begin with, the verbal and the vocal non-verbal utterances have to be carefully transcribed by a trained phonetician and not merely written down in an approximation to adult pronunciation. To prepare students for this task alone will take some major changes in the present day curriculum of many Linguistics Departments. Constructive research in language acquisition is hard work. Attention to detail is tedious, time-consuming, and unglam-

32 8 WALBURGA VON RAFFLER-ENGEL

orous. The aprioristic theories and short-cuts of the MIT ap­proach have not provided us with much insight.

The post-transformational theories are not very helpful either. Children combine interpretative semantics (e.g., word order) and generative semantics (contextual meaning clues) in their attempt to understand what they hear. The limits of case grammar are difficult to establish.

All linguistic theories, including transformational gram­mar , are interesting, but the study of child language is an interdisciplinary project. Karel Ohnesorg ten years ago sug­gested chairs of pedolinguistics and I would go so far as to suggest Interdepartmental Programs in pedolinguistics.

When we look at the masses of children that arrive at the first year of formal schooling far less well-prepared than their more fortunate middle class peers, and when we consider the complexities of making children bilingual in the South­west of the United States or in Canada, we can only weep over the money which during the sixties governments and private foundations have showered upon scholars sitting at their desk to formulate rules of transformational grammars, many of which are declared 'inoperativef before the ink is dry.

ON THE NATURE OF LANGUAGE AND MIND

UHLAN V. SLAGLE

1.0. There is a growing awareness of the interdependence of lin­guistics, philosophy, and psychology. For it is becoming more apparent all the time that it is impossible to settle any of the really funda­mental questions concerning the nature of mind and language without transcending the boundaries of the individual disciplines. Consequent­ly, since this article deals with the nature of the representation of experience in thought and language, it will necessarily be of an inter­disciplinary nature. Hopefully, the reader will be both tolerant and patient as the considerations involved are of profound significance to linguistics. Indeed, the evidence brought forth in this article strong­ly suggests that the rationalist viewpoint advanced by Chomsky is in­herently incapable of providing an adequate foundation for linguistic analysis.

In this study I shall argue that there is indeed a systematic cor­relation between the structure of immediate perceptual experience and the structure of meaning (both lexical and grammatical) and thought. In this context I shall attempt to show that the factors underlying categorization and the dynamics of thought are factors underlying the organization of perceptual experience itself, thus explaining the basis of the possibility of correlating the structure of perception with the structure of thought and meaning. Moreover, I shall suggest that valid objections to such a theory will not find a solid foundation in recent split-brain research.

2.0. Surprisingly enough, the needed experimental evidence for such a theory has been available for a number of years. For the key

330 UHLAN V. SLAGLE

lies in the brilliant experimental work of the Gestalt psychologists. Unfortunately, the intellectual climate of the times, and the fact that they were wrong in the specifics of their neurophysiological hypothesis, kept scholars from properly appreciating the truly revolutionary break­through made by the Gestalt psychologists. In fact their real break­through came in discovering that similarity and contiguity are factors underlying spontaneous unification in perceptual fields. The role of similarity as a principle of spontaneous unification in sensory fields is convincingly illustrated by the famous Wertheimer dots (cf. Wert-heimer 1923, Köhler 1940). Thus, for instance, given a perceptual 'field' of equidistantly distributed dots with half of the dots having one set of characteristics and the other half having a sharply differ­ent set of characteristics, one will perceive the perceptual field of evenly distribution dots as being organized into horizontal rows if the dots having the same characteristics are arranged in rows; if, on the other hand, the dots having the same characteristics are arranged in columns, the perceptual Yield will be perceived as being organized into vertical columns - even though, on the basis of the relative prox­imity of the dots, it should be equally easy to perceive the dots as being organized into either rows or columns in both instances; this is not, however, the case; indeed, if the factor of similarity is as strong as it is in Wertheimer's original examples, then it is difficult to overcome the factor of similarity and perceive the second theoretically possible (on the basis of contiguity) mode of organization.1 The func­tioning of similarity as a principle of spontaneous unification in sen­sory fields has been even more convincingly illustrated in conjunction with studies of stroboscopic movement (cf. Schiller 1933, Köhler 1940). In the case of stroboscopic movement, one perceives movement where it does not exist in the stimulus pattern but is instead a product of spontaneous perceptual unification. Hence, for example, if two lights are mounted a few inches apart, and one switches the left light on and off, and then quickly switches the right light on and off, the left

ON THE NATURE OF LANGUAGE AND MIND 331

l i gh t w i l l give the appearence of moving across the intervening space to the r i gh t . Now i f one conducts the same experiment on stroboscopic movement with three l igh ts arranged in a row at equidistant in te rva ls , switching the l i g h t in the center on and o f f , followed closely by the simultaneous switching on and of f of the l igh ts on ei ther s ide, one can p r e d i c t , on the bas is o f the f a c t o r of s i m i l a r i t y , i n which d i r e c t i o n the l i g h t i n the cen te r w i l l appear to move. Thus, i f the perceptual object in the center ( in th is case a l i gh t ) i s s i m i l a r to the p e r c e p t u a l o b j e c t on the l e f t (and not s imi lar to the one on the r i gh t ) in regard to color , s ize, and shape, move­ment towards the l e f t w i l l be p e r c e i v e d . I f , on the other hand, the perceptual object in the center i s the same as those on both s i d e s , a movement i n both d i r e c t i o n s w i l l tend to be per ­c e i v e d . Another s t r i k ing i l l u s t r a t i o n of the pr inciples of spontane­ous un i f ica t ion in sensory f i e lds at work l ies in the fact i f the short exposure of an upsidedown V-shaped f igure is fol lowed, at the proper time i n t e r va l , by the exposure of an upright V-shaped f igure situated beneath the f i r s t f i gu re , the f i r s t f igure w i l l be seen as f l i pp ing over into the reverse or ien ta t ion . Thus, what i n f a c t i s the ex­posure of two f i g u r e s w i t h o u t any movement a t a l l i s pe r ­ce i ved as one o b j e c t f l i p p i n g over th rough the t h i r d d imen­s ion (a dimension not given in the stimulus pattern i t s e l f ) . 2 As we have seen in regard to the Wertheimer dots, the same f i e l d effects are operative in s ta t i c perceptual f i e lds as in the perception of movement. This f i e l d ef fect in the perception of s ta t i c phenomena is underscored by the existence of the well-known Müller-Lyer i l l u s i on which shows that the perceived size of an object is strongly influenced by the larger perceptual f i e l d in which the object in question is perceived (cf . Pribram 1971:134). While the i r study of f i e l d effects in general was undoubtedly very valuable, the most important contr ibut ion of the Gestalt psychologists was, as we have already observed, the i r being able to demonstrate that s im i l a r i t y and cont igui ty are pr inciples of

332 UHLAN V. SLAGLE

spontaneous un i f icat ion in sensory f i e l d s . This functioning i s , as Wertheimer 1923 and Köhler 1947 repeatedly emphasized, not l imi ted to any given sense modality.

The signif icance of the f indings of the Gestalt psychologists in regard to the role of s im i l a r i t y and cont igui ty in ef fect ing perceptual un i f i ca t ion is put into proper perspective when one remembers that since the time of A r i s to t l e (cf . Süllwold 1964:36) s im i l a r i t y and cont igui ty have been considered factors underlying the recal l of past experience, and when one remembers that James (1950 [1890] I I , 641-51), Jevons (1890:121-23), J . S. M i l l (c f . Slagle 1974), Ehrenstein (1965:170-218), and many others have cogently argued that resemblance provides the ult imate foundation of log ic , with respect to both inference and c las­s i f i c a t i o n . Cons idered w i t h i n t h i s c o n t e x t , the expe r imen ta l f i n d i n g s of the G e s t a l t p s y c h o l o g i s t s s t r o n g l y i n d i c a t e t h a t the f a c t o r s u n d e r l y i n g the dynamics of memory and the process of c a t e g o r i z a t i o n are a l r e a d y o p e r a t i v e in the o r ­g a n i z a t i o n o f p e r c e p t u a l expe r ience i t s e l f . 3

Thus, i t is extremely interest ing to note that Köhler and Restorff 1935 suggested that s im i l a r i t y as a pr inc ip le of spontaneous grouping in sensory f ie lds underlies the process of recogni t ion, with s im i l a r i t y ef fect ing a functional union of ' t race ' and perceptual process. The fundamental v a l i d i t y of such a posit ion is brought c lear ly into focus when one examines such phenomena as Gestalt completion f igures. For they, as Wallach (1961:167) astutely observed, show that memory t r a c e s can become f u n c t i o n a l l y u n i f i e d w i t h the p e r c e p t u a l process i t s e l f , w i t h the memory ' t r a c e s ' a c t u a l l y s u p p l y i n g the m i ss i ng aspects of i ncomp le te f i g u r e s . When th is functional un i f i ca t ion of trace and perceptual process takes place, what at f i r s t appears to be a group of unconnected, unrelated elements in the visual f i e l d spon taneous ly r e o r g a n i z e s i t s e l f , with the apparently un­related parts sometimes almost l i t e r a l l y snapping into place, and the previously disparate elements being perceived as parts of a coherent

ON THE NATURE OF LANGUAGE AND MIND 333

whole representing something one has seen in the past, such as, for ex­ample, a telephone or a large merchant vessel .4 Now in o rde r f o r memory t r a c e s to be ab le to become f u n c t i o n a l l y u n i f i e d w i t h the p e r c e p t u a l process i n such a way as to be ab le to f u r ­n i sh the m iss i ng aspects o f the i ncomp le te s t i m u l u s p a t t e r n s r ep resen ted by G e s t a l t comp le t i on f i g u r e s , the sensory p r o ­cess u n d e r l y i n g the memory of a g iven expe r i ence must neces­s a r i l y be h i g h l y s i m i l a r to the sensory process u n d e r l y i n g the p e r c e p t u a l expe r ience of t h a t phenomenon (as we shall show, recent sp l i t -b ra in research of fers no substantive counter arguments). Since we know t h a t s i m i l a r i t y u n d e r l i e s spontaneous u n i f i ­c a t i o n in sensory f i e l d s , t h i s r e q u i r e d s i m i l a r i t y of the two sensory processes means t h a t s i m i l a r i t y as a p r i n c i p l e o f spontaneous u n i f i c a t i o n i n sensory f i e l d s must u n d e r l i e the spontaneous f u n c t i o n a l u n i f i c a t i o n of ' t r a c e ' and per ­cep tua l process e x e m p l i f i e d in G e s t a l t c o m p l e t i o n f i g u r e s . Thus, an examination of Gestalt completion f igures indicates that sim­i l a r i t y , functioning as an immanent organizational factor of sensory f i e l d s , underlies the process of recognition (as Köhler and Restorff 1935 suggested) and consequently the a b i l i t y to categorize, categori­zation being, of course, the process through which we relate present experience to s imi lar past experience, the sine qua non of success­f u l l y adapting to and cont ro l l ing one's environment. Here, i t should be noted that the experiments on stroboscopic movevent (where move­ment is perceived which does not ex is t in the stimulus pattern) have shown that spontaneous un i f ica t ion in sensory f i e lds can take place even when there are differences in the d is ta l st imul i involved in re­gard to s ize, or ientat ion ( e . g . , the f i r s t stimulus being upsidedown and the second upr igh t ) , etc. ( for de ta i l s , see Schi l ler 1933). More­over, the spontaneous f u n c t i o n a l u n i f i c a t i o n o f t r a c e and p e r c e p t u a l process e x e m p l i f i e d i n G e s t a l t compi e t i on f i g u r e s takes p lace even when no cor respondence in s i z e and o r i e n ­t a t i o n e x i s t s between the G e s t a l t c o m p l e t i o n f i g u r e s and

334 UHLAN V. SLAGLE

any previous perceptual experience of the phenomenon in question, thus explaining how we recognize new instances of given categories even though they may differ in a number of respects from any previous experience of that phenomenon. The stronger the factor of similarity is, the more readily spontaneous unification takes place in sensory fields; therefore, it seems reason­able to assume that the spontaneous unification of trace and percep­tual process is far more likely to occur under normal stimulus condi­tions represented by Gestalt completion figures where the stimulus pat­tern is often far less clearly defined. This assumption is supported by the fact that the more fully Gestalt completion figures are depicted, the easier they are to recognize.

There is then no need to posit a separate intellectual faculty, for the process of categorization can be fully explained in terms of the functional unification of trace and perceptual process effected by the functioning of similarity as a principle of spontaneous unifica­tion in sensory fields, with the arbitrary symbols of language becoming integrated into the representational system through the functioning of contiguity as a principle of spontaneous unification. Hence, for exam­ple, we can recognize new instances of the category 'dog' by virtue of the fact that the symbol dog has been previously experienced in con­junction with an appropriate instance of that category; this means that when the perceptual process underlying the experience of a new instance of a given category becomes functionally unified with the appropriate memory trace (as in Gestalt completion figures), the present perceptual process is normally also functionally united with the process underly­ing the memory of the experience of the relevant symbol (cf. Slagle 1973b).

3.0. Basic to the foregoing analysis is the assumption that the structure of thought is describable in terms of the structure of sen­sory experience. Indeed, the very fact that memory 'traces' can 'com­plete' the incomplete figures in Gestalt completion tests, shows that

ON THE NATURE OF LANGUAGE AND MIND 335

the structure of information 'stored' in the mind is describable in terms of the structure of perception itself. Such an interpretation finds powerful support in the parallels between the 'Aha-Experience' of thought and spontaneous structural reorganization on the perceptual level as represented by Gestalt completion figures and reversible figures. Often when looking at Gestalt completion figures, for example, at first one sees a group of disparate, seemingly unrelated items, and at the next moment the parts suddenly and spontaneously 'snap' into place and a unified whole is perceived (cf. Thurstone 1944:20). Revers­ible figures offer a similar experience. At one instant, one perceives one figure at the next instant the perceptual field spontaneously reorganizes itself and a different figure is perceived (cf. Krech 1969: 166). Both cases are, of course, strikingly similar to the 'Aha-Experi­ence' during which suddenly one 'sees' the light and understands the problem, with the seemingly disparate aspects of the problem falling into place and being 'seen' for the first time in their proper rela­tionship.

The parallels between the 'Aha-Experience' of imageless thought and structural reorganization on the perceptual level also support the assumption that similarity and contiguity as principles of spontaneous unification in sensory fields are determinants of the dynamics of thought and memory.

Before going on, I should like to emphasize that the stress on the purely 'perceptual' aspects should in no way be interpreted as meaning that I am unaware that the 'emotional' and motor faculties of the human organism play an extremely important role in controlling the attention processes which constitute a critical part of the perceptual process. This fact should be taken into consideration when judging the following discussion of the important role of perceptual focus and distinctive features.

Distinctive features offer no difficulty within the proposed theo­retical framework. Figure-ground differentiation is perhaps the most

336 UHLAN V. SLAGLE

ubiquitous mode of perceptual organization. Indeed, there is some form of figure-ground differentiation in all domains of sensory experience, and it is the interaction of the functioning of figure-ground differen­tiation (or perceptual focus) with that of similarity as a principle of spontaneous unification in sensory fields which underlies the distinc­tive features upon which classification is based. Generally speaking, any given segment of experience can be regarded as being similar to a number of different phenomena, depending upon which salient features constitute the focal point of attention. Hence, if we focus upon any given set of features, we help to establish the conditions which facil­itate the spontaneous functional unification of the perceptual process underlying the experience of that set of characteristics with its coun­terpart in the trace system, and when this functional unification of 'trace' and perceptual process is effected, the phenomenon in question is recognized as belonging to the appropriate category. Needless to say, if different characteristics were focused upon, different 'traces' would functionally interact with the perceptual process and consequently the phenomenon would be categorized differently.

4.0. Naturally, any theory of mind which does not address itself to the issue of problem solving is simply ducking the issues. Problem solving often takes the form of applying old knowledge in a new context. This can easily be explained within the suggested framework. The solu­tion to a given problem comes when the process underlying the appropri­ate past experience becomes functionally unified with the process un­derlying the experience of the criterial aspect of the present problem. Thus, for example, if we see something, such as a stout stick, which has certain characteristics in common with something, such as a lever, we have used, or seen used, in the past to overcome a similar difficul­ty, then the new phenomenon (the stick) can become 'imbedded' in the old context (the use of a lever) via the functioning of similarity as a principle underlying the spontaneous functional unification of trace and perceptual process. When this functional unification takes place,

ON THE NATURE OF LANGUAGE AND MIND 337

we can 'see' that the new item (the stick) can be used in the same way as something we are already familiar with (the lever) As has been pointed out in several recent studies, dreams, hypnagogic images, and hallucinations shed light on the perceptual process itself (cf. Neisser 1967:145-54; Pribram 1971:162). Thus it is not surprising that proof that spontaneous unification of the sort posited occurs is provided by the nature of dreams. For example, not too long ago I had difficulty parking on a driveway and I had also recently seen a miniature auto in a Laurel and Hardy movie. Well, last night in my dreams I was having similar difficulties parking on a similar driveway, one I had seen many, many times but had certainly never parked on (not having been able to drive at the time); moreover, I was driving the same sort of miniature auto I had seen in the Laurel and Hardy movie (but had certainly never actually driven). What had happened in this dream is that the processes underlying different previous experiences had be­come functionally unified in such a way as to create a to­tally coherent but completely new (as a whole) experience. The 'real' experience I had had was similar to this - it is just that most of the component parts, including the location, were different; and in each case the integration of those component parts into this experience can be explained in terms of similarity effecting their functional unification into this pattern. As the same 'faculty' which underlies the organization of dreams also underlies the organization of perception itself, such dreams show that our nervous system is ob­viously able to effect the posited interaction of memory trace and per­ceptual process and thereby provide the basis for man's ability to apply old knowledge in new contexts. This, of course, ties in beauti­fully with the suggested approach to the problem of categorization, which posits the functional unification of trace and perceptual process, as exemplified in Gestalt completion figures. In evaluating this ex­planation of the basis of problem solving, one should not forget the striking parallels between structural reorganization on the perceptual

338 UHLAN V. SLAGLE

level and the 'Ana-Experience' of thought (a topic we have already touched upon).

Contrary to what many might think» recent split-brain research does not invalidate the suggested theory. To begin with, while there is cer­tainly some lateralization of function, the language dominant hemisphere can, in general, perform the same tasks of perceptual discrimination as the minor hemisphere (cf. Gazzaniga 1970:105; Levy, Trevarthen and Sperry 1972; Hilliard 1973). Moreover, if the language dominant hemi­sphere were unable to distinguish between sensory phenomena in any given domain, then it would be unable to correctly classify the phenomena in question in terms of the semantic code of any given language; consequent ly, the other hemisphere would necessarily need to be involved in some way in any linguistic categorization of such phenomena. Of considerable importance in this context is the fact that the tasks of perceptual dis­crimination and classification that the language dominant hemisphere performs are far too complex to be explained in terms of atomistic dis­tinctive features - for what is necessarily involved in such classification is not only a set of distinctive features but also a systematic pattern of arrangement of those fea­tures (i.e., a Gestalt); otherwise one would not be able to distinguish between those distinctive features put together in any random arrangement and those features when put to­gether according to some meaningful pattern representing an instance of a given category. Finally, while the language dominant hemisphere is a 'poor artist', it would be very naive indeed to equate what one can draw with either the way one perceives or remembers any given phenomena, since children (and many adults as well) are often un­able to draw anything accurately whether from memory or not. Yet these same 'artists' have no problems in perceptual discrimination and are fully aware when confronted with an accurate depiction (such as a photo­graph) that their drawings are inaccurate renditions of what they were drawing from memory. Since a memory trace (of whatever nature) must underlie the ability of there poor artists to make this distinction in

ON THE NATURE OF LANGUAGE AND MIND 339

regard to the accuracy of a photograph versus their drawing, and since memory will suggest that the photograph is correct, then the photograph must come closer than the drawing to representing how past information is stored in the mind. Consequently, the fact that the major hemisphere is a 'poor artist' proves nothing of consequence to the arguments ad­vanced in this study.

5.0. If my viewpoint is correct, and the dynamics and structure of thought can be correlated with the dynamics and structure of perception, then this correlation should be reflected in the nature of grammatical meaning, since the 'laws of grammar' are commonly considered to reflect the laws of thought. I believe that there is in fact ample evidence to support this position. This evidence can perhaps best be put into pro­per perspective by a brief examination of Kant's schema concept.

Although Kant is often criticized for not dealing with the prob­lem of language, he did deal with the problem of just what constitutes the basis of applying logic to sensory experience (cf. Milmed 1961:26 to 28). Indeed, Kant's transcendental schemata can, in a sense, be re­garded as referential rules for the innate logical categories (cf. Kör­ner 1955:70-73; Beck 1969:480). Hence, given the intimate connection between logic and grammar, it should be of great interest to note that (1) Kant derived the transcendental schemata (i.e., the referential rules for the categories) from the nature of time as the highest level invariant of sense experience (cf. Paton 1936 II, 77), and (2) he spe­cifically related the modes of the spatial organization of sensory ex­perience to the possibility of applying the categories to experience; in fact, he suggested that the categories could be understood only in terms of their exemplification in the spatial aspects of sensory orga­nization (cf. B291-93). These insights represent a revolutionary ad­vance - a breakthrough Kant unfortunately did not follow through on.5

Implicit in Kant's position is the assumption that the function of thought is to organize sensory experience into a coherent, unified whole. Kant, of course, thought it necessary to posit a separate intel-

340 UHLAN V. SLAGLE

lectual faculty in order to explain the basis of organizing experience into a unified whole. Yet as we have seen, similarity as a factor ef­fecting spontaneous unification in sensory fields brings about the func­tional unification of memory traces (past experience) and perceptual process (present experience) which underlies the possibility of orga­nizing experience into a unified whole. This means that, as I have sug­gested elsewhere (Slagle 1974), if one reinterprets Kant's transcenden­tal schema doctrine in the light of the experimental findings of Gestalt psychology, then the need to distinguish between schema and category disappears since there is no dichotomy of thought and perception. With­in such a framework the meaning of grammatical and logical terms is, in general, based on the structural aspects of perception, such as re­semblance and the modes of psychological (i.e., perceived) space and time, with the function of thought being that of organizing experience into a coherent whole in terms of resemblance, cause and effect, etc. The mention of similarity as a structural aspect of sensory experience will surprise some. However, as Bertrand Russell (1940:347, 1959:172 to 174) observed, even resemblance is given in immediate perceptual ex­perience. In addition, it should be noted that even Whorf recognized the universality of the modes of consciousness which constitute per­ceived space and time within the framework of Gestalt psychology (see Slagle 1974).

I am suggesting then that the structural aspects of sensory ex­perience such as resemblance, causality, and the modes of perceived space and time, constitute, in general, the referential domain of gram­mar and logic. The evidence for such a position has been available for a long time; however, until the advent of Gestalt psychology it was im­possible to establish a sound basis for adequately interpreting this material.

In his New Essays Concerning Human Understanding (1710) Leibniz suggested that

ON THE NATURE OF LANGUAGE AND MIND 341

it will ... be well to consider this analogy between sensible and non-sensible things ... as, for example to, with, from, before, in, outside, by, for, upon, towards (à, avec, de, devant, en, hors, par, pour, sur, vers), which are all derived from place, from dis­tance, and from motion, and afterwards transferred to every sort of change, order, sequence, difference, agreement. (1961 [1765]: III.i.5) (Translation mine)

Thus, according to Leibniz, prepositions denoting logical relationships are derived from terms originally denoting sensory relationships. The same pattern Leibniz noted in regard to prepositions is also reflected in the derivation of morphological markers. This evidence is compelling enough to have led a number of first-rate scholars, among them Brugmann (1911:473), Delbrück (1901:133), and Whitney (1882:91), to the view­point expressed by Cassirer when he wrote (1964[1923]):

In general, it turns out that the Indo-Germani case forms served originally to express spatial, temporal or other perceptions, and that they only later gradually acquired an abstract sense. (Trans­lation mine)

This pattern, it was believed, constitutes a universally valid state of affairs. Now the dominance of terms denoting spatial relationships is such that some scholars advanced a purely local ist version of this hy­pothesis. I believe that Wundt was correct in suggesting that the dom­inance of spatial terms in this pattern of derivation came from the fact that spatial relationships constitute the highest level invariants of sensory experience and that therefore terms denoting spatial rela­tionships provide an ideal foundation for calling attention to the oth­er sensory relationships, which necessarily co-occur with spatial rela­tionships (Wundt 1900:73-75). Consequently, if grammatical meaning is ultimately based on sensory relationships, it would seem a natural state of affairs for case markers and prepositions to have often been derived from terms originally signifying spatial relationships. One need only examine the copious evidence presented in Cassirer (1964 [1923]) to realize that this pattern of derivation cannot easily be explained away. I believe then that this pattern, when interpreted in the light of

342 UHLAN V. SLAGLE

Wundt's insight and the experimental findings of Gestalt psychology, shows that the criterial conditions for the use of grammatical concepts are founded in the structural aspects of sensory experience. In this context it is important to recall the important role perceptual focus plays in the proposed theory. As I have suggested elsewhere (1974), I think Brown was correct in maintaining that the part-of-speech member­ship of a word operates as a perceptual filter which guides our atten­tion towards the relevant aspects of the given phenomena being classi­fied (see Brown 1957:3, 1958:249-53). The necessity of such a filtering function arises from the fact that actions and attributes, the charac­teristic denotata of verbs and adjectives, are not experienced in isola­tion. Thus, for instance, one will never experience 'playing' apart from someone doing the playing; nor will one ever encounter 'tallness' in isolation from someone or something tall. Hence, if we are to learn the denotative use of given terms in concrete situations we must be provided with some information on just which aspects of the phenomenon in question (e.g., a tall boy who is playing) the terms refer to. It is important to note that within this framework one can easily explain the shifts in form class (e.g., how verbs become or function as nouns, etc.) which often take place, with a shift in focus correlating with a shift in form class function (see Slagle 1974 for details).

6.0. From the foregoing considerations it is apparent that there is a substantial body of evidence which indicates that the suggested position on the nature of language and mind is correct.

Needless to say, the viewpoint advanced in this study differs to a considerable extent from the sort of rationalism advocated by Chomsky. While the suggested theory is based on phenomena which demonstrably exist regardless of one's theoretical frame of reference, the ratio­nalism of Chomsky and his followers is largely founded on the capacity of 'Cartesian theory' to explain the complex relationships existing between the hypothetical constructs (e.g., 'surface and deep struc­ture') posited by Chomsky and his fellow transformationalists.

ON THE NATURE OF LANGUAGE AND MIND 343

Amazingly enough, they suggest that the goal of linguistic theory is to explain these relationships (cf. Chomsky 1972) - relationships which in fact exist in many cases only in their collective imagi­nation. To be sure, they are touching upon crucial problems in their work. It is simply that the issues involved have become distorted. I believe that the sort of rationalism which seems inherent in transfor­mationalist theory has been out-of-date since 1781, namely, since the appearance of Kant's Critique of Pure Reason. Kant gave us the key to relating the formalism underlying logic (and thus grammar) to the struc­ture of experienced reality. Chomsky and his colleagues have simply avoided the issue, which is, of course, in keeping with the Cartesian tradition they represent. Unfortunately, this is a problem which cannot profitably be avoided. Consequently, I believe that Kantian insights, when fully modified in the light of Gestalt psychology, provide a far sounder foundation for linguistics than the Cartesian insights in which the transformationalists place their faith.6 For such a synthesis pro­vides the basis for correlating lexical and grammatical meaning not on­ly with the structure of thought but also with the structure of experi­enced reality.

NOTES

For convincing examples of 'Wertheimer' dots, see not only Werthei-mer 1923 but also Köhler (1940:61-62). For detailed information on the experiments which show that the same factors underlying unification in static sensory fields are operative in the structuring of stroboscopic movement, see especially Schiller 1933. Also useful are Koffka (1935:301) and Köhler (1940:60). Krech (1969:223) provides excellent illustrations of the points under dis­cussion in regard to stroboscopic movement; on the other hand, the adaptation of the Wertheimer dots in Krech 1969 is poor.

344 UHLAN V. SLAGLE

On the basis of closely related evidence, Walter Ehrenstein (1954: 317-18, 1956:72-73, 1965:207) has argued in considerable detail that the immanent organizational factors of perception are the immanent organizational factors of thought. The position advanced in this study differs from Ehrensteins's in that it incorporates a number of in­sights not represented in Ehrenstein's work since he left a number of crucial issues unexplored, among them the nature of the interaction of memory and perception - an issue of paramount importance to an ad­equate theory of mind. Examples of Gestalt completion figures are found in Thurstone 1944 and Krech (1968:75). Kemp Smith [1923] 1962 provides the best analysis of the implications of Kant's breakthrough; on this, see pp. 35-36, 181, 184-86, 335-40 of his Commentary, In this context, see also Smart 1955. On the role of psychological considerations in Kant's theory (a hotly debated is­sue), see Wolff (1963:177) and Vaihinger (1922:I,324). In this context it is very illuminating to note that Humboldt, about whom Chomsky has said a great deal, was far more Kantian than Carte­sian. On this, see Slagle 1973a.

REFERENCES

Beck, Lewis W. 1969. Early German Philosophy, Cambridge, Mass.: Harvard Univ. Press.

Brown, Roger W. 1957. "Linguistic Determinism and the Parts of Speech". Journal of Abnormal and Social Psychology 55.1-5.

. 1958. Words and Things, New York: Free Press. Brugmann, Karl. 1911. Lehre von den Wortformen und ihrem Gebrauch. (=

Grundriss der vergleichenden Grammatik der indogermanischen Spra­chen, 2.) Strassburg: . J. Trübner.

Cassirer, Ernst. 1964[1923]. Die Sprache, (= Philosophie der symboli­schen Formen, 1.) 4th ed. Darmstadt: Wissenschaftliche Buchgesell­schaft. (6th unchanged ed., 1973.)

Chomsky, Noam. 1963. "Perception and Language". Boston Studies in the Philosophy of Science ed. by Marx W. Wartofsky, vol.1.199-205. Dor­drecht/Holland: D. Reidel.

. 1966. Cartesian Linguistics: A chapter in the history of rationalist thought. New York & London: Harper & Row.

ON THE NATURE OF LANGUAGE AND MIND 345

Chomsky, Noam. 1970. "Problems of Explanation in Linguistics". Explana­tion in the Behavioural Sciences ed. by Robert Borger and Frank Ci-offi, 425-51. Cambridge: Cambridge Univ. Press.

1971. Problems of Knowledge and Freedom. New York: Pantheon Books.

. 1972. Language and Mind, 2nd enl. ed. New York: Harcourt-Brace-Jovanovich. (1st ed., 1968.)

Delbrück, Berthold. 1901. Grundfragen der Sprachforschung, mit Rück­sicht auf W. Wundt's Sprachphilosophie erörtert. Strassburg: . J. Trübner.

Ehrenstein, Walter. 1954. Probleme der ganzheitspsychologischen Wahrneh­mungslehre. 3rd ed. Leipzig: J. A. Barth.

. 1956. "Intelligentes Denken". Die Ganzheit in Wissenschaft und Schule: Johannes Wittmann zum 70. Geburtstag, 43-95. Dortmund: W. Crüwell.

. 1965. 'Probleme des höheren Seelenlebens. Munich: E. Rein­hardt.

Gazzaniga, Michael. 1970. The Bisected Brain. New York: Appleton-Centu-ry-Crofts.

Hilliard, Ronald. 1973. "Hemispheric Laterality Effects on a Facial Recognition Task in Normal Subjects". Cortex 9.246-58.

James, William 1950[1890]. The Principles of Psychology. 2 vols. New York: Dover Publications.

Jevons, William. 1890. Pure Logic and other Minor Works. London: Mac­millan & Co.

Kant, Immanuel. 1956[1781; 1787]. Kritik der reinen Vernunft. Critical ed. Hamburg: F. Meiner. [Contains the text of both the 1781(A) and the 1787(B) editions.]

Koffka, Kurt. 1935. Principles of Gestalt Psychology. New York: Harcourt, Brace & Co.

Köhler, Wolfgang. 1940. Dynamics in Psychology. New York: Liveright. . 1947. Gestalt Psychology. Ibid. , and H. von Restorff. 1935. "Zur Theorie der Reproduktion".

Psychologische Forschung 21.56-112. Körner, Stephen. 1955. Kant. Harmondsworth, Middlesex & Baltimore, Md.:

Penguin. Krech, David, and Richard S. Crutchfield. 1968. Grundlagen der Psycho­

logie. Transl. and ed. by Hans Werner Wendt and Otto M. Ewert. Ber­lin: J. Beltz. (3rd ed., 1971.)

346 UHLAN V. SLAGLE

Krech, David, and Richard S. Crutchfield. 1969. Elements of Pschology, 2nd ed. New York: A. A. Knopf.

Leibniz, Gottfried Wilhelm 1961[1765]. Neue Abhandlungen über den menschlichen Verstand, Transl. from the French and ed. by Wolf von Engelhardt and Hans Heinz Holz. 2 vols. Frankfurt/M.: Insel-Verlag. [Contains the original text as well.]

Levy, Jerre, Colwyn Trevarthen, and Roger W. Sperry. 1972. "Perception of Bilateral Chimeric Figures following Hemispheric Deconnexion". Brain 95.61-78.

Milmed, Bella. 1961. Kant and Current Philosophical Issues, New York: New York Univ. Press.

Neisser, Ulric. 1967. Cognitive Psychology, New York: Appleton-Century-Crofts.

Paton, Herbert J. 1936. Kant's Metaphysic of Experience: A.commentary on the first half of Kant's "Kritik der reinen Vernunft", 2 vols. London: Allen & Unwin.

Pribram, Karl. 1971. Languages of the Brain, Englewood Cliffs, N.J.: Prentice-Hall.

Russell, Bertrand. 1940. An Inquiry into Meaning and Truth, London: Al­len & Unwin.

. 1959. My Philosophical Development, New York: Simon & Schu­ster.

Schiller, Paul von. 1933. "Stroboskopische Alternativversuche". Psycho­logische Forschung 17.179-214.

Slagle, Uhlan V. 1973a. "The Kantian Influence on Humboldt's Linguistic Thought". Historioagraphia Linguistica 1:3.341-50 (Sept. 1974).

. 1973b. "The Relationship of the Structure of Meaning to the Structure of Experienced Reality". Linguistics 138.81-95 (15 Oct.1974).

. 1974. Language, Thought, and Perception: A proposed theory of meaning. The Hague: Mouton.

Smart, Harold. 1955. "Two Views on Kant and Formal Logic". Philosophy and Phenomenological Researcti 16.155-71.

Siillvold, Fritz. 1964. "Gedachtnistätigkeit und Vorstellungsverlauf". Lernen und Denken (= Handbuch der Psychologie, 1:2), 36-52. Göttin­gen: Verlag für Psychologie.

Thurstone, Louis Leon. 1944. A Factorial Study of Perception, Chicago: Univ. of Chicago Press.

Vaihinger, Hans. 1922. Kommentar zu Kants Kritik der reinen Vernunft, 2 vols. 2nd ed. Stuttgart: Union Deutsche Verlagsgesellschaft.

Wallach, Hans. 1961. "Some Considerations concerning the Relation be-

ON THE NATURE OF LANGUAGE AND MIND 347

tween Perception and Cognition". Documents of Gestalt Psychology ed. by Mary Henle, 164-71. Berkeley & Los Angeles: Univ. of California Press;

Wertheimer, Max. 1923. "Untersuchungen zur Lehre von der Gestalt". Psy­chologische Forschung 4.301-50.

Whitney, William Dwight. 1882. "General Considerations on the Indo-Eu­ropean Case-System". Transactions of the American Philological Asso­ciation 13.88-100.

Wolff, Robert. 1963. Kant's Theory of Mental Activity. Cambridge, Mass.: Harvard Univ. Press.

Wundt, Wilhelm. 1900. Völkerpsychologie. Vol.l: Die Sprache. Part 2. Leipzig: W. Engelmann. (3rd rev. ed., 1912.)

Addendum :

Smith, Norman Kemp. 1962[1923]. A Commentary to Kant's Critique of Pure Reason. New York: Humanities Press.

IV. E P I S T E M O L O G Y A N D H I S T O R Y OF L I N G U I S T I C S

EPISTEMOLOGICAL DILEMMAS AND THE TG PARADIGM

LYLE CAMPBELL

O.O. My purpose in this paper is to express my disen­chantment with certain basic assumptions of the Transforma­tional-Generative (henceforth TG) paradigm. I will preface my comments with some general dissent, and then turn to spe­cific epistemological dilemmas.

Indoctrinated, as a graduate student in TG dogma, to strain at gnats and swallow camels, I nonetheless could nev­er feel quite comfortable with such needlessly unreal ide­alizations such as the ideal speaker-hearer in a completely homogeneous speech community, or the instantaneous acquisi­tion of language. I had difficulty swallowing the Katz and Fodor (1963) gnat (or was it a camel?) of linguistic descrip­tion divorced from context (real-world or linguistics). I never quite understood either the basis for the assertion that kinds are universally constrained to learn the simplest grammar. Finally, I was defensive and ashamed that my sci­ence, like no other, needed an artificial simplicity metric to fix up the poor fit between theory and data.1 These doubts led to more deep-seated heresies, some of which fol­low.

1.0. Psychological vs. sociological r e a l i t y . My soul-searching over the nature of language leads me to believe

35 2 LYLE CAMPBELL

that there is an epistemological dilemma, common to all so­cial sciences, which stems from theoretical goals, i.e., from the kinds of questions one wants his theory to answer. The psychological reality goal of the TG paradigm requires an account of native speaker's knowledge of their language. What constitutes data relevant for answering this question and what the methods are for obtaining such data are obvi­ously different questions from those of someone with the goal of accounting for 'sociological reality' and who asks questions about how language is used and how linguistic vari­ation may be conditioned by aspects of society and culture. Those with the psychological reality goal find the locus of language in the individual (whether real or ideal), while those seeking sociological reality will attribute the locus of language to the community. Investigation will be centered on either the individual or the group, depending on interest and goals, and the data for one goal will be deemed inappro­priate for by those holding the other goal. The difference in goals and basic questions is then often misunderstood so that considerable squabbling in the literature between those of different goals boils down to so much methodological quib­bling. For example, Labov (1970) charges traditional TGists with failing to account for variation because intuition is the source of their data; on the other hand Kiparsky (1971) claims that the percentage attached to Labov's variable rules don't have psychological reality. Neither the psychological nor the sociological realists seem capable of answering the questions posed by the other.

The consequences of this dilemma become apparent when we consider how it has affected work within the TG paradigm. I will consider only TG work in dialectology. It is basically impossible to account for dialect variation while still main^

EPISTEMOLOGICAL DILEMMAS AND THE TG PARADIGM 353 taining the TG goal of psychological reality (cf. Campbell 1972, 1973 for details) .

Two basic approaches to dialectology have been taken in TG works. The first was proposed by Halle (1962) and fol­lowed by Keyser (1963), Klima (1964), Saporta (1965), Sledd (1966), Vasiliu (1966), Saltarelli (1966), and others. In this approach essentially a single grammar of the language is written, assuming nearly the same underlying forms for all dialects, with extension rules to account for interdia­lectal variation. For example, given the following two dia­lects (based on Quechua) in which two distinct underlying segments, one of which conditions a rule, have merged, they would be described as:

D I Underlying segments: /k/, /q/, /i/, /u/ Rule I: i e

u / q D II Underlying segments: same as D I

Rule I: same as R I in D I Rule II: q →

But this account misses the goal of psychological reality; neither the segment q nor the merger rule to get rid of it (R II) can have any psychological reality for speakers of D II, since q is never heard by children learning D II.

The other TG approach to dialectology has been taken by Becker (1967), King (1969), and others. In it individual grammatical descriptions are written for each dialect, as­suming no knowledge of the existence of other dialects in the process. This procedure places high priority on the psy­chological reality goal, but falls prey to the dilemma be­cause there is no basis within TG theory for subsequently comparing the independently written grammars of separate dialects. Independently written descriptions have underlying

354 LYLE CAMPBELL forms and rules appropriate only internally to a single gram­mar. Saussure's dictum that structural units are defined within an idiolect by the web of relationships locking them into the system hold for TG grammars as well; hence it is impossible to compare two dialects which have independently written grammars.

The example considered above will help to clarify this matter. In this approach D I would have:

Underlying segments: /1/, /u/, /k/, /q/ Rule I: (same as above) i e /

u Then D II with its independent account would have:

Underlying segments: /i/, /u/, /e/, //, // Rules: none

Since these two accounts have their parts defined by their relation within the individual grammars, there is no basis for comparison. Both have different underlying forms, differ­ent rules acting upon those underlying forms, and different surface forms. There is little more justification for com­paring these than for comparing grammars of Quechua and Chi­nese, since these two will have different underlying forms, different rules, and different surface forms. Thus the dia­lect variation is left unaccounted for.2

TG machinery is not adequate for both psychological re­ality and variation.

2.0. Static vs. dynamic description. A Siamese twin sister dilemma (in that it is different but not distinct) of the foregoing comes from diachronic wrinkles in synchrony and synchronic approaches to diachrony. Again it is common to all social sciences. The time when linguistics was con­sidered solely a diachronic enterprise will be easily re-

EPISTEMOLOGICAL DILEMMAS AND THE TG PARADIGM 3 55 called and contrasted with the modern leanings toward com­plete homogeneity, where even historical change is seen as successive static (synchronic) states. I will consider two aspects of the static-dynamic dilemma.

Weinreich, Labov, and Herzog (1968) have shown that the treatment of historical change as successive homogeneous syn­chronic grammars forces one to the counter-factual assumption that all change in a grammar is introduced from without (cf. King 1969, for confirmation of this in the TG paradigm). TG is not only committed to this view of change, but it is inca­pable of any other (cf. again Weinreich, et al. 1968). The change of q to in D II considered above is rather straight­forward - there was a grammar with q originally; a later re­structured grammar had only - but through what process did q become k? The transition problem (the intermediate grammar between original q and later k) is neglected.

The other side of this paradox is the difficulty of keep­ing historical considerations out of purely synchronic descrip­tions. The difficulty arises in attempting to generate semi-productive forms which are often difficult to distinguish from frozen, unproductive forms. Most languages have 'fossils' of past changes preserved in various layers from fairly reg­ular morphophonemic processes (e.g., English g-dropping in strong-stronger), to semi-transparent (e.g., the English vow­el-shift), to totally quixotic processes (e.g., umlaut mouse-mice, goose-geese, man-men; palatalization drink-drench, hake-batch, c o o l - c h i l l ; initial consonant cluster simplification know-acknowledge). Where do phonological rules stop? - All too often phonologies are written with rules which reflect compli­cated historical processes without even lip-service to exter­nal evidence for the synchronic psychological reality of the rules. It is my firm belief that child language-learners with

356 LYLE CAMPBELL different abilities and different experiences may learn dif­ferent processes. This seems to be supported in work by Sher-zer (1971), Ohala (1972), and others. How are we to account for the possibility that some may have learned more rules than others? That is, to take an example, what do we do with two children, one exposed to Latinate vocabulary early thus acquiring the vowel shift rule in some sense, the other ex­posed to vowel-shift illustrating lexical items quite late thus learning them piecemeal as separate lexical items? What is wrong with the alternative view of language acquisition that a child gradually acquires his grammar, modifying it minimally to accommodate new material and sometimes learning some aspects of it the hard way since he did not hit upon the simple solution until he already had a successful slightly less simple solution? Even Kiparsky (1971) has shown (using Hale's material) that Maoris apparently were perverse enough to fail to learn a simple phonological rule which forced the learning of complicated morphologically conditioned allomorphs for a variety of morphemes.

If it is true that children may acquire different gram­mars or non-simple ones, then much of the current noise about abstract analyses, global constraints, opacity, and unordered rules would seem to make little sense. One has to know what a rule is and for whom and how, before questions of how to order one can be answered.

Perhaps to solve problems of group vs. individual focus or 'fossils' of historical changes preserved as synchronic reflexes we must step outside the TG paradigm, reassess our goals and strategies for accomplishing those goals, and then begin again with new trowels and pickaxes.

357

NOTES

1 It seems unfair of me to take undocumented potshots at these assump­tions; however, I hope to have time to develop these criticisms with proper bibliography later. The evaluation metric based on a simplicity that must be defined ad hoc makes linguistics quite different from oth­er sciences. Others characteristically test hypotheses by devising cru­cial experiments which should resolve the issue. But since it is claimed that a number of grammars may account for the same data, we can create no crucial experiment. We are thus forced to resort to an artificial notion of simplicity designed to help eliminate some of the alternative grammars. If the fit between data (gotten from intuition in TG).and theory were better, we could need no such device. But since intuitions waver in just the cases where a crucial sentence might perform the func­tion of a crucial experiment, we are left leaning on a crutch that it­self is held up by only a shaky prop. 2 Bailey (1969) has proposed a pan-dialectal generative approach, sug­gesting that speakers develop a "poly-dialectal competence". If indeed speakers had competence based on the other dialects to which they had been exposed, then Bailey might have a viable alternative. However, Troike (1969) showed through perception and repetition tests that speak­ers who don't make certain distinctions made in other dialects are in­capable of perceiving them. For example, speakers were unable to per­ceive the difference between pen and pin or horse and hoarse if they did not happen to make these distinctions in their own dialect. This would seem to indicate that Bailey's pan-dialectal grammars do not have psychological reality, and are hence equivalent in effect to Halle's.

REFERENCES

Bailey, Charles-James N. 1969. "Implicational Scales in Diachronic Lin­guistics and Dialectology". Working Papers in Linguistics 1:8.123-to 138, and 10.245-49.

Becker, Donald A. 1967. Generative Phonology and Dialect Study: An in­vestigation of three modern German dialects. Ph.D. diss., Austin, Tex.: Univ. of Texas.

Campbell, Lyle. 1972. "The Dilemma of Generative Dialectology". From Soundstream to Discourse ed. by Daniel G. Hays and Donald M. Lance,

358 LYLE CAMPBELL

191-200. Columbia, Mo.: Univ. of Missouri; Linguistics Area Program. Halle, Morris. 1962. "Phonology in Generative Grammar". Word 18.54-72. Katz, Jerrold J., and Jerry A. Fodor. 1963. "The Structure of Semantic

Theory". Language 39.170-210. Keyser, Samuel Jay. 1963. Review of Hans Kurath and Raven I. McDavid,

The Pronunciation of English in the Atlantic States (Ann Arbor, Mich.: Univ. of Michigan Press, 1961). Language 39.303-16.

King, Robert D. 1969. Generative Grammar and Historical Linguistics. Englewood Cliffs, N.J.: Prentice-Hall.

Kiparsky, Paul. 1971. "Historical Linguistics". A Survey of Linguistic Science ed. by William Orr Dingwall, 596-642, 643-49 (Discussion). College Park, Md.: Linguistics Program, Univ. of Maryland.

Klima, Edward. 1964. "Relatedness between Grammatical Systems". Lan­guage 40.1-20.

Labov, William. 1970. "The Study of Language in its Social Context". Studium Generale 23.30-81.

Ohala, John J. 1972. On the Design of Phonological Experiments. Berke­ley, Calif.: Dept. of Linguistics, Univ. of California, mimeo.

Saltarelli, Mario. 1966. "Romance Dialectology and Generative Grammar". Orbis 15.51-59.

Saporta, Sol. 1965. "Ordered Rules, Dialect Differences, and Historic­al Processes". Language 41.218-24.

Sherzer, Joel. 1970. "Talking Backwards in Cuna: The sociological real­ity of phonological descriptions". Southwestern Journal of Anthropol­ogy 26.343-53.

Sledd, James H. 1966. "Breaking, Umlaut, and the Southern Drawl". Lan­guage 42:1.18-41.

Troike, Rudolph 1969. "Receptive Competence, Productive Competence, and Performance". Georgetown University Monograph Series on Languages and Linguistics 22.63-74.

Vasiliu, Emanuel. 1966. "Towards a Generative Phonology of Daco-Rumanian Dialects". Journal of Linguistics 2.79-98.

Weinreich, Uriel, William Labov, and Marvin I. Herzog. 1968. "Empirical Foundations for a Theory of Language Change". Directions for Histor­ical Linguistics: A symposium ed. by Winfred P. Lehmann and Yakov Malkiel, 95-195. Austin & London: Univ. of Texas Press.

THE PRE-WAR PRAGUE SCHOOL AND POST-WAR AMERICAN ANTHROPOLOGICAL LINGUISTICS

DELL HYMES

I should like to discuss the relation between the pre­war Prague School and post-war American anthropological lin­guistics, first from the standpoint of my own experience of them, and then from the standpoint of the place of each in the emergence of 'functional' linguistics. My own work in linguistics owes much to the coming together of influence from these two intellectual currents.

I apologize for writing of my own experience, but offer in excuse that it may be of interest as a case that does not quite fit common notions of the recent history of linguistics and that it has led to the comparison to be drawn in the next section between the schools of Prague and Yale.

The years 1950-55 are almost pre-historic for many younger linguists today, at least in the United States. They are years "B.C.", "Before Chomsky", before the publication of Syntactic Structures in 1957. A common image of those years would be of the dominance of the 'neo-Bloomfieldian' or 'Yale' school, a school characterized by hostility to 'mentalism', a neglect of functional problems, and hostili­ty or indifference to European traditions of linguistics in which 'mentalism' and 'functional' perspective were impor­tant. My own initiation into linguistics in those years was

360 DELL HYMES

different in each respect. There was something of an earlier 'Yale school', that of Sapir; something of mentalism and of functional problems; and something of the Prague School.

I was a graduate student at Indiana University, start­ing in anthropology in 1950 and receiving the doctorate in linguistics in 1955, with anthropology and folklore as minor subjects. These three fields, which Boas had combined in one person and department at Columbia University in the period of the founding of American academic anthropology, were still closely allied (cf. perhaps the connection between the pre­war Prague School und the work of the late Petr Grigor'evic Bogatyrev [1893-1971]). At Indiana the early unity of these three fields in the study of the American Indian was sus­tained as a dominant interest. The chairman and founder of the Department of Anthropology, F. Voegelin, was an Amer­icanist and linguist; his colleague, George Herzog was pri­marily an Africanist and othno-musicologist, but partly an Americanist and linguist. Two prominent figures in the Com­mittee which guided graduate training in linguistics, Harry Veiten und Thomas A. Sebeok, had both done some work with American Indian languages (Velten with Nez Perce in Idaho, Sebeok with Winnebago in Wisconsin, and also, through texts, with the South American Indian language, Aymara). In these years after the Second World War, much of American anthro­pology was turning away from the American Indian to study other parts of the world; folklore was academically margin­al in most universities, if it existed at all; and many Amer­ican anthropologists were finding their traditional obliga­tion to understand linguistics an increasing strain, when confronted with the new rigor of linguistic method; but at Indiana in those years the American Indian, folklore, lin­guistics and anthropology were a natural unity and ambience.

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 361'

The gods of this dispensation remained Boas, Sapir and Bloomfield, although all were dead. The tradition they re­presented was not transmitted wholly and purely - the view of it that is sketched in the next section of this paper is something I have come to understand only with recent re­search and is even yet incomplete. But the tradition was relevant, not rejected. In one seminar a professor asked, "Why had Boas been interested in grammatical categories?" The question reflected the climate of distrust of meaning, and turning of attention away from grammatical categories in

'advanced' circles; yet the asking of the question reflected also an honest puzzlement and an assumption that the answer was of interest. At the time I was shocked by the ignorance that the question seemed to betray. One had only to read Boas, especially his Introduction to the Handbook of Ameri­can Indian Languages (1911), to discover why he had been in­terested in grammatical categories. A little exploration quickly disclosed a continuous tradition, on the one hand, leading back to Steinthal, von Humboldt and Herder, and known as such to Boas, Sapir and Bloomfield, and on the oth­er, leading forward to Whorf. Whorf himself was clear that he followed in the footsteps of Boas and Sapir (the very phrase identified with Whorf, 'linguistic relativity', orig­inates with Sapir). But Whorf did not apparently remember Steinthal or von Humboldt; and after the Second World War, no one, it seemed for a while, remembered Boas or Sapir. Whorf was celebrated posthumously as the discoverer of 'lin­guistic relativity'. (The use of 'relativity' in this con­nection had originated with Sapir).l

1 Cf. Sapir in his article of 1924, "The Grammarian and his Language": "... a kind of relativity that is generally hidden from us by our naive acceptance of fixed habits of speech as guides to an objective under­standing of the nature of experience." (Quoted from Mandelbaum 1949: 159).

362 DELL HYMES There was then a sense of continuity with a past tra­

dition that offered alternatives to present doctrines, even though the tradition was conveyed mostly as great names and anecdotes about them - how much had been lost from view, how much the Second World War and the immediate post-war years changed the intellectual landscape in the United States, I have only recently come to realize. There was a sense of al­ternatives in space, as it were, as well as time. The late Harry Velten cast an amused and penetrating eye on problems of comparative and historical linguistics, and in some in­visible, yet real way helped inspire me to take up histori­cal linguistics in the first years after my degree. Thomas Sebeok kept reminding one of a rich, slightly risque world in which linguistics dealt with semantics, poetry, myth, and such things, as honest linguists. Of course such work was welcomed in the partly anthropological, partly neo-Bloom-fieldian, partly new-born American rationalist climate at Indiana. But in that climate such work seemed something that would have to start from scratch, and be done by linguists within the limitations of the descriptive linguistic method of the day. The other world seemed one in which linguistics had been, already at work, with methods that left out far less of what one knows semantics , poetry, myth and the like to con­tain. Velten and Sebeok, of course, were adherents of the tradition of the pre-war Prague School.

In the summer of 1952 Voegelin, aided by Sebeok, orga­nized a Conference of Anthropologists and Linguists. Its par­ticipants included Roman Jakobson and Claude Lévi-Strauss, and the resulting monograph was co-authored by all four. (I had prepared the digested transcript that constituted the re­mainder of the text, after the addresses by Lévi-Strauss and Jakobson, and was for a short time considered a co-au-

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 36 3 thor also. But Voegelin decided that co-authorship would be unfair, since it might expose me, a defenseless graduate student, to the ill will of those digested.) It would be dramatically pleasing to be able to say that the encounter, a decade before Lévi-Strauss had achieved his present pre­eminence, and a little before Jakobson's stature was fully appreciated in this country, had shaped my life. But it did not. I was not ready. The work in social structure wiiose in­spiration Lévi-Strauss attributed in important part to Ja­kobson and Trübetzkoy was something I admired, but no more. Of Lévi-Strauss' initial proposals for the analysis of myth (1955), I was intensely critical, especially of the way in which the parallel with language was drawn. (And might have said so had a projected contribution to the same special is­sue of the Journal of American Folklore been realized). I was to come to terms with Lévi-Strauss' use of the Prague tradition later, both as to what to emulate and what to ab­jure (cf. Hymes 1965, 1966, 1968, 1970b). A decisive impact was to come from Jakobson a few years later, and not at the University at which we were then colleagues (Harvard, 1955-1960), but again in Bloomington, at a conference on style organized b'y Sebeok.

At the conference on style (April 1958) Jakobson deliv­ered a report, published, under the title "Linguistics and Poetics", in which he justified linguistic attention to the poetic function, and in which he placed the poetic function in terms of a scheme of general functions of language. Where­as much of previous theory had begun with the linguistic sign but not gotten much beyond it, or had, rather arbitrarily, listed quite general sociological functions, whose relation to the linguistic sign was obscure, Jakobson proposed to be­gin with an analysis of the speech situation, placing the

364 DELL HYMES linguistic sign within it, and deriving an exhaustive typo­logy of functions naturally and logically, by primary focus (Darstellung) on each of the constituent factors of the speech situation in turn. (He maintained the Prague view that all the functions would be compresent in each case, on­ly differing in hierarchy). The six components of the speech situation were the message, channel, context, code, sender, receiver, and the six corresponding functions the poetic, contact, reference, metalinguistic, expressive, and direc­tive. (See Jakobson 1960).

It was this presentation that turned my thinking to a functionalist perspective, and that led, among other things, to the article published in Slovo a Slovosnost (Hymes 1970a). At first I remembered the number of Jakobsonian functions as five, not six as published, and when the approach had been thought through in an ethnographic framework, some fun­damental differences would emerge despite a discussion of them with Jakobson. But the debt to Jakobson, and to the functionalist approach of the Prague School, was fundamental too. The paper (1962), which begins my turn to a sociolin-guistic direction, and begins any distinctive contribution I may have made to sociolinguistics, is dedicated to Jakob-son.

The crucial contribution was to introduce a 'function­alist' perspective, and to do so in a way that suggested an empirical, manageable way of dealing with speech functions. To some, Jakobson's discussion might have seemed simply a variant on an 'information theory' model of the speech sit­uation, or even just an arbitrary classification, worked up for the occasion to give ad hoc organization to a host of examples. The presentation, however, was something more than a schema. There was the methodological spirit of linguistic

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 36 5 inquiry, showing the principles of commutation and permuta­tion, of paradigmatic and syntagmatic relations, at work. The host of examples showed linguistic form to depend not only on the relations in grammar in the broadest sense (en­compassing phonology and lexicon) , but also on the relations among components of speech events such as channel, context, sender, receiver. To attend to this larger domain was not to attend to variation and style in some endlessly quantitative sense, but to an additional realm of structure. The host of examples showed linguistic form to covary as the relations among components of the speech event covaried. And the no­tion of alternative hierarchies of function opened up' the prospect of treating function, not as given, but as problem­atic .

The impact of this perspective was gradual, or at least I can not now remember a particular hour of leaping out of the bath, shouting "Eureka!". The first public sign was two years later, when an invitation to give a talk at the Uni­versity of Pennsylvania on anthropology and education led to a discussion of functions of speech from an evolutionary perspective (summer 1960; published 1961). The talk went past most of the audience of educators and students of edu­cation, but the published paper gained wide attention among anthropologists through inclusion in a reprint series. When invited to speak in a series sponsored by the Anthropolo­gical Society of Washington that fall (November 1960), the functional framework was developed into a call for an ethno­graphy of speaking. As luck would have it, a presentation of the argument at the annual meeting of the American Anthro­pological Association that month was assigned to a general session on theory, scheduled at the same time as the major linguistic session of the meeting. The published essay came

366 DELL HYMES gradually to attention (even in disguised form, as when a student at Iowa wrote about "Durkheim's seven functions of speech"). With the rise of interest in ' sociolinguistics' , the essay has become established as part of the programmatic literature paving the way for sociolinguistic research, and since my coming to the University of Pennsylvania, led to a body of empirical, research that will just begin to appear this year. (Among the students involved: Regna Darnell, Mary Foster, H. Hogan, Dhanesh Jain, Edward Keenan, S. Phillips, P. and S. Seitel, Joel Sherzer, A. Strauss, K.M. Tiwary). A concern with the functions of speech has pervaded my theo­retical writing (beside references already cited, note Hymes 1967a, 1967b, 1968b, 1971a, 1972). In this respect, then, a perspective of the pre-war Prague School is, to use a cur­rent colloquial expression, "alive and well" in post-war American anthropological linguistics (or linguistic anthro­pology, as it is increasingly called).

A relevance of the essays by Jakobson and myself to pre­war discussions of langue and parole is pointed out by Uhlen-beck (1967:361). Functionalist perspective has undergone elaboration and criticism both in Prague and elsewhere, and, as noted, my own views depart from those of Jakobson. It would not be possible or appropriate to argue those views here, and a good deal of their nature can be seen in a re­cent article of mine (Hymes 1970a). An understanding of the relation between pre-war and post-war views, however, re­quires some comment on another pre-war perspective, that of Sapir.

Let me approach this by discussing the later history of the Sapir tradition in post-war American, linguistics.2 It

Some aspects of Sapir's intellectual history and the tradition he shaped are investigated in two studies, Hymes 1970b and 1971b.

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 367

was this subsequent history that defined the relation be­tween language and culture as a problem, to which a func­tionalist perspective could come to be recognized as an an­swer. And let me take as a text for this discussion a state­ment by Trnka and others in 1958:

"As to the study of the relationship between the relevant features of language and those of society, it must be admitted that all schools of structural linguistics - with the exception of that of Edward Sapir and his followers - have failed to develop it in a satisfactory way." (Quoted from Vachek 1964:477).

I can agree with the statement insofar as it is the tradition of Sapir and his followers to which I myself would point, as precedent in American linguistics for my own work. At Indiana, Voegelin and Herzog had both been associated with Sapir at Yale; Hoijer, who taught one summer at a Lin­guistic Institute, and with whom I studied for a year at UCLA, had been Sapir's student as had Swadesh, who visited Indiana, and whose historical interests, which shaped my own work for some years, were the following out of a program of historical research inspired by Sapir. In the immediate post-war years, when some used Bloomfield's name to define a very narrow conception of linguistics, and depreciated Sapir as a more intuitive genius without method, it was to Sapir that one looked as symbol of a wider conception of linguistics in which meaning and literature had a natural place. In 1959, on the 20th anniversary of Sapir's death, indeed, A. L. Kroeber organized at Berkeley a memorial meet­ing at which Yakov Malkiel spoke hopefully of a sign of a Sapir 'renascence'. Another friend, the anthropologist R. H. Lowie, edited Sapir's letters to him (see Lowie 1965). These things seem symptoms of a sense that American anthropology and linguistics had not fully come to terms with the heri-

3 6 8 DELL HYMES

tage of Sapir. Such at least was my situation at the time when Jakob-

son^ treatment of functions had its impact. In historical work on American Indian languages I identified with the aims of Sapir (and of his follower, Swadesh). A sometime poet, I rejoiced that Sapir had been a fairly successful one. A worker with Chinookan grammar and texts, and inheritor of some of Sapir's unpublished notes, I pored over plain evi­dence that his has been the most brilliant and accurate mind to touch that language. For sustenance in a dry season, for evidence that linguistics and anthropology could have intel­lectual stature, I reread his essays and book. And as a fledgling academic at Harvard, I was the protege of Clyde Kluckhohn, to whom Sapir had been a major influence, the embodiment (later joined by Lévi-Strauss) of the idea that linguistics offered anthropology a way to be rigorous and yet true to the patterned nature of its materials, a way to be scientifically exact without aping inappropriate methods of the natural sciences.

Of all this I was aware at the time. There was a new element of which I also became increasingly aware. In its Boas and Sapir-like unity of linguistics, anthropology and folklore, around the American Indian, Indiana University had been something of a refuge area. As linguistics became an independent academic discipline after the Second World War, a discipline no longer dependent on anthropological and lan­guage department hospitality, tracks which had seemed to run parallel began increasingly to diverge. Yet it was my pro­fessional responsibility, as a linguist in anthropology, to travel both. A sense of strain had begun to appear right after the war, signalled by articles re-assessing the rela­tionship between linguistics and ethnology. Anthropologists

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 369

not trained in structural linguistics often found its con­cepts alien. And linguistics had changed in scope in just a few years. In 1941 a memorial volume to Sapir (Spier et al. 1941) had contained articles expressing in a variety of ways the link between linguistics and anthropology. But the vol­ume did not serve to define a post-war tradition. After the war the center of the stage was taken by concerns purely in­ternal to descriptive linguistics. The contributors to the Sapir volume did not abandon their interests, but they and their interests seemed peripheral to a focus on meaning, to a 'young Turk' stance of washing away all previous work, 'contaminated' by meaning, 'metaphysics', and 'mentalism'.

In 1951 those at the heart of the methodological devel­opment seemed to feel that some sort of stable plateau had been reached. The Outline of English Structure by Trager and Smith, and Methods in Structural Linguistics by Zellig Harris, seemed to define the new approach rather satisfac­torily. A new period of reaching out from linguistics to a larger sphere of investigation began. The study of gesture and body motion, and the study of the use of the voice, were rechristened kinesics and paralinguistics, and launched anew. Models for the analysis of all culture were derived from linguistics by Trager and Smith, and by Kenneth Pike. (Pike, it should be stressed, was never part of the orthodoxy just discussed). At the same time a more generalized, less publi­cized use of linguistics was underway, in the work of Lévi-Strauss, inspired by the Prague School, and in the work of W. Goodenough and others, derived from the Yale School. No spe­cific model of linguistic analysis was taken over. Rather, the goals of linguistic analysis were taken over, and adapted to the exigencies of ethnographic material. In brief, one did not look for 'phonemes' of culture, or for other su-

3 70 DELL HYMES

perficial parallels of structure. One sought cultural units and structures that had the same methodological basis as phonemes, i.e., contrastive relevance within a locally val­id frame, and conceived ethnography, like grammar, as dis­covering a theory for the particular case.

Much of the development sketched here was exciting, all of it was interesting, and all of it had precedent in the work of Sapir, but it did not add up to an integration of the study of language with the study of culture. The two tracks might run parallel again for some, but they remained separate tracks. Linguistics was autonomous, and was what dominant linguists said it was. The only choices were to regard linguistics as unique, or to look for parallels to linguistics. The implicit assumption was that human life was manifested in series of separate domains. Behind the domains might lie the human mind, and thus some common struc­ture. But no one suggested that something might have been left out of account. Pike began his generalization of method (dedicated to Sapir) by discussing activities in which lin­guistic and non-linguistic features were integrated, but he went on to concentrate on linguistic analysis proper. The original purpose of developing a method to deal with behav­ior, linguistic and non-linguistic, in an integrated way, was left behind. The relation of language to society was discussed again, only in terms of parallel domains and of analogues between them.

The continuing influence of Sapir that was most dis­cussed was that mediated by Whorf. Here again the implicit assumption was that language was one domain, culture (or metaphysical assumptions) another, and that one should take the separation as given, seeking for correlations. Whorf's suggestion of 'fashions of speaking' was not developed into

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 3 7 1

a novel method of description, but used only as a principle of selection within grammars written in usual ways.

These efforts to relate language to the rest of cul­ture were put into the shade by a new development within linguistics itself, one which returned to Sapir somewhat in its 'mentalism' and morphophonemics, but that carried one trend of Sapir's thought to an extreme, and so severed its connection with the rest. The kind of structural linguistics on which the efforts were based was eclipsed by transforma­tional generative grammar. To continue the metaphor of 'tracks', linguistics and anthropology seemed hardly to run parallel anymore. They seemed almost to be oriented in op­posing directions. As Chomsky's views developed, they seemed to seal language off almost altogether from the sociocultur-al sphere. Chomsky's notions of the acquisition of language, of language use, of language change, and of the origin of language, all conspired, as it were, to deny experience or social life any constitutive role, or even relation. The thrust of linguistic speculation was reversed. Whereas those influenced by the Sapir tradition looked out from language to other institutions, those influenced by Chomsky looked in toward the brain. Structure was immanent to the mind and innate in the organism.

The work of Chomsky now seems to me the ultimate devel­opment, the 'perfection', as it were, of the dominant trend of linguistics in this century. It is the trend that moti­vated much of Sapir's work, and that informed the recurrent efforts under his influence to relate language to culture. Briefly put, the trend is that toward the isolation of lin­guistic structure as an object of study. When de Saussure, Sapir and Bloomfield wrote, it was necessary to insist on the separation of language as an autonomous object of study.

3 72 DELL HYMES

And it has been around that separation that modern linguistics has developed as a profession. The degree of separation, and the basis for it, however, have varied. In both the Sapir and the Prague traditions, recognition of the autonomy of linguistic structure was joined with recognition of the many connections of linguistic structure with social life. Chomsky severs the connections of linguistic structure, and in a way that motivates the severance. He does not, as Hjelmslev, sim­ply call for the independence of linguistics. He argues that the ultimate goal of linguistic theory, explanation, lies in the brain, not at all in social life. To such an appeal it is not enough to insist on a tradition of looking for struc­tural parallels, or of tracing the occurrence of language or linguistic features in social contexts. Even less does it suffice to insist on the practical value of sociolinguistic work. None of these positions counter the argument that sci­entific goals require solving the problems internal to the analysis of grammar, before secondary uses of those results can be seriously made. An alternative conception of scientif­ic goal, of explanation is required.

In the sphere of linguistic change, such an alternative conception of explanation has been put forward and substanti­ated by the work of Labov. In a theoretical article, Wein-reich, Labov and Herzog show themselves sympathetic indeed to the work of Mathesius and others of the Prague School (1968:167-69), while finding it necessary to go beyond it. In the sphere of linguistic description, my article in Slovo a Slovesnost (Hymes 1970a) indicates some of the reasons why I believe an alternative conception of explanation in the synchronic sphere to be necessary.

More generally, I would call Chomsky's notion of expla­natory adequacy a partial, 'essentialist' kind of explanation

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 373 Full explanation requires another kind, which may be called 'experiential' or 'existential' adequacy. The heart of the conception of 'experiential adequacy' is that linguistics has as its goal explanation not only of structure, but also of function. Explanation not only of the means of speech, but also of the relation of those means to the ends they serve. Such a conception, as the Prague tradition well knows, leads one to discover further aspects of structure in lan­guage - aspects of the organization of linguistic means in the service of expressive, stylistic, standard language and other functions. Chomsky's conception of explanatory ade­quacy effectively limits language to. grammar organized in terms of the single function of semantic reference, exclud­ing all the many other ways in which language functions. Chomsky's conception of the 'creative use' of language re­duces to a conception of language use as monologue. Although he speaks of 'creative' use of language in terms of appro­priateness to situation,'his method can deal only with nov­elty of utterance.

Appropriateness is a relation between utterance and situation. A new utterance in an old situation, an old ut­terance in a new situation, may equally be creative use of language. One must be able to analyse the structure of sit­uations, as well as the structure of sentences, and be able to establish the relations of appropriateness between the two, if the 'creative aspect of language use' is to be more than a slogan. In short, the Chomskyan conception appeals to our concern with the meaning of language in human life, but its definition of linguistic analysis will forever frus­trate an answer to that concern. If linguistic research is to serve human freedom, it cannot limit itself, as does Chomsky's work, to freedom from situation. It must also ad-

3 74 DELL HYMES

dress the use of language in context in order to account for freedom to master situations. Such an approach will re­quire a method of description that is both social and lin­guistic.

This critique of Chomsky's conception of explanation, and of the relation of language to social life, is not a defense of either the Sapir or Prague tradition in their pre-war form. Both traditions, like other lines of lin­guistic research, participated in the long-run trend of the century, the separating out of linguistic structure for study in its own right. Neither resolved the problem of a descriptive method encompassing both linguistic and social features. Both traditions, however, may be said to have based the separation of linguistic structure on a concep­tion of the autonomy of linguistic structure, not of the complete independence of linguistic structure. Chomsky has attempted to find a ground in the brain and biology for independence . When the inadequacy of that attempt is real­ized, when the goals of linguistics are understood to re­quire a broader base, it is to the traditions of Sapir and Prague that one can turn.

Ultimately both the Prague and Sapir traditions have become intertwined in my work. Looking back, I see in Sapir' 1925 paper, "Sound Patterns and Language", a fundamental methodological principle that applies successively to all levels of structure, and that embodies the point that func­tion is prior to structure. The development of the tradition of linguistic work in ethnography, to which Sapir and his students contributed so much, has led to the standpoint from which I criticized and revised Jakobson's formulation of components and functions of speech events, treating their number and kind as problematic, and needing to be determined

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 3 7 5

for the particular speech community. And I can see now in Sapir's work in the last years of his life (1931-1939) a development that quite anticipates the 'ethnography of speaking', and the perspective from which I have criticized Chomsky. In those years Sapir stated his dissatisfaction with the impersonal, generalized patterns of anthropology and sociology, and asserted the need to understand such patterns in terms of their meaning in personal lives. In­deed, he proposed that new kinds of organization of pat­terns would come into view from that standpoint. He spoke of the need to

bring every cultural pattern back to the living context from which it has been abstracted in the first place and, in parallel fashion, to bring every fact of personality formation back to its social matrix... The social psychology into which the conventional cul­tural and psychological disciplines must eventually be resolved is related to these paradigmatic studies as investigation into living speech is related to grammar.(Quoted from Mandelbaum 1949:592-93)

The quotation could be applied to our present situation, by speaking of "The sociolinguistics into which the conven­tional linguistic and social disciplines must eventually be resolved..." (cf. Hymes 1967a). Sapir did not develop this new perspective in empirical studies, and this aspect of his thought was not carried further after the Second World War. The dominant characteristics of the first Yale School, the Sapir tradition, before and after the war, continued to be shaped by the general situation of linguistics as a still emergent discipline in the United States, and by the anthro­pological context in which much of linguistics then grew. Briefly put, the main tasks were seen as:

(1) to develop the methods of structural linguistics, and to test their application in both exotic and well known

3 76 DELL HYMES

languages; (2) to sustain a profession of linguistics, where almost none existed; (3) to rescue knowledge of disappearing languages; (4) to pursue proof and establishment of genetic relationships among aboriginal languages; (5) to relate lin­guistic inquiry to other disciplines and subjects. (See dis­cussion in Hymes 1971b).

One can see here important points of contrast with the Prague tradition. The Sapir tradition recognized expressive and other functions in language, and occasionally described them, when they became salient in ordinary descriptive work, but mainly it was concerned to write basic grammars of lan­guages little known. The Prague School, on the other hand, worked to a considerable extent with languages well known to it, and could, given the appropriate perspective, pene­trate further into their functional complexity. The Sapir tradition worked mostly with languages of American Indians, whose societies showed relatively little differentation, and no class structure. Questions of a standard language did not arise. Questions of the language of verbal art might arise, and to some extent did, but not in the marked form in which they were posed in a European country. The Sapir tradition was at work in a native linguistic situation in which many questions of genetic relationship remained unresolved, and in which solution would be of considerable interest to col­leagues in anthropology. The Prague tradition worked mostly with language families whose relationships were established, and perhaps could more easily give attention to phenomena of convergence. Such phenomena were never excluded in the Sapir tradition, but they remained at the periphery, the main task appearing to be the working out of genetic classi­fication. Finally, despite Sapir's remarks on 'living speech' quoted above, there never developed an explicit attention to

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 3 77 variability or 'potentiality' such as was present in the Prague School. Again, the primary focus on basic grammars of American Indian languages may well have been responsible.

In sum, the work of the pre-war Prague School is more in keeping with the world linguistic situation today in a number of respects. The Prague tradition early shook off the limitations of what I have called the 'Herderian' perspec­tive of equating a language, a culture and a society. It addressed itself to the synchronic diversity of linguistic varieties within a speech community. It moved beyond con­structing the results of the diversification of language, to the affinities and specializations of function which lan­guages acquire within a common society or social field. These attributes are of prime importance for an adequate linguistics today.

If I may end on a personal note, it was a linguist from Prague, visiting Philadelphia, who pointed out to me the ab­sence from my own thinking of attention to linguistic situa­tions such as those that are familiar in Europe, and who made me aware of the unconscious limitation of the anthropo­logical tradition in which I had worked. I can not say, then, that the Sapir tradition, any more than any other, has devel­oped the study of the relationship between the relevant fea­tures of language and those of society in a satisfactory way. Much of the basis for such a development is present in the Sapir tradition; much of that basis is shared with the Prague tradition, and the Prague tradition has essential elements lacking in the Sapir tradition. To speak only for myself, it is in the fruitful combination of the two that I see the main hope for the progress of linguistics in the remainder of this century. We need not look back a century or two, but only a generation or two, for inspiration.

378

REFERENCES

Boas, Franz. 1911. "Introduction". Handbook of American Indian Lan­guages, Part I (= Bureau of American Ethnology; Bulletin 40:1), 1-83. Washington, D.C.: Government Printing Office. (Re-ed. Washington, D. C.: Georgetown Univ. Press, 1963.)

Chomsky, Noam. 1957. Syntactic Structures, The Hague: Mouton. (12th printing, 1972.)

. 1966. Cartesian Linguistics: A chapter in the history of rationalist thought. New York & London: Harper & Row.

Hymes, Dell. 1961. "Functions of Speech: An evolutionary perspective". Anthropology and Education ed. by Frederick C. Gruber, 55-88. Phila­delphia: Univ. of Pennsylvania Press.

. 1962. "The Ethnography of Speaking". Anthropology and Human Behavior ed. by Thomas Gladwin and W. C. Sturtevant, 13-53. Washing­ton, D.C.: Anthropological Society of Washington; repr. in Readings in the Sociology of Language ed. by Joshua A. Fishman, 99-138. The Hague: Mouton, 1968.

. 1965. "Some North Pacific Coast Poems: A problem in anthrop­ological philology". American Anthropologist 67.316-41.

. 1966. "Two Types of Linguistic Relativity". Sociolinguistics ed. by William Bright, 114-67. The Hague: Mouton.

. 1967a. "Why Linguistics Needs the Sociologist". Social Re­search 34:4.632-47.

. 1967b. "Models of the Interaction of Language and Social Setting". Journal of Social Issues 33:2.8-28. New York.

. 1968a. "The 'Wife' who 'Goes out' like a Man: Reinterpreta-tion of a Clackamas Chinook Myth". Social Science Information 7:3. 173-99.

. 1968b. "Linguistics - the Field". International Encyclopedia of the Social Sciences 9.351-71. New York.

. 1970a. "Lingvistická teorie a promulvové funkce". Slovo a Slovesnost 31:1.7-32.

. 1970b. "Linguistic Method of Ethnography". Method and Theory in Linguistics ed. by Paul L. Garvin, 249-311, 312-15 (Discussion), 315-25 (Bibliography). The Hague: Mouton.

PRE-WAR PRAGUE SCHOOL AND POST-WAR LINGUISTICS 379

Hymes, Dell. 1971a. "Sociolinguistics and the Ethnography of Speaking". Social Anthropology and Language ed. by Edwin Ardener, 47-93. London: Tavistock Publications.

. 1971b. "Morris Swadesh: From the First Yale School to world prehistory". The Origin and Diversification of Languages by Morris Swadesh, ed. by Joel Sherzer, 228-70. Chicago: Aldine-Atherton; Lon­don: Routledge & Kegan Paul, 1972.

. 1972. "Models of the Interaction of Language and Social Life". Directions in Sociolinguistics ed. by John J. Gumperz and D. Hymes, 35-71. New York: Holt, Rinehart & Winston.

Jakobson, Roman. 1960. "Concluding Review: Linguistics and Poetics". Style in Language ed. by Thomas A. Sebeok, 350-77. Cambridge, Mass.: MIT Press; New York: J. Wiley & Sons. (2nd printing, 1964.)

. 1963. "Efforts towards a Means-ends Model of Language in Interwar Continental Linguistics". Trends in Modern Linguistics ed. by Christine Mohrmann, et al., 104-08. Utrecht & Antwerp: Spectrum; repr. in Vachek 1964:481-85.

Lévi-Strauss, Claude. 1955. "The Structural Analysis of Myth". Journal of American Folklore 68.428-44.

, Roman Jakobson, Carl Frederick Voegelin, and Thomas Albert Sebeok. 1963. Results of the Conference of Anthropologists and Lin­guists, Bloomington: Indiana Univ. Press.

Lowie, Luella Cole, ed. 1965. Letters from Edward Sapir to Robert H. Lowie. Berkeley, Calif.: Editor.

Mandelbaum, David. G., ed. 1949. Selected Writings of Edward Sapir. Berkeley & Los Angeles: Univ. of California Press. (Repr., 1963.)

Pike, Kenneth L. 1954-60. Language in Relation to a Unified Theory of the Structure of Human Behavior. Preliminary ed., 3 vols. Glendale, Calif.: Summer Inst. of Linguistics. (Rev. ed. The Haugue: Mouton, 1967; 2nd printing, 1971.)

Spier, Leslie, A. Irving Hallowell. and Stanley S. Newman, eds. 1941. Language, Culture, Personality : Essays in memory of Edward Sapir. Menasha, Wisc.: Sapir Memorial Pub. Fund. (Repr. Salt Lake City: Univ. of Utah Press, 1960.)

Swadesh, Morris. 1948. "On Linguistic Mechanism". Science and Society 12.254-59.

Trnka, Bohumil, et al. 1958. "Prague School Linguistics". Philologia Pragensia 1.33-40. (Repr. in Vachek 1964:468-80.)

Uhlenbeck, E. M. 1967. Review of L'école de Prague d'aujourd'hui (= Travaux linguistiques de Prague, 1). Lingua 17.358-74.

380 DELL HYMES

Vachek, Josef, ed. 1964. A Prague School Reader in Linguistics. Bloom­ington & London: Indiana Univ. Press. (2nd printing, 1966.)

. 1966. The Linguistic School of Prague: An introduction to its theory and practice. Bloomington & London: Indiana Univ. Press. (2nd printing, 1970.)

Weinreich, Uriel, William Labov, and Marvin I. Herzog. 1968. "Empirical Foundations for a Theory of Language Change". Directions for Histor­ical Linguistics: A symposium ed. by Winfred P. Lehmann and Yakov Malkiel, 95-188. Austin, Texas & London: Univ. of Texas Press.

TRANSFORMATIONAL GRAMMAR AND THE PHILOSOPHY OF SCIENCE

ESA ITKONEN

1.0 By 'positivism' I understand that very influential trend in the philosophy of science which adheres to the so-called methodological monism, or the conception that all empirical sciences (= Wissenschaften) are characterized by a common method of explanation and prediction; a method that appears in its purest form within the natural sciences, es­pecially in mechanical physics.1 According to the positivist view, ex­planation and prediction are similar insofar as in both cases we have to deduce a sentence referring to a particular event from a whole con­sisting of one or more sentences referring to general regularities and one or more sentences referring to particular events (= so-called ante­cedent conditions).2 Sentences referring to regularities, or universal hypotheses, are tested by finding out whether or not logically indepen­dent evidence corresponds to predictions which have been deduced from these sentences, plus some sentences referring to antecedent conditions. The 'logical independence' of evidence means that the evidence must not be somehow contained in the antecedent conditions; in other words, the relation between antecedent conditions and evidence must not be logical, viz. conceptual, but empirical, viz. based on observation. If a pre­diction turns out to be empirically true, i.e., if logically indepen­dent evidence corresponds to it, then it confirms the universal hy­pothesis used in predicting. A particular event is explained by iden­tifying those regularities and antecedent conditions which are referred to by the sentences permitting the deduction of the sentence referring

382 ESA ITKONEN

to the event to be explained. This kind of event, (just) as well as any event to be predicted, must stand in an empirical relation to its own antecedent conditions.

It is also said that regularities are 'explained' by deducing the universal hypotheses referring to them from some other, more abstract universal hypotheses. However, this type of explanation is logically secondary with respect to the type discussed above, since it is partic­ular events which determine which universal hypotheses about general regularities are true.

For reasons that will become apparent in a moment, Ī have restric­ted my discussion to the 'deductive-nomological' (= D-N) model of ex­planation. This model can be represented more explicitly in the follow­ing manner:3

L1 ; L 2 ... L m = general laws or universal hypotheses explanans

Cj; C2 ... C n = statements of antecedent condition E = description of the event to be ex- explanandum

plained

The simplest case of a D-N explanation can be expressed in the follow­ing modus ponens sentence:4

[(x)(f(x) = g(x)). f(a)]=g(a) Or, represented in the form of the above model: (x)(f(x)=>g(x)) f (a)

~Ï(I7 This model explains the event referred to by "g(a)", provided the

explanans is true and "g(a)" cannot be deduced from "f(a)" alone. (If the latter condition is not fulfilled, the relation between the refer­ents of "f(a)" and "g(a)" is not empirical, but conceptual.) The ob­servations-report "f(a)". g(a)" confirms the universal hypothesis "(x)(f(x)=>g(x))", whereas the observation-report "f(a).~g(a)fl would falsify, i.e., conclusively disconfirm, it (cf. Hempel 1956b:39-40). —

TG AND PHILOSOPHY OF SCIENCE 383 The standard case of a D-N explanation is a causal explanation, and in such a case antecedent conditions and explanandum-events are identifi­able as causes and effects, respectively. However, there are also non-causal D-N explanations, (cf. Hempel 1965f:352-53), and in such cases the term 'antecedent condition' might be somewhat misleading.

Now it is quite obvious that the metascientific position of trans­formational grammar (henceforth TG) has always been positivist, in the sense defined above. Consider the following statements, which could easily be multiplied:

It is important to emphasize that the difference between the men-talistic subject matter of linguistics and that of, say, physics does not impose a logically different character on linguistic the­ories from that of physical theories, that is, does not impede the testing and confirmation of linguistic assertions as has some­times been claimed contra mentalistic linguistics.7 (Postal 1966: 154). A grammar of a particular language can be considered, in what seems to me a perfectly good sense, to be a complete scientific theory of a particular subject matter, and if given a precise enough form, a formalized theory. Any interesting scientific theory will seek to relate observable events by formulating general laws in terms of hypothetical constructs, and providing a demonstration that certain observable events follow as consequences of these laws. In a par­ticular grammar, the observable events are that such and such in an utterance and the demonstration that this observable event is a consequence of the theory consists in stating the structure of this predicted utterance on each linguistic level, and showing that this structure conforms to the grammatical rules or the laws of the theory. (Chomsky, 1955, Chap.I, §2.1).

D-N explanations are customarily distinguished from so-called (in­ductive-) statistical explanations (cf. Hempel 1965d). As the name in­dicates, it is the characteristic feature of the latter that the laws occurring in them are of statistical-probabilistic form. Different ob­jects of research - e.g., planets and gas molecules - require or may require different types of explanations and, therefore, it is an empir­ical question which type of explanation is appropriate in which (posi­tivist) science. That is to say, the adequacy of a given type of expla-

384 ESA ITKONEN

nation must be experimentally established. It is interesting to note that in this respect TG differs from all the natural sciences, since statistical considerations are excluded a priori from TG. In other words, the inadequacy of statistical explanations is not established on the basis of experimentation, but of i n s i g h t . 5 In any case, the rejec­tion of statistical explanations implies that if TG is to be an empi­rical science in the positivist sense, its 'explanations' must be of the D-N type. Nothing could show more clearly the insignificance of metascientific theorizing within TG than the fact that TG's leading ex­ponents have never explicitly discussed their conception of scientific explanation and have even failed to identify their explanations with D-N explanations. This identification has at least been made by such European linguists as Botha (1968:58-63) and Kanngiesser (1972:7-12, 30-31).

In this paper I will argue that TG is not an empirical science in the positivist sense, which means inter alia that TG 'explanations' are fundamentally different from D-N explanations, and that TG descriptions are not empirically testable in the same sense as theories of the natu­ral sciences are. It will be shown that, instead of explaining and pre­dicting observable events, TG descriptions, identifiable as instances of 'explication', articulate or reconstruct intuitive linguistic know­ledge much in the same way in which conceptual analysis, including log­ical analysis, articulates 'presystematic' or 'preanalytic' knowledge of different kinds. All this implies, at the same time, that the ideal of science presupposed by TG is 'hermeneutic', rather than positivist.6 That is to say, TG - and more generally every linguistic theory con­cerned with describing the native speaker's intuition - is not a natu­ral science, but a human science.

2.0. In this section I will examine certain philosophical assump­tions of TG which in spite of their fundamental importance have never been explicitly discussed by transformationalists.

The history of epistemology has been dominated by one powerful

TG AND PHILOSOPHY OF SCIENCE 385 tradition, which has been represented by mind - matter dualists like Descartes and Locke, subjective idealists like Berkeley, empiricists like Hume, and phenomenal ists like Russell and Ayer. The basic tenet of this line of thought (which I will in agreement with Saunders and Henze all 'traditionism') is the claim that subjective experiential data are the primary source of knowledge. According to this (egocen­tric) position, public things and qualities are somehow construed out of subjective experiences. Furthermore, the knowledge of other minds is supposed to be gained inductively on the basis of the so-called argument from analogy: when I perceive that bodies (construed out of my sense-impressions and) resembling mine behave under similar circum­stances in a way similar to the way my body behaves. I may infer with high probability that these bodies possess minds which think and feel similarly to mine. In this view, in contradistinction to the existence of my mind, the existence of other minds is a contingent fact; in ad­dition, the existence of other minds is only contingently connected with the regular behavior 'pointing to' it, which means more specifi­cally that psychological concepts like pain, knowledge, and intention are only contingently connected with their manifestations.

To reformulate the traditionist position in linguistic terms: or­dinary intersubjective languages, which refer to public things and qualities, must (or could) have been preceded by private languages re­ferring to subjective experiences only. By 'private language' I under­stand here a language logically prior to any intersubjective language.

As Apel (1972a) for example has pointed out, traditionism in the form of 'methodical solipsism' provides still today the epistemological foundations of the positivist philosophy of science. However, I think it is generally felt that the traditionist account of knowledge is somehow unnatural, and this account has in fact been rejected by such non-positivist, but otherwise rather dissimilar thinkers as Mead (1934, Part III), Heidegger (cf. Rossi 1972) and Wittgenstein (1958). I use here the Wittgensteinian approach.7

386 ESA ITKONEN

There are several, complementary ways of refuting traditionism. One of them concentrates on the internal inconsistencies of the analogy ar­gument. This argument can be represented more explicitly in the form of the following (invalid) inference:

i) If I have the experience A, I am (usually) in the situation and behave in the way

ii) (I perceive that) a body resembling mine is in the situation and behaves in the way C.

Therefore, it is probable that the body has the experience A and thus possesses a mind similar to mine; i.e., I have to do here with another person.

The premises, which ought to make the conclusion probable, are supposed to represent that primary state of knowledge where I know nothing but myself and my private experiences: this is the whole point of traditionism.8 The conclusion introduces the conception of someone else's experience, hence, of someone else. But it is quite easy to see that there can be no concept of I without the correlative concept of he (and you and we), just as, e.g., there can be no concept of father without the correlative concept of mother. Therefore, contrary to the basic assumption of traditionism, the notion of person (i.e., I and others) must be assumed from the outset, i.e., a priori, and all purely egocentric accounts of knowledge are inherently inconsistent.

This also refutes the possibility of private languages: I could not say that I, in the logically primary sense, privately invent a rule of language and then begin to follow it, since at that point I could not even have the concept of myself. On the other hand, once I learned an intersubjective language, I could try secondarily to invent a language of my own and to follow its private rules. But even in such a case I could not be said to be following a rule, since I could not have any independent checks on whether or not I would be following the rule correctly.

Since the world which I perceive is necessarily intersubjective,

TG AND PHILOSOPHY OF SCIENCE 387 the things in it are public things, not constructions out of my sense-impressions. This is not to deny the reality of sense-impressions, but to maintain that they are logically secondary with respect to public things and qualities. My knowledge of my experiences is interdependent with my knowledge of other people's experiences; but I can only come to know these through understanding other people's publicly observable be­havior; and this I can only understand if it does not vary at random, but conforms to more or less stable rules. Therefore, all experiences, including mine, are logically dependent upon observable, regular be­havior: "An 'inner process' stands in need of outward criteria" (Witt­genstein 1958 I, p. 580).9 On the other hand, realizing that experi­ences as possible objects of knowledge are conceptually dependent upon behavior, does not of course mean reducing experiences to behavior, or claiming that experiences do not exist (Wittgenstein has been commonly misunderstood in this way).

The refutation of private languages makes it clear, first, that rules must be intersubjective and, second, that mental states like know­ledge of a linguistic rule or, more generally, of language cannot be separated from outward criteria. Hence, language as well as the know­ledge of it is inseparable from the use of language, which follows intersubjective rules. Wittgenstein has illustrated this point with the following example: If I imagine that ' a b e d ' means "The weather is fine", does it follow that 'b' has now the meaning "weather"? No, just as it does not follow that there is a book on the table, if I imagine that there is. For there is no actual use or practice to back the claim that 'b' has this or that meaning. Meaning exists only in the context of use; and since there is no language without meaning, there is no language without use of language.

Now let us consider some implications of the preceding discussion for TG. First of all, notice that the question whether knowledge is primarily subjective or intersubjective is more fundamental than the question whether, supposing (falsely) knowledge to be primarily subjec-

388 ESA ITKONEN

tive, its acquisition necessitates a more intricate innate apparatus (= 'rationalism') or a less intricate one (= 'empiricism'). Therefore, in this more fundamental context, the well-advertised (in Chomsky 1966 and elsewhere) distinction between rationalism and empiricism disap­pears.

As is well known, TG attempts to describe the ideal speaker-hearer's competence within a totally homogeneous speech-community. For TG the similarity (or, in the above idealization, the identity) of the lin­guistic rules which different members of the same community follow is a contingent fact.10 To be sure, it is explained by a reference to the similar psychological und physiological constitution of the speakers, but this, again, is a contingent fact. Our treatment of the notion of rule shows, however, that the near-identity of the rules in a speech-community is due to their intersubjectivity, which in turn is a neces­sary characteristic of rules.

Furthermore, the distinction between competence and performance has been interpreted in the Chomsky-type TG11 in such a way that (know­ledge of) language is primary while (knowledge of) its use is secondary. This non-functional, or formal, conception of language has been expounded for instance by Moravcsik (1967:223), who claims that such communica­tive skills as announcing and requesting, which imply a reference to the (intentional) situation of language use, do not belong to the primary linguistic skills. And if it is permissible to speak of the use of lan­guage at all in the context of competence, it can supposedly only be said that one uses language to articulate one's physical and social en­vironment and to 'freely' express one's beliefs and thoughts. Humboldt seems to have held similar views concerning the primary characteristics of language, 12 and Hjelmslev voices several times the ideal that "sys­tem is primary with respect to process".13 But such a formal conception can easily be shown to be untenable.

A language into which the roles of speaker and hearer have not been built in as essential constituents of the concept of speech situ-

TG AND PHILOSOPHY OF SCIENCE 389

ation14 (which in turn is inseparable from the concept of language it­self), i.e., a language in which the intersubjectivity of rules is treated as theoretically secondary, is equivalent to a private lan­guage. Such a language bears no resemblance to natural languages; there­fore, a linguistic theory which either implicitly or explicitly endorses this private-language conception is inadequate on empirical grounds. This is the case of the Chomsky-type TG. But since private languages are not only intuitively unnatural but also logically impossible, the Chom­sky-type TG is, in this respect, not only empirically inadequate but also logically inconsistent.15 This inconsistency is well illustrated by Moravcsik's claim that since use is secondary, hearers could under­stand sentences without knowing the conditions for their correct use, and all speakers could be misusing language all the time (and could ap­parently also refuse to use language at all) but still remain speakers. This claim is so obviously false that I need not comment upon it.

Considerations more or less similar to the ones presented above have led Wunderlich, Habermas, Apel, and others to replace the formal Chomskyan notion of competence with a notion involving a pragmatic or communicative component. It is also interesting to note that even with­in the purely logical study of language the formal conception is being replaced by a more realistic one.16 Similarly proponents of generative semantics have slowly come to realize that Chomsky's approach is too narrow. Thus from the fact that felicity conditions on illocutionary acts must be described in a grammar as the presuppositional part of the meaning of corresponding performative verbs, Lakoff (1971:335-36) draws the conclusion that certain aspects of language use belong to the lin­guistic competence. However, he is apparently not aware of the more general logical reasons which make it imperative to adopt a functional or pragmatic conception of language.

3.0. The term 'rule' is employed in TG only in the sense of a theoretical-descriptive, viz. grammatical rule. I will use this term in its more sociological sense in which it is a subtype of norm;17 TG's

390 ESA ITKONEN

'rules' I will call 'grammatical rules'. On the present interpretation, rules 'govern' intentional social behavior and are in return manifested by this same (rule-governed) behavior. Rules are primary with regard to any particular instances of behavior subsumable under them,18 but they are interdependent with this relevant behavior as a whole: where there is w regular behavior, there are no rules either (cf. 2.O. above). -We have to distinguish rules from the sentences referring to them; the latter will be called 'rule-sentences'. It is, for example, a rule of English that the definite article the precedes (and does not follow) the noun. This rule is (correctly) referred to by the rule-sentence "In English the definite article precedes the noun".

Next we have to make the far-reaching distinction between regulari­ties (or uniformities) in nature and rules. It is an axiom of the phi­losophy of natural sciences that a universal hypothesis referring to a presumed regularity is falsified, if counter-instances occur to it (cf. Hempel 1965:39-40); the hypothesis "All pieces of metal expand when heated" is falsified, if we find a piece of metal that does not expand when heated. On the other hand a sentence referring to a rule is not falsified simply because there occur (apparent) counter-instances to it. Thus our rule-sentence "In English the definite article precedes the noun" is not falsified even if we should come across an utterance like *"Girl the came in": such an utterance is incorrect whereas rule-sen­tences are about correct utterances (and sentences) only. It is a re­markable fact that behavior violating a rule does not falsify the cor­responding rule-sentence.19 Furthermore, universal hypotheses can only be confirmed to a higher or lower degree, which means that they may eventually be assumed to be true; but they can never be conclusively confirmed, or verified, i.e., known to be true. But it is clear that our rule-sentence is known to be true with absolute certainty. To deny this is to claim that even when examining our own intuitive knowledge

of English, we can only 'observe' particular occurrences, e.g., 'the man', 'the woman', 'the boy', etc., 'confirming' the 'hypothesis' "In

TG AND PHILOSOPHY OF SCIENCE 391

English the definite article precedes the noun", without being able to know whether or not the 'hypothesis' is true. But such a claim is sure­ly absurd. - Remember that the refutation of private languages provided the logical justification for us to investigate the language of our community by investigating our own language, i.e., our own linguistic intuition. And in any case, TG has always been following this strategy (cf. its concept of 'totally homogeneous speech-community'). (The very real problems arising in connection with dialectology need not concern us here.)

According to another axiom concerning the nature of hypotheses, an existential hypothesis like "There are unicorns" can be verified by finding a confirmatory instance, but cannot be falsified, i.e., conclu­sively disconfirmed: if we do not find any unicorns, this does not prove that there are none (although it makes this highly probable). Now, if we make an analogous existential statement concerning (our dialect of) English, e.g., "In English there is a preposition 'blip' which is fully synonymous with 'on'", it is clear that we know its falseness with abso­lute certainty. To deny this is to argue, absurdly, that although we do not personally know of any such English preposition, this does not prove that it does not exist in our own dialect. And even if we happened to make an utterance like *"The book is blip the table", we could simply brush aside this example as irrelevant, since - once again - it is in­correct English.

Thus we see that in matters of verification and falsification em­pirical hypotheses of natural sciences and statements about language be­have in fundamentally different ways. In particular, the difference be­tween rules and regularities could be preliminarily summarized in the following way: rules determine which occurrences are correct, whereas occurrences determine which hypotheses about regularities are true.

At this point it might be objected that the rule-sentence "In En­glish the definite article precedes the noun" is actually false because of such (correct) forms as "Ivan the Terrible". This objection does not

392 ESA ITKONEN carry much weight, however. There are obyiously very few rules with no exceptions, but exceptions are known with just as much certainty as are the rules which they are exceptions to. Therefore» exceptions cannot falisfy rule-sentences and are thus in no way comparable to 'counter-instances' occurring within natural sciences. And since exceptions are known in advance, it is often unnecessary to enumerate them when formu­lating a rule-sentence referring to a particular rule.

It has often been argued that 'rules' (i.e., my 'rule-sentences') cannot have truth-value since they are equivalent to orders. But it is quite undeniable that though our rule-sentence about the place of the English definite article for example may have a prescriptive function, its primary function is nevertheless descriptive. That is to say, this sentence is true, and any of its negations, e.g., "In English the defi­nite article follows the noun", is false. Notice that an order to have truth-value, a rule-sentence must indicate the language (or, more gen­erally, the 'game') containing the rule which the rule-sentence pur­ports to refer to. Rule-sentences purporting to refer to rules which are not constituents of any specified existing games "hang in the air" (Wittgenstein 19581, p.380) and lack both truth-value and a determinate meaning.20

It is not only the case that rule-sentences are true or false. We have seen above that true rule-sentences are unfalsifiable, or known with absolute certainty to be true, which means that rule-sentences are either necessarily true or necessarily false.

Here I have to make a short digression in order to clarify the terms involved. It is customarily said that indicative sentences of the universal form can be divided into those which are necessarily true (or false) and those which are empirically true (or false). Necessary truths are further divided into analytic truths and synthetic a priori truths. Sentences like "All red roses are red" are explicitly analytic, viz. substitution-instances of logical principles, while sentences like "All bachelors are unmarried" are implicitly analytic, viz. analyzable

TG AND PHILOSOPHY OP SCIENCE 393

as explicitly analytic sentences. Sentences like "No surface is both red and green all over" are synthetic a priori viz. not reducible to analy­tic sentences. It is the defining property of necessary truths that it is impossible, or contradictory, to think and assert their falsity. On the other hand empirical sentences like "All swans are white" can be imagined to be false without contradiction; i.e., they are in principle refutable by experience.21 (But notice that in all standard textbooks of logic and philosophy of science empirical sentences of the universal form always refer to regularities, but never to rules.)

To return to my claim of the necessity of rule-sentences, it may seem at first sight that this claim is obviously false. It could be argued that though "In English the definite article precedes the noun" happens to be true (and even "irrefutably" so), it is nevertheless con­ceivable that it could be false; and in that case it is not necessarily true. Furthermore, since necessary truth is often associated with unin­formativeness (cf. "All red roses are red"), it may seem counter-intu­itive to claim that rule-sentences contained in school-grammars for instance are necessarily true: for one who does not know English it is certainly informative to hear that the English definite article precedes the noun and does not follow it, as is the case in Swedish for example22

- It will be seen that these objections overlook the peculiar nature of rules, as distinguished from regularities.

When we say that it is conceivable that "In English the definite article precedes the noun" could be false, what we mean is that it is conceivable that an utterance like "Girl the came in" could be correct. But in that case we have to assume that English has changed from what it is now. In other words the rule referred to by "In English the definite article precedes the noun" would have to be replaced by a new rule re­ferred to by "In English the definite article follows the noun"; and this new rule-sentence would be known to be necessarily true precisely in the same way in which the first-mentioned rule-sentence is today known to be necessarily true.23 But notice that from the synchronic

394 ESA ITKONEN

point of view (with which I am exclusively concerned in this paper) the assumption of a language change is illegitimate, or even contradictory. If I were speaking of regularities, instead of rules, such an assumption would mean that universal empirical hypotheses about regularities could not be falsified by single counter-instances, but only if the regulari­ties themselves changed. But besides being completely unnatural, this reasoning is fallacious as well, because the change of regularities would not bring about the falsity of the hypotheses in question: these were meant to refer to the regularities in their not-yet-changed status; and when the regularities changed, the hypotheses did not begin to fail to refer to them, but rather ceased to be hypotheses with a determinate referential function.

It is clear that terms figuring in a universal linguistic theory must have constant abstract meanings, but it is just as clear that the applications of these terms in the descriptions of particular languages have practically always different meanings. For example, "definite ar­ticle" has different particular meanings in the grammars of English and Swedish (or Finnish, for that matter). Now it can be seen that if the rule referred to by "In English the definite article precedes the noun" changes in the way described above, the meaning of "English definite article" changes, too. That is, today it is part of the concept '(cor­rect) definite article in English' that it precedes, and does not fol­low, the noun. And, analogously, if the rule in question changes, the concept 'English language', strictly speaking, changes too: there is no language over and above the rules of language;24 therefore, a language A whose rules are not identical with those of a language must be different from the language B. Since the meaning of rule-sentences is in more than one way dependent of the rules to which they refer, it is clear that rule-sentences cannot even in principle be falsified by a change in rules, in addition to the fact that changes are excluded from a synchronic context anyway. Moreover, rule-sentences can be falsified neither by incorrect forms nor by exceptions (cf. above). Hence rule-

TG AND PHILOSOPHY OF SCIENCE 395

sentences could apparently only be falsified by correct forms (or ac­tions). But since rule and correctness are correlative concepts, to know the correctness of a form is to know the rule(s) to which it con­forms; and to know a rule is to know - without any kind of verification or falsification - the truth or the falsity of the sentence purporting to refer to it. Hence true rule-sentences are always known to be true and cannot be falsified, whereas false rule-sentences, e.g., "In En­glish the definite article follows the noun", are known to be false and are thus necessarily 'falsified' by the correct forms, e.g., 'the man', 'the woman', etc.

As ordinary language philosophers like Cavell (1958) and Hare (1957) realized quite some time ago, the necessity of rule-sentences depends on who is uttering them. When one is learning a language (or a game), one's tentative rule-sentences are genuine hypotheses insofar as new evidence may"show that they are false. But after one has mastered a language or a game, one's knowledge of it is 'internal', i.e., intuitive and irrefutable, and the sentences which one (sincerely) uses to refer to this knowledge are necessarily true. Let it be added that there are objective criteria for distinguishing between learning a language (or a game) and mastering it. Cavell and Hare have explicitly claimed that due to the normativity of rules, rule-sentences formulated by a native speaker or a competent player are necessary, and they also point out that this is a type of necessity unknown to philosophers operating either with the trichotomy "analytic" - "synthetic a priori" - "empirical" or with the dichtomy "analytic" - "empirical".25 Hence it is by no means the case that the necessity of rule-sentences entails, or ought to en­tail, their uninformativeness: the existence of a game is contingent and not necessary; therefore, it is informative to hear that such and such a game is played. But given its existence, the rules within it (are known to) hold necessarily.

4.0. Regularities are instantiated by (physical) events, whereas rules are instantiated by (intentional) actions. Corresponding to these

396 ESA ITKONEN

two different types of phenomena there are two different methods of gaining knowledge , namely observation and understanding. Still today positivists often make the gross mistake of trying to reduce actions to events or, equivalently, to derive understanding from observation. But it is quite easy to see that a sentence referring to an action neither entails nor is entailed by any definite set of sentences referring mere­ly to movements and/or sounds. One and the same action(-type) may be performed in an indefinite number of ways so that knowing that a certain action has been performed does not entail knowing which particular move­ments have been made or which particular sounds have been emitted in per­forming this action. Contrariwise, from the mere fact that certain move­ments are made or certain sounds are emitted, it cannot yet be inferred that anything is being done intentionally. This is Brentano's well-known 'thesis of intentionality',26

Knowledge of rules is achieved by understanding that standards of correctness, i.e., right or wrong ways of acting, are involved. To know a rule is to know what is a correct action falling under it or, equiva­lently, to be able to correctly follow it oneself. In this respect reg­ularities are of course entirely different, as can be seen from the fact that in connection with regularities there is no use for the 'cor­rect - incorrect' distinction: if within the natural sciences we have to do with »incorrect behavior, it is of necessity the behavior of re­searchers , not of research objects (cf. Winch 1958:57-62, 83-86).

Rules which are not known (to exist) are not rules, at least not in the sense defined here. But regularities can certainly be said to exist even if no one knows them; and, in fact, it is an axiom of the philosophy of the natural sciences that genuine regularities can never be known with certainty (cf. pp. 390f., above). Regularities and the know­ledge of them do not coincide; therefore, when one is inquiring about regularities, one cannot be said to be inquiring about one's knowledge of them, except in the case of axiomatization (cf. pp. 224-25). But when one is inquiring about rules (after having learned them), one is in

TG AND PHILOSOPHY OF SCIENCE 397

fact inquiring about one's knowledge of them. The preceding discussion clearly shows that knowledge of rules is

different in kind from knowledge of regularities. That the agent's know­ledge differs from the observer's knowledge, is in effect a very old idea.27 When one investigates the rules of one's own behavior, includ­ing rules of language, one cannot help relying upon what Specht (1969: 132-33) has called 'immanent reflection': this term indicates that one need not take any new facts into consideration, but only reflect upon what one already knows. The fact that rules are attended to by immanent reflection, instead of external observation, is connected in an obvious way with the necessary character of rule-sentences (cf. above).

Without acknowledging it, TG investigates rules which are relevant to the concept 'correct sentence in a language L'. It is in this sense that TG is a peculiar form of conceptual analysis, rather than an empir­ical science. (Notice that all conceptual analyses are empirical in the sense that the concepts to be analyzed are empirically given.) This al­so makes it clear that TG is in no way different from such (sociological) disciplines which describe institutions of different societies by ana­lyzing the concept 'correct (religious/juridical/political/etc.) action in a society S'. Such disciplines are not empirical sciences making hypothetical descriptions about what is or will be done as a matter of fact; rather, they are systematizing descriptions attempting to calcu­late or generate all possible correct actions, irrespective of whether, and under what circumstances, such actions are actually performed. In this way we could construct 'generative grammars' for different insti­tutions, making the success of our 'grammars' depend on whether or not they are able to generate all and only (possible) correct actions of the relevant type. This means that Winch's 'aprioristic sociology' is the exact sociological counterpart of TG. It is interesting to note that in the domain of physics, too, there is a counterpart to TG, namely Lo­renzens 'protophysics', which, instead of studying actual physical events, defines the concept 'possible physical event' by starting from

398 ESA ITKONEN

the three distinct ways of measuring length, weight, and time (cf. Lo­renzen 1969). Although protophysics does not deal directly with rules, it is nevertheless comparable to TG and to Winch's theory, because the possible events it studies are results of correctly following the above-mentioned three ideal rules of measuring.

By now we are in a position to expose the untenability of the anal­ogy between TG and natural sciences which Chomsky postulated by claiming that TG investigates 'observable events' consisting in that "such and such is an utterance" (cf. p.383). First, that something is an utterance, is not an observable event. Second, it is also not an event. Thirdly, and most importantly, TG does not deal simply with utterances, but with correct utterances. And since particular utterances (or actions) are secondary with respect to rules which determine their correctness'or in­correctness (cf. note 18), it follows that TG is in fact concerned with rules or - since rules are inseparable from the knowledge of them - with the knowledge pertaining to rules, viz. linguistic intuition. By 'intu­ition' I understand a type of knowledge which can always be brought on­to the level of consciousness and which must thus be sharply distin­guished from the hypothetical 'tacit competence'. Linguistic intuition is (pace Postal 1966:3) fundamentally different from the quantifiable subject-matter of standard natural sciences and rather belongs together with intuitive knowledge of the sociological, philosophical, or logical kind, as it is investigated in hermeneutic or Winchian descriptions, philosophical explications, and logical analyses, respectively.

Given that TG deals with intuition, but aspires to be a natural science, it is not surprising that its official position on the rule -regularity dichotomy is rather confused. Since natural sciences do not have to recognize the existence of rules and norms, it is to be expected that TG professes to be investigating regularities, not rules. This is in fact what is asserted in most methodological statements, but it has also been claimed by Chomsky that (due to the degenerate quality of actual speech) there are no regularities in speech (see Chomsky 1966b:

TG AND PHILOSOPHY OF SCIENCE 399

32 and also 1965:207) - a claim which is easily contradicted for in­stance by Slobin's (1971:42, 53) observation that there are already regularities in the speech of 18~month-old children. On the other hand Chomsky has always maintained that TG examines some 'deep regularities' of language, and what he has in mind is probably that TG attempts to make generalizations concerning the correct sentences of particular languages. (Notice that all 'generalizations' need not be taken in the positivist sense; e.g., the philosophical analysis of epistemic concepts consists in making 'generalizations', i.e., in making as general and as simple an analysis as possible.) The only type of rule which TG allows for are rules of a theoretical description, i.e., gramr.atical rules, about which native speakers mostly have absolutely no antecedent intu­ition. This has several unnatural consequences. Although rule and cor­rectness are well-known correlative concepts, TG is forced to maintain that native speakers are aware of the correctness of sentences without being (able to become) aware of the rules determining their correctness. Hence, bound by its allegiance to positivism, TG ignores the existence of linguistic rules as socially given, normative phenomena and inter­prets languages simply as infinite sets of sentences (i.e., correct sentences); sentences are straightforwardly taken to be 'objects' of a certain kind, and they are investigated on the basis of methods which are supposed to be identical with those employed by natural scientists in the investigation of their own data.28 But this interpretation gives a wholly distorted, static and reified picture of language, an ergon in Humboldt's famous terminology, rather than an energia, an in­tellectual 'generative' activity. There is no way to represent the ob­vious fact that utterances, as tokens of corresponding sentence-types, are results of intentional rule-governed behavior, viz. speaking (cf. pp. 388-89 and 431-33). And within this positivist framework it re­mains incomprehensible why this particular set of objects (= sentences), as distinguished from sets of objects investigated in standard natural sciences, requires its own type of knowledge (= linguistic intuition).

400 ESA ITKONEN

Within a non-positivist or herraeneutic framework: however, linguistic intuition is seen to be a special case of the 'agent's knowledge', i.e., man's knowledge about his own actions and the rules governing them.

It is in this same positivist vein that Chomsky and Halle (1968: 331) felt obligated to make the 'simplifying assumption' that "all of the primary linguistic data must be accepted as 'correct'". But this assumption is not just simplifying. Rather, since it rules out any ex­plicit consideration of the basis for the correctness of utterances, it in fact obliterates one of the most crucial aspects of the distinction between natural and human sciences. And even when TG does examine 'un-grammatical' sentences, ungrammatically is not brought into relation with the more general, social phenomena of norm-consciousness and norm-breaking.

5.0. Since the rules of a language are known in an irrefutable way and referred to by necessarily true rule-sentences, TG-type linguistics might appear to be a yery trivial undertaking; however, this is not the case. Knowing the rules of a language is only a -precondition for writ­ing a theoretically interesting grammar of this language; a mere list of (necessarily true) rule-sentences is certainly not such a grammar. To put it briefly from the fact that I knowi something, it by no means follows that I also know2 how to describe my knowledgel in the best possible way. Here I distinguished between two different types or levels of knowledge, identifiable as the knowledge of the native speaker (= re­search object) and that of the linguist (= researcher). These two levels may be called 'atheoretical' and 'theoretical', respectively. Clearly there must be a constant mediation between atheoretical and theoretical because, yery simply, one cannot (theoretically) describe atheoretical knowledge unless one in fact possesses it. This means that it is a necessary precondition for human sciences that the scientist in a sense identifies himself with his research objects. ('Human' sciences are sciences which rely on understanding, not on observation and experimen­tation.) Schutz (1967:5-6) has elucidated this two-level character of

TG AND PHILOSOPHY OF SCIENCE 401

the human and social sciences in the following way:

The facts, data, and events with which the natural scientist has to deal are just facts, data, and events within his observational field but this field does not 'mean' anything to the molecules, atoms, and electrons therein. But the facts, events, and data before the social scientist are of an entirely different structure. His observational field, the social world, is not essentially structureless. It has a particular meaning and relevance structure for the human beings liv­ing, thinking, and acting therein. They have preselected and prein-terpreted this world by a series of commonsense constructs of the reality of daily life, and it is these thought objects which deter­mine their behavior, define the goal of their action, the means available for attaining them - in brief, which help them to find their bearings within their natural and socio-cultural environment and to come to terms with it. The thought objects constructed by the social scientists refer to and are founded upon the thought objects constructed by the common-sense thought of man living his everyday life among his fellow-men. Thus, the constructs used by the social scientist are, so to speak, constructs of the second de­gree, namely constructs of the constructs made by the actors on the social scene, whose behavior the scientist observes and tries to explain in accordance with the procedural rules of his science.

We have to distinguish between atheoretical observational knowledge pertaining to events and regularities and atheoretical intuitive know­ledge pertaining to actions and rules. I am concerned exclusively with the latter type of atheoretical knowledge, and the 'rules' which I have been discussing can now be more fully characterized as atheoretical (linguistic) rules. These rules are infersubjective and normative, and they are directly referred to by necessarily true rule-sentences. On the other hand, every attempt to systematize a greater or smaller num­ber of atheoretical rules produces a description containing theoretical rules.29 Since we have no antecedent intuition about the rules of a theoretical description it is clear that, unlike rule-sentences, these descriptions are 'disconfirmable' in some sense, namely in the non-po-sitivist sense in which Winchian descriptions and philosophical or log­ical analyses are also 'disconfirmable' (see below). And since theore­tical, or grammatical, rules are not known in advance, they cannot be normative.

402 ESA ITKONEN

Using the terms 'theoretical ' and 'grammatical' interchangeably at this theoretical level eliminates the necessity of making a distinction corresponding to the one between atheoretical rules and rule-sentences referring to them. A distinction I prefer not to make, because the onto­logical status of the referents of theoretical-grammatical rules seems unclear to me. To say that these referents must be features of a psy­chological and ultimately physiological mechanism,30 is to answer a different question. Another example: do all of the 'grammatical' rules needed in philosophy and logic 'exist' only in the sense that they ul­timately refer to something in our brain? And if this question is ir­relevant within philosophy (as it certainly is), why should it be re­levant within linguistics? This is not an attack on psycholinguistics; it is simply an assertion that a linguistic description does not in it­self force us into making the additional, psychological assumption that our description is somehow represented in the human brain. Actually, this is the course also taken by TG in practice, programmatic state­ments notwithstanding. As a matter of fact, plausible or acceptable TG rules have never been rejected just because available psycholinguistic evidence does not support them; this fact unmistakably points to a cer­tain de facto independence of linguistics from psychology. Similarly, no one would dream of rejecting, for example, a rule of deontic logic just because experimental psychological evidence either fails to sup­port it or contradicts it.31

The rule-conception operating with the atheoretical - theoretical distinction must not be confused with a conception which distinguishes between an 'internal' and an 'external' attitude towards rules in such a way that the former is characteristic of those who act according to rules and thus know them 'from inside', whereas the latter is charac­teristic of those who describe rules. Although this conception (which has recently been represented by Gumb 1972, for instance) rightly em­phasizes the social and normative aspect of rules, it is indefensible as should be evident from the quotation from Schutz above - there can

TG AND PHILOSOPHY OF SCIENCE 403

be no external attitude with respect to rules: one who has not ('inter­nally', intuitively) understood rules, cannot (theoretically) describe them; the most he can do is to describe observable regularities of sounds and movements. Therefore, description of rules can only build upon their atheoretical, 'internal' understanding; it cannot be sepa­rated from it as a totally different, unrelated attitude. This is also why Gumb and others are mistaken when they suggest that, from the ex­ternal point of view, the difference between rules and regularities is not methodological, but merely heuristic: if rule-governed behavior is really viewed externally - as recurrent pattern of events consisting in sounds and movements which are not (and could not be) understood - then there is absolutely no difference, not even a heuristic one, between rules and regularities; on the other hand, if one sets about to describe actions which one has understood by relating them to rules which one knows, then any 'external' point of view is a fiction (due, once again, to an uncritical imitation of the methodology of the natural sciences), and the difference between rules and regularities has certainly the greatest methodological importance.32 - Notice that, strictly speaking, we have to distinguish between describing behavior governed by rules and describing rules manifested by behavior. Behavior, whether inten­tional or not, is located in time and space; therefore, there are always instances of behavior, both past and future, which we cannot know. On the other hand, rules exist in a different way: once we have learned them, we know them, and what we have to do in order to describe them is not to look for new facts, but to reflect upon this (atheoretical) knowledge of ours (cf. pp.396-97; Specht 1969:132-33). As I have al­ready indicated several times before, TG descriptions belong to the latter category. But this does not change the fact that describing rule-governed behavior is fundamentally different from describing re­gularities in nature (cf. section 9 below).

In contrast to the one-level theories characteristic of the natural sciences, linguistics has a de facto two-level character, a point TG has

404 ESA ITKONEN

never been able to acknowledge because of its general positivist out­look. Indeed, TG has not been able to distinguish between atheoretical rules known by the native speaker and theoretical rules constructed by the linguist. This confusion has led to the generally accepted view that native speakers (consciously ) know the rules of their language only in an unreliable and incomplete way. This view is customarily justified by referring to the certainly incontestable fact that speakers do not know the rules contained in the (transformational) grammar, i.e., theoretical description, of their language. But this 'justification' has no validity whatsoever, because it is self-evident that knowing atheoretical rules does not entail knowing theoretical rules (cf. Schutz 1967:5-6 quoted above).

I maintain that people are able to know, or to come to know, the a-theoretical rules of their native language. (Cavell (1958) and Hare (1957) say as much in claiming rule-sentences are necessary). For example in English such rules as a definite article precedes a noun; or, apart from certain exceptions, -s is the correct plural ending; or the adverb 'voluntary' cannot be applied to events involving only inanimate objects; or words may begin with p-, p r - , p l~ , but not with r p - , lp-; etc. The claim that atheoretical rules are known is in fact valid in a fundamen­tal, albeit trivial, sense, since the meanings of all words are based on the rules of their use: surely it cannot be seriously doubted that a native speaker of English knows (or is able to come to know) the rules for the correct use of say, 'coca-cola' or 'movie star'? Knowing a rule does not entail being able to formulate the corresponding rule-sentence straight away. And, as Chomsky has pointed out, if someone should not immediately grasp a given rule which is obvious to us, "it may be necessary to guide and draw out the speaker's intuition in per­haps fairly subtle ways". (Chomsky 1965:24)33

Just as there can be no norm-breaking without a norm, so each time we say something incorrect, there is an atheoretical rule involved. (Here I am not concerned with establishing a distinction between gram-

TG AND PHILOSOPHY OF SCIENCE 405

maticality and acceptability.) Consider for example the obviously in­correct sentence *"Likes Joad philosophy?". When we utter it, we are not, as far as we know, violating any TG rules, in this case the 'aux­iliary incorporation transformation", which says inter alia that 'in­corporation' does not take place when the 'verbal' does not contain the 'feature' '<+copula>'.3U Hence, this (theoretical) TG rule is not nor­mative for us, though it may be profitably used to describe something which is normative for us. In other words we do know that we are vio­lating an (atheoretical) rule which we may not be able to formulate in any elegant way, but which could perhaps be expressed by the following rule-sentence: "In English, when a whole sentence is questioned, the sentence begins with some form of the verb 'do' and the main verb fol­lows the subject, except when the sentence begins with one of the fol­lowing verbs: 'be', 'must', etc."35 Therefore, in contradistinction to theoretical rules, atheoretical rules are normative; and it is a characteristic of rule-sentences referring to them that when they are understood, they are known to be either true or false, i.e., the truth of a rule-sentence cannot be doubted, but only accepted or rejected. This is a general characteristic of necessary sentences.

Most 'rules' of traditional school-grammars are more similar to my 'rule-sentences' than to my 'theoretical' or 'grammatical rules'. (Incidentally, the same is true of von Wright's (1963:6-7) 'rules of grammar', since they are normative.) Now it is clear that any com­prehensive grammar of this type is bound to contain hundreds or, more probably, thousands of more or less unrelated rule-sentences. These are known to be true, and it is precisely for this reason that they are theoretically uninteresting (although valuable to a foreigner try­ing to learn the language in question). Therefore, if we want to give a systematic and interesting description of a language, we have to resort to (grammatical) rules of a more theoretical kind. Such rules attempt to give a coherent account of all the data involved by discov­ering (or inventing) 'generalizations' which have not been known be-

406 ESA ITKONEN fore. In order to make generalizations or to find out whether they are valid or not, one has to be able to reflect in a creative way upon one's linguistic intuition and to master certain formal techniques to express the results of one's reflections. Some linguists are better in these exercises than others, and it is always possible, and even probable, that better and more revealing generalizations will be made. This is why no grammar will ever be (known to be) the true one or the 'final' one. But since these generalizations are based on atheoretical rules which are known with absolute certainty to exist or, to put it in another way, since they abstracted from an indefinite, perhaps extremely large num­ber of (potential) rule-sentences each of which is (known to be) neces­sarily true, it follows that they are different in kind from generaliza­tions made within the natural sciences. And, therefore, TG descriptions must be interpreted as conceptual analyses or explications, rather than empirical theories in the positivist sense (see sections 6 and 7 below).

Since, in principle, TG has to account for everything there is in a language, a grammar of a particular language must contain, not only grammatical rules expressing high-level generalizations about a large number of atheoretical rules, but also grammatical rules expressing quite low-level generalizations and, ultimately, referring to single atheoretical rules. As an example, consider "a certain grammatical rule of English" discussed by Chomsky, namely "a selectional rule that deter­mines the grammatical categories of the subject and object of the verb 'play'" (Chomsky 1961:234-35). Now it is self-evident that we must credit eyery normal English speaker with a (potentially) conscious knowledge about the atheoretical rule which says that the subject of 'play' must be animate ("John plays golf"), not inanimate (*"Golf plays John"). More generally, strict subcategorization rules and selectional rules are particularly clear cases of TG rules identifiable as rule-sentences.

We have seen that the self-imposed positivist methodology has pre­vented TG from recognizing the crucial distinctions between rules and

TG AND PHILOSOPHY OF SCIENCE 407

regularities, rules and rule-sentences, atheoretical rules and theoreti­cal rules, and rule-sentences and theoretical, viz. grammatical rules. But in addition I would like to suggest that the failure to distinguish between atheoretical and theoretical rules is to some extent due to the influence of formal languages. Since formal languages, too, are lan­guages, they must be explicable in terms of their usage (cf. pp. 387-88, above), and they are indeed used by logicians and mathematicians. Yet a formal language is created by establishing a set of explicit, 'gram­matical' rules, e.g., rules for inference or generation, and the use of this language consists in meticulously following or applying these grammatical rules. Therefore, in a formal language there is, properly speaking, no difference between atheoretical rules of use and theoreti­cal rules describing this use; rather, the grammatical rules, which are consciously used in 'speaking' this language, could be employed ve-flexively to describe the language itself. Now if we think the grammar of a formal language is defined as a set of explicit, grammatical rules, then it is quite plausible to define the language itself as an infi­nite set of sentences generated by the grammar (cf. Chomsky and Miller 1963:283-85), because there is nothing else by which the language could be defined. This conception may be acceptable as far as formal languages are concerned, but it does not apply at all to natural languages. Formal languages come into existence together with their grammatical rules, but natural languages, as human institutions, exist irrespective of whether or not they are codified into explicit generative grammars. The only natural way to interpret an institution is to interpret it in terms of its constituents, i.e., atheoretical, social rules or rules of inten­tional behavior, not by any set of single, unrelated actions or results of actions. But if it can be assumed (as I think it can) that TG has mechanically applied to natural languages that conception of language which it has once learned in connection with formal languages, then it becomes immediately understandable why TG has to ignore social rules of language (which constitute nevertheless the most conspicuous aspect of

408 ESA ITKONEN

language!) and to define a language as an infinite set of sentences gen­erated by a grammar composed of non-social, theoretical rules.

6.0. Since TG aspires to be a natural science, a TG description ought to make use of the methods of explanation, prediction, and con­firmation, as they have been specified by the positivist philosophy of science. But we have already seen that such an assumption is impossible. A theory of any natural science explains first of all particular occur­rences and is confirmed or disconfirmed by them. But a TG description must in the first place 'explain' (atheoretical) rules, not particular linguistic occurrences, since it deals with correct sentences and the correctness (or the incorrectness) of a sentence is determined by the relevant rules. The rules determine primarily the correctness of sen-tence-types, which means that in linguistics types are primary with respect to tokens, whereas - as Hempel (1965d:423) has explicitly stated - in natural sciences tokens, as objects of explanation, are primarywith respect to types.

Moreover, it is simply impossible to apply the D-N model to TG de-# scriptions, because the 'individuals', i.e., space-time points, a, b, c, etc. mentioned in D-N explanations have no role to play in TG de­scriptions.36 When we are describing institutions by analyzing the concept 'correct action' or 'correct sentence', we may with justifica­tion deal with actions or sentences which have never occurred in the intersubjective spatio-temporal world. We may 'explain' an action (or a sentence) which has never been performed (or uttered) by relating it to some general properties of correct actions (or sentences) of the same or of a different category. And we may 'confirm' our analysis by non-exis­tent actions (or non-uttered sentences), i.e., by showing that actions (or sentences) which according to our analysis ought to be correct are correct, irrespective of whether or not they have occurred or will ever occur. We can do this, because we know the rules which determine the correctness of actions (or sentences), and because knowing a rule means

TG AND PHILOSOPHY OF SCIENCE 409

knowing an indefinite number of possible correct actions (or sentences). Compare this situation to the one prevailing in natural sciences. It would be absurd to say that a natural scientist confirms his theory on the basis of non-existent events.

The reason why the above-mentioned facts about the nature of lan­guage have always been overlooked by TG (and by most other linguistic theories as well) can perhaps be seen in the peculiar character of ut­terances, conceived as actions. In the case of utterances, the distance from imagining an action to performing it is exceptionally small. Since we can utter all the (correct) sentences we can imagine (whereas we can­not straight off perform all the correct actions we can imagine), we are inclined to think, erroneously, that we are dealing with empirical, 'ob­servable' events only.

What I have said so far shows the inapplicability of the positivist notions of explanation, prediction, and confirmation to TG; however, for the sake of clarity, I will discuss this question in somewhat greater detail.37

According to TG's standard position, a grammar is constructed on the basis of a certain corpus and it is 'tested' by finding out whether or not the grammar 'predicts', i.e., generates, all and only correct sentences not included in the corpus with their correct structural de­scriptions. Sentences are 'explained' by deriving them, with correct structural descriptions, from the grammar. - The language (= 'L1') which I will consider in what follows is extremely simple, nevertheless it must possess some theoretical interest, since it has been used for illustrative purposes by Chomsky, Postal, Katz, McNeill, and others. TG's conception of confirmation can be represented as follows:

'Corpus': Grammar G1 : 'Prediction': "aabb is ab, aaaabbbb, R1. S → aSb a correct sentence of L1" aaaaabbbbb R2 • S → ab S

/l\ (R,) / s \ (R2) / \

410 ESA ITKONEN

The 'prediction' turns out to be true. Are we now entitled to say that our grammar Gl has been empirically confirmed? Obviously not. We must consider the full consequences of the fact that the corpus does not merely consist of actually observed utterances, but of actually observed correct utterances; utterances like *ba, *aab, *aba, etc. have been ex­cluded because we, as native speakers of Lļ, intuitively know that they are incorrect. Therefore, the utterances in the corpus are utterances instantiating an intuitively known atheoretical rule, which the average native speaker is able to reflect upon, and if he is suitably guided by the linguist (cf. note 33), he may even be able to describe this rule with the aid of the rule-sentence "The correct sentences of L1 consist of a certain number of a's followed by an equal number of b's". But this does not mean, of course, that the average speaker would also be able to describe this atheoretical rule with the aid of the theoretical or gram­matical rules R1 and R2 constituting the grammar G1. In any event, Gļ does not just describe a corpus of (correct) utterances, but rather lin­guistic intuition, i.e., an intuitively known rule which determines the correctness vs. incorrectness of all possible utterances. For this rea­son, establishing a corpus of factually made correct utterances is just an idle ceremony. And in actual practice no transformationalist ever bothers to base his discussion on a real corpus; their simple sentences are rather in the style of "Sincerity may frighten the boy".

Grammars attempt to describe rules by generating the sentences in­stantiating them; to know a rule by generating the sentences instantia­ting them; to know a rule is to know what kind of sentences instantiate it and thus are correct; G1 describes a rule which we know; therefore, G1 must generate or 'predict' only correct sentences, unless we have made a mistake at some stage of our description. That is to say, we know that our 'predictions' are true even before we have tried to 'test' them, and, therefore, they are not predictions. Consider the analogous case where, after constructing a grammar NP → the + N on the basis of the 'corpus' "the man", "the woman", I attempt to 'test' my grammar by

TG AND PHILOSOPHY OF SCIENCE 411

picking out the unit "girl" from the lexicon and making the 'prediction' that "the girl" is a correct form.

If I actually try to 'test' G1 by 'testing' whether the 'predicted' sentence "aabb" is correct or not, I use as the criterion of correctness that very same rule which I intuitively know and which I have formally described with the aid of my grammar G1 now generating or 'predicting' "aabb". Since one and the same linguistic intuition is both the subject-matter of the description and the criterion of its correctness, there can be no new, conceptually independent evidence which could empirically either confirm or disconfirm the description. The new sentence aabb does not constitute conceptually independent evidence, because its correctness is correlative with the atheoretical rule which I know and which is a-theoretically described by the (necessarily true) rule-sentence "The cor­rect sentences of L1 consist of a certain number of a's followed by an equal number of b's" and theoretically described by my grammar G1. In fact, the whole terminology of 'testing', which has been borrowed from the methodology of natural sciences, is out of place here. Natural events are outside us: we do not know for sure whether they will come out as predicted, and therefore, they may disconfirm our (hypothetical) descrip­tions. But when we are formalizing our own knowledge, which is always the case when we are conducting a conceptual analysis, there is no new, 'external' knowledge which could disconfirm our formalization, viz. ex­plication. (And there can be no such new external events either, because events are by definition irrelevant to a description of knowledge; cf. pp. 396-97, above.)

On the other hand, although we cannot speak of genuine disconfirma­tion here, it is clear that when we are dealing with formal descriptions of a complex intuitive knowledge pertaining perhaps to several hundreds of atheoretical rules, some formalizations are bound to be better than others. A grammar which merely lists the atheoretical rules is not, strictly speaking, contrary to fact, but it is uninteresting. On the other hand, a grammar which makes generalizations concerning all the a-

412 ESA ITKONEN

theoretical rules involved is interesting but likely to be contrary to fact. As Karl Popper has constantly emphasized, an increase in scientif­ic interest or generality is correlated with an increase in liability to falsification. Yet this is a universal truth about every kind of de­scription, including logical and mathematical analyses. If, for example, an epistemic analysis of the concepts of knowledge and belief fails to do full justice to the concept of anticipation for instance, this anal­ysis is either modified or rejected, but this does not mean that it has been disconfirmed or falsified in the positivist sense.

Suppose that the sentence "aaabbb", which is generated by Gļ, turns out to be incorrect. This then means that while making our description, we have neglected the fact (which we knew all along) that the previously mentioned atheoretical rule of does not hold without exception. In other words, we have made an over-hasty generalization, and we must mod­ify our description accordingly. But this modification is not due to new evidence, because all the facts were given in our linguistic intuition from the start. Rather, the deficiency of Gj is due to our inattention, viz. an insufficient introspection, and it is revealed by a more care­ful introspection. We could, for instance, observe that all sentences where the number of a's and fc's is divisible by 3 are incorrect; such a restriction should somehow be expressed in our grammar.

We have seen that there are two prinicipal reasons why a grammar as a formalization of atheoretical intuitive knowledge should be invalid: a grammar may contain defects which are due either to the linguist's lack of insight or to his inattention. Lack of insight is involved when the linguist knows the data perfectly well, but simply does not come up with any interesting generalizations; or when he is not able fully to exploit the theoretical possibilities of his formal techniques. On the other hand, we have to do with inattention in cases where the linguist overlooks, or is mistaken about, certain features of his own atheore­tical knowledge; also eventual misuses of formal techniques are mainly due to inattention. Our errors due to inattention are mistakes: we over-

TG AND PHILOSOPHY OF SCIENCE 413

look something that we know. Now, it is obvious that a natural scientist has not made a mistake, when his empirical prediction turns out to be false: he could not have possibly known beforehand (intuitively or oth­erwise), what would be the outcome of his experiment. But then he is in­vestigating regularities, not rules.

TG deals with sentences which ave correct, not with sentences which in particular circumstances seem to certain (test-) persons to be correct. If I am fantasizing and think that "The weather is fine" is not correct English whereas "Ugglubb" is, this fact may have some psycholinguistic interest, but from TG's point of view it is irrelevant. Analogously mathematics is not a science about what calculations or proofs, as a matter of fact, seem correct under different circumstances, but a formal discipline about those calculations and proofs which are correct (see, e. g., Wittgenstein 1967 I, 156-58, 161, 163, and II, 20, 70, 76, 79). This only goes to show that neither TG nor mathematics (nor conceptual anal­ysis in general) are experimental sciences.

Since confirmation and explanation are correlative notions (cf. my initial statements in this paper), the fact that grammars cannot be con­firmed entails that they cannot explain, either, i.e., they cannot con­tain universal empirical hypotheses which would be needed as part of the explanans if grammars were to fit Hempel's (D-N) model of explanation. Furthermore, the lack of any observable linguistic events which could be reasonably considered as constituting the subject-matter of grammars testifies equally against the possibility of confirmation and explana­tion within TG. Now if we for the sake of argument assume that in con­nection with G1 the explanandwn-event (or 'effect') is for instance the fact that "aabb" is a correct sentence, then it is hard to see what could possibly serve as the antecedent condition (or 'cause'). The cor­rectness of a sentence is supposedly explained by deriving it from the grammar. But the foregoing steps in a derivation can hardly be inter­preted as (expressions of) antecedent condition, because this would mean that the initial symbol 'S' is, in the way of an "unmoved mover",

414 ESA ITKONEN

the ultimate cause of the correctness of any sentence. Hence the preter­minal string "aSb" cannot be the antecedent condition of the correctness of "aabb". Nor can it be the preterminal string together with (the appli­cation of) the rule R2: S → ab, since this would mean that the rules applied have a causal or at least empirical relation to the results of the derivation or proof.

The only possible 'D-N explanation' which can be proposed here takes the following form:

i) All sentences of L] consisting of a certain number of a's followed by an equal number of b's are correct.

ii) aabb is a sentence of L] consisting of a certain number of a's, i.e., two a's, followed by an equal number of b's.

Therefore "aabb" is a correct sentence of L,.

It is self-evident, however, that the second premise and the con­clusion are not conceptually independent, since the truth of the former cannot be 'verified' without 'verifying' the truth of the latter. In other words, knowing that "aabb" has the (correct) structure mentioned in the second premise involves knowing that "aabb" is correct. There­fore, we have to do here with one and the same intuition, and the at­tempt to force linguistic description into Hempel's model amounts to an (impossible) attempt to split one and the same intuition into two. The same conclusion can be reached by considering the first premise. It expresses an atheoretical rule, and if it is true, it is necessarily true. And since, if true, it is a necessarily true universal implication, its specialized form is also necessarily true; to give an example (which involves, to be sure, necessary truth of the explicitly analytical sort): not only (x)(f(x). g(x)=»f(x)) but also f(a). g(a)=>f(a) is neces­sarily true.

Since the second premise and the conclusion of a purported lin­guistic D-N explanation are conceptually dependent in such a way that the former entails the latter, the former cannot of course serve as the basis on which the latter could be predicted. Hence prediction, as de-

TG AND PHILOSOPHY OF SCIENCE 415

fined by Hempel and others, is impossible within TG. To sum up, it seems to me that the transformationalists' view of

the testability of their grammars is based on the unexpressed (and false) assumption that each time we say something that we do not know for sure, we are making an empirical hypothesis. But this simple-minded conception would turn for instance philosophy into a natural science, because philosophers repeatedly make claims which may or may not be true. As for 'explanation', TG seems to use this term in some vague sense in which "to explain x" means roughly "to increase the under­standing of x". More specifically, in TG 'explanation' amounts to dis­covering generalizations in the atheoretical linguistic knowledge to be described; and a generalization is expressed when two or more grammatical rules are replaced by a single rule. Now, this notion may provide an expedient strategy for making linguistic descriptions, but it is defin­itely not the positivist notion of explanation, as defined by Hempel and others.

Finally, there is a sense in which one can genuinely speak of ex­planation in connection with linguistic data: grammars, and ultimately sentences, of particular languages could conceivably be causally ex­plained from outside, i.e., by deducing them from some independently established general cognitive capacities (= regularities) plus partic­ular situational contexts (= antecedent conditions). It can be plausi-bily argued that a child could not have learned all the veritable in­numerable atheoretical rules of its native language without some kind of innate psychological mechanism which either presupposes or establishes certain organizing generalizations in the linguistic data presented to the child. And since, on the other hand, a (theoretical) grammar attempts to discover generalizations of precisely this kind implicit in the a-theoretical rules, it could be argued that a child is innately equip­ped to construct such a grammar for itself. But these speculations have absolutely no bearing upon the question as to the metascientific status of TG as a linguistic theory. TG descriptions can be shown to be expli-

416 ESA ITKONEN

cations of intuitive linguistic knowledge, and this is the answer to the question concerning their metascientific or methodological status. Wheth­er or not there is an (unconscious, innate) psychological reality closely corresponding to these descriptions, is an entirely different question, which can only be answered on the basis of psychol inguistic experimenta­tion, and certainly not on the basis of these very same descriptions which have given rise to this psychological hypothesis in the first place.38 If we want to consider this other (hypothetical, empirical) question, then we may note that available psycholinguistic evidence hardly supports TG's thesis of the isomorphism between the linguist's grammar and the language learner's mental 'grammar'. But, to repeat, this fact has no direct relevance to the validity or invalidity of TG as a linguistic theory. As I noted earlier (p.402), no TG descriptions have ever been rejected just because they conflicted with experimental psycho!inguistic evidence. And since TG descriptions are factually in­dependent of psychological descriptions, which alone could function as a basis for causal explanations of linguistic data, TG cannot right­fully claim to be providing genuine explanations. Similarly mathematics and the psychology of mathematics are two different things: a proof is rejected by showing that it is incorrect, not by showing that it con­flicts with psychological evidence concerning (the learning of) mathe­matical operations.

I do not deny that the study of language is ultimately part of the study of mind. I only require that one ought to be aware of the differ­ence between linguistic description and psycholinguistic explanation. Since TG has often failed to make this distinction, it has been able to create the illusion that it is using the term 'explanation' in its customary sense.39

7.0. So far it has been shown that a TG description pertaining to a particular language is a conceptual analysis, i.e., an analysis of the concept 'correct sentence in a language L' or of some subconcept 'cor­rect subordinate clause in L', 'correct declined/conjugated from in L',

TG AND PHILOSOPHY OF SCIENCE 417

'correct sound combination in L', etc. More precisely, the conceptual analysis in question is an analysis of intuitively known (atheoretical) rules and rule-instances which are relevant to the concept 'correct sentence in L'. The analysis is represented in the form of a grammar, i.e., a finite set of grammatical rules recursively specifying the in­finite set of the sentences in L and assigning structural descriptions to them.

In other words, it has been shown (insofar as it is possible in this limited space) that, contrary to its own methodological self-un­derstanding, TG is a human science. But it seems to me possible to characterize TG descriptions still more narrowly by identifying them with explications in a precisely defined sense. To begin with, it can be said that all good definitions are 'explicative' or 'explicatory' in the sense that by making our vague or merely which have previously been hidden from us, and in this way they increase our understanding: explications turn atheoretical conceptual knowledge into theoretical knowledge.

Explication has always been practiced, either implicitly or ex­plicitly, by philosophers and scientists alike, but it has occupied a particularly prominent place within analytic philosophy. The notion of explication may be defined in a number of different ways. I will make use of the definition Pap develops in his excellent bookof 1958.40 According to this interpretation, in the course of explication an intuitive­ly known concept or conceptual system, i.e., explicandum, which is referred to by the corresponding explicandum-expression(s), is replaced by its redefined or reconstructed form, i.e., explication, which again is re­ferred to by the corresponding explicatum-expression(s). Explicandum-expressions mostly belong to the ordinary language, whereas explicatwn-expressions belong to some theoretical, formal language. Often it is needless to distinguish between explicanda or explicata and the cor­responding expressions. An explicandum is identified with the aid of so-called criteria of adequacy, i.e., sentences which are intuitively

418 ESA ITKONEN

known to be necessarily true and in which the explicandum (-expressi on) occurs as an essential feature. To give an example, if we explicate the concept of knowledge, we may start identifying our explicandum with the aid of the following sentence which is intuitively known to be necessar­ily true: "If a knows that p, then p is true." The actual process of ex­plication consists in transforming these necessary truths of the merely intuitive kind into necessary truths of the formal or analytical kind. In the sentence functioning as criteria of adequacy, this is achieved by replacing the explicandum, which mostly has no inner structure, by an appropriate explication which has an explicit, articulated inner struc­ture. And sometimes it is also necessary to change the whole sentence-structure of the initial criterion of adequacy. Now, if the new sentence resulting from replacing the explicandum by the explication is formally true, the explication is said to satisfy the criterion of adequacy in question. Since the criterion of adequacy was initially used to identify the explicandum and since it is now satisfied by the explication, it fol­lows that the explicandum and the explication are SIMILAR in the required sense. (Explication would of course lose its point if explicata could uncontrollaby differ from explicanda.)

To give an example one possible explication of the concept of knowing (cf. Pap 1958:295):

Explicandum: a knows that p (="q") Criterion of adequacy: if a knows that p, then p is true (="q=-t") Explicatum: a has good reasons to believe that p. a believes that

p. p is true (="r.s.t") Explication: a knows that p = def. a has good reasons to believe

that p. a believes that p. p is true (="q=def. r.s.t") Criterion of adequacy after explication: if a has good reasons to believe that p, and a

believes that p, and p is true, then p is true (="r.s. t=>t")

This account is oversimplified insofar as the whole sentence "q", viz. "a knows that p" is taken to be the explicandum, though it would be more accurate to say the explicandum is the concept of knowing, as

TG AND PHILOSOPHY OF SCIENCE 419

it occurs in the sentence "q". It is good to point out that if we take the criterion of adequacy

as our starting point, the same explication could be presented as fol­lows:

Explication: q t = , - r.s.t t def.

We thus establish a definitional equivalence between the criteri­on of adequacy and its new, analyzed (or reconstructed or explicated) form, "q t" is merely intuitively known to be necessarily true, where­as "r.s.t z" is a formally necessary or analytical truth. On the oth­er hand, as I just noted, the equivalence between "q t" and "r.s.t => t" is not an analytical truth, but rather a definition which - like all good definitions - is intuitively acceptable.

When we check whether a given explicatum satisfies its criteria of adequacy, it could be said that in a sense, we are 'testing' the expli­cation. It is clear, however, that what we have here is not empirical testing in the positivist sense. Rather, we are examining the consis­tency and the exhaustiveness of our formalization. In other words, when we are 'testing' an explication, we are looking over the implica­tions of our formal description to find out whether or not we have suc­ceeded in formalizing all, and only, that which we intended to formal­ize.

The question as to which of several competing explicata is the best one, is decided by finding out which one satisfies most of those criteria of adequacy which have been generally accepted as relevant. If two explicata satisfy the same criteria of adequacy, that one is better which is formulated in a simpler and more general way.

In the light of the preceding discussion it ought to be clear by now why TG descriptions are to be considered as instances of explica­tion: explicanda represent atheovetical knowledge (Pap and Hanna use the terms 'preanalytic' and 'presystematic', respectively), whereas explicata represent theoretical knowledge; and the atheoretical know-

420 ESA ITKONEN

ledge which explications deal with is of the intuitive kind. The explicandum of a linguistic explication is the language to be

described. More precisely, we describe a language L by analyzing the concept 'correct sentence in L'. Therefore 'language L' is really an abbreviation for "all the rules which are relevant for the correctness of sentences in L", and, for example, the sentence "In English the def­inite article precedes the noun" should read "As far as the correctness of sentences in English is concerned, the definite article precedes the noun". Notice that in the expression "correct sentence in English" the constituents "correct sentence" and "English" are so-called incom­plete symbols: neither of them can be analyzed in itself, but they can only be analyzed as a whole.

The explicandum of a linguistic explication is identified with the aid of an indefinite set of rule-sentences and rule-instance sentences. These are intuitively known to be necessarily true and thus function as criteria of adequacy. Now if the explicandum is the language L, then it is only natural to think that the explicatum should be the grammar of L. In other words, a grammar is a theoretical definition of a language. (This sense of 'definition' must not be confused with the sense in which atheoretical rules are said to 'define', i.e., constitute, a language.) By now it is evident that linguistic explication consists essentially in replacing the atheoretical reference to L (as formulated in rule-sentences) by a theoretical, grammatical reference. But at this point I have to mention certain peculiarities of linguistic explication Unlike rule-sentences, viz. criteria of adequacy, a grammar which gen­erates all and only correct sentences does not speak about sentences and their properties, but it shows them. Consequently we could say that while the criteria of adequacy for a grammar are metalinguistic, the grammar itself constitutes, in a sense, an object-language. This object-language contains both different grammatical symbols and, as the terminal strings of derivations, sentences of L. This means that the object-language is used to make derivations which show (the struc-

TG AND PHILOSOPHY OF SCIENCE 421

ture of) the sentences of L.41 But notice that this "grammatical ob­ject-language" must be carefully distinguished from L as a natural ob­ject-language, i.e., a language which is used to make statements, give orders, ask questions, etc. Now, if TG descriptions are to be explica­tions, we must be able to reformulate our initial criteria of adequacy as formally necessary truths: our grammar provides the required formal necessity (see below), but in order to speak about those linguistic units spoken about in the initial criteria of adequacy and, hence, to attain appropriate explicated counterparts of the latter, we need, not only an ('object-linguistic') grammar, but also a metalinguistic (meta-) theory of grammar, which both defines the general notions of grammati­cal rule and derivation (viz. generation) and speaks about our partic­ular grammar as well as the sentences derived or derivable from it.

Now, as a result of our linguistic explication, we may establish a definitional equivalence between atheovetical metalinguistic criteria of adequacy of the type "In L all x's are y's" and me ta theoretical meta­linguistic sentences of the type "The grammar of L which consists of the rules R\ ... Rn generates all x's, and only x's which are y's". In other words, we replace the explicandum-expression) 'L' by the expli-catum(-expression) 'grammar of L', and at the same time we change the initial sentence-structure of the criteria of adequacy in the way il­lustrated above. (Hence, our explicatum is, technically speaking, a 'contextual definition' of the explicandum.) Now as it stands, the explicated form of the criterion of adequacy is not yet a formally nec­essary truth, but it can in principle be shown to be such a truth. Since generative grammars are formal, quasi-deductive systems consist­ing of the axiom symbol 'S' and a definite set of explicitly given rules of derivation, it is not a contingent, but a necessary fact that such and such a grammar generates such and such sentences with such and such structural descriptions.42- (On the other hand, it is of course a contingent consideration whether the sentences and structural descriptions generated by a grammar are or are not indeed the correct

422 ESA ITKONEN

ones.) Since the object-linguistic modus ponens sentence "a. (a=>b)=>b" where 'a' and 'b' are meant to be sentence-constants is analytically true, so is also the following metalinguistic sentence: "'b' is derived from 'a' and 'ab' by modus ponens." Now although TG's rules of deri­vation are sui generis, it is quite clear that within TG the following expression for example is a 'formally necessary truth' insofar as the conclusion can be reached by a purely formal step-by-step procedure: (NP → the + N, N → girl) → the girl.43 And, therefore, also the corresponding metalinguistic (and metagrammatical) sentence is a for­mally necessary truth: "the grammar consisting of the rules R1: NP → the + N' and 'R2: N → girl' generates the form 'the girl'". The same holds true of, e.g., the following pair of sentences: "(S → aSb, S → ab) → aabb" and "the grammar consisting of the rules 'R1: S → aSb' and 'R2: S → ab' generates the form 'aabb'".

The mediating link between the two types of necessarily true sen­tences, i.e., atheovetical metalinguistic criteria of adequacy and their explicated, metatheoretical metalinguistic forms, is provided by the grammar which is a theoretical 'object-linguistic' description. There­fore, it is, strictly speaking, wrong to identify the grammar simply with the explicatum, since here the explication replaces the explicandum in metalinguistic sentences and is, hence, the grammar as spoken about in the metalanguage of the (meta-)theory of grammar.

If all rule-sentences and rule-instance sentences, which are in­tuitively necessary sentences about the language L, can be transformed in this way into formally necessary sentences about the generative capacity of the grammar of L, and vice versa, this means that the gram­mar satisfies its criteria of adequacy, and is the best possible one. On the other hand, if there are some (atheoretical) sentences about L which cannot be transformed into corresponding metatheoretical sen­tences, this means that the grammar fails to generate something which it ought to generate. And similarly, if there are metatheoretical sen­tences about the forms generated, and the properties assigned, by the

TG AND PHILOSOPHY OF SCIENCE 423

grammar which have no counterparts among atheoretical sentences, this means that the grammar generates something which it ought not to gen­erate.

In the above discussion a prima facie difference between philosoph­ical and linguistic explication has become apparent. In philosophy the analyticity of criteria of adequacy into which the explicatum is embedded is the goal of explication; once this goal has been reached, the expli­cation has ended. In (TG) linguistics, however, the form of grammars as quasi-deductive systems guarantees from the outset the analyticity of sentences containing the explicatum, i.e., the grammar. Moreover, the explication itself consists in writing such a grammar which accounts for all criteria of adequacy identifying the explicandum, i.e., the language. In other words, one proceeds from the analyticity of sen­tences containing the explicatum to the similarity between explicandwm and explication^ whereas in philosophical explication the direction is in a sense the opposite one. In spite of this (apparent) difference be­tween philosophical and linguistic explication, it is clear we have to do in both cases with the same fundamental operation.

Since it is impossible to describe, viz. explicate, all of a lan­guage at the" same time, one has to start by describing for instance a sublanguage A, and then to move over to a sublanguage B, and so forth. But it is obvious that when we are describing the sublanguage A (con­stituted perhaps by 200 atheoretical rules), we cannot see directly all the implications which our description of A will entail for the description of (constituted perhaps by 300 partly same and partly different atheoretical rules). Therefore, writing a grammar means, first, that one tries to bring the explications of the different sub-areas of a language to the point where there are no inconsistencies between them and, second, that after discovering generalizations im­plicit in particular subareas covered by corresponding explications, one tries to discover generalizations implicit in these different explications. When one is writing or refining a grammar, one has to

424 ESA ITKONEN

consider it simultaneously from the viewpoints of particular general­izations (instantiating different degrees of abstractness) and general consistency. Because of the factual complexity of language, it seems clear, in any case, that no 'final' explicatum will ever be achieved in the description of any language.

On several occasions transformationalists have been forced to ad­mit, implicitly, the explicative nature of TG descriptions. Consider for example the following statement by Bach (1964:151):

That is, we know ahead of time in some sense what we want to come out with as a result of our analysis. This is one sense in which it can be said that linguistic analysis tries to account for the linguistic intuition of the native speaker.

By contrast, it would be absurd to say of a natural scientist (which Bach in his self-consciously methodological remarks claims to be) that when he undertakes his investigation, he 'in some sense' al­ready knows its results, and that he merely wants to clarify or expli­cate this knowledge.

In an early passage Chomsky (1957:13) in fact admits that TG de­scriptions are explications:

... we assume intuitive knowledge of the grammatical sentences of English and ask what sort of grammar will be able to do the job of producing these in some effective and illuminating way. We thus face a familiar task of explication of some intuitive concept -in this case, the concept "grammatical in English", and more gen­erally, the concept "grammatical".

But since the difference between empirical hypothesis and expli-eatum is irreducible, Chomsky is here contradicting his own standard position.

Although explication is fundamentally different from empirical explanation, it is rather similar to axiomatization within natural sciences. In the case of axiomatization, too, we do not try to expand our knowledge through the testing of empirical hypotheses; rather, the universal hypotheses entailed by the theory to be axiomatized are as-

TG AND PHILOSOPHY OF SCIENCE 425

sumed to be true, and the logical relations between them are explicitly stated by deducing them from a few axioms. Axiornatization is thus sys­tematization of existing knowledge, and here as elsewhere systematisa­tion produces new, theoretical knowledge. On the other hand, in the case of linguistic explication the rule-sentences functioning as cri­teria of adequacy are known to be true, and they are 'deduced' from the grammar in the sense that it demonstrably generates all and only forms conforming to the atheoretical rules referred to by them.

Consequently a (generative) grammar may be considered as an axiom-atization within all other non-formal sciences, it does not speak about its subject matter within the limits of a deductive system, but rather shows it in the form of such a system. That is to say, the axi-omatization of biology for instance, viz. biological knowledge, does not consist in manipulating cells and genes, whereas a grammar as an axiomatization of linguistic knowledge consists, ultimately, in mani­pulating sentences and their parts. Consequently a grammar could be called an 'iconic' axiomatization. Yet it is also true, that axioma­tization of the grammar, in the sense of its metatheory, would be more directly comparable to axiomatizations within other sciences. It may be mentioned that Hjelmslev for one employs the term 'deduction' in such a manner that it applies to both language and linguistics (cf. Itkonen 1968:457-58).

Finally, it might seem that even if writing the grammar of one particular language amounts to an explication of the relevant intuitive knowledge, the resulting grammar is bound to contain empirical hypo­theses in the sense that it is an empirical question whether the prin­ciples of analysis applying to the description of the language in ques­tion also apply to the description of other languages. Hence, construct­ing a universal linguistic theory would be an empirical task in the positivist sense. However, this argument does not show that (TG) lin­guistics and natural sciences have the same logical structure. In oth­er words, it is a contingent fact that there are many languages in the

426 ESA ITKONEN

world that we still do not know. (On the other hand, notice that even if there were one single natural regularity in the world, we could nev­er come to know it in the way we can come to know the rules of any lan­guages; cf. sect. 3 above.) But when all the languages of the world are known, and when this knowledge has been written down in grammars pos­sessing varying degrees of generality, the (explicative) task facing the linguist corresponds, methodologically, to that which is facing him when he sets out to describe his native language or any group of lan­guages (intuitively) known to him.

8.0. As analyses of linguistic intuition, i.e., linguistic norm-conscience, TG descriptions have an unmistakable hermeneutic character. On the other hand, they are also closely similar to explications as practised within analytic philosophy. The question then arises: Can philosophical explications, too, be considered as hermeneutic descrip­tions? As especially Pap 1962 has emphasized, philosophy and logic can only start with an intuitive understanding of relations of necessity. Intuition cannot be banished altogether from logic, and Carnapian at­tempts to formalize all logic are doomed to failure from the outset. But it is also clear that after we have formalized our initial intu­itions, our formalization enables us to better understand them, which again leads to a modification of our formalization; and so on. It is easy to see that this method of 'successive redefinition' is practi­cally identical with the so-called hermeneutic circle (or spiral), which consists in a dialectical movement from an object of knowledge to its description, and back again.

On the other hand, Pap does not even raise the question as to the basis of this philosophical or logical intuition. We have seen that, in the case of linguistics, intuition is knowledge about our own ac­tivity, i.e., speaking, which follows certain rules. Now it seems to me that the situation is the same in philosophy. When we explicate intuitively known concepts like 'explanation', 'number', and 'infer­ence', it would be quite natural to think that our intuition is based

TG AND PHILOSOPHY OF SCIENCE 427

on and pertains to the activities of explaining, calculating, and in­ferring. These activities, which we perform in our every-day life, are certainly not accidental or arbitrary but follow definite (sets of) rules, which could even be said to constitute corresponding institutions, i.e., 'institutions' of explaining, calculating, and inferring. These rules are not merely rules of our language, but rather rules of our to­tal intelligent behavior of explaining, etc. On the other hand it is also clear that, as Wittgenstein has pointed out, rules of a 'language-game' are inextricably interwoven with rules of a 'form of life'; there is no way of separating the knowledge pertaining to the meaning of a word "A" referring to an action-type A from the knowledge pertaining to A. But one knows neither A nor the meaning of "A" well until one has learned to perform A. Thus the conclusions which I have drawn from the consideration of the nature of TG descriptions clearly support Winch's (1958:100) claim that "criteria of logic are not a direct gift of God, but arise out of, and are only intelligible in the context of, ways of living or modes of social life".

Our 'language-games1 of logic and mathematics (and of ordinary lan­guage, of course) are rooted in'our forms of life. And these in turn are (contingent) facts about man's 'natural history'. As Wittgenstein dem­onstrates by giving examples of imaginary tribes, it is possible that there could be other ways of living; hence, thinking than ours, but we cannot understand them; therefore, we tend to think that our logic is the only possible one.44

Within the philosophy of logic there is a trend called 'conven­tionalism' which maintains that necessarily true sentences are neces­sarily true by virtue of (empirically discoverable) rules of natural languages; this conception has recently been defended by Giannoni 1971. Although conventionalism is right in rejecting the myth of eternal truths of logic, it fails - as far as I can see - on the following counts: First, it is simply wrong to base logic exclusively on lan­guage, as can be seen from the investigations which Piaget and Furth

428 ESA ITKONEN

have made on the logical thinking in preverbal children and in the deaf. Secondly, for reasons indicated earlier, it is wrong to assume that the relevant rules (whose normativity is admitted) must be empirically dis­covered. And thirdly, there remains the question why, as far as neces­sary truth is concerned, different languages have the same rules. Gian-noni certainly does not answer this question when he stipulates that all natural languages are to be treated as one single (object-) language, with equivalence relations holding between the 'same' sentences taken from different languages (see Giannoni 1971:112-17). Any genuine answer to this question must make reference to the fact that everywhere people act (although they do not speak) in the same way because of a common natural history, i.e., common psychological constitution (with a pre­disposition to social life) and common physical environment. But this does not mean that we could literally explain our way of thinking by deriving it from our natural history, because in order to do this, we would have to transcend our own thinking (cf. the criticism of Katz and Chomsky below).

At this point it might be objected that my notion of explication could not be rightfully used to show the common, ultimately hermeneutic, nature of such disciplines as Winch's (1958) aprioristic sociology, Lo­renzen's (1969) protophysics, TG, and logic, because explication is al­so utilized by natural scientists. This objection results from a mis­understanding, however. In nature there are no necessary relations but only empirical relations. (At least this is assumed by the positivist philosophy of natural sciences.) Therefore, when a natural scientist makes use of necessary relations in selecting criteria of adequacy for his explications, he is in fact explicating concepts which, as typical­ly human products, are either invented or at least used by him. In oth­er words, far from invalidating my account of explication, the present 'objection' shows that the notion of explication is able to do justice to the fact that, though natural sciences deal with non-human nature, they are nevertheless made by man, and are thus accessible to a her-

TG AND PHILOSOPHY OF SCIENCE 429

meneutic description.45 Science is a social enterprise which follows its own rules, as has been noted by Peirce, Winch, Apel, and others.47

The preceding discussion of the relation between linguistics and logic may be profitably compared to Katz's frequent claims - echoed by Bierwisch for instance - that TG ought to provide the foundation for philosophy and logic (Katz 1966 passim). Now since Katz as an orthodox transformationalist takes it for granted that TG is a paradigmatic nat­ural science, it follows that, in his view, philosophy and logic should be natural sciences of some derivative type. But this does not make sense. In the same vein, Chomsky (1968 passim) has argued that TG em­pirically determines, or makes a contribution toward so determining, the limits of man's conceptual space, i.e., of what man can or cannot think; and such an achievement would supposedly be of the utmost im­portance for philosophers and logicians. But it is a glaring contradic­tion to assume that we could empirically discover what it is that we cannot think (and to assume, perhaps, that we could test over and over again, whether we really cannot think "it'; but what?). In their pos­itivist zeal transformationalists have failed to understand that a particular science, e.g., TG linguistics (whether or not it is falsely interpreted as a natural science) cannot answer genuinely philosophi­cal questions.47

9.0. Ferdinand de Saussure said, more than three generations ago, that language is a social institution. To be sure, it is a spe­cial type of institution which presupposes the possibility of communi­cation and language as the principal medium through which it comes into being. Hence it might be more adequate to call language a 'meta-institution' (cf. Radnitzky 1970 II, p.6), which label also points to the instrumental nature of language: in order to communicate within a given institution, i.e., a (partial) form of life, we need a language with which we communicate such and such things. But this way of commu­nicating, or language in itself, could not be a form of life.

Like any other institution, language only exists through the ac-

430 ESA ITKONEN

tions instantiating the rules which constitute it. By participating in this 'rule-governed activity', people acquire an internal, intuitive knowledge of language. TG must, of necessity, start at this level of linguistic Vorverständnis, and it attempts to formalize or explicate it in terms of theoretical rules incorporated into a sentence-generat­ing device. Notice that so-called understanding 'social sciences' (= verstehende Soziologie), which have an intimate connection with tradi­tional hermeneutic descriptions, concentrate precisely on the inter­pretation of different institutions (and also of non-institutional ac­tions) .

When we are describing actual human behavior which is not, or not only, rule-governed, we may have to resort to the use of causal, i.e., positivist, explanations. Furthermore, in dealing with social macro-phenomena, statistical explanations are unavoidable, though the sta­tistical point of view is not congenial to the hermeneutic approach. These facts have given rise to the thesis of the mediation between, or the complementarity of, the hermeneutic and the positivist approach within human sciences. It has even been argued that "[s]ince man is not fully transparent to himself, the mediation thesis applies to all human sciences - as they exist today" (Radnitzky 1970 II, p.68). 48 But now it seems that, far from being a paradigmatically positivist science, TG is an exceptionally pure hermeneutic science insofar as it contains no positivist elements whatever: First, TG a priori excludes all sta­tistical considerations. And second, instead of trying to answer the question, why such and such an observable linguistic event happened, TG concentrates upon the entirely different question: What is a pos­

sible correct sentence? When one describes an institution, i.e., a conceptual system, man­

ifested through social behavior, one's integral subject-matter comprises three different, logically interdependent components: (a) conceptual system (or rule system) itself, (b) intention of the agent, (c) action. Our concepts logically determine what we can intend to do (cf. Winch

TG AND PHILOSOPHY OF SCIENCE 431

1958:51-57, for example). Furthermore, the relation between the in­tended goal and the knowledge of a possible action capable of serving as a means to this end is not empirical but conceptual, since it is a part of the concept of intention. On the other hand, whether a chosen means will as a matter of fact lead to the intended goal, is of course an empirical question. The intention 'explains' (in the hermeneutic sense) the action, whereas the latter 'confirms' the occurrence of the former. The model through which particular (intentional) actions are 'explained' is called 'practical syllogism'.49 Chomsky and others have amply demonstrated that the occurrence of particular utterances cannot be causally explained through a simple stimulus - response mech­anism. But they have been utterly unable to specify the more complex causal factors supposedly determining when and what people speak. And although TG has not even tried to offer a positivist explanation of the occurrence of utterances, it a priori rules out the possibility of other types of explanation. However, since making an utterance is ob -viously not just an acoustical or physiological event, but also - or rather - an (intentional) action, a practical syllogism of the fol­lowing type offers a rather natural explanation for it:50

presupposition: "p" means p A intends] that knows p. A intends2 that recognizes that A intends] that knows p. A considers that he will not achieve his intentions] and 2, unless he utters "p".

Therefore A sets himself to utter "p".

Taken as a whole, this practical syllogism describes the primary, pragmatic dimension of language (cf. section 2): a rule-governed lin­guistic action is made which consists of A uttering "p" in order to get to know p by virtue of a correspondence between "p" and p. This dimension is given to us as such,51 and from it we secondarily ab­stract the semantic rule which says that in the standard case whenever "p" is uttered, the state of affairs p holds. After this we may pro-

432 ESA ITKONEN

ceed to distinguish, within the semantic rule, between syntax, i.e., "p", and semantics (or meaning), i.e., the capacity of "p" to refer to p. We may then analyze p and correspondingly distinguish between different elements of "p" capable of referring to different elements of p: this amounts to the analysis of the meaning of "p". Finally, "p" itself is subjected to syntactical, morphological, and phonological analysis; the first two types of analysis are practically inseparable from the aforementioned analysis of the meaning of "p".

It is a peculiarity of the acts of communication that they fail to be what they are meant to be, unless they are understood (by the hearer) as what they are, in other words, unless the intentions behind them are recognized. (On the other hand, in order to be what it is meant to be, the act of opening a window, for example, need not be un­derstood.) This idea, which is reflected in 'intention2' above, ties up with Schutz's 'mirror-effect1: Taking the role of the speaker entails taking (or anticipating) the role of the hearer, which in turn entails taking the role of the speaker, etc. In fact, one could argue (as Straw-son for instance has done) that A could also intend3 to recognize the intentions1 and 2 (for details, see Itkonen 1972b).

The above model also draws a new line between competence and per­formance, since it establishes a distinction between a (potentially) conscious competence, which is always present in intentional behav­ior, and performance, given that the notion of practical syllogism has been expressly devised to account for the unique combination of concep­tual and empirical elements, something which is characteristic of in­tentional behavior, namely while the relations between the conceptual system of a community, the intention of an agent, and the decided-on action are conceptual, the concrete action itself is subject to inter­ferences due to the physical environment and/or human limitations. This is not to deny that the notion of an unconscious, ultimately physiol­ogical, competence would, intev a l i a , be required for a c a u s a l , i.e., positivist, explanation of the occurrence of utterances. Unfortunately

TG AND PHILOSOPHY OF SCIENCE 433

(for TG, that is) such explanations have never been offered and are not 1ikely to be offered.

The practical syllogism clearly differs from the D-N model in that it contains no universal (empirical) hypotheses. Rather, its particular premises suffice logically to imply the conclusion. The reason is not far to seek. If a conceptual system determines that the concept A log­ically implies the concept (as, e.g., the concept of an intended goal implies the concept of some supposedly necessary means), it is clear that the occurrence of an instance of A logically implies the occur­rence of an instance of B. Therefore, it would be pointless to try to 'support' this conclusion by pointing out that it holds for every in­stance of A that its occurrence logically implies the occurrence of an instance of B: this is just another consequence of the fact that A logically imp!ies .

A reference to regularities, i.e., to all instances of, in our example, A and B, is the defining property of genuine D-N explanations. It was briefly suggested above why it is futile to attempt, in the pos­itivist vein, to reduce the explanation of intentional behavior to a D-N explanation.52 Notice also that when we describe or 'explain' (in a non-positivist sense) rule-governed intentional behavior (which is a special case of intentional behavior), we do not make a reference simply to all actions, but to all correct actions.

10. CONCLUSION

We have seen that, at the methodological level, Transformational Grammar has consistently refused even to consider the contributions of all those thinkers who have viewed language as a specifically human phenomenon. As far as I can see, this fact can only be explained by a reference to the nearly absolute hegemony of the positivist philosophy of science in the United States. Since positivist philosophers of sci­ence attempt to dictate the criteria of all empirical sciences, without making a distinction between natural sciences and human sciences, they have presented linguists with a fictitious choice: either to accept

434 ESA ITKONEN

that linguistics is a positivist science, or to admit that it is not an empirical science. It is understandable that most linguists - not only transformationalists! - have opted for the former alternative, but it is regrettable that they have not rejected these conditions of option in the first place.

NOTES

1 Nagel 1961 and Hempel 1965a are standard expositions of neopositivism. Wolfgang Stegmüller may be mentioned as leading European neopositivist.

2 A perfect symmetry between explanation and prediction, as initially postulated by Hempel and Oppenheim in 1948 (see Hempel 1965c), has proved to be an oversimplification; cf., e.g., Scheffler 1963:43-57. This fact is also acknowledged in Hempel 1965d. For the purposes of my exposition, however, the structural differences between explana­tion und prediction are irrelevant.

3 This model is also called the 'Hempe1-Oppenheim model' of explana­tion.

4 Both "f(x)" and "g(x)" can be taken as standing for a conjunction of predicates. The same example has been used by Hempel as an instance of explanation, "which surely is intuitively unobjectionable" (Hempel 1965c:275).

5 Cf. Chomsky (1957:16-17): "Evidently, one's ability to produce and recognize grammatical utterances is not based on notions statistical approximation and like. ... We see, however, that this idea is quite incorrect, and that a structural analysis cannot be understood as a schematic summary developed by sharpening the blurred edges in the full statistical picture" (italics mine). This feature already points to the de facto non-positivist nature of TG. It would not make sense to say of a physicist for instance that he 'sees' that events con­sisting in the movements of gas molecules are 'evidently' subsumable under statistical laws, but not under universal laws.

6 My notions of 'positivism' and 'hermeneutics' correspond exactly to the two 'contemporary schools of metascience' discussed by Radnitzky 1970. As far as I know, Radnitzky's book is the most comprehensive treatment of the controversy between these two philosophies of science. The same topic is discussed, in a more restricted context, also in von Wright 1971. - I have developed my thesis that TG is not a posi­tivist (or natural), but a hermeneutic (or human) science in a number of papers, including Itkonen 1969, 1970a, and 1972a. A much more ex­tensive account is given in Itkonen 1974, which should be consulted on all those questions which I have been unable to consider here.

7 In what follows, I rely mainly on Saunders and Henze's (1967) account; for more details, see Itkonen 1970b and, here as elsewhere, 1974.

436 ESA ITKONEN

Cf. for instance the following statement by Descartes :"...it is I who have sensations, or who perceive corporeal objects as it were by the sense. Thus, I am now seeing light, hearing a noise, feeling heat. These objects are unreal, for I am asleep; but at least I seem to see, to hear, to be warmed. This cannot be unreal; and this is what is properly called my sensation." Meditations II, p. 71).

9 Taylor 1964 makes a similar point when he notes that if the theoret­ical concepts of psychology are, in the standard neopositivi st fashion, 'partially interpreted* in terms of their manifestations, then the relation of the former to the latter cannot be empirical. This observation also undermines Chihara and Fodor's (1965) attempt to disprove Wittgenstein's 'logical behaviorism1.

10 The same point is also made in Strawson, p. 187. 11 The reason for this qualification will be seen later on. 12 Cf. for example the following passage: "Der wahre Vorzug einer Spra­che ist nur der, sich aus einem Princip und in einer Freiheit zu entwickeln, die es ihr möglich machen, alle intellectuelle Vermögen des Menschen in reger Thätigkeit zu erhalten, ihnen zum genügenden Organ zu dienen, und durch die sinnliche Fülle und geistige Gesetz­mässigkeit, welche sie bewahrt, ewig anregend auf sie einzuwirken. In dieser formalen Beschaffenheit liegt Alles, was sich wohlthätig für den Geist aus der Sprache entwickeln lässt." (Humboldt 1836:206; italics in the text.)

13 See Hjelmslev 1961:39-40, and for comments, Itkonen 1968:456. 14 To speak entails, in Mead's terminology, "taking the role of the hearer"; but it is part of the role of the hearer to anticipate the role of the speaker; and so on. This 'mirror-effect' is also one of the cornerstones of Schutz's 'phenomenological sociology'; see Schutz 1967:11-13 and 31-34.

15 Cf. also Strawson 1971, Habermas 1971, and Apel 1972b. All these authors distinguish between a formal and a functional, viz. prag­matic, conception of language. The former is represented by the early Wittgenstein and Chomsky, for instance while the latter is represented by Peirce, Mead, the late Wittgenstein, Austin, Grice, Searle, and others. My conception of language has been influenced most by the late Wittgenstein; as for the relevance of the Peircean approach, see Anttila (in this volume).

16 Cf. Hintikka (1969:16): "There is not a shred of reason, ... why the general structures exhibited by language in use could not also be studied by logical and mathematical means." And the language so studied is used "not just for its own sake but to some purpose".

17 For a philosophical discussion of norm and rule, see von Wright 1963.

TG AND PHILOSOPHY OF SCIENCE 437 1 8 Cf. Miller and Isard 1963: "Rules of practice are ... logically prior

to any particular behavior specified by the practice." The only re­markable thing about this obviously true remark is that it has been made by transformationally-minded psycholinguists, while TG usually pays no attention to the social aspect of rules (see below).

19 The reason is simple, of course: What one does has no direct rela­tion (although it certainly has some relation) to what one ought to do.

2 0 E.g., the sentence "The definite article precedes the noun" would be true in English, false in Swedish, and meaningless in Finnish.

2 1 This account is of course oversimplified in the extreme. For a more comprehensive account, cf., Pap 1958, where the same distinctions are discussed in over 400 pages. For the sake of completeness, it may be added that representatives of 'analytic philosophy' generally deny the existence of synthetic a priori truths as a special class. However, Pap argues convincingly against this position; cf. also note 25.

2 2 English: a girl - the girl; Swedish: en flicka - flickan. 2 3 Here I am only speaking of the time before and after the change. I

thus disregard the fact that during the change there will inevi­table be uncertainty about the correct form.

24 From this it is clear that I do not: accept TG's manner of defining a language as an (infinite) set. of sentences', cf. pp. 407-08 and 429-30 in the present paper.

2 5 However, our synthetic a priori knowledge may also be interpreted, in a "deeper" way, as a knowledge pertaining to those properties of things which we ourselves create or construct in the course of our intellectual activity; cf. Hintikka 1965 and 1967. On this inter­pretation, synthetic a priori knowledge is a case of the "agent's knowledge"; cf. here p. 12 and 22. In fact, it seems to me that there is a certain affinity between Hintikka's theory of the syn­thetic a priori and the notions of intuition and intuitive neces­sity which I have been outlining here; cf. especially section 8.

2 6 Cf. Taylor 1964:55-57 and von Wright 1971:110-31. From the irreduc-ibility of intentional to physical Quine draws the surprising con­clusion that the former must simply be ignored; Quine 1960:216-21. Schutz's following remark exposes the limits of any variants of physicalism: "Even an ideally refined behaviorism can ... merely explain the behavior of the observed, not of the observing behav­iorist" (Schutz 1961:54).

2 7 "But in the night of thick darkness enveloping the earliest antiq­uity, so remote from ourselves, there shines the eternal and never failing light of a truth beyond all. questions: that the world of civil society has certainly been made by men and that its principles

438 ESA ITKONEN

are therefore to be found within the modifications of our own human mind. Whoever reflects on this cannot but marvel that the philoso­phers should have bent all their energies to the study of the world of nature, which, since God made it, He alone knows; and that they should have neglected the study of the world of nations, or civil world, which, since men made it, man could come to know" (Vico 1961 [1744]:§331). In fact, this line of thought goes back to Greek phil­osophy; cf. Hintikka 1965.

28 However, it is possible that the influence of formal languages has been here at least as decisive as that of positivism as a general methodology of science.

29 The 'atheoretical - theoretical' distinction corresponds to Winch's (1958:89) distinction 'unreflective - reflective1. As Mehtonen 1971 has pointed out, the two-level account of human (including political) sciences has the far-reaching implication that practice determines the nature of theory. - Although the distinction between atheoreti-cal and theoretical is a necessary part of every human science, it is obvious that this distinction is not - and need not be - an ab­solute one.

30 This is TG's position on this ontological question, cf. Katz 1964 and Botha 1968:94-106.

31 This discussion is obviously related to the dichotomy between what Householder termed 'God's truth' vs. 'hocus-pocus' views of language. i.e., the views that the linguist either discovers the pre-existent structure of language or invents the structure and projects it into language. My own position lies somewhere between these two extremes. The linguist is not dealing with 'language' i abstracto, but with the (linguistic) consciousness of ordinary people (cf. the Schutz-citation). Therefore, where people think that there is structure, there is structure; in other words, the existence of that type of structure which is reflected in atheoretical rules cannot be doubted. On the other hand, as I have indicated above, the ontological status of those structures which are expressed in the theoretical rules constructed by the linguist is more unclear. Notice that the analogy with natural sciences is not very helpful, because regularities are not known but only assumed, and therefore both atheoretical and theoretical hypotheses are tested experimentally.

32 In any case, Gumb's attempt to reconcile philosophy of action and linguistics is laudable, and he correctly criticizes TG for its ne­glect of the social-normative aspect of rules. The internal - exter­nal distinction has also been endorsed, e. g., by Miller and Isard 1963, who, predictably, decide for the latter (fictitious) alternative.

3;3 To be sure, Chomsky is not speaking about rules, but of particular forms, but in this he is clearly mistaken, since it would be absurd to claim that no atheoretical rules are or can be known. This is a

TG AND PHILOSOPHY OF SCIENCE 439 question of principle: because no regularity can be known (cf.p.ll), it is indifferent whether 10 or 10,000 (atheoretical) rules can be known; and above I have given examples of rules which certainly can be known. As a positivist, Chomsky must have argued in the following way: Like any natural science, TG deals with regularities, not with rules; but regulatities cannot be known; therefore linguistic intu­ition can only pertain to particular linguistic forms.

3 4 The sentence and the TG rule describing it are taken from Jacobs and Rosenbaum 1968:124-27. The factual correctness of the rule is here irrelevant.

3 5 It is probable that even this sentence contains terms which are un­familiar to many English speakers, but the important thing is that the concepts expressed by these terms are not unfamiliar, i.e., the same concepts could be expressed by more down-to-earth circumlocu­tions comprehensible to everybody.

36 This point is related to the previous one. 3 7 In Itkonen 1974 I also illustrate my claims with the aid of more com­

plex artificial languages and representative examples of TG descrip­tions .

3 8 In effect, transformationalists often seem guilty of having committed this fallacy: they postulate a psychological mechanism on the basis of their linguistic descriptions, and they prove its existence on the basis of the same descriptions. For similar criticism, see Botha 1971, esp. pp. 169-70.

3 9 Apel 1972b offers a reinterpretation of TG which is in several re­spects similar to the one which I have been developing. However, his account is somewhat vitiated by his inability clearly to distinguish between the linguistic and the psychological aspects of TG.

1+0 A shorter, basically similar, account of explication is given in Hanna 1968. Cf. pp. 387-88 above, where it was shown that even formal languages, qua languages, must be used,

41 Cf., e.g., Pap 1962:104: "A metastatement to the effect that such and such a formula is a theorem in the system that is characterized by such and such postulates and such and such rules of deduction... is, if true, necessarily true.u

4 3 This means that, mutatis mutandis, the following 'law of tautological derivability1 (or 'theorem of deduction') also holds within TG: When a formula "q" is derivable form a set "p" of definite formulas (here: the axiom symbol 'S'), the formula "pq" (here: "S → ... —→- q") is a tautology.

1+4 Stroud 1965 provides an excellent short introduction to Wittgen­stein's philosophy of logic and mathematics.

440 ESA ITKONEN 4 5 Cf. Wittgenstein 1958, I 570: "Concepts lead us to make investiga­

tions; are the expression of our interest, and direct our interest." More recently, similar ideas concerning the role of concepts in science have been expressed by Feyerabend 1968 and Kuhn 1970, but they are not aware of the specifically hermeneutic dimension. Cf. also here note 25.

4 6 Combining ideas from Wittgenstein's philosophy and from the tradi­tional hermeneutics, Apel 1972a has shown that a community of com­munication and interpretation is an a priori presupposition of nat­ural sciences (not to speak of human sciences).

^7 With his above-mentioned descriptions of the life of imaginary tribes Wittgenstein has in fact given us hints about what we cannot think. But then he is not a natural scientist, as he himself has clearly stated: "If the formation of concepts can be explained by facts of nature, should we not be interested, not in grammar, but rather in that in nature which is the basis of grammar? - Our interest cer­tainly includes the correspondence between concepts and very gener­al facts of nature. (Such facts as mostly do not strike us because of their generality.) But our interest does not fall back upon these possible causes of the formation of concepts; we are not doing nat­ural science; nor yet natural history - since we can also invent fictitious natural history for our purposes; ..." (Wittgenstein 1958 II, p. xii). It is a pity that this is one of those not-so-few pas­sages which have somehow escaped the transformationalists' atten­tion; see also Apel 1972b.

4 8 This claim is clearly untenable, because what it says is, in fact, that each time when there is something that we de not immediately understand, we have met with positivist (or 'naturalistic') elements. But this view would turn, e.g., philosophy into a natural science and make all hermeneutic sciences impossible by definition.

4 9 This model is carefully defined and illustrated in von Wright 1971, chap.III. It is a subtype of the general teleological explanation; cf. Taylor 1964, esp. pp.54-62.

5 0 For a justification of this particular model, see Itkonen 1972b. Although Grice, Strawson, and Searle are all concerned with speech acts, they seem to have overlooked the possibility of describing and explaining them with the aid of practical syllogisms. - As it stands, the present syllogism applies to indicative sentences only. It may also be mentioned that the formula "'p' means p" is unsatis­factory on logical grounds, but this fact has no direct relevance here.

51 Pragmatic rules of language have been outlined Wunderlich 1972 and, earlier, Searle 1969.

52 Such attempts are made in Hempel 1965d and in many other contexts.

REFERENCES

Apel, Karl-Otto. 1972a. "Communication and the Foundations of Humani­ties". Acta Sociologica 15.7-26.

. 1972b. "Noam Chomskys Sprachtheorie und die Philosophie der Gegenwart". Neue Grammatiktheorien und ihre Anwendung auf das heuti­ge Deutsch (= Jahrbuch 1971 des Instituts für deutsche Sprache; Spra­che der Gegenwart, ed. by Hugo Moser, et al., 20), 9-54. Düsseldorf: Schwann.

Bach, Emmon. 1964. An Introduction to Transformational Grammars. New York: Holt, Rinehart & Winston. (2nd entirely recast ed., 1974.)

Botha, Rudolph P. 1968. The Function of the Lexicon in Transformational Generative Grammar. The Hague: Mouton.

. 1971. Methodological Aspects of Transformational Generative Phonology. Ibid.

Cavell, Stanley. 1958. "Must we Mean what we Say?". Inquiry 1.172-212. Chihara, Charles, and Jerry A. Fodor. 1965. "Operationalism and Ordinary

Language: A critique of Wittgenstein". American Philosophical Quarter­ly 2.281-95.

Chomsky, Noam. 1955. The Logical Structure of Linguistic Theory. Cam­bridge, Mass.: MIT, mimeo. (Printed, New York: Plenum Press, 1975.)

. 1957. Syntactic Structures. The Hague: Mouton. (10th print­ing, 1972.)

. 1961. "Some Methodological Remarks on Generative Grammar". Word 17.219-39.

. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.

. 1966a.Cartesian Linguistics: A chapter in the history of linguistic thought. New York & London: Harper & Row.

. 1966b. Topics in the Theory of Generative Grammar. The Hague: Mouton. (3rd printing, 1972.)

. 1968. Language and Mind. New York: Harcourt, Brace & World. (2nd enl. ed., 1972.)

Chomsky, Noam, and Morris Halle. 1968. The Sound Pattern of English. New York & London: Harper & Row.

Chomsky, Noam, and George A. Miller. 1963. "Introduction to the Formal

442 ESA ITKONEN

Analysis of Natural Languages". Handbook of Mathematical Psychology ed. by Robert Duncan Luce, Robert R. Bush, and Eugene Galanter, vol. 2.269-321. New York: J. Wiley & Sons.

Descartes, René. 1954. Philosophical Writings. Transl, into E. by G. Elizabeth M. Anscombe and Peter Thomas Geach. Edinburgh: Nelson.

Feyerabend, Paul K. 1968. "How to Be a Good Empiricist: A plea for tol­erance in matters epistemological". The Philosophy of Science ed. by Peter Harold Nidditch, 12-39. London: Oxford Univ. Press.

. 1970. "Consolations for the Specialist". Criticism and the Growth of Knowledge ed. by Imre Lakatos and Alan Musgrave, 197-230. Cambridge: Cambridge Univ. Press.

Grice, Geoffrey Russell. 1957. "Meaning". Philosophical Review 66.377-to 388.

. 1968. "Utterer's Meaning, Sentence-Meaning, and Word-Mean­ing". Foundations of Language 4.225-42.

Gumb, Raymond D. 1972. Rule-Governed Linguistic Behavior, The Hague: Mouton.

Habermas, Jürgen. 1971. "Vorbereitende Bemerkungen zu einer Theorie der kommunikativen Kompetenz". Theorie der Gesellschaft oder Sozialtech­nologie: Was leistet die Systemforschung? by Jürgen Habermas and Nik­las Luhmann, 101-41. Frankfurt/Main: Suhrkamp.

Hanna, Joseph F. 1968. "An Explication of 'Explication111. Philosophy of Science 35.28-44.

Hare, R. M. 1957. "Are Discoveries about the Use of Words Empirical?". The Journal of Philosophy 54.741-50.

Hempel, Carl G. 1965a. Aspects of Scientific Explanation and Other Es­says in the Philosophy of Science. New York: The Free Press.

. 1965b. "Studies in the Logic of Confirmation". In Hempel 1965a:3-51.

. 1965c. "Studies in the Logic of Explanation". Ibid., 245-to 295.

. 1965d. "Aspects of Scientific Explanation". Ibid., 331-493. Hintikka, Jaakko. 1965. "Kant's 'New Method of Thought' and his Theory

of Mathematics". Ajatus: Yearbook of the Philosophical Society of Finland 27.37-47.

__. 1967. "Kant on the Mathematical Method". The Monist 51.352-to 375.

. 1969. "Epistemic Logic and the Methods of Philosophical Analysis". Models for Modalities by J. Hintikka, 3-19. Dordrecht/ Holland: D. Reidel.

TG AND PHILOSOPHY OF SCIENCE 443

Hjelmslev, Louis. 1961. Prolegomena to a Theory of Language. Transl. from the Danish by Francis J. Whitfield. 2nd rev. ed. Madison, Wis.: Univ. of Wisconsin Press.

Humboldt, Wilhelm von. 1836. ¡Jeher die Verschiedenheit des menschlichen Sprachbaues und ihren Ein fluss auf die geistige Entwickelung des Men­schengeschlechts. Berlin: F. Dümmler. (Facs.-repr., Bonn: F. Dümmler, 1960, 1968, and 1972.)

Itkonen, Esa. 1968. "Zur Charakterisierung der Glossematik". Neuphilolo­gische Mitteilungen 69.452-72.

. 1969. Semantics as an Analysis of Language-independent Re­ality. Cambridge, Mass.: MIT, unpublished MS.

. 1970a. "Zwei verschiedene Versionen der Bedeutungskomponen­te". Linguistics 59.5-13.

. 1970b. "An Epistemological Approach to Linguistic Semantics". Ajatus: Yearbook of the Philosophical Society of Finland 32.96-142.

. 1972a. "Concerning the Methodological Status of Linguistic Descriptions". Derivational Processes (= KVAL PM, Ref. No. 729.), ed. by Ferenc Kiefer, 31-41. Stockholm.

. 1972b. "On Grice's, Strawson's, and Searle's Concept of Mean­ing". Ajatus 34.149-54.

. 1974. Linguistics and Metascience. (= Studia Philosophica Turkensia, 2.) Kokemäki: Risteen Kirjapaino.

Jacobs, Roderick A., and Peter S. Rosenbaum. 1968. English Transforma­tional Grammar. Waltham, Mass.: Blaisdell.

Kanngiesser, Siegfried. 1972. Aspekte der synchronen und diachronen Lin­guistik. Tübingen: M. Niemeyer.

Katz, Jerrold J. 1964. "Mentalism in Linguistics". Language 40:1.124-37. . 1966. The Philosophy of Language. New York & London: Harper

& Row. Kuhn, Thomas S. 1970. The Structure of Scientific Revolutions. 2nd enl.

ed. Chicago: Univ. of Chicago Press. (1st ed., 1962.) Lakoff, George. 1971. "Presupposition and Relative Well-Formedness".

Semantics : An interdisciplinary reader in philosophy, l i n g u i s t i c s , and psychology ed. by Danny D. Steinberg and Leon A. Jakobovits, 329-40. New York & Cambridge: Cambridge Univ. Press.

Lorenzen, Paul. 1969. "Wie ist die Objektivität der Physik möglich?". Methodisches Denken by P. Lorenzen, 142-51. Frankfurt/Main: Suhrkamp.

McNeill, David. 1970. The Acquisition of Language: The study of develop­mental psycho Unguis tics. New York: Harper & Row.

444 ESA ITKONEN

Mead, George Herbert. 1934. Mind, Self, and Society. Chicago: Univ. of Chicago Press.

Mehtonen, Lauri. 1971. "Some Reflections on Theory and Practice in Pe­ter Winch's 'Idea of Science'". Ajatus: Yearbook of the Philosophi­cal Society of Finland 33.274-79.

Miller, George A., and Stephen Isard. 1963. "Some Perceptual Conse­quences of Linguistic Rules". Journal of Verbal Learning and Verbal Behavior 2.217-28.

Moravcsik, J. M. E. 1967. "Linguistic Theory and the Philosophy of Lan­guage". Foundations of Language 3.209-33.

Nagel, Ernest. 1961. The Structure of Science. New York: Harcourt, Brace & World.

Pap, Arthur. 1958. Semantics and Necessary Truth. New Haven, Conn.: Yale Univ. Press.

. 1962. An Introduction to the Philosophy of Science. Glencoe, I11.: The Free Press.

Postal, Paul Martin. 1966. Review of Andre Martinet, Elements of Gener­al Linguistics (London: Faber & Faber; Chicago: Univ. of Chicago Press, 1964). Foundations of Language 2.151-86.

Quine, Willard Van Orman. 1960. Word and Object. Cambridge, Mass.: MIT Press.

Radnitzky, Gerard. 1970. Contemporary Schools of Metascience. 2nd rev. ed. Stockholm: Akademiforlaget.

Rossi, Sirpa. 1972. "Ueber das Problem der Verdinglichung bei Lukacs und Heidegger". Ajatus 34.155-61.

Saunders, John Turk, and Donald Henze. 1967. The Private-Language Prob­lem. New York: Random House.

Scheff1er, Israel. 1963. The Anatomy of Inquiry. New York: A. A. Knopf. Schutz, Alfred. 1967. Collected Papers I: The Problem of Social Reality.

The Hague: M. Nijhoff. Searle, John R. 1969. Speech Acts: An essay in the philosophy of lan­

guage. Cambridge: Cambridge Univ. Press. Slobin, Dan I. 1971. Psycholinguistics. Glenview, I11.: Scott, Foresman

& Co. Specht, Ernst Konrad. 1969. The Foundations of Wittgenstein's Late Phil­

osophy. Transl, into E. by D. E. Walford. Manchester: Manchester Univ. Press.

Strawson, Peter Frederick. 1971a. "Intention and Convention in Speech Acts". Logico-Linguistic Papers by P. F. Strawson, 149-69. London: Methuen.

TG AND PHILOSOPHY OF SCIENCE 445 Strawson, Peter Frederick. 1971b. "Meaning and Truth". Op. cit., 170-

to 189. Stroud, Barry. 1965. "Wittgenstein and Logical Necessity". The Philos­

ophical Review 74.504-18. Taylor, Charles. 1964. The Explanation of Behaviour. London: Routledge

& Kegan Paul. Vico, Giambattista. 1968[1744]. The New Science. Rev. transl, of the

3rd Italian ed. by Thomas Goddard Bergin and Max Harrold Fisch. Ithaca, N.Y.: Cornell Univ. Press.

Winch, Peter. 1958. The Idea of a Social Science. London: Routledge & Kegan Paul. (5th printing, 1965.)

Wittgenstein, Ludwig. 1953. Philosophical Investigations. Transl. by G. Elizabeth M. Anscombe. Oxford: B. Blackwell. (2nd ed., 1958.)

. 1967. Remarks on the Foundations of Mathematics. Transl, by G. E. M. Anscombe. 2nd ed. Oxford: B. Blackwell.

Wright, Georg Henrik von. 1963. Norm and Action. London: Routledge & Kegan Paul.

. 1971. Explanation and Understanding. Ibid. Wunderlich, Dieter. 1972. "Sprechakte". Pragmatik und sprachliches Han­

deln by D. Wunderlich and Utz Maas, 69-188. Frankfurt/Main: Athenäum.

Addendum:

Giannoni, Carlo Borromeo. 1971. Conventionalism in Logic: A study in the linguistic foundation of logical reasoning. The Hague: Mouton.

BIOGRAPHICAL NOTES

RAIMO ANTTILA (1935- ) received his degrees in English, German, Latin and Greek from the University of Turku, Finland, in 1959 and 1961. Af­ter a year of study as a Canada Council Fellow at the University of To­ronto (1961-62), he studied linguistics at Yale University where he re­ceived his Ph.D. in 1966. Since 1965 he has taught Indo-European and historical linguistics at the University of California, Los Angeles, and spent the years 1972-74 as a Professor of General Linguistics at the University of Helsinki. His major areas of research are reflected in his many publications, of which only the books are mentioned here: Vusimman äännehistorian suunnasta ja luonteesta [Tendencies and char­acter of newest historical phonology] (Turku: Univ. of Turku, 1969), Proto-Indo-European Schwebeablaut (Los Angeles: Univ. of California, 1969), An Introduction to Historical and Comparative Linguistics (New York: Macmillan, 1972), and Analogy (Helsinki: Univ. of Helsinki, 1974). He is the Linguistics Editor of The Journal of Indo-European Studies (1973ff.), and a member of various international linguistics boards and committees. DWIGHT L. BOLINGER (1907- ) was a Professor of Romance Languages and Literatures at Harvard University from 1963 till his retirement in 1973. He earlier taught at the University of Colorado, Boulder, the University of Southern California, Los Angeles, and Washburn University of Topeka, Kansas. Although his preparation was mainly in Spanish literature, his interests turned.to linguistics almost from the start, and he has done most of his research in syntax, semantics, and language teaching, the latter particularly as applied to the Romance languages. Within seman­tics he has been concerneh especially with the effects of fundamental frequency as shown in intonation and accent; compare his Interrogative Structures of American English: The direct question (University, Ala.: Univ. of Alabama Press, 1958), Generality, Gradience, and the Ali-or­­one (The Hague: Mouton, 1961), or That's That (Ibid., 1972). His many other publications include: Forms of English: Accent, morpheme, order (Tokyo: Hokuon, 1965); Aspects of Language (New York: Harcourt, Brace & World, 1968); The Phrasal Verb in English (Cambridge, Mass.: Harvard Univ. Press, 1971), and Degree Words (The Hague: Mouton, 1972), to men­tion only a few of his monographs. He was President of the American Association of Teachers of Spanish in Portuguese (1960) and of the Lin­guistic Society of America (1972); he was a Fellow of the Center for Advanced Study in the Behavioral Sciences and is a Fellow of the Amer-

448 BIOGRAPHICAL NOTES

ican Academy of Arts and Sciences. LYLE R. CAMPBELL (1942- ) spent his undergraduate years at Brigham Young University (B.A. in Archaeology, 1966), and received his linguis­tics degrees from the University of Washington, Seattle (M.A., 1967), and the University of California, Los Angeles (Ph.D., 1971). From 1971 till 1974, he was an Assistent Professor in Behavioral Research and An­thropology at the University of Missouri, Columbia, and is currently an Assistent Professor at the State University of New York at Albany. His main research interests are phonological theory, historical and anthro­pological linguistics, American Indian (especially Central American) languages, and Finnish. His publications include: Review of Robert D. King, Historical Linguistics and Generative Grammar (Englewood Cliffs, N.J., 1969), Language 47:1.191-209 (1971); "Is a Generative Dialectol­ogy Possible?", Orbis 21.289-98 (1972); "Mayan Loan Words in Xinca", UAL 38.187-90 (1972), and "Phonological Features: Problems and propos­als", Language 50:1.52-65 (1974). BRUCE L. DERWING (1938- ) is an Associate Professor of Linguistics at the University of Alberta, Edmonton, Canada. He received undergraduate degrees from the University of Southern California, Los Angeles, both in Slavic Studies (1961) and, earlier, in the Natural Sciences and Math­ematics (1960), and graduate degrees (M.A., 1965; Ph.D., 1970) from In­diana University, Bloomington. He received his first linguistic train­ing at the Summer Institute of Linguistics in Oklahoma and later he be­come a Graduate Intern in Linguistics at the Foreign Service Institute in Washington, D.C. He held National Defense Foreign Language Fellow­ships in Russian and a fellowship for advanced study in linguistics from the American Council of Learned Societies. He taught Russian at Indiana University (1962-65) an English as a Foreign Language at Max­well A.F.B., Alabama (1965-66). He was a Visiting Professor of Slavic Languages at the University of Alberta (1968-70) before joining the De­partment of Linguistics there. His publications include Transforma­tional Grammar as a Theory of Language Acquisition: A study in the em­pirical j conceptual and methodological foundations of contemporary lin­guistics (Cambridge Univ. Press, 1973), and papers on experimental lin­guistics and language acquisition. PETER R. HARRIS (1946- ) is a doctoral candidate in linguistics at the University of Alberta, Edmonton, Canada. He received a B.A. in English from Acadia University, Wolfville, Nova Scotia, and an M.Sc. in Linguis­tics from the University of Alberta. He is currently completing a dis­sertation on problems with explanation in linguistics. HSIN-I HSIEH (1939) was born in Taiwan and received his B.A. and M.A. degrees from Taiwan University in Taipei. In 1966 he came to the United States to study at the University of California, Berkeley, where he re­ceived his Ph.D. in Linguistics in 1971. From 1970 to 1971 he taught at the University of Detroit; in 1972 he was an assistent research linguist at the Phonological Laboratory of the University of California, Berkeley. Since 1972 he has been an Assistant Professor of Chinese at the Univer-

BIOGRAPHICAL NOTES 449

sity of Hawaii at Manoa. His recent research has focused on historical linguistics and psycholinguistics; his publications include: "The Psy­chological Reality of Taiwanese Tone Sandhi Rules", Papers from the Sixth Regional Meeting of the Chicago Linguistic Society, 489-503 (1970); "The Time Variable in Phonological Change" (with Matthew Chen), Journal of Linguistics 7:1.1-13 (1971); "Lexical Diffusion: Evidence from child language acquisition", Glossa 6:1.89-104 (1972); "A New Method of Sub-grouping", Journal of Chinese Linguistics 1:1.64-92 (1973), and "Time as a Cause of Phonological Irregularities", Lingua 33:3.253-64 (1974). DELL H. HYMES (1927- ) received his A.B. from Reed College and his M.A. and Ph.D. in Linguistics (with minors in Anthropology and Folklore) from Indiana University in 1953 and 1955, respectively. He taught at Harvard University (1955-60) and the University of California at Berke­ley (1960-65) before going to the University of Pennsylvania, where he is currently Professor of Folklore and Linguistics. He was for a number of years an Associate Editor of the International Journal of American Linguistics and is Editor of Language and Society (since 1973). He has been elected to the American Academy of Arts and Sciences, and is cur­rently co-chairman of the Committee on Sociolinguistics of the Social Science Research Council as well as a Director of the Center for Applied Linguistics. His major areas of research and writing have included Amer­ican Indian languages, linguistic anthropology and sociolinguistics, and the history of linguistics. He has edited the following books (to mention only these, a collection of his own papers having just appeared under the title Foundations in Sociolinguistics): Language in Culture and So­ciety: A reader in linguistics and anthropology (New York: Harper & Row, 1964); (together with John J. Gumperz), The Ethnography of Communication (Menasha,/ Wis. : American Anthrop. Assoc, 1964); (together with William E. Brittle), Studies in Southwestern Ethnolinguistics (The Hague: Mouton, 1967); Pidginization and Creolization of Languages (London: Cambridge Univ. Press, 1971), and Studies in the History of Linguistics: Tradi­tions and paradigms (Bloomington & London: Indiana Univ. Press, 1974). ESA ITKONEN (1944- ) studied German, French, Latin, and General Lin­guistics at the University of Helsinki where he received his doctorate in 1974 with a dissertation on the epistemological foundations of gen­eral linguistic theory. In 1968-69 he was a Research Fellow at the Mas­sachusetts Institute of Technology; since 1971 he has been Researcher at the Academy of Finland; he is currently acting professor in General Linguistics at the University of Helsinki. His main interests concern the methodology of linguistics and the philosophy of language; his pub­lications include: "Zur Charakterisierung der Glossematik", 'Neuphilolo­gische Mitteilungen 69.452-72 (1968); "Un conflit entre facteurs phoné­tiques et facteurs fonctionnels dans un texte latin mérovingien", ibid. 70.471-84 (1969);' "Semantiikan ja syntaksin suhteesta" [On the relation­ship between semantics and syntax], Virittäjä 73.217-37 (1969); "An Ep­istemologica! Approach to Linguistic Semantics", Ajatus: Yearbook of the Philosophical Society of Finland 32.96-142 (1970); "Zwei verschiedene Versionen der Bedeutungskomponente", Linguistics 59.5-13 (1970); "On

450 BIOGRAPHICAL NOTES

Grice's, Strawson's, and Searle's Concept of Meaning", Ajatus 35.149-54 (1972), and Linguistics and Metascience, which appeared as volume 2 of "Studia Phi'losophica Turkuensia" (Kokemäki: Risteen Kirjapaino, 1974). MICHAEL J. KENSTOWICZ (1945- ) earned his degrees from San Jose State College and the University of Illinois at Urbana-Champaign (Ph.D., 1971), joining the faculty of the latter institution in 1970. He is currently an Associate Professor of Linguistics and Slavic Languages. As reflec­ted in his publications, his chief area of interest has been phonolog­ical theory, in particular issues in rule ordering and universals. He has co-authored with Charles W. Kisseberth a number of articles in this area, e.g., "Rule Ordering and the Asymmetry Hypothesis", Papers from the Sixth Regional Meeting of the Chicago Linguistic Society, 504-519 (1970), and "The Multiple Application Problem", Studies in Generative Phonology ed. by Charles W. Kisseberth (Edmonton, Alberta: Linguistic Research, Inc., 1974). Together with the latter he has also edited Is­sues in Phonological Theory: Proceedings of the Urbana Conference on Phonology (The Hague: Mouton, 1973). Kenstowicz has also published a number of papers in the field of Baltic and Slavic Linguistics, includ­ing "On the Notation of Vowel Length in Lithuanian", Papers in Linguis­tics 3.73-114 (1970); "The Lithuanian Third Person Future", Studies pre­sented to Robert B. Lees ... ed. by Jerrold M. Sadock and Anthony L. Va-nek, 95-108 (Edmonton, Alta.: Ling. Research, 1970), and "The Morphopho­nemics of the Slovak Noun", Papers in Linguistics 5.550-67 (1972). E. F. K. KOERNER (1939- ) studied German and English philology, philo­sophy and history of art at the universities of Göttingen, Berlin, Edin­burgh, and Giessen before he specialized in general linguistics (Ph.D., Simon Fraser University, 1971). He was a Social Scientist Research Asso­ciate in linguistics at the University of Texas at Austin and a Visiting Research Associate at the Center for the Language Sciences at Indiana University, Bloomington (1972-73), and is currently a Habilitand in Gen­eral Linguistics at the University of Regensburg. He is editor of Histo-riographia Linguistica and general editor of the series "Amsterdam Stu­dies in the Theory and History of Linguistic Science". His publications include: Bibliographia Saussureana, 1870-1970 (Metuchen, N.J.: Scarecrow, 1972); Contribution au debat post-saussurien sur le signe linguistique (The Hague: Mouton, 1972); Ferdinand de Saussure: Origin and development of his linguistic theory in western studies of language (Braunschweig: Vieweg; Oxford & Elmsford, N.Y.: Pergamon, 1973; 2nd printing, 1974), and The Importance of Tec'hmer's "Zeitschrift für Allgemeine Sprachwissen­schaft" in the Development of General Linguistics (Amsterdam: Benjamins, 1973). ROBERT K. KROHN (1937- ) received his degrees in linguistics at the Uni­versity of Michigan, Ann Arbor, and also studied at the Freie Universtä't of Berlin. Since 1970, he has been on the faculty of the University of Hawaii, where he is an Associate Professor in the Department of English as a Second Language. His research interests include English phonology and the application of linguistics to problems of second language teach-

BIOGRAPHICAL NOTES 451

ing. Among his publications are: "The Vowel Shift Rule in English", Working Papers in Linguistics 2.141-54 (Univ. of Hawaii, 1970); English Sentence Structure (with the Staff of the English Language Institute ; Ann Arbor, Mich.: Univ. of Michigan Press, 1971); "On the Sequencing of Tautosegmental Features", Papers in Linguistics 5.114-23 (1972); "The Vowel Shift Rule and its Productivity", Language Sciences 20.17-18(1972); "Underlying Vowels in Modern English", Glossa 6.203-24 (1972), repr. in Essays on the Sound Pattern of English ed. by D. L. Goyvaerts and G. K. Pullum (Ghent: Story-Scientia, 1975); "How Abstract is English Vowel Phonology?", Towards Tomorrow's Linguistics ed. by Roger W. Shuy and Charles-James N. Bailey, 220-25 (Washington, D.C.: Georgetown University Press, 1974), and "Is There a Rule of Absolute Neutralization in Nupe?", Working Papers in Linguistics 6:3.105-13 (Univ. of Hawaii, 1974). LEONHARD LIPKA (1938- ), currently a Professor of English Linguistics at the University of Frankfurt, did his undergraduate work in English and Romance philology, completing his studies with a dissertation on English and German word-formation at the University of Tübingen in 1965. He pursued his contrastive work on German and French and on verb-parti­cle constructions in English and German; compare his Habilitationsschrift of 1971, Semantic Structure and Word-Formation: Verb-particle construc­tions in contemporary English (Munich: Fink, 1972). Besides publishing original work, e.g., "Assimilation and Dissimilation as Regulating Fac­tors in English Morphology", Zeitschrift für Anglistik und Amerikanistik 17.159-73 (1969), and "Grammatical Categories, Lexical Items, and Word-Formation", Foundations of Language 7.211-38 (1971), he translated Uriel Weinreich's Explorations in Semantic Theory into German (Tübingen: Nie­meyer, 1970) and was co-author of the Festschrift for Hans Marchand, Wortbildung^ Syntax und Morphologie (The Hague: Mouton, 1968). J. PETER MAHER (1933- ) received his degrees from Harpur College (B.A., 1955), The Catholic University of America, Washington (M.A., 1958), and Indiana University (Ph.D. in linguistics, with minors in Slavic and Latin, 1965), and has since 1964 been on the faculty at Northeastern Illinois University, Chicago, where he was promoted to Professor of Linguistics in 1970. He has recently been invited to fill the Chair of English Lin­guistics at the University of Hamburg which he accepted. He is a member of the Editorial Advisory Board of The Journal of Indo-European Studies and has served as Associate Bibliographer in the Comparative and Histor­ical Linguistics Section of the MLA Bibliography Committee. His research interests concern the integration of linguistic theory within the areas of socio-ethnolinguistics, historical-comparative, and general linguis­tics. His publications include: "More on the History of the Comparative Method: The tradition of Darwinism in August Schleicher's work", Anthro­pological Linguistics 8:3.1-12 (1966); "The Paradox of Creation and Tra­dition in Grammar: Sound pattern of a palimpsest", Language Sciences 7. 15-24 (1969); "Italian mostaccio", Zeitschrift für romanische Philologie 87.320-33 (1971); "Generative Phonology and Etymology in Traditional Lex­icon", General Linguistics 11.71-98 (1971); "Neglected Reflexes of Proto-Indo-European *pet- 'fly': Greek petros 'stone' ...", Lingua e Stile 8:3.

452 BIOGRAPHICAL NOTES

403-17 (1973), and "The Situational Motivation of Syntax and the Syntac­tic Motivation of Polysemy and Semantic Change: Spanish-Italian bravo, etc.", Diachronie Studies in Romance Linguistics ed. by Mario Saltarel­li and Dieter Wanner (The Hague: Mouton, 1973). ADAM MAKKAI (1935- ) did his undergraduate work in Romance as well as Slavic languages and literatures, first at the University of Budapest (1954-56), then at Harvard University (1957-58), majoring in Russian and minoring in French. After a two-period as a foreign language teacher at Iolani College Preparatory School in Honolulu, Hawaii, he entered Yale Graduate School in 1960 on a Ford Foundation Fellowship, receiving his M.A. and Ph.D. degrees in general linguistics in 1962 and 1965, respec­tively. During the academic year 1963-64 he was a Visiting Assistant Professor at the University of Malaya, Kuala Lumpur, and during 1965-66 he did research in computational linguistics at the RAND Corporation in Santa Monica, California. After a one-year appointment as assistant pro­fessor of Russian, English, and linguistics at Occidental College and California State College at Long Beach, California (1966-67), he joined the linguistics department of the University of Illinois at Chicago Cir­cle, where he is currently a full professor. His diversified research interests are reflected in his publications, which include many artic­les on literary themes as well as poetry in English and Hungarian. His Idiom Structure in English of 1965 (printed, The Hague: Mouton, 1972) was the first stratificational dissertation written in terms of Sydney Lamb's theory. With David G. Lockwood, he edited Readings in Stratifi­cational Linguistics (University of Alabama Press, 1973), and has re­cently advocated his own brand of stratificationalism (Pragmo-Ecolog-ical Grammar) in Language Sciences 27.9-23 and 31.1-6 (1973-74). He is co-founder, together with J. Peter Maher, Robert J. DiPietro, and oth­ers, of The Linguistics Association of Canada and the United States (LACUS), the first yearbook of which is scheduled to appear this spring (Columbia, S.C.: Hornbeam Press, 1975). EDWARD R. MAXWELL, Jr. (1943- ) completed his Ph.D. in Linguistics with a dissertation on semantic structures at Northwestern University, Evans-ton, in 1972 and has since been an Assistant Professor at Northeastern Illinois University, Chicago. His publications include: "Performatives in Korean" (with Hong Bae Lee), Papers from the Sixth Regional Meeting of the Chicago Linguistic Society, 363-79 (1970); "Aspects of Lithuani­an Complementation", Papers in Linguistics 4:1.169-95 (1971); "A Compu­terized Lexicon of English" (with Raoul N. Smith), Proceedings of the 1973 International Conference on Computers in the Humanities (Edinburgh Univ. Press, 1974), and "An English Dictionary for Syntactic and Seman­tic Processing" (with Raoul N. Smith), Proceedings from the 197'3 Inter­national Conference on Computational Linguistics (Pisa: Univ. of Pisa, 1974). JOHN ODMARK (1942- ), a former NDEA Fellow at the University of Oregon, Eugene (1966-69), is currently a lecturer of English at the University of Regensburg.

BIOGRAPHICAL NOTES 453

FRED . PENG (1934- ) came to North America from China, studied at the University of Toronto, the University of Buffalo, Cornell Universi­ty, the University of Texas at Austin, and completed his doctoral dis­sertation at the State University of New York at Buffalo, where he was the first to receive a Ph.D. in anthropology and linguistics. During the academic year 1971-72, he was a visiting faculty member in the De­partment of Anthropology, Tulane University, New Orleans; since 1966 he has been on the faculty of the International Christian University, Tokyo, where he is now a Professor of Linguistics in the Division of Languages. His current research interests, besides Chinese and Japanese linguistics, include the language of the Ainu in Northern Japan, child language with emphasis on the comparison of Japanese-speaking and En­glish-speaking children, and Japanese sign language with due consider­ation of non-verbal communication in general. His many publications in­clude: "Some Comments on Affixation in Standard Chinese", Studies in Linguistics 18.49-57 (1964-66); "Amoy Phonology: Phonemicization of three nasals", Archiv Orientâlni 34.411-16 (1966); "A Note on 'A Note of Mandarin Phonology1: A critique of generative phonology", Monumenta Serica 26.175-201 (1967); "A Note on 'Hokkaido Ainu Morphophonemics'", Studies in Linguistics 21.95-100 (1969-70); "A Grammar of Ainu Number Names" (with Barron Brainerd), Lingua 25.381-97 (1970); "Morphological Agreement: A special trait in Ainu", Ching Hua Hsueh Pao 9.271-89(1970); "La Parole of Japanese Pronouns", Languages Sciences 25.36-39 (April 1973), and "Communicative Distance", ibid. 31.32-38 (Aug. 1974). WALBURGA VON RAFFLER ENGEL (1920- ), a native of Munich, Germany, re­ceived her Doctor of Letters degree in Classics at the University of Turin (1947), and her Ph.D. in General Linguistics at Indiana University in Bloomington (1953), and is currently an Associate Professor of Lin­guistics at Vanderbilt University, Nashville, Tennessee, and chairman of the Committee of Linguistics of Nashville University Center. In addition she is Secretary of the International Child Language Association as well as Chairman of the board of editorial advisors of the Journal of Child Language. She has developed a theory of language acquisition by levels of abstraction and was one of the first linguists to challenge the con­cept of competence in transformational-generative theory on empirical grounds. Her publications, which have appeared in six languages, deal mainly with questions concerning child language, bil inguai ism, and kin­es i c s ; her numerous studies include: Il -prelinguaggio infantile (Brescia: Paideia, 1964); "Child Bil inguai ism and some Current Problems in Speech Perception and Linguistic Theory", Word 23:1.423-42 (1967[1969]) ; "Com-portamento linguistico e teorie generativo-transformazionaliste", Archi­vio Glottologico 54.238-40 (1969); "The Inadequacy of the Transformation­al approach to the Analysis of Child Language", Word 26:3.395-401 (1970 [1974]); "Competence — a term in search of a concept", Hommage à Eric Buyssens (Brussels: Edit, de l'Inst. de Sociologie, 1970), 271-85; "The LAD, our Underlying Unconsciousness, and more on 'Felt Sets'", Language Sciences 13.15-18 (Dec. 1970); "Some Phono-Syntactic Features of Black English", Phonetica 25.53-64 (1972), and "Neki metodoloski problemi u

454 BIOGRAPHICAL NOTES

USVajanju maternjeg jezika" [Some methodological problems in first lan­guage acquisition], Symposium on Child Language (Belgrade: Univ. of Bel­grade, 1973), 53-58. J. HOWARD SHAW (1944™ ) graduated with Honours in German at the Univer­sity College of Wales, Aberystwyth, in 1967, after having studied at Ab­erystwyth and Munich. He was awarded a State Studentship which enabled him to study at the University of Manchester, where he received his M.A. degree in General Linguistics in 1969, the title of his dissertation be­ing "Transparency in German and English Vocabulary". After five years teaching at schools in England and West Germany he went to the Universi­ty of Regensburg, where he is a lecturer in the English department and a doctorand. ROYAL SKOUSEN (1945- ) did his undergraduate work in English and math­ematics at Brigham Young University (B.A. "magna cum laude", 1969), and completed his graduate work at the University of Illinois at Urbana with a Ph.D. in linguistics. Since 1972 he has been an Assistant Professor in the Linguistics Department of the University of Texas at Austin. His pub­lications include: "Consonant Gradation in Finnish", Studies in the Lin­guistic Sciences 1.67-91 (1971); "Consonant Alternation in Fula", Stu­dies in African Linguistics 3.77-96 (1972); "On Capturing Regularities", Papers from the Eighth Regional Meeting of the Chicago Linguistic Society, 567-77 (1972); "Empirical Restrictions on the Power of Transformational Grammars", Papers in Linguistics 5.250-69 (1972); "Evidence in Phonology", Studies in Generative Phonology ed. by Charles W. Kisseberth, 72-103 (Ed­monton, Alberta: Linguistic Research, 1973); "On Limiting the Number of Phonological Descriptions", Glossa 7.167-78 (1973), and "Finnish Vowel Harmony: Rules and conditions", Issues in Phonological Theory ed. by M. J. Kenstowicz and C. W. Kisseberth, 118-29 (The Hague: Mouton, 1973). UHLAN V. SLAGLE (1937- ) received his B.A. in German from Indiana Uni­versity in 1964, and his Ph.D. in Linguistics from Georgetown University, Washington, D.C., in 1972. During 1969-70 he held a Research Fellowship from the Thyssen-Stiftung for study in West Germany; he is currently a Younger Humanist Fellow of the National Endowment for the Humanities (1974-75). His main research interests are the philosophy of language and the history of linguistics. His publications include: Language, Thought, and Perception (The Hague: Mouton, 1974); "The Kantian Influ­ence on Humboldt's Linguistic Thought", H i s t o r i o g r a p h i a Linguistica 1:3. 341-50 (1974), and "A Viable Alternative to Chomskyan Rationalism", First LACUS Forum ed. by Adam Makkai and Valerie Becker Makkai, 177-93 (Colum­bia, S.C.: Hornbeam Press, 1975). DANNY D. STEINBERG (1931- ) studied psychology and sociology at the Uni­versity of British Columbia (B.A., 1960) and the University of Toronto, receiving his graduate degrees in Experimental Psychology from the Univer­sity of Hawaii (M.A., 1964; Ph.D., 1966). He did research in psycholin-guistics on a NIMH post-doctoral fellowship with Charles E. Osgood at the University of Illinois at Urbana-Champaign (1967-69). Since 1969 he has

BIOGRAPHICAL NOTES 455

been an Associate Professor teaching psycholinguistics in the Department of English as a Second Language, University of Hawaii, Honolulu. During the summer of 1974, he taught at Simon Fraser University, Vancouver, B. C., Canada. His major research interests are semantics, reading, and phonology. Together with Leon A. Jakobovits, he edited Semantics: An in­terdisciplinary reader in philosophy, linguistics, and psychology (Cam­bridge Univ. Press, 1971; paperback ed., 1974). His publications include: "Natural Class, Complimentary Distribution, and Speech Perception", Jour­nal of Experimental Psychology 79:2.195-202 (1969); "Analyticity, Amphi-gory, and the Semantic Interpretation of Sentences", Journal of Verbal Learning and Verbal Behavior 9.37-51 (1970); "Current Issues in Psycho­linguistics", Pacific Speech 4.36-50 (1970); "Negation, Analyticity, Am-phigory, and the Semantic interpretation of Sentences", Journal of Ex­perimental Psychology 84:3.417-23 (1970); "Analyticity and Amphigory", Language Sciences 20.19-25 (April 1972), and "Phonology, Reading, and Chomsky and Halle's Optimal Orthography", Journal of Psycholinguistic Research 2:3.234-58 (1973).

INDEX OF NAMES

A. Adams, V.: 183 Andersen, H.: 263, 267, 271-74,

276, 279, 293 Anglin, J. M.: 267, 272, 293 Anscombe, G. E. M.: 442, 445 Anttila, R.: 263-96, 447 Apel, K. 0.: 385, 389, 429, 436n.l5, 439n.39, 440n.46+47, 441

Ardener, F.: 379 Aristotle: 332 Austin, J. L.: 436n.l5 Ayer, A. J.: 385

B. Bach, E. W.: 38, 56, 84, 87, 105, 173, 424, 441

Bailey, C. J. N.: 141, 236, 357+ n.2

Baron, N.: 267, 271, 294 Bartsch, R.: 294 Beatty, J.: 26 Beck, L. W.: 339, 344 Becker, D. A.: 353, 357 Bergin, T. G.: 445 Berkeley, G.: 385 Berko, J.: 110, 141 Berne, E.: 38 Bever, T. G.: 147, 173, 266, 294, 308, 312

Bierwisch, M.: 265, 296 Bikson, .: 263n.l Binnick, R. I.: 84 Birnbaum, H.: 59, 84

Bloomfield, L.: 87, 95, 104n.8, 105, 145-49, 152, 159, 172, 173, 361, 371 Boas, F.: 46, 360, 361, 378

Bogatyrev, P. G.: 360 Bolinger, D. L.: 3-35, 447-48 Borger, R.: 345 Botha, R. P.: 176-78, 183, 384,438

.30, 439.38, 441 Brame, M. .: 30 Brêal, M.: 267, 275-76, 279n4, 290 Brekle, H. E.: 176 Brentano, F.: 396 Brown, R. W.: 342, 344 Brugmann, .: 341, 344 Bush, R. R.: 313, 442

. Campbell, L. R.: 361-58, 448 Carroll, J. .: 235n.l Cassirer, E.: 341, 344 Cavell, S.: 395, 404, 441 Chafe, W. L.: 19, 87, 105 Chao, Y. R.: 105 Chapin, . .: 176 Charbonneau, R.: 295 Chen, M.: 109n.l Cheng, . : 141 Cheng, R.: 110, 139., 141 Chomsky, : 120, 141, 233, 256-57, 258

Chomsky, N. : 4, 38-42, 55, 73, 84, 87, 98, 104.10, 105, 178-79, 181-to 183, 185-86, 227n.l, 229+n.20, 233-58, 263, 264, 272-73, 278, 297-313, 320-23, 342-43, 344-45,

458 INDEX OF NAMES

Chomsky, N. (cont.): 359, 371-to 375, 378, 383, 388-89, 398-400, 404, 406-07, 409, 424, 428-29, 431, 435n.5, 436n.l5, 438-39n.33, 441

Cioffi, F.: 345 Clark, E. V.: 120, 141, 264-67, 272-73, 294

Collinder, .: 170, 173 Coseriu, E.: 176, 178 Crothers, J.: 109n.l Crutchfield, R. S.: 345, 346 Cust, Mrs. H.: 294

D. Dahlstedt, . . : 294 Darnell, R. D.: 366 DeCamp, D.: 287, 294 Delbrück, .: .: 341. 345 Denison, N. : 281, 294 Descartes, R.: 385, 436n.8, 442 Derwing, B. L.: 297-314, 448 Dingwall, W. 0.: 21, 109n.l, 141,

173, 296 Dokulil, M.: 176

E. Ehrenstein, W.: 332, 344n.3, 345 Erben, J.: 176 Ervin-Trípp, S. M.: 271, 294

F. Feyerabend, . .: 440n.45, 442 Fisch, M. H.: 445 Fischer, .: 324 Fishman, J. A.: 282, 284, 295, 378

Fleischer, W.: 176 Fodor, J. A.: 43, 84, 87, 93, 101, 104n.4, 105, 305, 307-08, 310, 313, 351, 436n.9, 441

Foster, G.: 169, 173 Foster, M.: 169, 173, 366 Fromkin, V. A.: 110, 142 Fürth, H. G.: 427

G. Galanter, E.: 313, 442 Garrett, M.: 305, 307-08, 310 Gauger, H. M.: 176 Gazzaniga, M.: 338, 345 Geach, P. T.: 442 Geis, M. L.: 34n.2, 35 Giannoni, . .: 427, 428, 445 Giglioli, P. P.: 283, 284, 295 Gladwin, T.: 378 Goodenough, W.: 369 Goodman, N. : 277, 295, 315, 320 Grice, G. R.: 436n.l5, 440n.50, 442

Gross, M.: 315-16, 320 Gruber, F. : 378 Gruber, J. S.: 176 Gumb, R. D.: 402, 438n.32, 442 Gumperz, J. J.: 379

H. Habermas, J.: 389, 436n.l5, 442 Hakulinen, L.: 194, 200, 209, 213,

214, 221, 228n.8, 229n.l4 Hale, K. E.: 356 Halle, M.: 97, 105, 141, 175-83, 185-86, 227n.l, 229, 233-58, 297, 304, 306, 313, 357n.2,358, 400, 441,

Halliday, M. A. K.: 64 Hallowell, A. I.: 379 Halpern, A. M.: 167-68, 173 Hanna, J. F.: 419, 439n.40, 442 Hansen, .: 176 Hare, R. M.: 395, 404, 442 Harms, R. .: 170+n.4, 173, 229+ n.12

Harris, P. R.: 297-314, 448 Harris, Z. S.: 379 Hatcher, A. G.: 176 Hayes, D. G.: 357 Hayes, J. R.: 314 Heidegger, M.: 385 Hempel, C. G.: 382-83, 390, 408, 414, 435n.l, 2+4, 440n.52, 442

Henie, M.: 347

INDEX OF NAMES 459

Henze, D.: 385, 435n.7, 444 Henzen, W.: 176 Herder, J. G.: 361 Hertzler, J.: 286, 287, 295 Herzog, G.: 360, 367 Herzog, M. I.: 355, 358, 372, 380

Hilliard, R.: 334, 345 Hintikka, J.: 436n.l6, 437n.25, 438n.27, 442

Hjelmslev, L.: 372, 388, 436n.l3, 443

Hockett, C. F.: 87, 93, 94-96, 104n.6+8, 173

Hogan, H.: 366 Hoijer, H.: 173, 367 Hook, S.: 273, 278, 292, 295 Householder, F. W.: 438n.31 Houston, S. H.: 141, 295 Hsieh, H. I.: 109-44, 253, 258,

448-49 Humboldt, W. von: 279n.4, 361,

388, 399, 436n.l2, 443 Hume, D.: 271, 272, 385 Huxley, J.: 285, 286, 295 Hyman, L. H.: 141 Hymes, D. H.: 295, 359-80, 449

I. Isard, S.: 437n.l8, 438n.32, 444 Itkonen, E.: 200, 202, 228n.8, 229

Itkonen, Esa: 263, 264, 275, 292, 293, 295, 381-445, 449-50

Itkonen, T.: 198, 230

J. Jackendoff, R. S.: 38, 318, 320 Jackson, F.: 233n, 254n.2 Jacobs, R. A.: 183, 439n.34, 443 Jain, S.: 366 Jakobovits, L. A.: 320 Jakobson, R.: 362-64, 366, 368, 374, 379

James, W.: 332, 345 Jespersen, 0.: 23, 176 Jevons, W.: 332, 345 Johnson, C. D.: 228n.6s 230

Joki, A. J.: 200, 202, 228n.8, 229

K.

Kachru, B. B.: 173 Kanngiesser, S. : 384, 443 Kant, I.: 339-40, 344n.5, 345 Kastovsky, D.: 176, 180-81, 181n.3, 183

Katz, J. J.: 38, 43, 87, 91, 93, 101, 104n.4, 105, 301, 313, 351, 358, 409, 428-29, 438n.30, 443

Keenan, E.: 366 Kendon, A.: 324 Kenstowicz, M. J.: 145-74, 450 Kettunen, L.: 206, 215, 228n.l3, 230

Key, M. R.: 324 Keyser, S. J.: 254, 258, 353, 358 Kim, C. W.: 109n.l King, R. D.: 353, 355, 358 Kiparsky, P.: 142, 143, 147, 156, 158, 159, 228n.6, 256,259, 267, 352, 358

Kisseberth, C. W.: 146, 158, 170+ n.4, 173, 186

Klima, E. S.: 259, 353, 358 Kloss, H.: 282, 284, 295 Kluckhohn, C.: 368 Knight, T. S.: 268-69, 291-92, 295 Kobayashi, L.: 233 Köhler, W.: 330, 332-33, 343n.l+2 Koerner, E. F. .: 450 Körner, S.: 339, 345 Koffka, .: 343n.2, 345 Koutsoudas, A.: 142, 153, 173, 230 Koziol, H.: 176 Krech, D.: 335, 343n.2, 344n.4,

345, 346 Kroeber, A. L.: 367 Krohn, R. .: 109n.l, 142, 233-59,

450-51 Kuhn, T. S.: 440n.45, 443

L. Labov, W.: 30, 35, 142, 352, 355,

358, 372, 380 Ladefoged, .: , 141

460 INDEX OF NAMES

Lakatos, I.: 442 Lakoff, G.: 38, 317, 320, 389, 443

Lakoff, R.: 30, 35 Lamb, S. M.: 66, 77, 84, 310, 313

Lance, D. M.: 357 Langendoen, D. T.: 266, 294 Lees, R. B.: 74, 173, 176 Lehmann, W. P.: 358, 380 Lehrer, A.: 88, 105 Leibniz, G. W.: 340-41, 346 Lenneberg, E. H.: 294 Lentin, A.: 315-16, 320 Lévi-Strauss, .: 362, 363, 368, 369, 379

Levy, J.: 338, 346 Liao, . .: 110, 142 Lieberson, S.: 295 Liljencrantz, J.: 276, 279, 295 Lindblom, .: 276-77, 295 Lipka, L.: 175-84, 451 Ljung, M.: 176 Locke, J.: 385 Lockwood, D. G.: 38, 54, 57, 60, 66, 84

Lorenzen, P.: 397, 398, 428, 443 Lowie, L. C : 367, 379 Lowie, R. H.: 367 Luce, R. D.: 313, 442

M. Maas, U.: 445 Mackey, W. F.: 294 Mäher, J. P.: 110, 142, 252, 259,

263, 271, 290, 295, 296, 451-52 Makkai, A.: 37-85, 259, 452 Makkai, V. .: 85 Malkiel, Y.: 176, 358, 367, 380 Mandelbaum, D.: 375, 379 Marchand, H.: 176, 177n.l, 178-79,

180, 183 Marshall, J. C.: 296 Maxwell, E. R.: 315-20, 452 McCawley, J. D.: 38, 87, 91, 97-102,

104n.10+11, 105, 164, 165n., 173, 180, 184, 186, 227-28n.5, 6, 7 +9, 317

McNeill, D,: 267, 273, 274, 296, 409, 443

Mead, G. H.: 385, 436n.l4, 436n. 15, 443

Mehtonen, L.: 438n.29, 444 Mihailovic, L.: 26, 35 Mill, J. S.: 332 Miller, G. A.: 301, 302-04, 307, 437n.l8, 438n.32, 441

Milmed, .: 339, 346 Mohrmann, .: 379 Moravcsik, J. M. E.: 388, 389,

444 Morciniec, N. : 176 Moser, H,: 441 Moskowitz, . A.: 109n.l, 110,

235, 263n. Motsch, W.: 176, 179, 184 Mott, E.: 233n. Musgrave, A.: 442

N. Nadzhip, È. N.: 171, 172, 174 Nagel, E.: 435n.l, 444 Neisser, U.: 337, 346 Neuhaus, H. J.: 176, 178, 184 Newman, S. S.: 379 Newmeyer, F. J.: 77, 85 Noll, . .: 142, 153, 173

0. Ogden, . .: 279n.4 Ohala, J. J.: 109n.l, 110, 142, 189, 229n.l7, 230, 259, 356, 358

Ohala, M.: 109n.l, 110, 142 Ohnesorg, K.: 328 Oldfield, R. .: 296 Oppenheim, F.: 435n.2 Ornstein, J.: 294 Osgood, . .: 324

Pap, A.: 417, 418, 419, 426, 437n. 42, 444

Paton, H. J.: 339, 346 Peirce, C. S.: 268, 271, 272, 275,

INDEX OF NAMES 461

Peirce, C. S. (cont.): 290, 429, 436n.l5,

Peng, F. . .: 87-106, 453 Pennanen, E.: 176, 184 Penttila, A.: 198, 230 Peters, P. S.: 38, 299, 313 Phi—ips, S.: 366 Piaget, J.: 427 Pike, . L.: 369, 370, 379 Plath, W. J.: 312 Popper, K. R.: 412 Postal, P. M.: 43, 85, 150, 174,

185, 230 Posti, L.: 193, 194, 212, 214,

221, 230 Pribram, K.: 331, 337, 346 Prideaux, G. D.: 299, 314

Q. Quine, W. V. 0.: 437n.26, 444

R. Radnitzky, G.: 429, 430, 435n.

6, 444 Raffler-Engel, W. von: 263, 266,

271, 296, 321-28, 453-54 Rapóla, M.: 198, 206, 207, 212,

215, 217, 221, 228n.l2, 231 Raun, A.: 193, 231 Reilly, F. E.: 268, 269, 270,

296 Restorff, H. von: 332, 333, 345 Richards, I. A.: 279n.4 Rigault, A.: 295 Ritchie, R. W.: 38 Robinson, J.: 235, 259 Rohrer, .: 176 Rosenbaum, P. S.: 439n.34, 443 Ross, J. R.: 38 Rossi, S.: 385, 444 Russell, .: 340, 346, 385

S. Saareste, A.: 193, 231 Saltarelli, M.: 353, 358 Sanders, G. A.: 141, 153, 173 Sapir, E.: 150-56, 158, 163-64,

174, 252, 259, 361+n.l, 366n.2,

Sapir, E. (cont.): 367-77, 379 Saporta, S.: 353, 358 Saunders, J. T.: 385, 435n.7, 444 Saussure, F. de: 185, 222, 223,

229n.l6, 231, 354, 371, 429 Schane, S. A.: 227n.l, 231, 233 Scheffler, I.: 435n.2, 444 Schiller, P. von: 330, 333, 343n.2,

346 Schlachter, P.: 30, 35 Schreiber, P. A.: 27, 35 Schutz, A.: 400, 404, 432, 436n.l4,

437n.26, 438n.31, 444 Schwarcz, R. M.: 311, 314 Schwartz, A.: 19, 20, 35 Searle, J. R.: 436n.l5, 440n.50+51,

444 Sebeok, T. A.: 360, 362, 379 Seitel, P. and S.: 366 Shapiro, M.: 263n., 289, 296 Shaw, J. H.: 454 Sherzer, J.: 356, 358, 366, 379 Shibatani, M.: 109n.l Siegel, F. M.: 176 Skousen, R.: 185-228, 454 Slagle, U. V.: 329-47, 454 Sledd, J. H.: 353, 358 Slobin, D. I.: 246, 265, 271, 399,

444 Smart, H.: 344n.5, 346 Smith, H. L.: 369 Smith, N. K.: 344n.5, 347 Specht, E. .: 397, 403, 444 Sperry, R. W.: 338, 346 Spier, L.: 369, 379 Stampe, S.: 229n.l9, 231 Stegmüller, W.: 435n.l Stein, G.: 176, 178, 184 Steinberg, D. D.: 109n.l, 110, 142,

233-59, 309, 314, 454-55 Steinthal, H.: 361 Stevens, . N.: 304, 313 Strauss, A.: 366 Strawson, P. F.: 432, 436n.10+15,

440n.50, 444, 445 Stroud, .: 439n.44, 445 Sturtevant, E. H.: 274, 278, 285,

287, 288, 296 Sturtevant, W. C.: 378

462 INDEX OF NAMES

S ü l l w o l d , F . : 332 , 346 Sv7adesh, M. : 1 5 6 - 6 7 , 174 , 3 6 7 ,

368 , 379

T. Taylor, C : 436n.9, 437n.26, 440n.50, 445

Teilhard de Chardin, P.: 285, 289

Thurstone, L. L.: 335, 344n.4, 346

Tiwary, K. M.: 366 Toivonen, Y. H.: 198, 200, 228

.8, 229n.l4, 231 Topping, D.: 233n Trager, G. L.: 369 Traugott, E.: 273. Trevar then, : 338, 346 Trnka, .: 367, 379 Troike, R. .: 357.2, 358 Trubetzkoy, N. S. : 363 Tuomi, T.: 191, 231

U. Uhlenbeck, E. M.: 366, 379

V. Vachek, J.: 367, 379, 380 Vaihinger, H.: 344n.5, 346 Vasiliu, E.: 353, 358 Velten, H.: 360, 362 Vendryes, J.: 279n.4 Vennemann, T.: 142, 294 Vico, G.: 437-38n.27, 445 Voegelin, C. F.: 156-67, 174,

360, 362, 367, 379

W. Walford, D. E,: 444 Wallach, H.: 332, 346-47 Wang, W. S. Y.: 109n.l, 120

Wardhaugh, E.: 233, 259 Wartofsky, M. W.: 344 Watt, W. G.: 308, 314 Weinreich, U.: 101, 106, 176, 178+n.2, 181, 184, 355, 358, 372, 380

Wertheimer, M.: 330, 332, 343n. 1, 347

Wescott, R. W.: 272, 284, 285, 287, 288, 296

Whatmough, J.: 294 Wheeler, . I.: 279n.4 Whinnom, .: 287-88, 296 Whitney, W. D.: 341, 347 Whorf, . L.: 361, 370 Wiik, .: 228.6, 7 + 9, 231 Winch, P.+ 396-98, 427, 429,

430-31, 438.29, 445 Wittgenstein, L.: 264, 385, 387,

413, 427, 428, 436n.9+15, 440 n.45-47, 445

Wolff, R.: 344n.5, 347 Wright, H. G. von: 405, 435n.6,

436n.l7, 437n.26, 440n.49, 445 Wunderlich, D.: 389, 440n.51,

445 Wundt, W.: 341, 347 Wurzel, W. U.: 179, 184

Y. Yngve, V. H.: 39-40, 45, 65, 85

Z. Zimmer, K. E.: 142, 176

* * * * *