49
What’s wrong with grammaticalization? Lyle Campbell University of Canterbury, Department of Linguistics, Christchurch, New Zealand Abstract Claims about grammaticalization, and especially about ‘‘grammaticalization theory,’’ are assessed. It is argued that grammaticalization is derivative, that is, that it has no independent status of its own, but rather relies on other processes and mechanisms of linguistic change which are independent of grammaticalization but which provide the explanations for the phenomena involved in grammaticalization. This raises the question of whether grammaticalization has any value at all. The position taken here is that it does, that while the phenomenon of grammaticalization is interesting and attention on it has provided a range of valuable information, there are serious problems with so-called ‘‘grammaticalization theory.’’ 7 2000 Elsevier Science Ltd. All rights reserved. Keywords: Grammaticalization; Language change; Typology; Directionality; Reanalysis 1. Introduction The goal of this paper is to assess the nature and value of grammaticalization. 1 I attempt to show that grammaticalization is derivative, that is, that grammaticalization has no true status of its own, but rather relies on other processes and mechanisms of linguistic change which exist independently of grammaticalization but which provide the explanations for the phenomena involved in grammaticalization. This raises the question of whether grammaticalization has any value of its own. I suggest that it does have, but only heuristic value — not theoretical or explanatory — in that it focuses attention on an interesting range of phenomena. That is, while the phenomenon of 0388-0001/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved. PII: S0388-0001(00)00019-X Language Sciences 23 (2001) 113–161 www.elsevier.com/locate/langsci 1 I thank Alice Harris and Muriel Norde for helpful comments on an earlier draft of this paper.

What’s wrong with grammaticalization?

Embed Size (px)

Citation preview

Page 1: What’s wrong with grammaticalization?

What's wrong with grammaticalization?

Lyle Campbell

University of Canterbury, Department of Linguistics, Christchurch, New Zealand

Abstract

Claims about grammaticalization, and especially about ``grammaticalization theory,'' areassessed. It is argued that grammaticalization is derivative, that is, that it has no

independent status of its own, but rather relies on other processes and mechanisms oflinguistic change which are independent of grammaticalization but which provide theexplanations for the phenomena involved in grammaticalization. This raises the question of

whether grammaticalization has any value at all. The position taken here is that it does,that while the phenomenon of grammaticalization is interesting and attention on it hasprovided a range of valuable information, there are serious problems with so-called

``grammaticalization theory.'' 7 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Grammaticalization; Language change; Typology; Directionality; Reanalysis

1. Introduction

The goal of this paper is to assess the nature and value of grammaticalization.1

I attempt to show that grammaticalization is derivative, that is, thatgrammaticalization has no true status of its own, but rather relies on otherprocesses and mechanisms of linguistic change which exist independently ofgrammaticalization but which provide the explanations for the phenomenainvolved in grammaticalization. This raises the question of whethergrammaticalization has any value of its own. I suggest that it does have, but onlyheuristic value Ð not theoretical or explanatory Ð in that it focuses attention onan interesting range of phenomena. That is, while the phenomenon of

0388-0001/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.

PII: S0388 -0001 (00)00019 -X

Language Sciences 23 (2001) 113±161

www.elsevier.com/locate/langsci

1 I thank Alice Harris and Muriel Norde for helpful comments on an earlier draft of this paper.

Page 2: What’s wrong with grammaticalization?

grammaticalization is interesting and not really in question, there are seriousproblems with so-called ``grammaticalization theory'' and these are the subject ofthis paper.

A number of the de®nitions of grammaticalization, which are surveyed in theIntroduction to this volume, play a role in the discussion in this paper. Principalamong these are Meillet's (1912) version which sees grammaticalization primarilyas lexical> grammatical change, Kuryl/owicz' (1965) of lexical> grammatical andless grammatical>more grammatical, and the prototypic (or core) de®nition mostfamiliar today: some linguistic element>more grammatical.

2. Some paradigm examples

It will be helpful to have some standard examples of grammaticalization inmind before we turn to an assessment of the status of grammaticalization. Thefollowing are very typical, and most are repeated frequently in thegrammaticalization literature.

[1] English will `future'< `want' (cf. German cognate will `want' [Present.Sg]).English will originally meant only `want', but was grammaticalized as anauxiliary verb meaning `future'. It was ``semantically bleached'' (lost the `want'sense), and has undergone ``phonetic erosion'' in the form of contraction, as inI'll, she'll, and so on. (For details, see Bybee et al., 1994.)

[2] Meillet's (1912, p.145) Greek example, `to want' > `future'. Modern Greek ya`future' began as Classical Greek theÂlei `want'. The in®nitive construction wasreplaced with a subordinate clause construction as in (la) and (lb):(la) theÂlo- hina graÂpho- `I want to write' [literally `I want that I write'](lb) theÂlei hina graÂphei `he/she wants to write' [`he/she wants that he/she writes'].The main verb theÂlei continued to mean `want' but also came to mean `future',as in, for example (2):(2) theÂlo- hina graÂpho-, meaning either `I want to write' or `I will write'. Later,`future' became restricted to the `third person' form only, /yeli/ (from theÂlei ),and /yeli hina/ changed to /ya/, as in (3):(3) /yeli hina/> /yeli na/> /ye na/> /ya na/> /ya/ `future'. This is illustratedin [4], where Modern Greek /ya/ is the fully grammaticalized `future' marker,semantically bleached and phonetically reduced from original theÂlei `(he/she/it)wants':(4) Modern Greek /ya ]ra fo/ `I will write'. (For details see Joseph, 1990, thisvolume.)

[3] English be going to> `future' (cf. gonna ). (See discussion below.)[4] English examples illustrating lexical> discourse marker are: deed `a doing, act'

> indeed; while `a time' (concrete noun)> `temporal conjunction' (While Iwaited, they played ) > `concessive conjunction' (While it may be troubling toconsider, it is not expensive ). (cf. Traugott and KoÈ nig, 1991; Hopper andTraugott, 1993, pp. 4, 82±85).

L. Campbell / Language Sciences 23 (2001) 113±161114

Page 3: What’s wrong with grammaticalization?

[5] Romance languages mente `in mind' (<ablative of Latin mens `mind') wasgrammaticalized as an adverbial clitic (in Spanish) or su�x (in French):absoluta mente `in absolute mind'> Spanish absolutamente, French absolument`absolutely'.

[6] Latin amare habe-o- [to.love have-I] > French j'aimerai, Spanish amare (earlieramar he [he< habeo- ]) `I will love'.

Examples [5] and [6] illustrate the cline which is generally accepted ingrammaticalization literature: free (unbound ) morpheme> clitic> su�x.

[7] Swedish (and Scandinavian generally) -s `passive, impersonal' <Old Norse sik`3rd person accusative re¯exive pronoun' (cf. Swedish sig ), as in SwedishdoÈrren oÈppnas `the door opens', hoppas `it is hoped, one hopes'). (For details,see Norde 1997 and this volume; cf. Hopper's 1998, pp. 154±156 discussion ofthe Danish -s passive.)2

[8] In some Balto-Finnic languages `comitative' case< postposition `with'. Forexample, compare the Estonian construction in (1), which has a case su�x,with the Standard Finnish construction in (2), with its postposition, whichactually corresponds to the older form from which the Estonian comitativecase developed:(1) Estonian poja-ga [boy-COMITATIVE] `with the boy' (Estonianorthographic g is [k])(2) Finnish poja-n kanssa [boy-GEN with] `with the boy'Finnish dialects have shapes such as kans, ka:s, kah, etc. for the postposition,and in some it has become a clitic and even a case su�x, just as in Estonian.For example, several Upper Satakunta and Savo dialects developed the`comitative/instrumental' clitic or case, -ka(h), -ka:n, e.g. isaÈnka `with father'(isaÈ `father'); koiranka:n `with the dog', (koira `dog'); in TytaÈ rsaari (and theKymenlaakso area) it undergoes vowel harmony: autonkaa `with the car',kaÈrrinkaÈaÈ `with the cart', clear evidence of its status as a su�x there.(Kettunen, 1930, p.29; Oinas, 1961; Laanest, 1982, pp.174±175.)

2 The history of the Scandinavian (medio) passive is actually a bit more complicated than this descrip-

tion of Swedish suggests, as Bergman (1973, p.25) explains: ``Characteristic, too, of the Scandinavian

languages is the development of a special re¯exive verb, which later became a passive. This was formed

by the su�xation of reduced forms of the re¯exive pronouns seÂR (dat.) `oneself' and sik (acc.) `oneself'

to various verbs; seÂR (>-sR> -ss ) was used in East Norse (Swed. and Dan.), and sik (>-sk, later> -st )

in West Norse (Norw. and Icel.). Thus, from OEN [Old East Norse] kalla seÂr comes Swed. kallas,

Dan. kaldes, from OWN [Old West Norse] kalla sik, OIcel. and ONorw. kallask, later Icel. kallast,

Norw. Nynorsk kallast''. (For details, see Norde 1997.) I thank Muriel Norde for pointing these facts

out to me.

L. Campbell / Language Sciences 23 (2001) 113±161 115

Page 4: What’s wrong with grammaticalization?

The comparisons in Table 1 exhibit the cognate forms in several Balto-Finniclanguages, where some have the postposition, others only a case su�x, and somehave both, illustrating the typical grammaticalization of postposition > case. (Forlists of typical grammaticalization changes, see Campbell, 1998/1999, pp.239±241;Heine and Reh, 1984; Hopper and Traugott, 1993.)

3. The status of grammaticalization

I now come to the question of whether grammaticalization has any independentstatus of its own or whether it is merely derivative of other already well-understood mechanisms and kinds of change (see also Janda, Joseph, andNewmeyer, this volume). Heine et al. (1991, p.23) raise this same question andsuggest it is a common one:

Another question, which has repeatedly aroused the interest of linguists, iswhether the principles underlying grammaticalization are the same as those tobe observed in other areas of linguistic evolution. Various attempts have beenmade to demonstrate that grammaticalization forms a process that in no waydi�ers from other kinds of language change.

I will argue that indeed grammaticalization does not have any independentstatus of its own, but rather is derivative of other kinds of language change. (Seesimilar arguments in Joseph, this volume; Newmeyer, 1998, this volume). Inresponse to this quote from Heine et al., I would say that it has been not so mucha matter of attempts being made to show that grammaticalization has noindependent status of its own, but rather that some supporters ofgrammaticalization have, usually without full cheer, recognized that there is

Table 1

Postpositions and case su�xes meaning `with' in Balto±Finnic Languagesa

Finnish Karelian Veps Estonian Votic Livonian Proto±Balto±Finnic

kanssa kanssa -ka -ga ka:sa ka:zu �kans(s)a-`with'b

`Postp' `Postp' `Com' `Com' `Postp' `Postp' `Postp'

-ka: -ka/-k/-ks

`Com' `Com'

a Com=comitative case, Postp=Postposition.b (Ultimately from �kansa(-ssa ) `(in the) people'.

L. Campbell / Language Sciences 23 (2001) 113±161116

Page 5: What’s wrong with grammaticalization?

nothing unique about the mechanisms and kinds of changes that are associatedwith grammaticalization, while at the same time others proclaim extraordinarilystrong explanatory value for grammaticalization. Thus we encounter two kinds ofcontrasting claims:

1. Supporters' claim: Grammaticalization can predict and has explanatory powerin its own right.

2. Detractors' claim: Grammaticalization has no independent status of its own;it merely involves other kinds of changes and mechanisms of change whichare well understood and are not limited to cases involvinggrammaticalization: sound change, semantic change, and reanalysis.

From the supporters we encounter claims of the following sort: ``a theory ofgrammar gains in explanatory power once it incorporates ®ndings ongrammaticalization and reanalysis'' (Heine and Reh, 1984, p. 264).``Grammaticalization studies are not only a means of relating present languagestates to past situations, rather by proposing generalizations on past developmentthey also allow us to predict future developments'' (Heine, 1993, p. 124). ``Thethings that happen in grammaticalization do so in an orderly fashion which notonly predicts what changes can occur but also puts constraints on whatsynchronic grammatical systems are found . . . it opens up a way of explaininggrammatical phenomena that has largely been neglected in post-Saussureanlinguistics'' (OÈ sten Dahl, LINGUISTIC 7.1170) (see also Heine, 1997, p. 6).

Among expressions of concern about the status of grammaticalization we notethe following. ``More and more things are recognized as belonging to generalcharacteristics of language change (instead of being speci®c togrammaticalization)'' (Lessau, 1994, p. 219). Hopper (1991, p. 21) sees as a``defect'' that the principles which Lehmann (1995[1982]) and he [Hopper] give tocharacterize grammaticalization ``are not distinctive for grammaticalization,'' butare applicable also to changes which ``are not distinctively examples ofgrammaticalization.'' He says, ``if grammaticalization is not already a given, theprinciples do not in fact identify it unambiguously'' (Hopper 1991, p. 31).Traugott and Heine (1991, p. 7) frankly acknowledge that ``grammaticalization isa subset of phenomena occurring in change,'' noting that ``a number ofmechanisms of language change [metaphorical transfer, metonymic transfer,analogy, borrowing] [are] . . .relevant to grammaticalization'' (see also Hopper,1991, p. 19; Joseph, this volume). Ramat and Hopper (1998, p. 3) note, ``anextreme formulation . . . is that ultimately grammaticalization is not separatelyde®nable from the concept of change in general. Such a position has in fact beenclaimed by Hopper (1991).'' They ask ``whether there are cases of language changethat are not part of the phenomenon of grammaticalization.'' They discuss somekinds of change that are good candidates, but they seem to come to no clearanswer to the question. Bybee et al. (1994, pp. 5±6) admit that the types of changeinvolved in grammaticalization are also known for their extensive application

L. Campbell / Language Sciences 23 (2001) 113±161 117

Page 6: What’s wrong with grammaticalization?

outside of grammaticalization and that no type of change is unique togrammaticalization, but, nevertheless, they suggest there is something specialabout the way these types of change are associated in grammaticalization changes:``The events that occur during this process [grammaticalization] may be discussedunder rubrics of semantic, functional, grammatical, and phonological changes,though we will argue that these processes are intimately connected with oneanother.'' We will see what they mean about interconnections below when we takeup claims about semantic bleaching, phonetic erosion, and metaphor, and the roleof ``emergent grammar.''

In what follows, I will lay open some of the claims and assumptions aboutvarious attributes thought to be diagnostic of or especially characteristic ofgrammaticalization, starting with that of semantic bleaching or loss. Ultimatelythis will reveal why so many see grammaticalization as having no independentstatus of its own.

3.1. Semantic change

Grammaticalization is associated in the literature with semantic ``bleaching''or loss (talked about variously also as desemanticization, fading, semanticattrition, semantic decay, semantic depletion, semantic impoverishment,weakening, generalization of semantic content, and abstraction) (cf. Meillet,1912, p. 132, 139; Lehmann, 1995 [1982], p. 127; Lessau, 1994, pp. 74±76). InLehmann's (1995 [1982], p. 129) rendition, ``grammaticalization rips o� thelexical features until only the grammatical features are left.'' For Heine and Reh(1984, p. 67), ``the more grammaticalization processes a given linguistic unitundergoes, . . . the more does it lose in semantic complexity, functionalsigni®cance and/or expressive value.'' In like vein, Matiso� (1991, p. 384)reports that ``bleaching . . .nicely captures the partial e�acement of a morpheme'ssemantic features, the stripping away of some of its precise content so it can beused in an abstracter, grammatical-hardware-like way.'' Haiman's (1991, p. 154)version is: ``semantic reduction, or bleaching, occurs as a morpheme loses itsintention: From describing a narrow set of ideas, it comes to describe an everbroader range of them, and eventually may lose its meaning altogether.'' Heineet al. (1991, pp. 108±109) call this the ``Bleaching model'': ``According to themost common interpretation, the mechanism involved may be conceived as of asa ®ltering device that bleaches out all lexical content and retains only thegrammatical content of the entity concerned'' (see also Heine et al., 1991, p. 40).

It is necessary to ask whether semantic bleaching is in any way diagnostic ofgrammaticalization, that is, whether it is an empirical property that can be tested.Clearly it is not, since it is an artifact of the de®nition of grammaticalization asthis is viewed by many, inherently and inseparably built into the very de®nition of

L. Campbell / Language Sciences 23 (2001) 113±161118

Page 7: What’s wrong with grammaticalization?

grammaticalization as a shift from more lexical meaning to more grammaticalmeaning (more precisely, a shift of lexical or grammatical to more grammatical).The change lexical > grammatical seems to imply by de®nition loss of lexicalsemantics. If this were all there were to it, we could comfortably conclude thatsemantic bleaching is just a design feature of grammaticalization, inherent in itsde®nition and not really open for empirical testing. However, not everyone seesgrammaticalization as limited precisely in this way.3 There are several kinds ofphenomena considered to be instances of grammaticalization which do notnecessarily require any shift in meaning at all, such as the changes in cohesion inthe cline free lexical> clitic> a�x (e.g. from the examples in Table 1, the shift inBalto-Finnic languages postposition > case su�x, both meaning `with' (see otherexamples in Newmeyer, this volume).4 A serious problem of a di�erent sort is thatthe kinds of semantic change that take place in association withgrammaticalization are not limited to instances of grammaticalization, but arefamiliar kinds of semantic changes (metaphor, metonymy) that take place alsobroadly outside of grammaticalization (see below for examples and discussion).

In short, semantic bleaching (loss) is neither su�cient (since the kinds ofsemantic change in grammaticalization operate in lexical change in general, notjust in grammaticalization) nor necessary (given grammaticalizations which do notinvolve meaning change or loss) for de®ning grammaticalization.

However, the emphasis on semantic bleaching has perhaps been overstated,since, as already mentioned, in grammaticalization forms can also take on new(albeit grammatical) meanings, and the source of the grammaticalization canremain with its former meaning along side the new grammaticalized form (as inthe case of be going to, directional verb and `future'). Rather than semantic loss,some scholars stress metaphor and metonymy, with possible polysemy as theoutcome of grammaticalization (i.e. `layering') (see Sweetser, 1988, 1990; Traugott,1988; Hopper and Traugott, 1993; cf. Heine et al., 1991, p. 72). As Heine et al.(1991, p. 109) indicate, ``other approaches have emphasized . . .thatgrammaticalization involves not only loss but also gain''; they call Sweetser'sapproach the ```loss-and-gain' model'' (Heine et al., 1991, p. 110). Traugott (1988,

3 For example, the notion of `layering' (that a form can acquire a new meaning without necessarily

losing its old meaning, becoming polysemous with both the old and new meanings surviving along side

each other) allows for instances of lexical > grammatical where after the change there is in fact no

necessary loss of lexical meaning, but there is a gain of new grammatical meaning. The be going to

example illustrates this; before grammaticalization it meant only a motion verb (with purpose) (e.g. I'm

going to the store ), but after grammaticalization maintained its lexical meaning (I'm going to the store

still exists and still means `go', a motion verb), but also acquired a grammatical meaning, `future' (e.g.

I'm going to go to the store `I will go to the store').4 See also Hopper and Traugott (1993, p. 62): ``the subset of processes that are particular to gramma-

ticalization are those that over time render more independent elements less independent.''

L. Campbell / Language Sciences 23 (2001) 113±161 119

Page 8: What’s wrong with grammaticalization?

p. 407) believes that ``bleaching and grammaticalization must be uncoupled''; sherecognizes that ``certainly, bleaching does occur,'' but argues that this takes place``only in the later stages of grammaticalization.'' In Hopper and Traugott's (1993,p. 68) view, it is in later stages of grammaticalization that ``meaning loss orbleaching takes place,'' where ``the initial phase of grammaticalization involves ashift in meaning, but not loss of meaning'' (1993, p. 89). For them, the meaningchanges in early stages of grammaticalization are pragmatic and associative(metonymic), arising in the ``context of the ¯ow of speech'' (1993, p. 68).5 Theyspeak of ``pragmatic enrichment'' and ``strengthening (of informativeness)'' at thebeginning of grammaticalization, rather than semantic loss (Hopper and Traugott,1993, p. 88; see also Traugott and KoÈ nig, 1991; Heine et al., 1991, p. 70); ``there isno doubt that over time, meanings tend to become weakened during the processof grammaticalization. Nevertheless, all the evidence for early stages is thatinitially there is a redistribution or shift, not a loss, of meaning'' (Hopper andTraugott, 1993, p. 88).

A related, but ultimately very di�erent view is that of Haspelmath (1999, p.1062), who expresses doubt ``that semantic grammaticalization is as central to theprocess as has generally been assumed.'' Instead, he sees semantic bleaching/generalization as a prerequisite to grammaticalization, ``in a sense the cause of theother processes of grammaticalizations.'' He thinks that:

Semantic bleaching or generalization . . .does not seem to be a consequence ofroutinization, unlike the phonological and syntactic changes of grammaticalizeditems, but a prerequisite for it . . .a lexical item can become grammaticalizedonly if it is used in a basic discourse function, because otherwise it would notincrease signi®cantly in frequency. For instance, the semantic change in Englishgoing to from the spatial sense to the future sense made it possible for the itemto become reduced to gonna and to develop further properties typical of afunctional category . . . it is not so much that semantic bleaching andphonological reduction go hand in hand, but semantic generalization is in asense the cause of the other processes of grammaticalization . . .semanticgeneralization or bleaching is usually a prerequisite for use in a basic discoursefunction, that is, for the increase in frequency that triggers the other changes(Haspelmath, 1999, p. 1062).

This, of course, begs the question of whether there are any substantiveconsequences which might distinguish this view with semantic shift as outside but

5 For many others, metaphor plays a central role in the semantic changes associated with grammati-

calizations. However, while Hopper and Traugott (1993, pp. 80±81) admit metaphor as one process

sometimes involved, they hold metonymy to be more important: ``since reanalysis, not analogy, has for

long been recognized as the major process in grammaticalization at the structural, morphosyntactic

level, it would be surprising if metaphor, which is analogical, were the prime process at work pragmati-

cally and semantically.'' Moreno Cabrera (1998), in contrast, sees grammaticalization as a metaphorical

process but lexicalization as a metonymical process.

L. Campbell / Language Sciences 23 (2001) 113±161120

Page 9: What’s wrong with grammaticalization?

prerequisite to grammaticalization from others which see it as part of thegrammaticalization process itself, though perhaps only the initial stage of it. It isunlikely that anything of crucial empirical consequence hinges on these di�erentoutlooks.

To summarize: through a di�erent route, Haspelmath has arrived at the sameview as that of detractors of grammaticalization, namely, that the semantic changeis not a crucial part of the de®nition of grammaticalization, since it operates freelyalso outside of grammaticalization and can have e�ect even when nogrammaticalization is involved at all. Given this independence, semantic loss is notdiagnostic for grammaticalization.

3.2. Phonological reduction (phonological attrition, phonological weakening, phoneticerosion, phonetic loss, Abnutzung )

Many also see phonological reduction as inherently connected withgrammaticalization and some consider it a diagnostic trait of grammaticalization(see, for example, Bybee et al., 1994), though others do not. Typical is Heine andReh's (1984, p. 67) outlook, that ``the more grammaticalization processes a givenlinguistic unit undergoes, . . .the more does it lose in phonological substance''.Heine (1993, p. 106) elaborates: ``reduction, or erosion, . . . is in fact predicted bygrammaticalization theory: Once a lexeme is conventionalized as a grammaticalmarker, it tends to undergo erosion; that is, the phonological substance is likely tobe reduced in some way and to become more dependent on surrounding phoneticmaterial.''

Like semantic bleaching, phonological reduction (erosion) ``is neither anecessary or a su�cient property of grammaticalization'' (Lessau, 1994, p. 263).Grammaticalization can take place with no phonetic reduction, and erosion ofform is not unique to grammaticalization, but is normal phonological change.Phonological reduction processes apply to items in appropriate phonologicalcontexts generally, not just to items involved in grammaticalization. As Lehmann(1995 [1982], p. 127) acknowledges, ``it is obvious that phonological attrition isomnipresent in linguistic change. It plays its role not only in grammaticalization,but a�ects, in the long run, practically every sign'' (See also Heine and Reh, 1984,p. 25). Heine (1994, p. 109) makes very clear that phonological reduction is notnecessary in grammaticalization: ``conceptual grammaticalization precedes[phonetic] erosion. This means that at a given stage, a morpheme may become®rmly established as a grammatical marker although its phonetic substance is stilluna�ected by this process.'' His examples and discussion leave little doubt on thisscore:

The High German auxiliaries haben, `have,' and sein, `be,' had assumed theirgrammatical functions as Perfect or Passive markers many centuries ago at thestage of Old High German . . . , yet their phonetic shape is still identical withthat of their respective lexical items . . .erosion is in no way con®ned to

L. Campbell / Language Sciences 23 (2001) 113±161 121

Page 10: What’s wrong with grammaticalization?

grammaticalization . . . ; rather it is a natural process to be observed in all kindsof language development and it is not always possible to decide unambiguously

which kind of forces were involved in a given case. To conclude, while theBybee/Perkins/Pagliuca hypothesis [that the development of grammatical

material is characterized by the dynamic coevolution of meaning and form, see

below] holds true on statistical grounds, the reader must not be surprised to®nd numerous exceptions once he or she looks at individual instances of

grammaticalization. (Heine, 1994, p. 109.)

However, for some the role of phonological reduction in grammaticalization is

not so straightforward, and several scholars hold there is an inherent connectionbetween phonetic erosion and semantic bleaching. For example, this connection is

central to Whaley's (1997, p. 285) de®nition of grammaticalization: ``a process of

language change by which a free lexical morpheme becomes semanticallygeneralized and phonologically reduced.'' Bybee and Pagliuca (1985, p. 76) call

this the Parallel reduction hypothesis: ``As the meaning generalizes and the rangeof uses widens, the frequency increases and this leads automatically to

phonological reduction and perhaps fusion.'' Bybee et al. (1994, p. 6) lay out thisnotion in the following way:

Certain components of meaning are lost in this process and so it can also be

called semantic reduction, in explicit parallel to the phonological reduction

which grammaticizing material undergoes . . .Parallel to semantic reduction,phonological reduction continues to take place throughout the life of a gram

[grammatical morpheme] (emphasis in the original).

They say:

It therefore seems natural to look for a direct, and even causal, link betweensemantic and phonetic reduction in the evolution of grammatical material,

beginning with the earliest stages of development from lexical sources andcontinuing throughout the subsequent developments grams undergo. Our

hypothesis is that the development of grammatical material is characterized by

the dynamic coevolution of meaning and form (Bybee et al., 1994, p. 20).

They conclude: ``our results . . .strongly support the hypothesis that form andmeaning covary in grammaticalization'' (Bybee et al., 1994, p. 21; cf. also

Pagliuca, 1994, p. ix.) Givo n (1975, p. 96), apparently the inspiration, at least inpart, behind the ``parallel reduction hypothesis,'' linked semantic reduction and

phonetic reduction; he saw this as the explanation for unidirectionality (seebelow): ``when a verb loses much of its semantic content and becomes a case

marker, in due time it also loses much of its phonological material, becomes a

bound a�x, and eventually gets completely eroded into zero. It is thus unlikelythat a more crucial portion of the information content of the utterance . . .will be

entrusted to such a reduced morpheme.'' Haspelmath (1999, p. 1050) also links

L. Campbell / Language Sciences 23 (2001) 113±161122

Page 11: What’s wrong with grammaticalization?

semantic and phonological reduction, elaborating Givo n's conception of it; hesays, ``as an item is desemanticized, it is also formally reduced, and nobody wouldexpect an element to become formally reduced but semantically enriched.''Haiman (1991, pp. 153±154) maintains ``there are two kinds of change which arealways associated with grammaticalization: phonetic reduction and semanticbleaching or generalization,'' which he says ``frequently but not invariably gotogether.''

Some attempted ``explanations'' for phonetic erosion in the context ofgrammaticalization have been put forward. For example, Heine et al. proposetwo, quantity iconicity and frequency:

There are at least two possible factors that have been held responsible forerosion. One is described by Givo n (1990) as the quantity principle, a principleof iconic coding according to which a larger chunk of information will be givena larger chunk of code: since lexical forms contain more information thangrammatical forms, the chunk of code employed for their expression is likely tobe reduced when they are grammaticalized. The second factor relates to relativefrequency of use: the higher frequency of use of grammatical morphemes favorswhat von der Gabelentz ([1891] 1901) has called the Abnutzung (abrasion) oftheir phonetic substance (Heine et al., 1991, p. 214).

Frequency may play a role, but by no means a very de®nitive one. The mostfrequent lexical verbs of English exhibit little tendency towards phonetic reduction(do, go, come, see, think ). If iconicity (resulting in quantity considerations) hasany deterministic or causal role to play, it is at such an abstract level and usuallytakes so much time to become implemented (if ever), that it becomes essentiallyimpossible to demonstrate a real causal connection, except perhaps in some roughstatistical sense of grammatical markers tending to be shorter than typical lexicalmorphemes. Still, this provides no reliable gauge when it comes to theconsideration of individual forms in particular languages. I would argue, rather,that phonological reduction often follows grammaticalization (sometimes onlyafter a very long time interval) precisely and mostly only because it is then thatthe conditions favorable to the sound changes which result in phonologicalreduction ®rst come about, e.g. where forms (with little or no independent lexicalmeaning) come to be in relatively unstressed positions Ð but this is not reallyabout grammaticalization; it is about standard phonological change of any formthat happens to satisfy the environment of phonological reduction processes (seeNewmeyer, this volume, for additional discussion).

In summary, like semantic bleaching, phonological reduction cannot beconsidered diagnostic of grammaticalization, since as seen in the examplesconsidered above it is neither necessary (given grammaticalizations where nophonological reduction has taken place) nor su�cient (since much phonologicalreduction takes place in total absence of grammaticalization, e.g. various sound

L. Campbell / Language Sciences 23 (2001) 113±161 123

Page 12: What’s wrong with grammaticalization?

changes result in loss of phonological material but are totally independent ofgrammar and meaning Ð to cite just one example, the well-known sound changein Spanish which led to loss of phonetic material: f> h>é as in ®lo> hilo> /ilo/`thread').

3.3. Unidirectionality

Givo n (1975, p. 96) is said to be the ®rst to consider unidirectionality expressly.Unidirectionality is the hypothesis that grammaticalization is irreversible, that``grammatical elements do not turn back in the direction of the lexicon'' (Ramat,1998, p. 115). As Lessau (1994, p. 885) points out, ``unidirectionality is certainlyone of the strongest claims about grammaticalization and [is] often cited as one ofthe basic principles of grammaticalization.'' Ramat (1998, p. 115) a�rms the samething: ``The claim that grammaticalization is irreversible, that allgrammaticalization involves shifts . . .from lexical item to grammatical item, orfrom less grammatical to more grammatical item . . . is one main tenet ofgrammaticalization theory.'' A fact not usually recognized explicitly bygrammaticalization enthusiasts is that unidirectionality is essentially built into thede®nition of grammaticalization. Herring (1991, p. 253) spells this out:

One of the foremost tenets in diachronic grammaticalization theory today is thenotion of unidirectionality, according to which change in meaning from less tomore grammatical is viewed as a linear and irreversible process. De®ned thus ingeneral terms, unidirectionality might appear to be a simple description of,rather than a claim regarding the nature of, the grammaticalization process.

In brief, because grammaticalization is de®ned as changes of lexical >grammatical, or grammatical > more grammatical, any change not going in thisdirection can be considered as outside of grammaticalization, and therefore,unidirectionality becomes not an empirical hypothesis that can be tested, but anartifact of the de®nition itself. (See the other papers in this volume.)

Nevertheless, views concerning unidirectionality in the grammaticalizationliterature di�er considerably, where some treat it as a hypothesis (an empiricalproperty) and others take it for granted (as an axiom, a de®ning property). Forexample, some de®nitions of grammaticalization merely assert unidirectionality:```Grammaticalization' . . .refers to the dynamic unidirectional historical processwhereby lexical items in the course of time acquire a new status as grammatical,morphosyntactic form'' (Traugott, 1988, p. 406). ``Grammaticalization as viewedfrom the diachronic perspective is hypothesized to be prototypically aunidirectional phenomenon'' (Hopper and Traugott, 1993, p. 94; see alsoHaspelmath, 1998). Heine et al. (1991, p. 4) consider unidirectionality a``characteristic that is implicit'': ``an intrinsic property of the process is thatgrammaticalization is unidirectional, that is, that it leads from a `less grammatical'to a `more grammatical' unit, but not vice versa.'' Hopper and Traugott (1993, p.

L. Campbell / Language Sciences 23 (2001) 113±161124

Page 13: What’s wrong with grammaticalization?

126) add, ``the strong hypothesis of unidirectionality claims that allgrammaticalization involves shifts in speci®c linguistic contexts from lexical togrammatical item, or from less to more grammatical item, and thatgrammaticalization clines are irreversible. Change proceeds from higher to lower,never from lower to higher on the cline'' (see also Bybee et al., 1994; Fischer,1997, p. 173; Heine, 1993, p. 53; Heine et al., 1991, p. 212; Hopper and Traugott,1993, p. 6; Traugott and Heine, 1991, p. 4; Traugott, 1995). The opposite extremeis represented by Janda's paper (this volume), with its long list of counterexamples to unidirectionality. Some other treatments fall between these extremes,acknowledging exceptions, but often playing down their signi®cance; e.g.``Extensive though the evidence of unidirectionality is, it cannot be regarded as anabsolute principle. Some counterexamples do exist. Their existence, and theirrelative infrequency, in fact help de®ne our notion of what prototypicalgrammaticalization is'' (Hopper and Traugott, 1993, p. 126). ``Counter-examplesare rare, but not as rare as some scholars seem to assume'' (Ramat, 1998, p. 115)(see also Ramat and Hopper, 1998, p. 6). ``Unidirectionality in grammaticalchange is a tantalizing idea that has both substantial empirical plausibility andappealing theoretical simplicity, but it lacks tooth in its current formulationbecause of a number of hard-to-dismiss counterexamples'' (Tabor and Traugott,1998, p. 229).

It is revealing to consider how counterexamples to the unidirectionality claimhave been treated (see below for examples). Most typically they have been treatedthe way accident claims are handled by many insurance companies: denied orignored in hopes they will go away. Lehmann (1995 [1982], p. 19) asserts that ``nocogent examples of degrammaticalization [reversal of directionality] have beenfound.'' He asserts that ``for degrammaticalization [more grammatical > lessgrammatical] to obtain, analytical forms would have to be historical continuantsof synthetic forms; but this actually never occurs'' (Lehmann, 1995, p. 20).However, this is just wrong; a number of examples of bound forms becomingunbound are known (several are cited in the examples of exceptions tounidirectionality in this paper; see Newmeyer, this volume for others). Heine et al.(1991, pp. 4±5) admit that examples of ``degrammaticalization [present when thedirection of grammaticalization is reversed, that is, when a more grammatical unitdevelops into a less grammatical one] . . .have been observed to occur,'' but theyassert that ``they are statistically insigni®cant and will be ignored in the remainderof this work.'' ``Examples like these [the Estonian given below and others of`degrammaticalization'] are, however, rare and will be ignored'' (Heine et al.,1991, p. 52). It is interesting that earlier, Heine and Reh (1984, pp. 74±76)recognized that their ``approach is based on the assumption that languagedevelopment is uni-directional'' but nevertheless noted that ``there is evidence tosuggest that this evolution is not without exceptions: under certain circumstances,basic processes can be reversed.'' They discuss several examples ``which suggestthat the uni-directionality principle may be violated'' under the categories of ``re-semanticized'' linguistic units (cf. Hopper and Traugott's ``lexicalization''),``decliticization'' (found only in ``hypothetical data'' involving an Indo-European

L. Campbell / Language Sciences 23 (2001) 113±161 125

Page 14: What’s wrong with grammaticalization?

reconstruction), and where ``an in¯ectional structure assumes a lessgrammaticalized function'' (e.g. tense> subjunctive) (see also Ramat, 1992).

Haspelmath would seem to want his cake and to eat it too, seeminglyequivocating between denying the existence of such counterexamples and assertingthat they exist but are of no relevance. For example, on one occasion he says, ``itis an undeniable empirical fact that such changes [the reverse ofgrammaticalization changes] do not occur, and that change in grammar isoverwhelmingly in one direction'' (1998, p. 53). But in the same article he says,``there are at most a few cases . . .I see no reason to regard these isolated cases asthreats to the robust empirical generalization that grammaticalization isoverwhelmingly unidirectional'' (Haspelmath, 1998, p. 80). Ramat's (1998)treatment is somewhat similar to Haspelmath's. She discusses a number of kindsof counterexamples which, she says, represent ``a serious challenge to theunidirectionality hypothesis'', but nevertheless she opts to maintainunidirectionality. She says, ``for conceptual clarity they [the kinds ofcounterexamples to unidirectionality she discusses] should be kept apart fromcases of grammaticalization as a process proceeding from lexical forms togrammatical forms along a unidirectional cline'' (Ramat, 1998, p. 123). Sheconsiders the ``possibility [which] would be to reject unidirectionality as anecessary condition to decide what changes a grammaticalization theory shouldexplain,'' but rejects this because:

I feel . . .what we call ``grammaticalization theory'' would be left with a toovague de®nition of its ®eld, including almost every instance of change. Theunidirectionality of changes from lexical categories to grammatical (functional)categories constitutes a signi®cant constraint on possible language changes. Inthe light of this constraint, possible counterexamples can be excluded becausethey do not adhere to the sequence of changes entailed in grammaticalization(Ramat, 1998, p. 123).

See Moreno Cabrera (1998, p. 224) for a similar view; Tabor and Traugott's(1998) approach is also similar, but much more complicated; see below. To me,Ramat's decision is curious. She recognizes counterexamples to unidirectionality ifthis is taken as an empirical hypothesis, but opts in spite of these to maintainunidirectionality because of its appeal for de®ning grammaticalization, thus givingup any hope of empirical underpinnings for the concept.

It is easy to appreciate the predicament scholars face with respect tounidirectionality and counterexamples to it. On the one hand, unidirectionality isbuilt into the de®nition of grammaticalization (i.e. it is a de®nitional property). Inthe core prototypical de®nition of grammaticalization as instances of lexical >grammatical, or grammatical > more grammatical, any potential counterexample(possible instances of grammatical > lexical, or more grammatical > lessgrammatical ) is simply shut out by de®nition Ð if examples of these non-licensed

L. Campbell / Language Sciences 23 (2001) 113±161126

Page 15: What’s wrong with grammaticalization?

changes exist and if this de®nition of grammaticalization is maintained, then suchcases are not considered real counterexamples to grammaticalization, but ratherjust fall outside the universe of discourse. That is, in this conception, theunidirectionality claim has no empirical content whatsoever; it cannot be testedbut rather is inherent in the very de®nition of grammaticalization. As Lessau(1994, p. 886) explains, ``if `unidirectionality' is an empirical property ofgrammaticalization, it follows, that any case of grammaticalization which runs inanother direction quali®es as a counterexample to the `unidirectionality'principle . . .If, however, `unidirectionality' is a de®ning property ofgrammaticalization, it follows, that any case of linguistic change which runs inanother direction is simply not a case of grammaticalization by de®nition.''

(More speci®c attempts to explain away the exceptions to unidirectionality areconsidered below, after examples of these exceptions have been presented.)

3.3.1. Counterexamples to unidirectionalityIn order to give the discussion more substance, let us turn now to some cases of

counterexamples to the unidirectionality claim, before considering furtherproposed explanations of these and what they mean for grammaticalization ingeneral. The following are some of the better known counterexamples.

(1) English genitive -'s: Jespersen's (1894, pp. 317±318) account is revealing:In modern English . . .(compared to OE) the -s is much more independent: itcan be separated from its main word by an adverb such as else (as insomebody else's hat ), by a prepositional clause such as of England (as in thequeen of England's power ), or even by a relative clause such as I sawyesterday (as in the man I saw yesterday's car ) . . .the English genitive is infact no longer a ¯exional form . . .historically attested facts show us in themost unequivocal way a development Ð not, indeed, from an originally self-existent word to a mere ¯exional ending, but the exactly oppositedevelopment of what was an inseparable part of a complicated ¯exional systemto greater and greater emancipation and independence (Jespersen's emphasis)(see also Janda, Norde, this volume).

(2) Middle Swedish -s `genitive in¯ection'>Modern Swedish phrasal clitic (seeNorde, 1997, pp. 1±3, 8, this volume).

(3) Lapp (Saami) case su�x> independent word: In Northern Lapp, -taga theabessive case su�x (`without') inherited from Proto-Finno-Permic (and still asu�x in all the other Finno-Permic languages), has become a clitic. Its statusas a clitic (and not an a�x) is shown by the facts, among others, that (1) itgoverns the genitive case attached to the preceding noun phrase, (2) itpermits conjunction reduction, and (3) it attaches outside possessive enclitics.In EnontekioÈ Lapp it has become a postposition and an independent adverb,no longer requiring a host for cliticization, as in mun baÂhcen taga [I gowithout] `I remain without', i.e. an instance of case a�x> clitic >postposition> independent adverb (Nevis, 1985, 1986).

L. Campbell / Language Sciences 23 (2001) 113±161 127

Page 16: What’s wrong with grammaticalization?

(4) Estonian former -pa/-paÈ `emphatic' (bound)> ep `emphatic' (free): Iparaphrase Ariste's (1973, p. 37) account: Estonian has an independenta�rmative adverb ep `yes, indeed, just so, then', which developed from thebound a�rmative su�x -p, -pa, -paÈ, as in seep `that is it indeed'> see ep,pealle-p `on top of indeed' (from paÈ aÈ llaÈ -paÈ ) > peall ep , and after this theword ep could change its syntactic position and precede in the sentence, e.g.ep see, ep peall (Campbell, 1991).

(5) Estonian question marker: -s> es `question marker' (su�x> independentword ), in the same way as in example (4) (Campbell, 1991).

(6) In Japanese bound grammatical markers> free discourse markers(clause-®nal -ga concessive subordinate detached and developed into a clause-initial weakly adversative conjunction) (Matsumoto, 1988; see Newmeyer, thisvolume).

(7) Ilokano to/nto in¯ectional future> separate lexical particle, a�rmativeresponse to a question (in response to `Come tomorrow, OK?', To `I'll dothat') (Rubino, 1994; see Newmeyer, this volume).

(8) Irish ®rst person plural su�x -mid/-muid was freed from being restricted as abound su�x to become an independent pronoun, replacing the earlierindependent ®rst person plural pronoun sinn (Bybee et al., 1994, pp. 19±20).

(9) Dutch -tig> tig `many times' (segmented from forms such as zestig `60',negentig `90', etc., like English umpteen ) (Norde, 1997, pp. 3±4, this volume).

(10) A much repeated example is English down, in, out, up (prepositions) > verb,noun, adjective, e.g. verb to down (as in they downed one [in hunting, sportsevents]), noun a down (in football, the act of computers going down, anegative turn/trend), adjective down (as in down time, down side ); the in thing,that's really in; to out someone (reveal that someone is homosexual), an out(in sports, an excuse/solution); to up (increase the price, etc.). A compellingexample of this sort is from a dialogue in Toni Morrison's Song of Solomon(1977, p. 161): ``you can o� [= kill] anybody you don't like. You can o� me.''``We don't o� Negroes.''6 (cf. Hopper and Traugott, 1993, p. 49; Ramat,1998, pp. 115±116; Newmeyer, this volume; Itkonen, 2000; see also Janda,Norde, this volume; see below.)

(For discussion and several additional kinds of counterexamples, see Joseph andJanda, 1988; Ramat, 1998; Moreno Cabrera, 1998; cf. also Heine, 1998, pp. 94±96, and the other papers of this volume. Janda, this volume, cites more than 70works critical of unidirectionality.)

Some attempts to explain away some of these exceptions have been made in ane�ort to save unidirectionality (and consequently certain versions ofgrammaticalization). I ®nd many of these attempts unsuccessful, moreterminological than substantive. Lehmann (1995 [1982], pp. 16±19) discussedseveral examples of what he calls ``degrammaticalization,'' i.e. counterexamples to

6 I thank Alice Harris for this example.

L. Campbell / Language Sciences 23 (2001) 113±161128

Page 17: What’s wrong with grammaticalization?

unidirectionality, and attempted to show that, for one reason or another, all haveproblems which eliminate them as proper examples of `degrammaticalization'. Hedid successfully show that several putative examples have probably beenmisinterpreted and are not legitimate counterexamples to unidirectionality.However, he was unaware of most of the counterexamples presented above (andelsewhere in this volume; see especially Janda, and Newmeyer), except (1), whichhe attempts to explain away as involving a reinterpretation based on thehomophony between possessive -'s and reduced forms of possessive pronoun his(e.g. Adam's son reinterpreted as Adam his son ). However, this interpretation hasbeen shown to be wrong. Through an extensive examination of Middle and EarlyModern English texts, Cynthia Allen (1997) demonstrated conclusively that theability of the genitive ending to be separated from its original nominal head (e.g.as in the king's of England daughter > the king of England's daughter ) could nothave been due to a reanalysis of the (separated) genitive based on the pronounhis, as some (including Lehmann's account) have suggested, but rather was anorthographic variant of the old in¯ectional genitive, which was reanalysed as aclitic Ð a true counterexample to the unidirectionality claim (see Allen, 1997 fordetails; see also Tabor and Traugott, 1998, p. 40; see also Janda more detaileddiscussion of the reanalysis, and Norde, this volume).

Hopper and Traugott (1993, p. 127) treat some of these counterexamples[especially (4) and (10) above], seeing them as instances of reanalysis which resultin what they call `lexicalization,' ``a shift from grammatical to lexical structure''(1993, p. 49), implying that this is a di�erent process and therefore not a matter ofcounterexamples to unidirectionality (see also Traugott and Heine, 1991, pp. 6±7).They say:

Probably the most often cited putative counterexamples [to unidirectionality]are those involving the lexicalization of grammatical items, as in to up theante . . . [see example (10) above]. Since lexicalization is a process distinct fromgrammaticalization, is not unidirectional, and can recruit material of allkinds . . .data of this kind can and, we believe, should be considered examplesof the recruitment of linguistic material to enrich the lexicon. In other words,they are not part of grammaticalization as a process, but rather oflexicalization'' (Hopper and Traugott, 1993, p. 127).7

7 The term ``lexicalization'' has been used elsewhere in the grammaticalization literature with a di�er-

ent meaning. Lehmann (1995 [1982], p.136) uses ``lexicalization'' to refer to cases he considers ``the last

phase of grammaticalization,'' fossilized forms which fall out of a former paradigm, such as English

where, how, why, etc., ``which are the fossilized remnants of cases that have fallen out of the pronominal

paradigm.'' These fossilized forms, he says, ``can indeed no longer be obtained according to the rules of

the grammar, but must be listed in the lexicon'' (p. 136) (see also Heine and Reh, 1984, p. 35; Traugott

and Heine, 1991, p. 7). Moreno Cabrera (1998, p. 214) de®nes ``lexicalization'' as ``the process creating

lexical items out of syntactic units'' Ð ``a lexically-oriented or lexicotelic process,'' which ``feeds the lex-

icon and bleeds the syntax'' (p. 218).

L. Campbell / Language Sciences 23 (2001) 113±161 129

Page 18: What’s wrong with grammaticalization?

They believe the Estonian examples (4) and (5) (above) are cases of``lexicalization.''8

Heine et al. (1991, p. 50) elaborate this notion. They say, ``while development ingrammatical morphemes is unidirectional, leading from `more concrete' to `moreabstract' meanings . . .developments in the lexicon do not undergo such aconstraint.'' They illustrate the distinction they are attempting to make with theSwahili example in which mbele `breast' > `frontside, front part' > locative (`infront [of]') and temporal (`before'). They see these as grammaticalizing changes,which are unidirectional, ``becoming increasingly `more abstract'.'' But thisexample also illustrates change in the other direction: ``it [mbele ] has assumed a`more concrete' meaning in that it was used as the vehicle of a taboo metaphorand has acquired the meaning (`the front' > `the front part'>) `male sexualorgans'.'' That is, developments involving mbele go both directions, lexical >grammatical (as in `breast, frontside' > `in front of') and grammatical > lexical(`in front of' > `male sexual organs'), but nevertheless, this example is stillconsidered to obey unidirectionality and the change which seems to be in theproscribed direction ( grammatical > lexical ) is somehow exempted as acounterexample because it has been given the label `lexicalization'. However, theact of labeling the exceptions (as `lexicalization') does not change the fact thatthey are exceptions.9

Moves such as these would make the unidirectionality claim essentiallyuntestable and devoid of empirical content. Unpacked, the claim seems to go likethis: changes of lexical > grammatical are called `grammaticalization' and areunidirectional, by de®nition; changes of grammatical > lexical are called`lexicalization', and while they would appear to go against the unidirectionalassumption of `grammaticalization', because they are given a di�erent name,`lexicalization', they can be considered, again by de®nition, not really to be

8 Traugott and Heine (1991, p. 7) say that these examples ``can be regarded as instances of reanalysis''

and therefore ``strictly speaking not cases of grammaticalization.'' The logic of this conclusion escapes

me; they a�rm that reanalysis is the principal mechanism of grammaticalization, yet decide that

because reanalysis is involved in these examples they therefore are not cases of grammaticalization.9 Ramat (1998, p. 121) argues instead that ``grammaticalization'' and ``lexicalization'' are not clearly

distinct: ''I argue that . . .the limits between grammatical elements and lexemes may be blurred, and pro-

pose to rethink the traditional view according to which grammaticalization and lexicalization are quite

distinct, even opposite processes. Rather, they seem to be complementary or overlapping processes of

change . . .grammatical materials may become lexical through a number of developments.'' For Hopper

(1998, p. 148), lexicalization and grammaticalization are also not distinct, but because he sees grammar

and grammaticalization as united under the notion of ``emergent grammar'' which then brings most

aspects of language change together under grammaticalization. This is discussed later in this paper. (See

Norde, this volume, for a di�erent view which does distinguish between ``degrammaticalization'' and

``lexicalization.'' See Ramat, 1992 for discussion and examples of ``lexicalization.'')

L. Campbell / Language Sciences 23 (2001) 113±161130

Page 19: What’s wrong with grammaticalization?

counterexamples to the unidirectionality claim. True, how languages recruit newlexical items is important, and not all of the means utilized involvegrammaticalization; however, instances of the conversion of grammatical materialinto lexical items are clearly against the unidirectionality claim, regardless ofwhether such examples are given a new name (`lexicalization') or not.

In any case, there are counterexamples which do not involve lexicalization, andthis makes calling upon lexicalization as a ploy to get around counterexamplesseem like even more of a red herring. For example, there are cases which do notresult in lexicalization which go in the wrong direction according to theunidirectionality claim on various of the grammaticalization clines. The exampleof Estonian question markers in (5) above is a clear case: -s [bound clitic]`question marker' > es [independent word] `question marker'. Here, both aregrammatical (not lexical content) morphemes, with essentially the same meaning/function, but the change goes against the direction predicted by the ``cline ofgrammaticality'': content item (lexical) > grammatical word > clitic > a�x(Hopper and Traugott, 1993, p. 7). Several other of the counterexamplespresented above are similar [e.g. (1), (4), (6), (7), and (8)].10

Hopper and Traugott's (1993, pp. 128±129) general conclusion is not totallydismissive of the counterexamples, though they are clearly not eager to accordmuch importance to them:

Unidirectionality is a strong hypothesis. The evidence is overwhelming that avast number of known instances of the development of grammatical structuresinvolved the development of a lexical item or phrase through discourse use intoa grammatical item, and then into an even more grammatical item, and thatthese changes were accompanied by decategorization from a major to a minorcategory. Counterexamples are few. All are of a speci®c type: moregrammatical items become less so. To date there is no evidence thatgrammatical items arise full¯edged, that is, can be innovated without a priorlexical history in a remote (or less remote) past.

Though this is something of an aside Ð though one relevant to the generalclaims being made Ð it is not strictly true that there is no evidence ofgrammatical items which arise without prior lexical history, if by this they mean,as I assume they do, the ``cline of grammaticality'' with ( full ) lexical item >grammatical item as the lexical history involved. Consider examples such as thefollowing:

10 It should be noted that not all instances claimed to involve unidirectionality involve the same thing,

or, stated di�erently, it is claimed that there is unidirectionality in di�erent spheres, e.g. in semantics

(concrete> abstract, lexical > grammatical), in syntax (free> bound), in pragmatics (pragmatic > syn-

tactic), and in phonology (fuller more>more eroded) (cf. Lessau, 1994, pp. 888±890).

L. Campbell / Language Sciences 23 (2001) 113±161 131

Page 20: What’s wrong with grammaticalization?

[1] Swedish ni `you' (plural, formal): from Old Swedish I `you' (cognate to Englishye ), which often appeared after verbs which ended in the -n `plural agreementmarker', and the -n+I combination was reinterpreted together as being thepronoun ni: veten I> vet ni `you know'; vissten I> visste ni `you knew'(Wesse n, 1969, p. 219).

[2] Colloquial French question particle ti: this began with subject-verb inversionquestions, as in: (1) est-il? `is he?', (2) dort-il? `does he sleep?', (3) aimet-il?`does he love?', (4) aimerait-il? `would he love?' In these, where the subjectwas il `he', the ®nal l eroded in colloquial pronunciation, leaving these endingphonetically in [ti ], a combination of the frequent t at the end of many verbs[sometimes part of the verb root, as in (1) and (2), and sometimes part of theverbal morphology, as in (3) and (4)] + the i which remained from thepronoun il. This ti was reanalysed as a question marker and later extended,used in questions in general, not only in those with third person masculinesubjects, as in (5) and (6): (5) les ®lles sont ti en train de dõÃner? [the childrenare QUESTION in way of to dine] `Are the children eating dinner'? (6)tu vas ti? [you go QUESTION] `Are you going?' (Harris and Campbell, 1995,pp. 55±56; mentioned in another context by Newmeyer, this volume, and alsoby Haspelmath, 1998).

[3] Finnish dialects, Estonian, Votic, and Livonian `past' -si: this new `past tense'marker comes from the -t which is part of the root of many verbs + -i the old`past' morpheme, where a phonological rule in these languages turns t before iinto s (t4 s/_i ). Some examples from Finnish dialects are: astu-s(i ) `stepped',istu-s(i ) `sat'; compare Standard Finnish astu-i, istu-i (by analogy to thoseverbs with t in the root which legitimately would have s before the past tensemarker i, e.g. kielti4 kielsi `forbad', makati4makasi `lay down')(Ravila, 1975, p. 86; Laalo, 1988, pp. 5, 47±49).

In e�ect, the grammatical morphemes in examples such as these have no ``priorlexical history,'' but rather are combined of phonological parts of more than onemorpheme (some of which were not themselves `lexical' morphemes to start with,but were grammatical).

Finally, in this context, it is valuable to point out that I have been discussingunidirectionality as though it were understood basically in the same way bygrammaticalizationists; however, the notion of unidirectionality has been appliedin very di�erent ways to di�erent phenomena, which include the following (seeTabor and Traugott, 1998, p. 230): (1) semantics/pragmatics: more referential >less referential meaning, not vice versa (Traugott and KoÈ nig, 1991); (2) semantics:less abstract > more abstract meaning (Sweetser, 1990; Heine et al., 1991; Bybeeet al., 1994); (3) phonetic form: phonetic substance is reduced, not increased(Heine and Reh, 1984; Bybee et al., 1994; Haspelmath, 1998); (4) statisticalnature: frequency of occurrence increases, not decreases (Bybee et al., 1994;Haspelmath, 1999; Heine et al., 1991); (5) structural type cline: discourse > syntax

L. Campbell / Language Sciences 23 (2001) 113±161132

Page 21: What’s wrong with grammaticalization?

>morphology (cf. free lexical item> clitic > a�x) (Givo n, 1979; Lehmann, 1995[1982]). It is perfectly plausible that the directionality posited for one of theseareas may be in con¯ict with that postulated for another of these.

While not acknowledged as such, di�erent views can result in violations of thespirit of unidirectionality. For example, as in Haspelmath's (1998, 1999) view, ifthe drive towards ease of perception can result in new ``semantic-syntactic''material being added (by way of new analytic constructions), then clearly thesecases go against the general directionality in grammaticalization postulated forphonetic reduction and of free forms becoming more bound, though they doinvolve new grammatical elements coming into the language. (This point is takenup again below.) More compelling counterexamples to the spirit ofgrammaticalization of a di�erent sort are found in what Gaeta calls the``phonological way to grammaticalization'' or ``grammaticalization from below.''These are primarily instances where phonological alternations take onmorphological functions, e.g. Germanic umlaut, originally a phonological rulefully phonologically conditioned (by following front vowels), which took ongrammatical functions, such as signaling `plural' and certain verbal morphemes.While Gaeta considers these phonology > morphology cases to be examples ofgrammaticalization (under the de®nition that anything that gives rise togrammatical formatives is grammaticalization), from the core view of lexical >grammatical or less grammatical > more grammatical, examples of this sort areclearly exceptions to unidirectionality, where morphemes can become grammatical,lose their meaning, and leave mere phonological residue which no longer has anygrammatical function, but where phonological alternations do not become imbuedwith grammatical functions Ð ``the phonological path to grammaticalization hasbeen shown to represent the . . .reverse path with respect to . . .clines singled out intraditional studies of grammaticalization'' (Gaeta, 1998, p. 102) (see also Janda,this volume).

To summarize the discussion of unidirectionality to this point: some proponentsof grammaticalization do recognize counterexamples to the unidirectionality claim,though many consider them so infrequent as to be of no great relevance. Eventhose who are critical of grammaticalization recognize that instances of changefrom less grammatical to more grammatical are the expected norm and faroutnumber changes in the opposite direction. In a sense, then, there is basically nodisagreement about the overall tendency towards directionality in changesinvolving grammatical elements; rather what is at stake is (1) whetherunidirectionality is a (testable) hypothesis at all, and (2) how the recognizeddirectional tendency is to be explained, and, in particular, whether any of theproposals involving grammaticalization are appropriate for explaining thistendency. With regard to (1), the unidirectionality claim in its strong form Ð thatexceptions do not exist Ð is clearly wrong, in light of counterexamples such asthose presented in this paper. This does not negate a tendency towardsdirectionality in grammaticalization changes, but does falsify the strong claim thatthe directionality is absolute or inviolable. With regard to (2), I argue below thatthe directionality tendency observed in changes discussed in the

L. Campbell / Language Sciences 23 (2001) 113±161 133

Page 22: What’s wrong with grammaticalization?

grammaticalization literature is derived as an expected by-product of generalproperties of linguistic change and requires no special appeal togrammaticalization itself to explain it. With this preview in mind, let us turn nowto attempts to explain unidirectionality.

3.3.2. Explanations for unidirectionalityVarious kinds of explanation have been o�ered for unidirectionality (see also

Newmeyer, this volume). Each is taken up here in turn, beginning with metaphor.Heine et al. (1991, p. 29) elevate metaphor to a central role in

grammaticalization, independent of semantic loss, and ultimately they deriveunidirectionality from it. Matiso� (1991, p. 384) has asserted that``grammatization may also be viewed as a subtype of metaphor (etymologically`carrying beyond'), our most general term for a meaning shift . . .Grammatizationis a metaphorical shift toward the abstract.'' Heine et al. (1991, p. 65) interpret``the process underlying grammaticalization as the result of a problem-solvingstrategy whereby `abstract' concepts are understood in terms of `less abstract'concepts.'' They speak of ``conceptual transfer'' and consider it ``a creative act.''They discuss with approval work which involves ``the principle of creativelanguage use according to which there is a constant attempt to express the same(grammatical) meaning in other words'' (Heine et al., 1991, p. 30). They con®netheir notion of creativity as problem solving to just one aspect:

the ability to conceptualize abstract domains of cognition in terms ofconcrete domains Ð for example, the domain of space in terms of that ofphysical objects, the domain of time in terms of spacial concepts, thedomain of logical relations in terms of temporal concepts, etc. Thelinguistic result of this creative act is that lexical structures are employedto express grammatical meanings, and grammatical structures serve toexpress even more grammatical meanings. Thus, the kind of creativity weare concerned with here is unidirectional, leading from concrete to abstractconcepts and from lexical to grammatical expressions (Heine et al., 1991,p. 31).

They say that ``it is abstraction of the metaphorical type, rather than any other,that underlies grammaticalization'' (Heine et al., 1991, p. 43), and they ``try todemonstrate that metaphorical transfer forms one of the main driving forces inthe development of grammatical categories; that is, in order to express more`abstract' functions, concrete entities are recruited'' (Heine et al., 1991, p. 48).They add: ``Much of what happens when source structures . . .develop intogrammatical structures can be described in terms of a few basic categories that, onaccount of their relative degree of metaphorical `abstraction,' can be arrangedalong the following scale, or chain . . .PERSON > OBJECT > ACTIVITY >SPACE>TIME>QUALITY'' (Heine et al., 1991, p. 48).

Haspelmath's explanation of unidirectionality ``consists of two parts,'' metaphor

L. Campbell / Language Sciences 23 (2001) 113±161134

Page 23: What’s wrong with grammaticalization?

and phonetic reduction, and later he brings ``expressivity'' into the picture. Hecalls on ``the general tendency for humans to conceptualize highly general abstractnotions in terms of speci®c concrete notions by means of metaphor'' (Haspelmath,1998, p. 53). Directionality begins to enter the picture, in his belief, because ``theuse of a highly general abstract expression for a speci®c concept would not be ofany help and thus does not occur in normal speech'' (Haspelmath, 1998, p. 53).But this is not right, as has been long known from traditional classi®cations ofsemantic changes, where, though many do go from concrete to abstract (bymetaphor), many others undergo what they are asserted here not to, namely``narrowing,'' ``litotes'' (exaggeration by understatement), and other sorts ofsemantic change that go from more abstract to more concrete. Ubiquitousexamples of these in the historical linguistic handbooks include meat `food' >`¯esh, meat', hound `dog in general' > `species of dog (long-eared hunting dogwhich follows its prey by scent)', and so on. As a ready contemporarycounterexample to this claim, consider perhaps the most general and abstract ofEnglish verbs, do, which is nevertheless used to express the ``speci®c concepts'' ofcopulation and killing, as in Barbarella said she would do him (if only she could gethim into bed with her ) or Barbarella said she had to do him (because he came at herwith a gun ). In short, ``the use of abstract expression for a speci®c concept'' doesindeed ``occur in normal speech.'' There is no strict directionality constraint inchanges by metaphor; therefore, metaphor cannot be a su�cient explanation fordirectionality in grammaticalization.

Phonetic reduction is the second part of Haspelmath's explanation, and it isalso a major consideration in others' ideas about what lies behind the assumeddirectionality of grammaticalization changes. Haspelmath (1998, p. 54) holds that``the formal side of grammaticalization shows a similar unidirectional tendency forphonetic reduction and merger.'' Reasons for this, reports Haspelmath, includethe forces of ease of perception and ease of production [called motivations of``clarity'' and ``economy,'' respectively, in Haspelmath (1999), Langacker's (1977,p. 105) ``perceptual optimality'' and ``signal simplicity'']. Haspelmath ®nds the keyto directionality in an asymmetry he perceives in these two forces:

The phonetic dimension is open-ended toward the pole of ease of production,whereas the syntactic-semantic dimension is open-ended toward the pole of easeof perception . . .when perception is unimportant or the speaker is under greatpressure to keep the cost of articulation low, linguistic expressions may bereduced almost inde®nitely . . .However, when ease of perception is important,there is a limit on the detail with which a sentence can be pronounced(Haspelmath, 1998, p. 55).

However, the interaction of these forces is insu�cient to explain directionality, asI will try to show. Haspelmath asserts that:

The unidirectionality of grammaticalization is a direct consequence of this[the asymmetry between ease of perception and ease of production]. In all

L. Campbell / Language Sciences 23 (2001) 113±161 135

Page 24: What’s wrong with grammaticalization?

languages, some elements with an abstract, general meaning prove to beparticularly useful in many contexts and increase in frequency. Since easeof perception is less important with frequent items, they are subject togreater phonetic reduction (Haspelmath, 1998, p. 56).

Janda, this volume, discusses related matters from a di�erent perspective.Haspelmath's claim is, apparently, that the phonetic pole is closed, so that new

phonological material outside the existing variants is never added in order toachieve greater ease of perception, and therefore syntactic-semantic means must beresorted to for any innovation which would increase ease of perception: ``In orderto achieve greater salience, the second dimension of variation, the syntactic-semantic dimension, must be employed'' (Haspelmath, 1998, p. 55). As anexample of what he means by this, he asserts that ``it is not possible to introduceadditional vowels or consonants to make an utterance more salient and easier toperceive (attempts to do this . . .are totally unacceptable)'' (Haspelmath, 1998, p.55). But this is erroneous; it denies the many kinds of sound change whosemotivation is very often assumed to be precisely to facilitate ease of perception(though in some instances such changes can also be seen as facilitating ease ofproduction), e.g. epenthesis (VCCV > VCVCV), prothesis (CCV > VCCV),anaptyxis (e.g. VCCV > VC3CV), dissimilation (VktV > VxtV), paragoge (CVC>CVCV), gemination (e.g. VtV>VttV), lengthening, strengthening, and so on.

Haspelmath (1998, p. 54) sees this asymmetry as related to the fact ``thatlanguage change originates in synchronic variation,'' and indeed sound changesoften do leave synchronic variants in a language whose choice is governed bysociolinguistic or stylistic factors, where the choice of one variant may fostereasier perception, while selection of another variant may contribute more to easeof production. For example, even in German (from where Haspelmath has takenhis main examples), choice among phonetic variants such as hab' [hap] vs habe[hab3] `I have' (where the former illustrates the phonetic change of loss of the ®nale ), governed stylistically and sociolinguistically, has an impact on ease ofperception (the longer variant habe being easier to perceive). This is scarcelydi�erent in its e�ect on ease of perception from the sort of example which requireschoice between sociolinguistically and stylistically governed semantic-syntacticvariants, as in, for example, between German kanst jetzt essen [can.2sg noweat.INFINITIVE] (more colloquial) and Du kanst jetzt essen [2sg.PRONOUNcan.2sg now eat.INFINITIVE] (more standard) `you can eat now'. Haspelmathclaims, essentially, that to achieve greater ease of production, only new options onthe phonetic dimension can be created, but that the syntactic-semantic pole isclosed, and therefore, for example, ``we cannot normally introduce innovationssuch as (3sg) [�Kann du mal komm(en )? [can 2sg.PRONOUN come?], from Kannstdu mal kommen [can.2sg 2sg.PRONOUN come?], where in¯ections (representing``semantic-syntactic'' stu�) are dropped. But again, examples are common inhistorical linguistics of precisely this sort, where greater ease of production isachieved (apparently) by innovations resulting in reduction or loss of syntactic-

L. Campbell / Language Sciences 23 (2001) 113±161136

Page 25: What’s wrong with grammaticalization?

semantic material; e.g. certain in¯ections have been dropped in many languages Ðand not just as the by-product of phonological change Ð e.g. in the history ofEnglish (as with the fate of -st `2sg', exactly parallel to the German case, �Kann dumal komm(en )?, which Haspelmath says cannot be introduced). Norde (thisvolume) calls such loss of in¯ectional categories ``de¯exion'' and she presentsseveral examples from the history of Swedish.

In brief, the engine driving unidirectionality for Haspelmath is his belief thatphonetic changes result in reductions which help to achieve greater ease ofproduction, while semantic-syntactic (metaphor-driven) changes, unlike phoneticones, can add new material and in this way achieve greater ease of perception; butease of production, he believes, cannot be achieved by adding phonetic materialand ease of perception cannot be fostered by loss of semantic-syntactic material.This brings us back full circle to where we started: since only the reductive sortsof changes (those involving semantic bleaching, phonological reduction) qualifyunder his de®nition as instances of grammaticalization, grammaticalization mustbe unidirectional, he unselfconsciously concludes that ``the irreversibility ofgrammaticalization is one of the most important constraints on possible languagechanges.''

In spirit Haspelmath's view results in violation of the notion ofunidirectionality. If drive towards ease of perception can result in new ``semantic-syntactic'' material being added (by way of new analytic constructions), thenclearly these cases go against the general view of grammaticalization as involvingphonetic (and semantic) reductions, though they do involve new grammaticalelements coming into the language, often from constructions involving formerlexical morphemes which come to have a grammatical function. The only way tointerpret such developments as not being counterexamples to unidirectionality ofgrammaticalization is to consider these new analytic constructions as examples ofsomething else which by de®nition falls outside the universe of discourse. But thenunidirectionality becomes a tautology: changes which happen to be in the rightdirection are unidirectional; changes that happen to be in the wrong direction areexempt from being counterexamples to unidirectionality.

Haspelmath (1999) adds the notion of ``extravagance'' (also called ``the principleof the salience of novelty'') to this explanation of unidirectionality:

The overall picture seems to be beyond reasonable doubt: The unidirectionalityof grammaticalization is an indirect e�ect of general principles of humancognition and behavior Ð the principle of metaphorical transfer from concreteto abstract domains, the principle of the salience of novelty, and theasymmetric orientation of the continua of phonetic and syntactic-semanticvariation (Haspelmath, 1998, p. 56).

[Interestingly, in contrast to this view, Heine et al. (1991, p. 60) assert thatexpressive metaphors ``are irrelevant for de®ning grammaticalization.'']

L. Campbell / Language Sciences 23 (2001) 113±161 137

Page 26: What’s wrong with grammaticalization?

Haspelmath associates this extravagance/novelty also with changes in pragmaticforce. ``In changes from a full personal pronoun to an agreement marker, from anemphatic negation marker to a simple negation marker, or from a demonstrativeto a de®nite article, the denotation remains more or less the same, but thepragmatic value is changed: The new meaning is less emphatic, less markedpragmatically. The unidirectionality of this development must probably beattributed to the tendency for humans to associate pragmatic force with novelty''(Haspelmath, 1998, p. 53). He suspects that ``the tendency for semantic changefrom concrete to abstract can somehow be subsumed under this unidirectionalchange from pragmatically marked to unmarked'' (Haspelmath, 1998, pp. 53±54).

This notion of ``extravagance'' comes from Rudi Keller's (1990) maxims ofaction, ``Extravagance: Talk in such a way that you are noticed'' (Haspelmath,1999, p. 1055), and is related to the notion of ``expressivity'' (cf. Hopper andTraugott, 1993, p. 65: ``new and innovative ways of saying things are broughtabout by speakers seeking to enhance expressivity''). Haspelmath's ``extravagance''and others' ``expressivity'' have certain similarities to Harris and Campbell's(1995, pp 54, 72±75) ``exploratory expressions.'' However, the motivation for these``exploratory expressions'' is not necessarily limited to any drive on the part ofspeakers to be noticed; rather, they might arise also for purposes of clarity, play,aesthetics, as afterthoughts, and so on. ``Extravagance'' plays a key role inHaspelmath's view of how grammaticalization changes are initiated, which is in aprocess that ``starts out with individual utterances of speakers who want to benoticed and who choose a new way of saying old things. Since they can onlyfreely manipulate the lexical end of the lexicon-grammar continuum, the newexpression necessarily involves a lexical category'' (1999, p. 1057). This is followedby Keller's Maxim of Conformity: ``an individual speaker's innovation must beadopted by other speakers,'' who thus are themselves being both extravagant andconforming (Haspelmath, 1999, p. 1057).

Frequency also plays a role:

Lexical items that ful®ll a frequent discourse function will then increase infrequency because they are very often useful . . .Increased frequency also meansincreased predictability, and the more predictable an item is, the lessphonologically salient it needs to be. Speakers can now a�ord a slurredpronunciation (by the maxim of economy) because the danger ofmisunderstanding (the maxim of Clarity) is not particularly high. In this wayfrequent items get reduced phonologically well beyond the average reduction(Haspelmath, 1999, p. 1058).

Haspelmath derives the assumed irreversibility of grammaticalization from this,starting with Keller's ``hypermaxim'': ``Talk in such a way that you are sociallysuccessful, keeping the cost as low as possible'' (Haspelmath, 2000, p. 14). He saysthat the maxims of Clarity, Economy, Conformity, and Extravagance can be

L. Campbell / Language Sciences 23 (2001) 113±161138

Page 27: What’s wrong with grammaticalization?

derived from this hypermaxim. He tells us that ``the opposite of extravagance isConformity,'' but ``while Extravagance may lead to behavior IIIa [essentially theuse of a lexical element in a context where a grammatical element was available],conformity does not lead to the opposite behavior of replacing lexical elements byfunctional elements. This asymmetry of the opposites extravagance andconformity seems to lie at the root of the unidirectionality of grammaticalization''(Haspelmath, 1999, p. 1059). Notice, however, that even in his own account,Conformity is not strictly opposed to Extravagance, since speakers are assumed toadopt the innovations of others in order to be extravagant, but in so doing areexhibiting conformity Ð here, Extravagance and Comformity are not oppositesand so do not really result in the asymmetry Haspelmath sees, but are sequential,where one in e�ect feeds the other.

Haspelmath gives a second reason for the asymmetry he proposes in theassertion:

Lexical elements are freely manipulable by speakers (more or less) accessible toconsciousness, whereas functional elements are processed automatically andunconsciously. So even if a speaker had some motivation for replacing a lexicalitem by a functional item, s/he would not be able to do this because functionalelements cannot be used outside their proper places (Haspelmath, 1999,p. 1059).

Notice that this argument rests on sheer assertion. Where is the evidence thatfunctional elements are processed only unconsciously and are not freelymanipulable by speakers? Because there are many cases of functional elementswhich are manipulated consciously by speakers and numerous examples violatingthe assumption of unidirectionality (see above), this claim fails. Examples ofextension after reanalysis involving grammatical categories show that grammaticalcategories can be manipulated by speakers, for otherwise the extensions would nottake place (examples given below; for details, see Harris and Campbell, 1995, pp.97±119).

Haspelmath (1999, pp. 1059±1060) gives in culmination of this line of argumentthe following account:

If by some miracle the ®rst two obstacles were overcome and a speaker were touse a grammatical item in the place of a lexical item (say, with for `tool'), thenthe next steps in the hypothetical functional-to-lexical development wouldapparently not be impossible. Thus, because the meaning of the item is nolonger basic to discourse, the item would drop in frequency . . .and it wouldbecome less predictable . . .Next, speakers might want to pronounce it moreclearly because the element is no longer highly predictable . . . , and they coulddo this, for instance, by lengthening the vowel . . . , strengthening the

L. Campbell / Language Sciences 23 (2001) 113±161 139

Page 28: What’s wrong with grammaticalization?

consonants . . . introducing additional syllables . . . , and so on. Such changes ofcourse do not occur.

But, in summarizing all this, Haspelmath asserts that the other steps in thisaccount (though conceivably possible) ``have no opportunity of ever occurring''because ``the reverse of step (a) [lexical-to-grammatical/functional] is impossible''(Haspelmath, 1999, p. 1060). With all this verbiage, we might wonder whether wehad missed something here: can this really mean what it seems to, thatHaspelmath's ``explanation'' boils down to this assertion: that the reason for whygrammaticalization is irreversible is because grammaticalization is irreversible? No,more careful reading reveals that this is precisely what is being said. I repeat,then, if we take unidirectionality as an empirical matter rather than as an axiomof grammaticalization, built into the very de®nition, then unidirectionality failsbecause counterexamples exist, Haspelmath's denial of counterexamplesnotwithstanding.

The ®nal attempt at explanation of unidirectionality to be mentioned is that ofTabor and Traugott (1998). They explore the hypothesis that cases of```grammaticalization' persistently involve increase of structural scope (de®ned asC-command Scope Increase), not decrease in structural scope'' (Tabor andTraugott, 1998, p. 231). For them unidirectionality is clearly an empirical matter,not just a de®nitional property of grammaticalization: ``it is clear that we cannottake structural unidirectionality as a de®nition of the subject matter [ofgrammaticalization], for structural unidirectionality is a hypothesis which needs tobe formalized and tested before it can be presupposed . . .we feel unidirectionalityis best treated as a hypothesis, not a de®ning characteristic.'' While investigationof possible regularities which may save some or all of the unidirectionalityhypothesis are clearly in order, their approach is a somewhat curious one, for theyreadily ``acknowledge that C-command Scope Increase Hypothesis is not likely toprove universal'' (Tabor and Traugott, 1998, p. 260) and they cite cases ofcounterexamples to their own hypothesis even among their own examples. Thus,their hypothesis does not adequately explain unidirectionality and su�ers from theproblem of counterexamples that plagues the more conventional orientations tothe unidirectionality claim.

In summary, unidirectionality is either a de®nitional property ofgrammaticalization and irrelevant or it is an empirical property and false. If it istaken as a de®nitional property of grammaticalization, it is then void of empiricalcontent, since in this view, non-conforming examples are not counterexamples, butjust fall outside the universe of discourse of grammaticalization altogether. If it istaken as an empirical property, then it has been falsi®ed by the numerouscounterexamples which exist. In either case, the grammaticalization process is, infact, reversible. Finally, to the extent that there is a tendency to directionality ingrammaticalization changes, this is explained by the intersection of the e�ects ofother sorts of linguistic change which are operative in language (see the followingsections of this paper). Unidirectionality neither explains nor is explained by

L. Campbell / Language Sciences 23 (2001) 113±161140

Page 29: What’s wrong with grammaticalization?

grammaticalization. Therefore no independent status for grammaticalization canbe underpinned by assumptions of unidirectionality.

4. Mechanisms and explanation of grammaticalization changes

Most agree that grammaticalization is not a mechanism of change in its ownright, but relies primarily on reanalysis (and also extension).11 This being the case,it will be helpful to de®ne these notions before examining what supporters have tosay about their relation to grammaticalization itself. I will give a de®nition andbrief description of each of these before going on to consider the role of reanalysisin grammaticalization and what this implies for its status.

Harris and Campbell (1995) argue that there are only three mechanisms ofsyntactic change: reanalysis, extension, and borrowing; grammaticalizations cancome about by any of the three, though reanalysis predominates.

4.1. Reanalysis

Reanalysis changes the underlying structure of a grammatical construction, butdoes not modify surface manifestation. The underlying structure includes (1)constituency, (2) hierarchical structure, (3) grammatical categories, (4)grammatical relations, and (5) cohesion. Surface manifestation includes (1)morphological marking (e.g. morphological case, agreement, gender), and (2)word order. An axiom of reanalysis is that reanalysis depends on the possibility ofmore than one analysis of a given construction becoming available. Harris andCampbell's de®nition of reanalysis incorporates Langacker's (1977, p. 58) ideathat reanalysis is characterized as ``change in the structure of an expression orclass of expressions that does not involve any immediate or intrinsic modi®cationof its surface manifestation.'' Hopper and Traugott's (1993, p. 32) treatment issimilar; they say:

Reanalysis modi®es underlying representations, whether semantic, syntactic, ormorphological, and brings about rule change. Analogy [which they also call`rule generalization'], strictly speaking, modi®es surface manifestations, and initself does not e�ect rule change, although it does e�ect rule spread eitherwithin the linguistic system itself or within the community.

11 Some scholars consider other mechanisms in addition to reanalysis to be relevant to grammaticaliza-

tion; e.g. Traugott and Heine (1991, p. 7) list also metaphorical transfer, metonymic transfer, analogy,

and calquing. Calquing is included in grammatical borrowing (one of Harris and Campbell's mechan-

isms). Metaphor and metonymy are kinds of semantic change. Traugott and Heine's notion of analogy

largely ®ts Harris and Campbell's ``extension.'' Signi®cantly, they recognize that ``all these are mechan-

isms that make change possible, but none are restricted to grammaticalization'' (Traugott and Heine

1991, p. 7).

L. Campbell / Language Sciences 23 (2001) 113±161 141

Page 30: What’s wrong with grammaticalization?

They say reanalysis `is not directly observable', but `analogy makes theunobservable changes of reanalysis observable' (Hopper and Traugott, 1993, p.61); the behavior of their ``analogy'' is like Harris and Campbell's ``extension.'' Ina slightly di�erent view, Heine et al. (1991, p. 210) talk about ``reanalysis'' or``(metaphorical) transfer'' being followed by ``metaphorical extension''; the latternotion is not for them limited just to grammar (as Harris and Campbell's notionof ``extension'' is), but is primarily exempli®ed in semantics.

The be going to `future' grammaticalization example (cited above) illustratesreanalysis (see Hopper and Traugott, 1993, pp. 2±3, 61±62). It began with adirectional verb in a purposive clause, as in [1]:

[1]I am going to marry Bill.I am goingVerb ofMotion to marry Bill.

There was then an inference of futurity from purposives: if I am traveling in orderto marry, the marriage will be in the future. The purposive be going (to ) wasreanalysed as a future auxiliary with a verb of activity, as in [2]:

[2]I am going to marry Bill.I [am going]Future Auxiliary marry Bill.

Notice that [1] is still fully grammatical also in the directional/purposive sense.Though a reanalysis produced [2], the surface manifestation remained unchangedÐ [1] and [2] are identical in form, but not in internal structure or meaning.

4.2. Extension

Extension results in changes in surface manifestation, but does not involveimmediate modi®cation of underlying structure.

Further developments in the be going to example (just mentioned) illustrateextension. After reanalysis, there was an extension so that be going to in its futuresense could appear with other verbs, not just limited to those which could appearas complements in the purposive and motion verb constructions, which wasformerly the situation. Hopper and Traugott say of this:

The reanalysis is discoverable, that is, is manifest, only when the verb followingbe going to is incompatible with a purposive meaning . . .I am going to like Bill,I am going to go to London. In other words, the reanalysis is discoverable onlybecause the contexts in which be going to can occur have been generalized, oranalogized [extended], to contexts that were unavailable before (Hopper andTraugott, 1993, p. 3; see also p. 61).

In the case of will, where a main verb was reanalysed as a `future' auxiliary(also cited above), extension is exhibited in that the auxiliary which formerly

L. Campbell / Language Sciences 23 (2001) 113±161142

Page 31: What’s wrong with grammaticalization?

required human subjects (e.g. the king will . . . ) was extended to permit all kinds ofsubjects (e.g. the weather will )(cf. Ramat and Hopper, 1998, p. 3).

In brief, these central examples of grammaticalization are described andexplained as perfectly ordinary cases of reanalysis followed by extension (seeHarris and Campbell, 1995, pp. 50±54).

4.3. Borrowing

Grammatical elements can be borrowed from other languages. This in e�ectconstitutes a kind of ``grammaticalization,'' i.e. the inception of new grammaticalmaterial in the language, without prior lexical source, although it does not ®t thestandard notion of grammaticalization (as lexical > grammatical ). I will cite justone example, related indirectly to one of the counterexamples to directionalitypresented above. Balto-Finnic languages have borrowed the morpheme -pa, -paÈ`intensi®er' from Baltic languages (e.g. Finnish sinaÈ-paÈ sen sanoit [you-EMPHATIC it you.said] `it was you that said it'; cf. Latvian -ba, Lithuanian baÁ,beÁ `intensi®ers') (Hakulinen, 1968, p. 195; Laanest, 1982, p. 294); for otherexamples and more discussion see Harris and Campbell, 1995, pp. 120±150). (Fordiscussion of the role of borrowing and language contact in grammaticalization,and for other examples, see Bisang, 1998; Keesing, 1991; Ramat, 1998, pp. 116±117; Ramat and Hopper, 1998, p. 7.)

4.4. Reanalysis and grammaticalization

In the typological literature, the central role of reanalysis as the primarymechanism behind grammaticalization is generally acknowledged. For example,Hopper and Traugott consider the ``mechanism by which grammaticalizationtakes place'' to be ``reanalysis primarily, and analogy [extension] secondarily''(1993, p. 32). They say, ``unquestionably, reanalysis is the most importantmechanism for grammaticalization, as for all change'' (1993, p. 32). Heine andReh (1984, p. 95) go so far as to equate reanalysis and grammaticalization:``Reanalysis thus appears as a concept which is largely synonymous with our term`grammaticalization'.'' Hopper and Traugott represent the more prevalent view ofthe relationship between grammaticalization and reanalysis: ``It is best, then, toregard grammaticalization as a subset of changes involved in reanalysis, ratherthan to identify the two'' (Hopper and Traugott, 1993, p. 50); ``reanalysis andanalogy are the major mechanisms in language change. They do not de®negrammaticalization, nor are they coextensive with it, but grammaticalization doesnot occur without them'' (Hopper and Traugott, 1993, pp. 60±61). For Traugott(1995, p. 1), reanalysis is ``the mechanism that brings about particular instances ofgrammaticalization'' (cf. Hopper, 1992, p. 80).

As illustrated above, central examples of grammaticalization are typically

L. Campbell / Language Sciences 23 (2001) 113±161 143

Page 32: What’s wrong with grammaticalization?

described and explained in the grammaticalization literature as perfectly ordinarycases of reanalysis, sometimes followed by extension. Nevertheless, the question ofhow reanalysis ®ts with grammaticalization has received other, more unusualinterpretations, in particular by Heine and associates and by Haspelmath. Forexample, Heine and Reh (1984, pp. 95±97) separate reanalysis fromgrammaticalization, essentially on the basis of the unidirectionality principle,which they say is an inherent property of the latter but not necessarily of theformer (cf. also Heine et al., 1991, p. 215). However, Heine et al. (1991, p. 217)appear to see the two as somehow interconnected and both as the result of moreabstract notions being expressed in terms of less abstract ones (via metaphor):``reanalysis accompanies grammaticalization . . .when reanalysis takes place, this islikely to involve . . .grammaticalization . . .both grammaticalization and reanalysisare the result of one and the same strategy, namely the one that aims atexpressing more `abstract' concepts in terms of less `abstract' ones.'' However,Heine et al. do intend the two to be understood as separate; they say:

There is reason to assume that, while both grammaticalization and reanalysisappear to be inseparable twins, they must, nevertheless, be kept strictly apart,in particular because of the following considerations: (a) Whereasgrammaticalization is essentially a unidirectional process, reanalysis is not . . .(b)Grammaticalization need not be accompanied by reanalysis. Thus when ademonstrative is grammaticalized to a de®nite article (this man> the man ), thede®nite article to a nongeneric article . . .or the numeral `one' to an inde®nitearticle (one man > a man . . .), then we are dealing with the grammaticalizationof a demonstrative or numeral where no reanalysis is involved: the syntacticstatus of the determiner-head phrase remains unchanged (Heine et al., 1991, p.219).

Their interpretation of reanalysis, however, does not conform to the standardde®nition which is generally accepted, and in fact it is the mechanism of reanalysisthat motivates all the examples of grammaticalization in (b) (above) which theycite. In the case of demonstrative> de®nite article and of numeral `one'> inde®nitearticle, the underlying structure of the syntactic construction (the grammaticalcategory) has changed, but the surface manifestation is not modi®ed (precisely asin the standard de®nition of reanalysis, cited above). Grammaticalization is alwaysthe result of reanalysis (except instances involving borrowing, if these are even tobe included in grammaticalization). Moreover, as seen above, there arecounterexamples to the supposed unidirectionality of grammaticalization, contraryto (a). Thus, Heine et al.'s reasons for assuming there to be instances ofgrammaticalization without reanalysis do not hold up.

Hopper and Traugott (1993, pp. 48±56) argue compellingly that reanalysisand grammaticalization are not equal, but not separate. They recognize thatgrammaticalization changes typically involve reanalysis, but not all cases ofreanalysis are instances of grammaticalization. Indeed, many reanalyses do not

L. Campbell / Language Sciences 23 (2001) 113±161144

Page 33: What’s wrong with grammaticalization?

involve grammaticalization, e.g. many word-order changes, changes from onesyntactic structure or morphological category to another of equal or lessgrammatical strength, and all the instances of counterexamples tounidirectionality mentioned above Ð i.e. any reanalysis which does notinvolve lexical > grammatical, grammatical > more grammatical; or discoursestructure > morphosyntactic. (For numerous examples, see Harris and Campbell,1995.)

The most idiosyncratic position concerning the relation between reanalysis andgrammaticalization is that of Haspelmath. Haspelmath (1998, p. 49) attempts tochallenge the generally accepted notion ``that syntactic changes involvinggrammaticalization are best accounted for by invoking the notion of diachronicreanalysis.'' He admits that ``the current mainstream view in the functionalistgrammaticalization literature appears to be that . . .most grammaticalizationprocesses are manifested in reanalyses, or that reanalysis is one of the mainmechanisms of grammaticalization change'' (cf. Hopper and Traugott, 1993, p.32). However, in his view, ``the large majority of syntactic changes are instances of`pure' grammaticalization and should be explained within the framework of atheory of grammaticalization, without reference to reanalysis. A minority ofsyntactic changes are due to reanalysis, and they must be explained in di�erentterms. Grammaticalization and reanalysis are disjoint classes of phenomena.'' Headds, ``my impression is that grammaticalization and analogical extension [Harrisand Campbell's ``extension''] account for the large majority of syntactic changes,whereas reanalysis changes form a small minority'' (Haspelmath, 1998, p. 61).However, it is possible to entertain this interpretation of the role of reanalysis ingrammaticalization only by rede®ning reanalysis in ways not accepted by orrecognized at all by other scholars. This is what Haspelmath does. He imposes thefollowing distinguishing characteristics on grammaticalization and reanalysis, andon this basis concludes that the two are sharply distinguished (Haspelmath, 1998,p. 60):

grammaticalization reanalysis

loss of autonomy/substance no loss of autonomy/substancegradual abruptunidirectional bidirectionalno ambiguity ambiguity in the input structuredue to language use due to language acquisition

But none of these assumed characteristics/de®ning traits is actually true ofreanalysis as normally de®ned (see Harris and Campbell, 1995 for discussion), andseveral listed as diagnostic for grammaticalization are not true either.

Even Haspelmath's application of these idiosyncratic criteria to speci®c cases inattempts to distinguish grammaticalization from reanalysis shows why this

L. Campbell / Language Sciences 23 (2001) 113±161 145

Page 34: What’s wrong with grammaticalization?

approach fails, since the instances he presents of grammaticalization which hebelieves ``do not involve reanalysis at all'' are themselves seriously ¯awed(Haspelmath, 1998, p. 62). He cites examples from Heine and Reh (1984)involving changes in which ``a verb turns into a preposition or a complementizer''(Haspelmath, 1998, p. 62). For example, he illustrates the application of hiscriterion (of loss of autonomy for grammaticalization, no loss of autonomy forreanalysis) with (1) a case of ``desemanticization,'' where the Ewe verb originallymeaning `give' comes to introduce ``recipients of all kinds and is much moreabstract than the original''; and (2) the verb `say' which became a complementizerwhich introduces also ``complements to non-speech verbs, e.g. verbs of cognitionand propositional attitude'' (Haspelmath, 1998, p.62). He says of these, ``bothprepositions and complementizers are closed classes of grammatical items, so it isclear that we are dealing with grammaticalization'' (Haspelmath, 1998, p. 62). But,notice that no one has claimed that the only mechanism involved ingrammaticalization is reanalysis, only that it is the principal one. Extension andborrowing can also be involved in aspects of grammaticalization changes. Thus,an expansion from prototypical recipients of `give' to ``recipients of all kinds''may well also involve extension. However, the reinterpretation of the verb as apreposition clearly involves reanalysis of the sort embodied in the standardde®nition Ð a change in grammatical categories, in grammatical relations Ð inspite of Haspelmath's interpretation of this as not involving reanalysis. That is,that extension also takes place in these cases does not mean that reanalysis wasnot also involved Ð it was.

Next, Haspelmath calls on his assumption that grammaticalization isunidirectional but reanalysis is not, saying of these examples (just cited) that ``it isevident that the changes are irreversible: complementizers and prepositions cannotbecome verbs'' (Haspelmath, 1998, p. 63). Here he is much in¯uenced by Heineand others (cf. Heine et al., 1991, p. 215; Heine and Reh, 1984, pp. 95±97). But,leaving aside for the moment that there are counterexamples which challenge theunidirectionality of these grammaticalization changes (and that some subdivisionsof reanalysis, or of grammaticalization for that matter, could be unidirectionalwithout requiring all to be), we see that the speci®c claim in this case is wrong. Tosee that prepositions can and do become verbs (and nouns and adjectives), weneed look no further a®eld than English: outPrep > outNoun (`an out', as in sports,a way to avoid consequences), outPrep > outVerb (`to out', as in to out someone byidentifying him/her to others as homosexual), outPrep > outAdjective (as in `that's',in bad taste or not popular).12 More speci®cally, Haspelmath (1998, p. 63) asserts

12 Complementizers can, it seems, also become verbs, if we consider the cases where a demonstrative

becomes a copula (`to be, to exist'), as in several Caribbean creoles and Gullah, where the copula da,

de historically from the became the copula from constructions which formerly had `zero' copula (e.g.

He da carpenter [he é the carpenter]> [he is (a) carpenter], where the demonstrative also had a comple-

mentizer-like function as well, e.g. He da huntin' [he is hunting]< [he é the.one (who is) hunting].

L. Campbell / Language Sciences 23 (2001) 113±161146

Page 35: What’s wrong with grammaticalization?

``word-class changes invariably turn content-words into function-words. Not onlyis the reverse excluded, but also changes from one content-word class to another

class are unattested or rare'' (Haspelmath, 1998, p. 63). He claims that the

following are the ``unattested (or rare) changes'': V > N, A>N, V> A, A> V,N > A, N > V.13 But, again, English readily gainsays this claim, and in fact

examples of all these supposedly unattested changes are well known: V > N: `arun, a kill'; A > N: `the red will do' (<`red one'), `I'll take the Bordeaux'

(<`Bordeaux wine'); V > A `the stun gun, the kill[ing] master' (not to mentioncases of verbs in participle form used as adjectives generally; there is a close

connection typologically between Verb and Adjective cross-linguistically); A > V:

`to green, to white it out'; N > A: `orange' (fruit > color); N > V: `to club, towater, to master'. Such examples are well-known; they are called `lateral'

conversions by Ramat and Hopper (1998, pp. 5±6, who present several examplesinvolving various languages), and they consider them to be outside of

grammaticalization (see also Tabor and Traugott, 1998, p. 236). It is revealing

that ``categorial downshifting to prepositions (e.g. Italian preposition durante`during' < verb dura- `to last' + -nte `present participle'), conjunctions, and

complementizers is typically regarded as grammaticalization'' (Ramat andHopper, 1998, p. 6), but cases of ``upshifting,'' such as outpreposition > outverb or noun

and others presented here, are ignored or denied, since they are counterexamples

to the assumed unidirectionality of grammaticalization.

Finally, calling upon his criterion of gradualness (limited in his account togrammaticalization only, where reanalysis is asserted to be abrupt), Haspelmath

merely asserts that ``the changes in question [verb to preposition or

complementizer] are gradual rather than abrupt. Verbs do not turn intoprepositions and complementizers overnight, but lose their verbal properties and

acquire the properties of their new word-class step by step'' (Haspelmath, 1998, p.63; see Heine et al., 1991, p. 65 for background to Haspelmath's view on

gradualness; see also Haspelmath, 2000, pp. 3±4). But once again the test fails. It

fails most seriously because there is no such distinction, and the implementationof reanalyses need not be abrupt, but rather typically is gradual in just the way

Haspelmath imagines for grammaticalization (see Harris and Campbell, 1995, pp.48±49). But it also fails more speci®cally in that it is merely an assertion that

verbs do not turn into prepositions or complements suddenly Ð the reanalysis of

category (so a form comes to have a new or di�erent lexical or grammaticalcategory from what it originally had) may very well even require an abrupt

change in individual speakers' conceptualization of it in order for the newinterpretation to be available. One way that someone might interpret this as

involving a gradual change might be to see an original sense gradually getting lost

13 Haspelmath (1999, p. 1046) restates this claim as: ``lexical categories do not turn into other lexical

categories . . .a verb does not become an adjective or a noun, a noun does not become a verb or an

adjective, and an adjective does not become a noun of a verb. Thus, when a lexical category changes its

categorial a�liation . . . , we can predict that it will become a functional category.''

L. Campbell / Language Sciences 23 (2001) 113±161 147

Page 36: What’s wrong with grammaticalization?

slowly after a grammaticalization. It might only be in later changes that theoriginal verb sense slowly gets lost, after the reanalysis produced a prepositionconstruction from the verb with both the verb and preposition from the sameoriginal source coinciding in the language until the verb sense is lost. This scenario®ts precisely the demands of reanalysis: reanalysis can take place only when morethan one interpretation becomes available; however, there is no requirement thatthe lexical source (the verb sense) be lost after the grammaticalized option (thepreposition) becomes available through reanalysis, though this sometimes doeshappen. In fact, such layering (as in the case of be going to as both original verbof motion and grammaticalized future, both in the language simultaneously) isconsidered by many to be a typical characteristic of grammaticalization. Thus itwould appear that the gradualness criterion is at best a red herring.

Tellingly, Haspelmath (1998, p. 64) reports that ``the gradualness of word-classchanges is recognized by some of the authors who describe these changes asreanalysis,'' but adds that ``clearly these authors use the term reanalysis in a muchbroader way than I, meaning roughly `any change in the structural description ofa phrase'. I have no disagreement with these authors, but I would prefer toreserve the term for the narrower concept.'' In essence, with these wordsHaspelmath confesses that it is he who is out of step with nearly all othersworking in grammaticalization and in syntactic change generally, and they do notaccept his de®nition of reanalysis. Haspelmath (1999, p. 1046) also admits that theview of reanalysis versus grammaticalization in terms of an abrupt reanalysis or agradual process is controversial.

In brief, Haspelmath himself has exposed the true nature of his project: ®rstrede®ne reanalysis (and grammaticalization) in a personal (not generally shared)way; then show why this unusual view of reanalysis is not consistent withgrammaticalization. In sum, Haspelmath has jousted with a straw man of his owncreation, but one which bears scant resemblance to the real reanalysis which hewould have us believe that he has discredited. Moreover, while his criteria (non-gradualness, ambiguity, etc.) are not true of the standard view of reanalysis, theyeven fail in the applications to which they are put in the individual examples hepresented as evidence for the argument. The real reanalysis (the standardreanalysis of others) is indeed the principal mechanism of grammaticalizationchanges, as so many other grammaticalizationists recognize explicitly and discussopenly, Haspelmath's straw man and red herrings notwithstanding.

Haspelmath (1998, p. 78) concludes:

If reanalysis is de®ned as a change resulting from the assignment of a di�erentstructural description to a given string in the process of language acquisition,then this notion plays a much less important role in diachronic syntax than iscommonly assumed. The main mechanism of syntactic change isgrammaticalization, i.e. the gradual unidirectional change that turns lexicalitems into grammatical items and loose structures into tight structures,subjecting frequent linguistic units to more and more grammatical restrictionsand reducing their autonomy.

L. Campbell / Language Sciences 23 (2001) 113±161148

Page 37: What’s wrong with grammaticalization?

In response to this it is necessary only to point out: (1) This is only his de®nition,

not accepted or shared by others. (2) It denies all changes which are not in the

direction of lexical to grammatical and of looser to tighter, though such changes

clearly exist. (3) It is a tautology; it seems to read: if we were to reduce the scope

of reanalysis signi®cantly, then it will be less important. Only with such a

rede®nition is it possible to envisage grammaticalization as outside the scope of

`reanalysis'; however, unfortunately for Haspelmath's view, others do not see

reanalysis this way.

Another extreme, but di�erent stance with regard to the relationship between

reanalysis and grammaticalization comes from Heine (1993, pp. 117±119), who

urges throwing the term `reanalysis' out altogether. He o�ers four considerations

in favor of this, mostly having to do with the fact that the term is not, he claims,

used consistently with the same meaning by all writers, making it vague and ill-

de®ned.

In his ®rst consideration, Heine says that ``the relevance'' of the label

`reanalysis' ``to grammaticalization theory is not entirely clear . . .the term has been

used in so many divergent ways.'' It is for this reason, he says, that:

Considering the wide range of uses the label reanalysis has found, and to

avoid that it therefore becomes vacuous, Heine and Reh (1984:95�)

propose to reserve it for instances such as (e) [That the constituent

structure [main verb-complement] is reinterpreted as another constituent

[auxiliary-main verb]], that is, to the reinterpretation of syntactic and

pragmatic constituents'' (Heine, 1993, p. 119).

However, some variation in the de®nition and use of terms is true of most major

concepts in linguistics (as well as in other disciplines). In Lessau's (1994, p. 384)

apt words, ``as is typical of central notions, . . .meaning appears to be a rugged

battle®eld of scattered polysemous fragments.'' So, if we followed Heine's

recommended practice of casting out such terms, the ®eld would be left with very

little Ð in particular, on this criterion ``grammaticalization'' itself would become a

casualty with far more wounds [from di�ering opinions about how it is to be

de®ned and to what it is to be applied] than ``reanalysis'' would. There is a very

widely accepted core meaning of ``reanalysis'' (see Langacker, 1977, p. 64; Harris

and Campbell, 1995), generally accepted by most working in grammaticalization

and in syntactic change, and this should be su�cient for any typologically

oriented linguist to help him/her see past the distracting noise that grows louder

the further one strays from the core, prototypical meaning by which this concept

in understood by the majority of practitioners.

Heine's second reason is because reanalysis has been used by some as ``strictly

belonging to the vocabulary of synchronic grammar.'' This is related to the ®rst

claim, about multiple uses of the term making its meaning unclear. Since almost

L. Campbell / Language Sciences 23 (2001) 113±161 149

Page 38: What’s wrong with grammaticalization?

no one in diachronic linguistics takes the synchronic use of the term as relevant togrammaticalization or considers this confusing, this reason is without force.

His third reason is, ``reanalysis is essentially a concept of cognitive psychology,yet so far it has not been described adequately in terms of psychologicalparameters'' (Heine, 1993, p. 119). Again, this objection goes to the question ofwhether the term has been properly de®ned. If we took this as a strictrequirement, that linguistic terms with cognitive implications must be describedadequately in terms of psychological parameters, then most of linguistics wouldhave to be abandoned. Moreover, some scholars ®rmly believe that thepsychological parameters of reanalysis are adequately described (involvingassociation, generalization, and transfer).

Heine's fourth reason involves the vexed matter of unidirectionality. He says:

While it [reanalysis] has been employed by some as a central item ofgrammaticalization terminology, it is hard to reconcile in at least one importantrespect with grammaticalization theory in that the principle of unidirectionaldevelopment, which is generally considered to be a canonical property ofgrammaticalization, is not inherently associated with reanalysis (Heine, 1993,p. 119).

This argument also is without force. First, if the unidirectional claim is false (asargued in this paper and others in this volume), then nothing follows fromunidirectionality for how or whether grammaticalization and reanalysis arerelated. Second, if, as argued here and as viewed by a majority of practitioners,grammaticalization relies on reanalysis as its primary mechanism but not all casesof reanalysis involve grammaticalization, then it is logically possible that thesubset of reanalyses that are grammaticalizations could be unidirectional, whileother kinds of reanalysis need not be. Third, it is commonly recognized that thereare various kinds of linguistic change which are unidirectional but are notinstances of grammaticalization, for example various kinds of sound change (seeHimmelmann, 1992, p. 6; Lehmann, 1995 [1982]). This being the case, thepresence of unidirectionality is no guarantee of the presence ofgrammaticalization. Even some changes involving reanalyses withoutgrammaticalization are unidirectional, e.g. certain word order changes.

In short, Heine's four reasons for abandoning the term `reanalysis' are not wellfounded at all, and it is therefore no surprise that most grammaticalizationistscontinue to recognize the relevance and role of reanalysis as the primarymechanism behind grammaticalization changes.

To summarize the role of reanalysis in grammaticalization: it is generally agreedthat reanalysis is the primary mechanism behind grammaticalization changes,though there are instances of reanalysis which are not grammaticalization. SinceHaspelmath's idea of reanalysis does not correspond to that accepted generallyboth by supporters of grammaticalization and by historical linguists generally, itcan be discounted. Similarly, none of Heine's reasons for avoiding the term`reanalysis' Ð and even he maintains the concept Ð are warranted. In short,

L. Campbell / Language Sciences 23 (2001) 113±161150

Page 39: What’s wrong with grammaticalization?

then, reanalysis (also sometimes extension) is the determining mechanism thatexplains grammaticalization and without appeal to these mechanisms,grammaticalization has no explanatory power of its own.

4.5. ``Grammaticalization theory'' and explanation

To summarize the discussion of the status of grammaticalization: it is safe toconclude it has no independent status of its own. It is not necessary for the kindsof changes most commonly encountered in grammaticalization to be present inorder for a change to qualify as an instance of grammaticalization, and thesekinds of changes are encountered commonly also in instances of changes whichhave nothing to do with grammaticalization. Semantic ``bleaching'' (or loss) isneither a necessary nor a su�cient attribute of grammaticalization; phonologicalreduction (or erosion) is also neither necessary nor su�cient. The unidirectionalityhypothesis, if taken as an empirical property of grammaticalization, is falsi®ed bynumerous counterexamples. If taken as a de®nitional property, unidirectionalitybecomes diagnostic of grammaticalization, but trivially so, since non-conformingcases are not explained, but merely excluded on the basis of de®nition(terminology) and unidirectionality in this sense has no empirical content. Thoughgrammaticalization changes do tend to be in the direction of less grammatical >more grammatical and not the reverse, this fact is not explained by anything aboutgrammaticalization itself, but rather by the properties of linguistic change ingeneral and how di�erent kinds of changes interact with one another. Anyexplanatory value involved is provided by reanalysis (and also extension andborrowing when they are involved), the mechanism(s) of change lying behindgrammaticalization. In short, grammaticalization is derivative, epiphenomenal,and has no independent status of its own. ``Grammaticalization theory'' has noexplanatory value because what it claims to explain is explained already by otherwell-understood mechanisms which lie behind it and, as is generally agreed, itcannot ``explain'' without appeal to these other mechanisms and kinds of change.(See Janda, Joseph, Newmeyer, this volume, who reach the same conclusion.)

At this point, someone might ask if there is not possibly an air of reductionismin this. It might be claimed that even if explanations based on the mechanismsand changes lying behind grammaticalization are readily available, that would notnecessarily invalidate other sorts of explanations at a less abstract level. Consideran analogy (from Wright 1976:44). To answer the question, `why did the windowbreak?' with `because John slammed it' is a completely adequate answer/explanation, even if shock waves and the molecular characteristics of the glassmay lie behind the breaking at some other level. There are contexts in which ananswer of `because of a certain causal factor x' is correct and adequate, even ifthere may be deeper, more mechanistic causal things one could point to atanother level. Thus, one might assert that those who argue thatgrammaticalization is derivative and has no explanatory power of its own are justtoo reductionist, that while reanalysis, phonological change, and semantic change,and their interaction, explain the events at a deeper level (analogous to the shock

L. Campbell / Language Sciences 23 (2001) 113±161 151

Page 40: What’s wrong with grammaticalization?

waves and molecular structure in the broken window example),grammaticalization may itself be explanatory on another level (analogous toJohn's slamming causing the window to break). While this is an importantpossibility Ð and I have considered it at length Ð I do not ®ndgrammaticalization to have an explanatory value of its own at a di�erent level;this is so because almost unanimously proponents of grammaticalizationthemselves appeal to reanalysis and other changes as the explanatory mechanismsupon which cases of grammaticalization depend. That is, in some sense, bothfriends and foes of grammaticalization theory alike acknowledge and rely onfactors beyond grammaticalization itself for explanation. Reductionism, then, isnot the problem, and appeal to it will not, I think, rescue the claims ofexplanatory power for grammaticalization in and of itself.

A ®nal point to consider is the claim in grammaticalization theory thatgrammaticalization allows us to ``predict'' and therefore to explain, whereexplanation is contingent upon being able to predict in forms of the philosophy ofscience accepted by a majority of researchers in the ®eld. Some grammaticalizationtheorists might argue that there is a certain predictive power ingrammaticalization which may be seen as giving it a certain explanatory power.They might say that there are constraints on what can and cannot grammaticalize:not just any lexical item can grammaticalize as any grammatical meaning/function, and there are not as many grammatical meanings/functions availablecross-linguistically as there are semantic meanings of lexical items which could bedrawn from if there were no constraints. Something needs to explain all this.

This is all true, but much hinges on the form in which the claims ofpredictability are stated. In the strong form in which they are often stated, theclaim is clearly wrong. When told we can predict in grammaticalization, we mustask, why can a single lexical source develop in multiple directions, giving variousdi�erent grammatical ``targets'' as a results, while at the same time a singlegrammatical target (an outcome of grammaticalization) can come about indi�erent languages from numerous di�erent lexical sources? An example of the®rst is the various grammatical categories (`targets') known to develop fromgrammaticalizations of lexical sources with a single meaning: `come' > `future',`near past', `venitive', `necessive', `resultative', `agent marker of passive', `proximaldeictic', etc.; `say' > `quotative', `subordinate conjunction', `complementizer',`evidential marker', `purpose question', etc. An example of the second is the manydi�erent lexical sources which can be grammaticalized to result in `future':FUTURE < `want', `have', `go', `come'; subjunctive; adverbs [`quickly',`tomorrow', `then', `afterwards']. The former has been called``polygrammaticalization,'' where a single form develops di�erent grammaticalfunctions in di�erent constructions (see for example, Craig, 1991; Hopper andTraugott, 1993, p. 112) or ``split'' or ``splitting'' (see Hopper, 1992, p. 80; Heineand Reh, 1984, pp. 57±59; Hopper and Traugott, 1993, pp. 112±113). The latterhave been called ``merger,'' when the same grammatical result comes fromdi�erent lexical sources (cf. Hopper and Traugott, 1993, p. 113; Heine et al., 1991,p. 108). As Heine et al. (1991, p. 38) say, ``it is by now well established . . . that one

L. Campbell / Language Sciences 23 (2001) 113±161152

Page 41: What’s wrong with grammaticalization?

source concept can give rise to more than one grammatical category and that,conversely, a given grammatical category may be historically derived from morethan one source concept or structure.'' In short, since we have multiple lexicalsources converging on a single grammatical target and a single lexical source givenmultiple grammatical targets (results), strong claims for the predictive power ofgrammaticalization are clearly exaggerated; lacking predictability,grammaticalization theory thus lacks explanatory power.

But what about the possibility of a weaker or more general sort ofpredictability, going back to the recognition of constraints, that not all lexicalitems are as likely to undergo grammaticalization and that there are generally onlya limited number of grammatical outcomes? Might we entertain some notion ofprototype theory which says we expect (weakly predictive) that, for example, most`future' markers, if they are the results of grammaticalization, will come fromlexical sources originally mostly meaning only `want', `have', `go', `come', andperhaps a few other things, or that if a lexical item meaning `come' undergoesgrammaticalization, we expect the outcome most of the time to be `future', `nearpast', etc., and usually not something too far a®eld from these? It is true that wedo need to recognize and explain these ``constraints'' on what can grammaticalizeand what the result of grammaticalization can be. It is even conceivable that wemight envisage a prototype approach to this, seeing grammaticalizations of themore commonly recurrent sort as core, somehow more prototypical, with othersas further removed from the core and less prototypical. However, it is not at allclear that the explanation of these patterns and constraints (prototypes) wouldmost naturally reside in ``grammaticalization theory''. It seems more likely thatthe explanation of these facts will more naturally be found in the domain ofsemantic change, reanalysis, and grammatical structure in general.

Also, the emphasis placed on the recurrent grammaticalization channels as anapproach to prediction and hence to explanation would seem overly optimisticand would seem to fail to take into account the great number of very unusual anduncommon grammatical markers which arise through grammaticalizations and themany unusual and unexpected lexical sources which end up as grammaticalmarkers encountered in so many languages from all around the world. It becomesa matter of where to place the emphasis: should we stress the recurrent cases atthe expense of the many unexpected and unusual cases, or should we stress theunusual and unexpected cases, allowing this to divert attention from the caseswhich recur in language after language in grammaticalizations? Perhaps the onlyacceptable answer is that we cannot a�ord to miss either the common or theunusual. The main points, however, are: (1) The existence of the ``exotic''grammaticalizations calls into question any claims about strong predictability (andhence explanation) based essentially only on the ``common'' ones. (2) Theexplanation for the ``constraints'' on what lexical sources grammaticalize and whatgrammatical meanings/functions can be the outcomes of particular lexical sourceswill almost certainly require recourse to semantic change and grammar in general,and will not be explained by looking merely inside grammaticalization theoryalone. Thus, even if the constraints were to allow for weak predictability,

L. Campbell / Language Sciences 23 (2001) 113±161 153

Page 42: What’s wrong with grammaticalization?

grammaticalization remains derivative in that the explanation behind these factswould still lie outside grammaticalization itself.

5. Is grammar grammaticalization?

To this point we have been considering the prototypical view ofgrammaticalization, which involves diachronic change, changes of lexical >grammatical and grammatical > more grammatical. We now take up the di�erentview which conceives of grammaticalization as synchronic as well. This startedwith Paul Hopper's notion of ``emergent grammar,'' with its very di�erent andmuch more abstract view of what grammaticalization is:

Because grammar is always emergent but never present, it could be said that itnever exists as such, but is always coming into being. There is, in other words,no `grammar' but only `grammaticalization' Ð movements towards structurewhich are often characterizable in typical ways (Hopper, 1987, p. 148).

As Heine et al. (1991, p. 4) point out, this outlook makes ``grammaticalization''nearly synonymous with ``grammar.'' This is a very controversial view (cf. Lessau,1994, p. 384). Others have pointed out the two di�erent outlooks:``Grammaticalization can be described alternatively as a diachronic or asynchronic phenomenon'' (Heine et al., 1991, p. 2; cf. also Lehmann, 1986; Heineet al., 1991, p. 4). ``According to DuBois (1988, p. 11), there are two distinct typesof models that grammaticalization draws on: the `substance' and the `patternmodel'. The present framework, like most other works on the subject, is based onthe former. The main proponent of the `pattern model' is Paul Hopper with hisnotion of emergent grammar'' (Heine et al., 1991, p. 77). ``A grammaticalizationchain has both a diachronic and a synchronic dimension; diachronic in that it isthe result of language change, and synchronic in that it consists of a range ofsynchronically de®ned uses'' (Heine, 1993, p. 53).

More speci®cally, Hopper gives the characterization:

The `Emergence of Grammar' Ð attitude Ð has come to view grammar as thename for a vaguely de®ned set of sedimented (i.e. grammaticized) `recurrentpartials' whose status is constantly being renegotiated in speech and whichcannot be distinguished in principle from strategies for building discourses(Hopper, 1988, p. 118).

``The concept of grammar as emergent suspends provision for ®xed structure, andsees all structure as in a continual process of becoming, as epiphenomenal, and assecondary to the central fact of discourse'' (Hopper, 1992, p. 366). ``The study ofgrammaticalization itself has revealed no clear boundaries to the conceptgrammar'' (Ramat and Hopper, 1998, p. 4). Himmelmann (1992, p. 12) callsemergent grammar a ``radicalized discourse-based approach to grammar'' (cited in

L. Campbell / Language Sciences 23 (2001) 113±161154

Page 43: What’s wrong with grammaticalization?

Lessau, 1994, p. 248). (For other comments, see Bybee et al., 1994, pp. 1±2;Himmelmann, 1992, p. 2; Hopper and Traugott, 1993, p. 18; Lessau, 1994, p. 422;Lichtenberk, 1991, p. 76.)

With this view in mind, it begins to become clear why Lehmann (1995[1982],p. x) saw grammaticalization as ``such a pervasive process and therefore such acomprehensive notion that it is often di�cult to say what does not fall under it'',and why Hopper and Traugott (1993, p. 31) ask researchers to address thequestion, ``what phenomena in language are not examples of grammaticalization?''It is apparently this sort of thinking, which equates the grammar, or replaces thegrammar, with the process of grammaticalization, which leads Haspelmath (1998,1999) to see grammaticalization as broader and more inclusive, with reanalysis assomehow subordinate to grammaticalization. Thinking in somegrammaticalization circles of this sort brings into sharp focus the question of thestatus of grammaticalization. As we will see, it is confusion between the core viewof grammaticalization as diachronic, as change to more grammatical, and thisview, in which grammaticalization is synchronic and becomes largely synonymouswith grammar, which is behind some of the misunderstanding of whatgrammaticalization is and does and behind some of the misrepresentationsconcerning the explanatory power of grammaticalization. It is also behind theconfusion which places grammaticalization as a superordinate categoryencompassing many kinds of linguistic change which under the core view ofgrammaticalization fall far outside grammaticalization. Hopper (1998, p. 148)makes this clear:

It is worth emphasizing that because there is no accepted de®nition ofgrammar, there can be no distinct process of grammaticalization. A widerview of grammaticalization demands a modi®cation in our perspective ongrammar, one which sees structure in language as intrinsically un®xed andunstable, in other words as emergent (Hopper, 1987, 1988). From thisperspective, there would be no neat distinction between lexicalization andgrammaticalization, or any other aspect of change (emphasis added).

This constitutes a panchronic view of grammaticalization which a�rms anisomorphism between diachronic change and typology, which allows synchronicthings to be talked of diachronically and diachronic things to substitute forsynchronic grammar (cf. Tabor and Traugott's, 1998, pp. 263±264 discussion).

Lessau (1994, pp. 218±219) correctly perceives the di�culty of reconciling thesetwo di�erent views of what grammaticalization is:

For several reasons, including the following, it is a quite tricky job to identify acase of grammaticalization.(a) More and more things are recognized as belonging to grammar and togrammaticalization. The theoretical scope of `grammaticalization' broadenedconsiderably and the range of phenomena studied as `grammaticalization cases'is not restricted to morphology any more but includes diverse phenomena like

L. Campbell / Language Sciences 23 (2001) 113±161 155

Page 44: What’s wrong with grammaticalization?

metaphorization, syntacticization, etc. . . .Quite a number of authors includecognitive processes, thus clearly leaving behind the scope of most of traditionallinguistics.(b) Less and less [fewer and fewer] things are recognized as belonging clearly togrammar . . .(c) More and more things are recognized as belonging to general characteristicsof language change (instead of being speci®c to grammaticalization) . . .(d) More and more things are recognized as belonging to the diachroniccontext of grammaticalization.

The two views di�er greatly. For most, ``grammaticalization is essentially a kindof historical morphology while grammar is concerned with the synchronicfunctioning of the language system.'' However, for some, this relationship is ine�ect reversed, where ``G[rammaticalization] T[heory] itself may be conceived as atheory of grammar (or, somewhat less ambitiously, as contributing substantially toa theory of grammar)'' (Himmelmann, 1992, p. 1). Those who follow Hopper'snotion of ``emergent grammar'' are among the latter group. In the ®rst, betterknown view, grammaticalization is but one type of change, a subset of the kindsof changes which a�ect grammar. However, there are some scholars, e.g. Hopper,1991; Lehmann, 1995; Shibatani, 1991; Haspelmath, 1998, 1999), and severalothers, who see grammaticalization as a broader category which encompassesgrammatical (syntactic) change in general (including reanalysis, word orderchange, etc.). (See Himmelmann, 1992, p. 1; Lessau 1994, pp. 396±397; andRamat and Hopper, 1998, p. 3 for discussion.)

Still other scholars who follow the Hopper line are concerned more exclusivelywith synchronic aspects of language, especially with the grammar±discourse±pragmatics interface, which is what grammaticalization means to them. For themchange is secondary; their emphasis in on how grammatical things function, noton how they came to exist. The following is representative of this outlook:

The term grammaticization as used in this work is thus not just referring to thehistorical process whereby a lexical item becomes grammatical or a lessgrammatical item becomes more grammatical (Heine et al., 1991, p. 3), butalso, it refers to the more fundamental ontology of grammar, i.e. to the ways inwhich grammar can be said to exist: grammar is `a vaguely de®ned set' ofgrammaticized recurrent patterns of discourse . . .Thus, in order to studygrammar we need to look at discourse and search for grammaticized recurrentpatterns (Helasvuo, 1997, p. 2).

From a grammaticization perspective we could say that languages resort tovery di�erent strategies in trying to handle the various discourse needs thatspeakers are faced with, such as the need to express events or track participantsof an event. We need careful discourse studies on individual languages in orderto build a crosslinguistic understanding of how grammar works (Helasvuo,1997, p. 186).

L. Campbell / Language Sciences 23 (2001) 113±161156

Page 45: What’s wrong with grammaticalization?

This is not the classical, core ``grammaticalization'' we have been discussing;rather, this view is concerned primarily with the function and use of ``grammar''written large to include discourse and pragmatics, with less direct concern for thediachronic events which are the focus of the more standard view ofgrammaticalization. While there is nothing inherently wrong with this di�erentoutlook, it seems unnecessary to equate it with ``grammaticalization.'' In one sensethis use of the term ``grammaticalization'' is like a substitute for ``grammar'' so asto be able to avoid the connotations the term ``grammar'' has in other, moreformal theories of linguistics. In another sense it is in large measure equivalent tothe typological-functional approach to linguistics in general. If``grammaticalization'' is to be used in these senses, then classical``grammaticalization'' (of the less grammatical> more grammatical sort) will needdi�erent terminology to designate it. I see little reason for using``grammaticalization'' for these senses; terms such as ``grammar,'' ``emergentgrammar,'' and ``typology'' cover this terrain nicely enough.

6. Conclusions

Let me conclude with some pointed questions (problems?) forgrammaticalization theory and some brief answers based on what has beendiscussed in this paper.

1. How crucial is unidirectionality to the de®nition/success ofgrammaticalization? To the extent that unidirectionality is considered an empiricalissue, it cannot be diagnostic of grammaticalization, since there are signi®cantcounterexamples to the unidirectionality claim. To the extent that unidirectionalityis a de®nitional property of grammaticalization (built into the core de®nition ofgrammaticalization as lexical> grammatical or grammatical>more grammatical ),it cannot be tested, but results in anything not conforming to the assumeddirectionality being considered outside the universe of discourse ofgrammaticalization. Such a move o�ers no insights or explanations. To the extentthat changes tend to go lexical> grammatical and not the other way around, thisis explained not by grammaticalization itself, but by the interaction of the kinds ofchange which lie behind grammaticalization (phonological, semantic, reanalysis).

2. How crucial are semantic bleaching and phonological reduction (erosion) togrammaticalization? Neither is necessary or su�cient, since there are cases ofgrammaticalization which involve no semantic bleaching or no phonologicalreduction, and since there are changes which are not instances ofgrammaticalization which do involve one or the other (or both) of these.Moreover, the semantic or phonetic change is often temporally separated from thegrammaticalization with which it is thought to be associated. For example,sometimes semantic changes precede morphosyntactic changes, sometimes theyaccompany them, and sometimes they follow them. Phonological reductionusually takes place after the grammatical changes, sometimes long after, if at all,but in some instances prior phonological change creates conditions which foster

L. Campbell / Language Sciences 23 (2001) 113±161 157

Page 46: What’s wrong with grammaticalization?

shifts in the status of grammatical items. Moreover, as should be obvious fromthe discussion in this paper and from general knowledge about linguistic change,the semantic changes and the phonetic reductions associated withgrammaticalization, the ``parallel reduction hypothesis'' notwithstanding, areusually fully independent of one another and usually do not take place at thesame time or ``proceed at the same pace'' in grammaticalization changes (cf.Haboud, 1997, pp. 202±203).

3. About the claims for the explanatory value of grammaticalization: (1) is it infact possible to ``explain'' based on grammaticalization theory? And, (2) why cana single lexical source develop in multiple directions, giving various and di�erentgrammatical ``targets'' as a result, while on the other hand a single grammaticaltarget (an outcome of grammaticalization) can come about in di�erent languagesfrom numerous di�erent lexical sources? An example of the ®rst is the variousgrammatical categories (`targets') known to develop from grammaticalizations of`come' as their source (see discussion above)?

The real explanatory power lies behind grammaticalization, not ingrammaticalization itself, but in the kinds of linguistic changes that are at play ingrammaticalization, in the factors and forces that drive phonological change,semantic change, and in particular in reanalysis. Grammaticalizations involvereanalysis, but reanalysis is a much more powerful mechanism of grammaticalchange and is not limited to nor co-extensive with grammaticalization. Similarly,sound change and semantic change apply to many things in addition togrammaticalizations. These ``explain'' instances of grammaticalization, butgrammaticalization itself explains nothing without ®rst calling upon these kinds ofchanges and the explanations they a�ord.

Is grammaticalization derivative and therefore lacking a status of its own? Ihave argued in this paper that grammaticalization has no independent status of itsown. Cases of grammaticalization are explained adequately by the othermechanisms of linguistic change, and grammaticalization explains nothing by itselfbut must rely on these other mechanisms and kinds of linguistic change. Does thismean that grammaticalization has no value? I would say the answer is ``no.''While grammaticalization does not have the theoretical value that some haveattributed to it, and while it lacks any status of its own independent of otherkinds of linguistic changes and mechanisms of change, it does have heuristicvalue. The enthusiasm for grammaticalization has brought forth a range ofexamples (especially the lexical > grammatical, less grammatical > moregrammatical sort) which is important for providing a broad and useful databaseand for focussing attention on a particular set of changes within the broaderpicture of linguistic change. Work in grammaticalization has given us a very richassembly of examples we would be unaware of had it not been done. These resultsinform us signi®cantly about various natural typological arrangements cross-linguistically and about certain kinds of recurrent changes. This is all valuableinformation, even if not of the kind of theoretical or explanatory value manygrammaticalizationists have assumed.

L. Campbell / Language Sciences 23 (2001) 113±161158

Page 47: What’s wrong with grammaticalization?

References

Allen, C., 1997. The origins of the `group genitive' in English. Transactions of the Philological Society

95, 111±131.

Ariste, P., 1973. Eesti roÄ humaÈ aÈ rsona ep (The a�rmative adverb ep in Estonian). Journal de la Socie teÂ

Finno-ougrienne 72, 33±37.

Bergman, G., 1973. A Short History of the Swedish Language. Berlingska Boktryckeriet, Lund

(Translated and adapted by F.P. Malgoun Jr, H. KoÈ keritz).

Bisang, W., 1998. Grammaticalization and language contact, constructions and positions. In: Ramat,

A.G., Hopper, P.J. (Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp. 13±58.

Bybee, J., Pagliuca, W., 1985. Cross-linguistic comparison and the development of grammatical

meaning. In: Fisiak, J. (Ed.), Historical Semantics, Historical Word Formation. Mouton, Berlin, pp.

59±83.

Bybee, J., Pagliuca, W., Perkins, R., 1994. The Evolution of Grammar: Tense, Aspect, and Modality in

the Languages of the World. University of Chicago Press, Chicago.

Campbell, L., 1991. Some grammaticalization changes in Estonian. In: Traugott, E.C., Heine, B. (Eds.),

Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 285±299.

Campbell, L., 1998. Historical Linguistics: An Introduction. Edinburgh University Press, Edinburgh

(1999: MIT Press, Cambridge, MA).

Craig, C., 1991. Ways to go in Rama: a case study in polygrammaticalization. In: Traugott, E.C.,

Heine, B. (Eds.), Approaches to Grammaticalization, vol. II. Benjamins, Amsterdam, pp. 455±492.

DuBois, J. 1988. Discourse as pattern model for grammar: the possessor=ergator a�liation. Paper

presented at the Symposium on Grammaticalization, University of Oregon. (Not seen, cited by

Heine et al., 1991 p. 77).

Fischer, O., 1997. On the status of grammaticalisation and the diachronic dimension in explanation.

Transactions of the Philological Society 95, 149±187.

von der Gabelentz, G., [1891] 1901. Die Sprachwissenschaft: Ihre Aufgaben, Methoden und bisherigen

Ergebnisse, 2nd ed. Weigel Nachf, Leipzig.

Gaeta, L., 1998. Some remarks on analogy, reanalysis, and grammaticalization. In: Ramat, A.G.,

Hopper, P.J. (Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp. 89±105.

Givo n, T., 1975. Serial verbs and syntactic change: Niger-Congo. In: Li, C.N. (Ed.), Word Order and

Word Order Change. University of Texas Press, Austin, pp. 47±112.

Givo n, T., 1979. On Understanding grammar. Academic Books, New York.

Givo n, T. 1990. Isomorphism in the grammatical code: cognitive and biological considerations.

Unpublished manuscript, Department of Linguistics, University of Oregon. [Cited by Heine et al.,

1991, p. 214, not seen].

Haboud, M., 1997. Grammaticalization, clause union, and grammatical relations in Ecuadorian

Highland Spanish. In: Givo n, T. (Ed.), Grammatical Relations: A Functionalist Perspective.

Benjamins, Amsterdam, pp. 199±231.

Haiman, J., 1991. From V/2 to subject clitics: evidence from Northern Italian. In: Traugott, E.C.,

Heine, B. (Eds.), Approaches to Grammaticalization, vol. II. Benjamins, Amsterdam, pp. 135±158.

Hakulinen, L., 1968. Suomen kielen rakenne ja kehitys. Ottava, Helsinki.

Harris, A.C., Campbell, L., 1995. Historical Syntax in Cross-linguistic Perspective. Cambridge

University Press, Cambridge.

Haspelmath, M., 1998. Does grammaticalization need reanalysis? Studies in Language 22, 49±85.

Haspelmath, M., 1999. Why is grammaticalization irreversible? Linguistics 37, 1043±1068.

Heine, B., 1993. Auxiliaries: Cognitive Forces and Grammaticalization. Oxford University Press,

Oxford.

Heine, B., 1997. Possession: Cognitive Sources, Forces, and Grammaticalization. Cambridge University

Press, Cambridge.

Heine, B., Reh, M., 1984. Grammaticalization and Reanalysis in African Languages. Buske, Hamburg.

Heine, B., Claudi, U., Htinnemeyer, F., 1991. Grammaticalization: A Conceptual Framework.

University of Chicago Press, Chicago.

L. Campbell / Language Sciences 23 (2001) 113±161 159

Page 48: What’s wrong with grammaticalization?

Helasvuo, M.-L. 1997. When discourse becomes syntax: noun phrases and clauses as emergent syntactic

units in Finnish conversational discourse. Ph.D. dissertation, University of California, Santa

Barbara.

Herring, S.C., 1991. The grammaticalization of rhetorical questions in Tamil. In: Traugott, E.C.,

Heine, B. (Eds.), Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 253±284.

Himmelmann, N. 1992. Grammaticalization and grammar. (Arbeitspapier 16). Institut fur

Sprachwissenschaft, UniversitaÈ t KoÈ ln, Cologne.

Hopper, P.J., 1987. Emergent grammar. Berkeley Linguistics Society 13, 139±157.

Hopper, P.J., 1988. Emergent grammar and the a priori grammar postulate. In: Tannen, D. (Ed.),

Linguistics in Context. Ablex, Norwood, NJ, pp. 117±134.

Hopper, P.J., 1991. On some principles of grammaticization. In: Traugott, E.C., Heine, B. (Eds.),

Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 17±35.

Hopper, P.J., 1992. Grammaticalization. In: Bright, W. (Ed.), International Encyclopedia of

Linguistics, vol. 2. Oxford University Press, Oxford, pp. 79±81.

Hopper, P., 1998. The paradigm at the end of the universe. In: Ramat, A.G., Hopper, P.J. (Eds.), The

Limits of Grammaticalization. John Benjamins, Amsterdam, pp. 147±158.

Hopper, P.J., Traugott, E.C. (Eds.), 1993. Grammaticalization. Cambridge University Press,

Cambridge.

Itkonen, E., 2000. MitaÈ on `kieliopillistuminen'? Analogiasta ja typologiasta. Suomen Kirjallisuuden

Seura, Helsinki (In press).

Jespersen, O., 1894. Progress in Language: with Special Reference to English. Swan Sonneschein,

London.

Joseph, B., 1990. Morphology and Universals in Syntactic Change: Evidence From Medieval and

Modern Greek. Garland, New York.

Joseph, B., Janda, R.D., 1988. The how and why of diachronic morphologization and

demorphologization. In: Hammond, M., Noonan, M. (Eds.), Theoretical Morphology. Academic

Press, San Diego, pp. 193±210.

Keesing, R.M., 1991. Substrates, calquing and grammaticalization in Melanesian pidgin. In: Traugott,

E.C., Heine, B. (Eds.), Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 315±

342.

Keller, R., 1990. Sprachwandel: von der unsichtbaren Hand in der Sprache. Francke, TuÈ ibingen

(English translation, 1994: Language Change: the Invisible Hand in Language. Routledge,

London.).

Kettunen, L. 1930. Suomen murteet II: murarealueet. [Finnish dialects II: dialect areas.] (Suomalaisen

Kirjallisuuden Seura, Toimituksia 188.) Suomalaisen Kirjallisuuden Seura, Helsinki.

Kuryl/owicz, J., 1965. The evolution of grammatical categories. Diogenes 51, 55±71.

Laalo, K., 1988. Imperfektimuojojen ti0 Si- vaihtelu Suomen kielessaÈ . Suomen Kirjallisuuden Seura,

Helsinki.

Laanest, A., 1982. EinfuÈ hrung in die ostsee®nnischen Sprachen. Buske, Hamburg.

Langacker, R.W., 1977. Syntactic reanalysis. In: Li, C.N. (Ed.), Mechanisms of Syntactic Change.

University of Texas Press, Austin, pp. 59±139.

Lehmann, C., 1986. Grammaticalization and linguistic typology. General Linguistics 26, 3±23.

Lehmann, C. 1995[1982]. Thoughts on grammaticalization. Lincom Europa, Munich.

Lessau, D.A., 1994. A Dictionary of Grammaticalization. Brockmeyer, Bochum.

Lichtenberk, F., 1991. On the gradualness of grammaticalization. In: Traugott, E.C., Heine, B. (Eds.),

Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 37±80.

Matiso�, J., 1991. Areal and universal dimensions of grammaticalization in Lahu. In: Traugott, E.C.,

Heine, B. (Eds.), Approaches to Grammaticalization, vol. II. Benjamins, Amsterdam, pp. 383±454.

Matsumoto, Y., 1988. From bound grammatical marker to free discourse marker: history of some

Japanese connectives. In: Axmaker, S., Jaisser, A., Singmaster, H. (Eds.), Berkeley Linguistics

Society 14 (Parasession on Grammaticalization). Berkeley Linguistics Society, Berkeley, pp. 340±

351.

Meillet, A. 1912. L'e volution des formes grammaticales. Scientia 12 (26) (Milan). (Reprinted 1951:

Linguistique historique et linguistique ge ne rale. C. Klincksieck, Paris, pp. 130±148).

L. Campbell / Language Sciences 23 (2001) 113±161160

Page 49: What’s wrong with grammaticalization?

Moreno Cabrera, J.C., 1998. On the relationship between grammaticalization and lexicalization. In:

Ramat, A.G., Hopper, P.J. (Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp.

211±227.

Morrison, T., 1977. Song of Solomon. Knopf, New York.

Nevis, J., 1985. Language-external evidence for clitics as words: Lappish particle clitics. Chicago

Linguistic Society 21, 289±305.

Nevis, J., 1986. Decliticization and dea�xation in Saame: abessive taga. In: Joseph, B.D. (Ed.), Studies

on Language Change (Working Papers in Linguistics, 34). Ohio State University, Department of

Linguistics, Columbus, pp. 1±9.

Norde, M., 1997. The History of the Genitive in Swedish: A Case Study in Degrammaticalization.

Vakgroep Skandinavische taal- en letterkunde, Amsterdam.

Oinas, F.J. 1961. The development of some postpositional cases in Balto±Finnic languages. (MeÁ moires

de la Socie te Finno-Ougrienne, 123.) Suomalais-Ugrilainen Seura, Helsinki.

Pagliuca, W., 1994. Introduction. In: Pagliuca, W. (Ed.), Perspectives on Grammaticalization.

Benjamins, Amsterdam, pp. ix±xx.

Ramat, A.G., 1998. Testing the boundaries of grammaticalization. In: Ramat, A.G., Hopper, P.J.

(Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp. 107±127.

Ramat, A.G., Hopper, P., 1998. Introduction. In: Ramat, A.G., Hopper, P.J. (Eds.), The Limits of

Grammaticalization. Benjamins, Amsterdam, pp. 1±11.

Ramat, P., 1992. Thoughts on degrammaticalization. Linguistics 30, 549±560.

Ravila, P., 1975. Johdatus kielihistoriaan. Suomen Kirjallisuuden Seura, Helsinki.

Rubino, C., 1994. Against the notion of unidirectionality in lexeme genesis. Linguistica Atlantica 16,

135±147.

Shibatani, M., 1991. Grammaticalization of topic into subject. In: Traugott, E.C., Heine, B. (Eds.),

Approaches to Grammaticalization, vol. II. Benjamins, Amsterdam, pp. 93±134.

Sweetser, E., 1988. Grammaticalization and semantic bleaching. In: Axmaker, S., Jaisser, A.,

Singmaster, H. (Eds.), Berkeley Linguistics Society 14 (Parasession on Grammaticalization).

Berkeley Linguistics Society, Berkeley, pp. 389±405.

Sweetser, E., 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic

Structure. Cambridge University Press, Cambridge.

Tabor, W., Traugott, E.C., 1998. Structural scope expansion and grammaticalization. In: Ramat, A.G.,

Hopper, P.J. (Eds.), The Limits of Grammaticalization. Benjamins, Amsterdam, pp. 229±272.

Traugott, E.C., 1988. Pragmatic strengthening and grammaticalization. In: Axmaker, S., Jaisser, A.,

Singmaster, H. (Eds.), Berkeley Linguistics Society 14 (Parasession on Grammaticalization).

Berkeley Linguistics Society, Berkeley, pp. 406±416.

Traugott, E.C., 1995. The role of the development of discourse markers in a theory of

grammaticalization. Paper given at the 12th International Conference on Historical Linguistics;

Manchester; 13±18, August, 1995.

Traugott, E.C., Heine, B., 1991. Introduction. In: Traugott, E.C., Heine, B. (Eds.), Approaches to

Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 1±14.

Traugott, E.C., KoÈ nig E, 1991. The semantics-pragmatics of grammaticalization revisited. In: Traugott,

E.C., Heine, B. (Eds.), Approaches to Grammaticalization, vol. I. Benjamins, Amsterdam, pp. 189±

218.

Wesse n, E., 1969. Svensk spraÊ khistoria I: ljudlaÈ ra och ordboÈ jningslaÈ ra. Almqvist & Wiksell, Stockholm.

Whaley, L.J., 1997. Introduction to Typology: the Unity and Diversity of Language. Sage, Thousand

Oaks, CA.

Wright, L., 1976. Teleological Explanation. University of California Press, Berkeley.

L. Campbell / Language Sciences 23 (2001) 113±161 161