Studies In Stemmatology II

Studies in Stemmatology II

ltDOCINFO AUTHOR TITLE Studies in Stemmatology IISUBJECT KEYWORDS SIZE HEIGHT 220WIDTH 150VOFFSET 4gt


Edited by

Pieter van Reenen

August den Hollander

Margot van Mulken

With the assistance of Annelies Roeleveld

John Benjamins Publishing Company

AmsterdamPhiladelphia

The paper used in this publication meets the minimum requirements8 TM

of American National Standard for Information Sciences ndash Permanenceof Paper for Printed Library Materials ansi z3948-1984

Publication of this volume was financially supported by the Netherlands organization

for scientific research (NWO)

Library of Congress Cataloging-in-Publication Data

Studies in Stemmatology II edited by Pieter van Reenen August denHollander and Margot van Mulken

p cmIncludes bibliographical references and indexes

1 Manuscripts 2 Manuscripts Medieval 3 Transmission of texts IReenen Pieter Th van II Hollander A A den III Mulken Margot van

Z105S782 2004091-dc22 20040100159isbn 90 272 3222 9 (Eur) 1 58811 535 6 (US) (Hb alk paper)

copy 2004 ndash John Benjamins BVNo part of this book may be reproduced in any form by print photoprint microfilm orany other means without written permission from the publisher

John Benjamins Publishing Co middot PO Box 36224 middot 1020 me Amsterdam middot The NetherlandsJohn Benjamins North America middot PO Box 27519 middot Philadelphia pa 19118-0519 middot usa

Table of contents

Prologue vii

I Stemmatological methods and techniques

Parallels between stemmatology and phylogenetics 3Christopher Howe Adrian Barbrook Linne Mooneyand Peter Robinson

Problems of a highly contaminated tradition the New TestamentStemmata of variants as a source of a genealogy for witnesses 13

Gerd Mink

Kinds of variants in the manuscript tradition of the Greek New Testament 87Klaus Wachtel

How shock waves revealed successive contamination A cardiogramof early sixteenth-century printed Dutch Bibles 99


The manuscript tradition of the Cligeacutes of Chreacutetien de TroyesA stemmatological approach 113

Margot van Mulken

II Textual variation

Genealogy by chance On the significance of accidentalvariation (parallelisms) 127

Ulrich Schmid

Constructing initial binary trees in stemmatology 145Evert Wattel

Trouble in the trees Variant selection and tree constructionillustrated by the texts of Targum Judges 167

Willem F Smelik

Table of contents

Scribal variations When are they genealogically relevant ndash and whenare they to be considered as instances of lsquomouvancersquo 207

Lene Schoslashsler

The effects of weighting kinds of variants 227Matthew Spencer Linne R Mooney Adrian C Barbrook BarbaraBordalejo Christopher J Howe and Peter Robinson

Cluster analysis and the Three Level Method in the study of theGospels in Slavonic 241

Dina Mironova

Different kinds of tradition in Targum Jonathan to Isaiah 269Alberdina Houtman

Valentin and Namelos discover their parentage Narrative elementsin the family tree of an international medieval tale 285

Annelies Roeleveld Erika Langbroek and Evert Wattel

Index 305

Prologue

The publication of this volume of Studies in Stemmatology is the second in aseries Its predecessor was published in 1996 and opened the most actual stateof the art in stemmatology to a broad audience1 That volume not only aimed atgiving scholars access to modern stemmatological methods and techniques butalso at illustrating how profitable the application of these methods might be fortheir future work The first volume was very well received by stemmatologistsall over Europe and also gave an impulse to new research as several articles inStudies in Stemmatology II clearly illustrate

The contributions to this present book partly proceed from those of thefirst volume Most of them are the result of the on-going scholarly debate onstemmatology of recent years Several of the contributions to this volume werepresented on 13ndash14 April 2000 during the NOSTER-conference at the Nether-lands Institute of Advanced Studies (NIAS) at Wassenaar and on 13 October2000 during the Stemmatology Conference at the Vrije Universiteit in Ams-terdam Some others are the result of the annual colloquia of stemmatologistsheld at the Vrije Universiteit in Amsterdam2

The object of this second volume of Studies in Stemmatology is the evalua-tion of the most recent methods and techniques in the field of stemmatologyas well as the development of new ones The book is largely interdisciplinaryin character it contains contributions from scholars from classical historicalbiblical (Smelik Houtman Den Hollander) medieval and modern languagestudies as well as from mathematical and computer scientists (Wattel) and bi-ologists (Howe ea Spencer ea) Various manuscript traditions are dealt withhere some of them within one field of language (Van Mulken Schoslashsler) somemultilingual (RoeleveldLangbroekWattel) the last group of course requir-ing a special methodological approach to the establishment of variants Othertraditions were very extensive eg the New Testament manuscripts (MinkWachtel) and the Old Church Slavonic manuscripts (Mironova) The contri-butions in the book have been divided into two sections The first section dealswith various stemmatological methods and techniques The second section fo-cusses more specifically on the various problems concerning textual variation

Prologue

Stemmatological methods

Not a bifurcating tree

Christopher Howe Adrian Barbrook Linne Mooney and Peter Robinsonpresent a relatively new stemmatological approach They explore the similar-ities between the evolution of DNA sequences and the changes occurring inmanuscript traditions They show how the techniques of evolutionary biol-ogy can be applied to stemmatic analysis and how a number of features ofmanuscript traditions have clear parallels in genetics They conclude that theprocess of incorporation into DNA mirrors the incorporation of changes intothe manuscripts It follows that programs for phylogenetic analysis of sequencedata can be exploited for stemmatic analysis of manuscript tradition For thisapproach the Splits Tree program is used which has the advantage that it doesnot presuppose as many methods do that the tree is a bifurcating one

Local stemmata

Methodologically new and very promising is the contribution of Gert MinkIn his contribution he broaches the problem of a text tradition of many hun-dreds of manuscripts in which hardly any type-2 variants are found and con-tamination is the rule This is the case in one of the catholic letters of theNew Testament the Epistle of James In such a tradition existing methodscannot be applied Instead Mink has two working hypotheses on which hisapproach is based

a If more than one exemplar was consulted by a scribe the exemplars areclosely related

b Variants are analysed one by one in local trees

Within the local trees the direction of the changes can be determined they canbe oriented by establishing which variant derives from which other variantWhen groups of local trees are oriented in the same direction parts of globaltrees can be constructed

Reduction of witnesses

Klaus Wachtel also deals with this extremely large textual tradition In his con-tribution he shows how the number of manuscripts to examine can be reducedbefore the structuring of a stemma The number of extant manuscripts of the

Prologue

New Testament is so large that any reasonable form of reduction of quanti-ties must be accepted before the building of a stemma starts By distinguishingtwo groups of manuscripts in the New Testament tradition Wachtel succeeds indoing so The two groups distinguished are the Majority group and the Byzan-tine group If two or more manuscripts are almost alike there is no need forfurther analysis By applying this approach Wachtel succeeds in reducing thenumbers considerably without the risk of excluding manuscripts which con-tain crucial textual information The resulting group forms the input of GertMinks analysis

Dealing with successive contamination

A illuminating example of how profitable the application of modern stemma-tological tools can be is given in the contribution of August den HollanderOne of the complex problems a philologist has to deal with is a contaminatedtext tradition In the first volume of Studies in Stemmatology Wattel and VanMulken offered the instrument of the so-called shock waves (cardiograms) asa help to reveal successive contamination in a text tradition which is ratherentangled In his contribution Den Hollander shows how the application ofthis instrument indisputably revealed successive contamination in the textualtradition of early sixteenth-century printed Dutch Bibles

In her contribution to the present volume Margot van Mulken showsthat the output of the quire separator developed by Wattel (see first volume)may have serious consequences for the further treatment of the stemmatolog-ical process When the separator indicates successive contamination as in thecase of the Cligeacutes it may be necessary to presuppose a multiple orientationof the stemmata However in the case of the Cligeacutes all the archetypes can befound in the neighbourhood of one manuscript which fortunately reduces thecomplexity of this operation

Textual variation

Accidental variation

Ulrich Schmid explores the phenomenon of accidental variation (parallelism)His contribution is a reaction to the recent study of B J P Salemans whosystematically reviewed various types of variant readings used in genealogicalstudies and offered strict text-genealogical rules in order to exclude possible

Prologue

variants caused by accidental variation3 In his contribution Schmid illustratesthe implications of applying Salemansrsquo rules to a text tradition on the onehand they would exclude too much leaving out many genealogically lsquovalidrsquovariants on the other hand even the variants that would be included on thebasis of Salemansrsquo rules still contain parallelistic readings Therefore Schmidconcludes no safe line can be drawn without proper statistical evaluation

No reduction of variants

Evert Wattel also writes as a reaction to the dissertation of Salemans Textualscholars do not generally agree on which type of variant readings are suitablefor the construction of a stemma and which are not In his contribution EvertWattel argues for the acceptance of as many version formulas as possible in ad-dition to expressing the reliability of the variants by adding more or less weightto the so-called version formulas His main focus is on the computational prob-lems of constructing a stemma on the basis of the collective formulas Alongwith a more general methodological discussion he dwells on specific problemssuch as lacunary version formulas and the computational complexity

Categories of variants

In his contribution Willem Smelik deals with variant selection and tree con-struction in the text tradition of Targum Judges The core of this study consistsof observations on the phenomenon of random variation in the manuscript re-production To identify random or coincidental variation he suggests a trans-parent verifiable categorisation of variant readings Secondly he discusses thepossible genealogical information of these various types of variants in great de-tail Further Smelik draws stemmata for each type or group of types of variantsFinally comparison of these stemmata reveals which types of variants turnedout to be genealogical relevant in his textual tradition and which not

Lene Schoslashsler compares the categorised variants of two closely relatedmanuscripts of the Perceval and four (five) manuscripts of the Charroi deNicircmes The first two were copied by the same scribe with perhaps a differencein time Assuming that the scribe copied twice from the same exemplar it isremarkable that the variations found between the two manuscripts and thosein the tradition of the Charroi de Nicircmes are hardly different In other wordswhether the same scribe copies the same manuscript twice or different scribescopy a manuscript may not necessarily result in more variants

Prologue

Weighting variants

In their contribution Matthew Spencer Linne Mooney Adrian Barbrook Bar-bara Bordalejo Christopher Howe and Peter Robinson attempt to increasethe chance of reconstructing correct stemmas by categorizing variants intoten different kinds such as ldquoline changed completelyrdquo ldquoword change affect-ing rhymerdquo ldquoword variant changes meaningrdquo ldquominor word added or omittedwithout changing meaningrdquo On the assumption that not all kinds of variantsare equally reliable the more a category of variants is reliable the more weight itis assigned On comparison between stemmata of the 55 manuscripts and threeprinted versions of Lydgatersquos Kings of England the choice of weights appearedrelatively unimportant However the authors expect that this may be differentin larger textual traditions The method used to reconstruct the stemmata wasneighbour-joining a simple clustering algorithm which sequentially separatespairs of manuscripts from an initially unresolved stemma

Dina Mironova deals with the problem of a textual tradition of manymanuscripts She compares two different formal genealogical methods in herstudy of the Gospels in Slavonic cluster analysis (Alexeev) and the Three LevelMethod (Wattel) Her research comprises no fewer than 531 manuscripts stillpresenting however a rather stable text The large number of witnesses im-pelled her to work with groups of manuscripts as a way of reduction Alexeevrsquosmethod turned out be less accurate but more economical since it is still eas-ier to apply to large traditions Wattelrsquos method is however more accurateand forces the scholar to formulate precise classifications or explicit philolog-ical labellings Despite the difficulties with extremely large textual traditionsaccording to Mironova his method is to be preferred when variants should beevaluated (weighted)

Exclusion of variants

Dineke Houtman studies the textual history of Tosefta Targum Jonathan anextended Aramaic Bible commentary She focuses on the question of how todeal with this type of text in stemmatological research especially when com-paring it with the Hebrew Bible text and its paraphrasing Aramaic Targumtext All three types of text represent different stages in the textual history Thetext of the Targum remains close to the Hebrew Bible the text of Tosefta Tar-gum however gives a more free rendering Houtman concludes that includingtextual variants from the Tosefta Targum may introduce a lot of bias in the re-

Prologue

sults of the stemmatological research and should therefore be done with greatprecaution

Alternative classification of variants

Annelies Roeleveld and Erika Langbroek in cooperation with Evert Watteldeal with the text tradition of lsquoValentin and Namelosrsquo Its extant manuscriptsare written in no fewer than four languagesdialects Middle Dutch MiddleLow German Middle Mid German and Old Swedish Other differences be-tween the texts are also considerable some are in verse others in prose someare much more lengthy and elaborate in their descriptions than others somehave been preserved in fragments only The problem in this tradition is whatsort of variant to group in comparable units The authors develop a classifica-tion in which the number of incidents the order of incidents and the detail inthe description of incidents play a part Although this notion of variant goesbeyond the traditional view their approach shows that satisfactory results canbe obtained The resulting stemma plausibly shows that the Middle Dutch ver-sions are the more original Rhyme analysis had already pointed in the samedirection the original language was most probably Middle Dutch

The editors hope that this second volume of Studies in Stemmatology will in-spire scholars like the first volume did and stimulate the development of newmethods and strategies aiming at further control of variation and contamina-tion in (large) text traditions

Pieter van ReenenAugust den Hollander

Margot van Mulken

Notes

Studies in Stemmatology (1996) Pieter van Reenen amp Margot van Mulken (Eds) Amster-damPhiladelphia John Benjamins Publishing Co

At this place we wish to thank NIAS NOSTER and the Faculty of Arts of the Free Uni-versity for their willingness to support the various stemmatological meetings in the pastyears

B J P Salemans (2000) Building Stemmas with the Computer in a Cladistic Neo-Lachmannian way The Case of Fourteen Text Versions of Lanseloet van Denemerken (dissNijmegen) Nijmegen

P I

Stemmatological methods and techniques

Parallels between stemmatologyand phylogenetics

Christopher Howe Adrian Barbrook Linne Mooneyand Peter RobinsonUniversity of Cambridge University of Maine Orono De Montfort University Leicester

Introduction

The work and ideas we discuss here are part of a project entitled ldquoSTEMMA ndashStudies on Textual Evolution of Manuscripts by Mathematical Analysisrdquofunded by the Leverhulme Trust This project aims to apply the techniquesof evolutionary biology to the analysis of manuscript traditions In particu-lar we are interested in the application of computer programs developed forevolutionary biology to the study of manuscripts In this paper we explore thesimilarities between the evolution of DNA sequences and the changes occur-ring in manuscript traditions We will show how the techniques of evolutionarybiology can be applied to stemmatic analysis and how a number of features ofmanuscript traditions have clear parallels in genetics Another paper in thisvolume (Spencer et al) to which this chapter should serve as an introductiondiscusses more specific issues in this work

The computer programs we are using were developed for research in biol-ogy so their application to manuscript stemmatics requires some knowledgeof the underlying biology A more detailed discussion can be found in biologi-cal textbooks (eg Voet et al 1999) DNA (deoxyribonucleic acid) is comprisedof four kinds of unit These are collectively called nucleotides (more fully de-oxyribonucleotides) and the four kinds are adenosine guanosine cytidine andthymidine deoxyribonucleotides They are more conveniently designated A GC and T DNA molecules are composed of chains of such nucleotides and theorder in which individual nucleotides comes carries the information used in

Christopher Howe et al

biological systems It specifies the order of amino acids in proteins and thusthe structure and function of proteins which are the main functional entitiesin cells For example enzymes which catalyse the reactions of metabolismare proteins

In most systems the DNA chains come in pairs ndash forming the double helixmade famous by Watson and Crick (Watson amp Crick 1953) The sequence ofnucleotides in one strand determines the sequence in the other Thus A in oneis always opposite T in the other G in one opposite C in the other As cells di-vide their genetic information has to be duplicated This process is termedDNA replication and the aim is to make an identical copy of the parentalmolecule However errors can occur in the replication process so a parentalmolecule with sequence say

ACGGTACTAGTGCCATGATC

might give rise to two daughters one of which had the same sequence and theother of which had a different sequence say

ACGGCACTAGTGCCGTGATC

Here the T in the fifth position in the upper strand has been replaced by a Cand in the lower strand an A has been replaced by a G We say that a lsquomutationrsquohas occurred and the information in the DNA has been altered

Recovering phylogenetic trees

As biological species evolve and give rise to new species they accumulate mu-tations in their DNA The longer it is since two species had a common ancestorthe more different (in general) is their DNA sequence So we can use the differ-ences in DNA sequence among species as a way of inferring their evolutionaryrelationships We can recover something akin to a family tree showing whichspecies share a common ancestor to the exclusion of others This is called a phy-logenetic tree Recovering a phylogenetic tree using a given DNA sequence fora group of species requires both a model for sequence evolution and a methodfor tree recovery

Models of sequence evolution are in effect a set of assumptions about howthe sequences change They may be explicit (in which case the computer pro-gram being used will require them to be specified or use default settings) or

Parallels between stemmatology and phylogenetics

Suppose we have species W X Y and Z with the following sequences (in reality muchlonger sequences would be used)

W AAAAAAAAX GGAAAAAAY CCTTTTAAZ CCTTCCAA

The distance matrix would be

W X Y ZW ndash 2 6 6X ndash ndash 6 6Y ndash ndash ndash 2

and the tree inferred would be

4

1 1

11

W

X

Y

Z

Figure 1 Hypothetical example of a distance matrix analysis

they may be implicit in which case one may be unaware that these assump-tions are being made It is particularly important to realize that even if invalidassumptions are being made it will still usually be possible to recover a phy-logenetic tree However the tree may be incorrect as a result of making thewrong assumptions Examples of the assumptions that may be made as partof a model include the relative frequencies of certain kinds of mutation (oftentermed the transitiontransversion ratio) independence of mutations (ie thatmutation at one position does not affect the chance of a mutation at anotherposition) or identical distribution (that the same regions of a sequence arepotentially able to mutate in all the organisms being considered)

Evolutionary biologists use several methods for tree recovery includingdistance matrix parsimony maximum likelihood (reviewed by Felsenstein1988 Beanland amp Howe 1992) and split decomposition methods (Huson 1998)

With distance matrix methods one calculates a matrix showing the num-ber of differences between pairs of sequences and determines the tree whichhas the best fit to this matrix An example is given in Figure 1

With parsimony methods we prefer the tree which requires the fewest mu-tations This is in effect a cladistic analysis So for example with the followingDNA sequences (showing only one of the two strands)


W AAGCCAATX TAGCCAATY CGCTTGGTZ GGCTTGGT

Positions 2ndash7 would require fewest mutations if species W and X shared mostrecent common ancestry to the exclusion of Y and Z Then the most parsimo-nious tree would be as shown in Figure 2A

Maximum likelihood attempts to find the tree that has the highest prob-ability of generating the data (ie sequences) observed This approach can bevery computationally intensive With split decomposition as implemented inthe program SplitsTree we consider possible splits between the sequences eg(WX YZ) or (WY XZ) etc and look at the number of nucleotide positionsconsistent with each split Split decomposition has a number of potentially de-sirable features It does not presuppose as many methods do that the tree is abifurcating one It allows recovery of trees where one individual can have manydescendents It also allows conflicting information to be shown ndash where thereare some positions in the data which are not consistent with the preferred treeThese appear as boxes in the output giving a network rather than a tree So theexample given in Figure 2B would arise where there was some signal linking Wwith Y and X with Z as well as the signal linking W with X and Y with Z

W

X

Y

Z

Figure 2A Other tree-recovery methods 2A shows results of parsimony analysis usingthe hypothetical data shown in the text

W

X

Y

Z

Figure 2B Other tree-recovery methods 2B shows the result of a hypothetical splitdecomposition analysis where there is support both for WX YZ groupings and for WYXZ groupings


Stemmatic analysis

The techniques which are used to recover evolutionary trees can be appliedin principle to datasets derived from manuscript traditions using changes be-tween texts in the same way as evolutionary biologists use changes between theDNA sequence of different organisms (Platnick amp Cameron 1977 Cameron1987 Lee 1989 OrsquoHara amp Robinson 1989 Barbrook et al 1998) We are apply-ing this to a range of texts such as John Lydgatersquos 15th century poem Kings ofEngland which exists in over 30 manuscript versions comprising a set of stan-zas describing the Kings of England from William the Conqueror onwardsThus we have variants such as

Worthy to stand among the worthy nyneAble to stand among the worthi nyneAble to stande among the worthyes nyne

for the same line in different texts (These are variants of line 96 of the poemin Oxford Bodleian Rawlinson C316 fol 122v Cambridge Jesus CollegeQG8 (56) fol 47v and Ipswich County Hall Deposit C44 Percyvalersquos GreatDoomsday Book Bk VI fol 239v) We can encode this information to producea dataset resembling a nucleotide sequence (but using any letters or numbersrather than just A C G or T) to represent different readings at a given positionand use this directly as an input to a phylogenetic program such as SplitsTreeThus we recently produced a stemmatic analysis of the extant manuscript ver-sions of Lydgatersquos Kings of England based on evolutionary biological techniqueswhich replicated stemmatic analysis done by traditional methods (Figure 3)

Many of the manuscripts grouped together in the traditional stemmaticanalysis were also grouped together in the phylogenetic analysis Some weregrouped together in the former but not in the latter This might simply reflecta need for more information in the dataset used All groupings in the phyloge-netic analysis were consistent with those in the traditional analysis This studyis described in more detail elsewhere (Mooney et al 2001)

It is remarkable how many parallels there are between the evolution of ge-netic material and the changes occurring in manuscripts (Howe et al 2001)These include recombination convergent evolution and transposition We willlook at these parallels in turn

Genetic recombination is the process whereby two different copies of agene come together to produce a hybrid called a recombinant The first partof the recombinant comes from one precursor gene and the second fromthe other The parallel in manuscript traditions is the change of exemplar de-


Harley 2261

Lambeth 306

Peterborough

Buhler 17

Dublin 516CUL Ad6686

RawlC316Ashmole 59

Nottingham

Harley 7333Egerton 1995

Harley 2251BL Ad 34360

Titus DXXTCC 601 2

TCC 601 1

001 distance unitsLincLat 129

Fairfax 16Bodley 912

Bodley 48Bodley 686

Harley 372Pynson

LeidenLansd 699

Jesus 56Caius 249

Figure 3 SplitsTree diagram of relationships among 27 most complete manuscriptsand early printed copies of Lydgatersquos Kings of England Bold indicates group A italics Bunderlined C suggested by manual analysis (Mooney et al 2001)

scribed as ldquosuccessive contaminationrdquo (Wattel amp van Mulken 1996) when ascribe changed the text from which he was copying part of the way throughWe found examples of this in the analysis of the Prologue to the Wife of BathrsquosTale in that the position of some texts on the stemmatic tree changed when ananalysis performed on the first part of the data was compared with one done onthe second (Barbrook et al 1998) Programmes are being developed to iden-tify recombination breakpoints in genes and these may be useful in stemmaticanalysis (eg Holmes et al 1999) As well as generating hybrids where thereis a clear breakpoint recombination can give rise to more complex productswhere the resulting DNA sequence is a mosaic of the two parental versions ofthe sequence (eg Medgyesy et al 1985) This process is analogous to simulta-neous (where the scribe used several exemplars simultaneously) and incidentalcontamination (where the scribe used a single exemplar first and then modifiedthe text afterwards with other exemplars) (Wattel amp van Mulken 1996)

Transposition of genetic information is when material from one source isinserted somewhere quite different This is a feature of the life cycle of someviruses for example If the source of inserted DNA is a different species it would


be regarded as lsquolateral gene transferrsquo We have found parallels in the Kings ofEngland tradition where there is contamination of one tradition with materialfrom another Thus the verse relating to William I begins typically

This myghti William Duk of NormandyeAs bokes old makith menciounBy just title and by his cheualryeMade kyng by conquest of brutes Albyoun (British Library Harley 2251)

There exists another fifteenth-century poem on the Kings of England not byLydgate which typically has this for William I

At Westmyster William icrowned wasThe furst day of CristemasA gret thyng after he dude thanneMade the kyng of Skottys his legeman (Bodleian Ashmole Rolls 21)

However within this second tradition is a text with a clear example of transpo-sition from the first

This myghtty William duke of NorthmandyThat by juste tytill and also by chyualeryConquered this land and king bycomeAnd the kyng of Scotts he made his legeman (Bodleian Bodley 131)

Convergent evolution is when the same change occurs independently in differ-ent lineages Thus for example an AT base pair might replace a GC base pairat the same position independently in two different species If this occurs fre-quently enough evolutionary tree-building may be misled and species with alarge number of convergent changes will be grouped artefactually closely Con-vergent evolution is comparable to convergence or parallelism in manuscripttraditions (Salemans 1996) So for example scribes working independentlybut in the same geographical area might alter words to fit their own dialectAn example of this might be the substitution of lsquokirkrsquo for lsquochurchrsquo in north-ern England and Scotland Thus the same change may happen in two or moremanuscripts not as a result of common ancestry but as a result of having beenproduced in the same part of the country


Conclusions

The process of incorporation of changes into DNA mirrors the incorporationof changes into manuscripts For this reason programs for phylogenetic anal-ysis of sequence data can be exploited for stemmatic analysis of manuscripttraditions and we believe the SplitsTree program has particular advantagesJust as phenomena such as recombination transposition and convergent evo-lution may pose problems for the evolutionary biologist there are closelyparallel problems in stemmatic analysis We hope that the development of tech-niques in one discipline to deal with these problems will help in their solutionelsewhere

Acknowledgements

We thank Matthew Spencer and Barbara Bordalejo for helpful discussions andthe Leverhulme Trust for financial support

References

Barbrook A C C J Howe N Blake amp P Robinson (1998) ldquoThe phylogeny of theCanterbury Talesrdquo Nature 394 839

Beanland T amp C J Howe (1992) ldquoThe inference of evolutionary trees from molecular datardquoComp Biochem Physiol 102B 643ndash659

Cameron H D (1987) ldquoThe upside-down cladogram problems in manuscript affiliationrdquoIn H M Hoenigswald amp L F Wiener (Eds) Biological metaphor and cladisticclassification an interdisciplinary perspective (pp 227ndash242) London Frances Pinter

Felsenstein J (1988) ldquoPhylogenies from molecular sequences inference and reliabilityrdquoAnn Rev Genet 22 521ndash565

Holmes E C M Worobey amp A Rambaut (1999) ldquoPhylogenetic evidence for recom-bination in dengue virusrdquo Mol Biol Evol 16 405ndash409

Howe C J A C Barbrook M Spencer P Robinson B Bordalejo amp L R Mooney (2001)ldquoManuscript evolutionrdquo Trends Genet 17 147ndash152

Huson D H (1998) ldquoSplitstree a program for analyzing and visualizing evolutionary datardquoBioinformatics 14 68ndash73

Lee A R (1989) ldquoNumerical taxonomy revisited John Griffith cladistic analysis and StAugstinersquos Quaestiones in Heptateuchemrdquo Studia Patristica 20 24ndash32

Medgyesy P E Fejes amp P Maliga (1985) ldquoInterspecific chloroplast recombination in aNicotiana somatic hybridrdquo Proc Natl Acad Sci USA 82 6960ndash6964


Mooney L R A C Barbrook C J Howe amp M Spencer (2001) ldquoStemmatic analysis ofLydgatersquos Kings of England a test case for the application of software developed forevolutionary biology to manuscript stemmaticsrdquo Revue drsquoHistoire des Textes 31 275ndash297

OrsquoHara R amp P Robinson (1993) ldquoComputer-assisted methods of stemmatic analysisrdquo In NBlake amp P Robinson (Eds) The Canterbury Tales project occasional papers Vol 1 (pp53ndash74) London Office for Humanities Communication Publications

Platnick N I amp H D Cameron (1977) ldquoCladistic methods in textual linguistic andphylogenetic analysisrdquo Syst Zool 26 380ndash385

Salemans B J P (1996) ldquoCladistics or the Resurrection of the Method of Lachmannrdquo In Pvan Reenen amp M van Mulken (Eds) Studies in Stemmatology (pp 3ndash55) AmsterdamJohn Benjamins

Voet D J G Voet amp C W Pratt (1999) Fundamentals of Biochemistry New York JohnWiley amp Sons

Wattel E amp M J P van Mulken (1996) ldquoShock waves in text traditionsrdquo In P van Reenenamp M van Mulken (Eds) Studies in Stemmatology (pp 105ndash121) Amsterdam JohnBenjamins

Watson J D amp F H C Crick (1953) ldquoMolecular structure of nucleic acids a structure fordeoxyribose nucleic acidrdquo Nature 171 737ndash738

Problems of a highly contaminated traditionthe New TestamentStemmata of variants as a source of a genealogyfor witnesses

Gerd MinkWestfaumllische Wilhelms-Universitaumlt Muumlnster

Introduction

It is a well-known fact that the textual tradition of the New Testament posesa formidable challenge in the way of textual criticism and edition techniqueThe number of manuscripts that have come down is large some 5600 knowncopies so far although most of the older ones in particular have been lost Thereal cause of the problems however is the vast degree of contamination Forsuch a textual tradition existing methods of reconstructing the tradition arenot sufficient and other approaches have to be developed This study attemptsto show a new way (i) of finding and evaluating the genealogical data that canbe used to construct a stemma of such texts and (ii) of constructing a stemmathat reflects all genealogical data1 The concept of coherence (cf paragraph411) will be essential for the analysis of genealogical relationships The methodis based on two design choices

(1) Instead of trying to start with the construction of overall structures of therelations between witnesses the first step is to construct local stemmata ofvariants Local stemmata consist of trees based upon just one place of varia-tion and not more than one If possible a local tree is constructed for eachplace of variation

These local stemmata are oriented as far as possible before being used as abasis of (sub)stemmata of witnesses or textual states (not of manuscripts) Thefollowing example illustrates how a local stemma can be constructed

Gerd Mink

James 41210-12 (εις εστιν ο νομοθετης) και κριτης(ldquoone is the lawgiverrdquo) ldquoand judgerdquo

variants a και κριτης ldquoand judgerdquob και ο κριτης ldquoand the judgerdquoc κριτης ldquojudgerdquod omission

local stemma of variantsa

b d

c

Figure 1 A local stemma of variants

Following the custom of the Editio Critica Maior (ECM cf the next para-graph) the places of variation will be referred to by chapter and verse as wellas word address2 Lower-case letters indicate the variants In the example fromJames 41210-12 the best hypothesis is that variant a represents the originaltext since the word κριτής (lsquojudgersquo) is very important for the authorrsquos argu-ment in the context The omission in variant d however is easily explained ascaused by homoioteleuton3 Variant b occurs in rather unimportant witnessesand only adds the article to variant a4 Variant c occurs in a single witness only5

this witness is quite distant from even its closest relatives6 which are part of thed attestation7 Even more distant are the relatives that belong to the a attesta-tion Apparently variant c is flawed as the word καί (lsquoandrsquo) which is in factindispensable in the context is missing The stemma rests on the assumptionthat variant c is based upon variant d and corrected the omission in accordancewith variant a In that process καί would have been overlooked a typical errorduring revisions However it is also possible (hence the interrupted edge fromd) that the only basis was variant a and καί was omitted on account of thesimilarity of its initial letter with that in κριτής (homoioarkton) yet the closestrelatives of the variant c witness (631) are all in the d attestation8

(2) The contamination in the tradition is viewed as a process The assumption isthat if contamination occurs it emerges from those texts which were at thedisposal of the scribe ie texts in his direct environment ie texts which arefor the most part closely related with each other

Problems of a highly contaminated tradition

James 5745 ldquoThe farmer waits for the precious fruit being patient about itrdquo(εως λαβη προιμον και οψιμον)(ldquountil he receives an early and late onerdquo ldquountil heit receives something early and something laterdquo)

variants a (εως λαβη) omission (προιμον και οψιμον)(ldquountil he receives an early and late onerdquo ldquountil heit receives something early and something laterdquo)

b (εως λαβη) υετον (προιμον και οψιμον)ldquo(until heit receives early and late) rainrdquo

c (εως) υετον (λαβη προιμον και οψιμον)ldquo(until heit receives early and late) rainrdquo

d (εως λαβη) καρπον (προιμον και οψιμον)ldquo(until he receives early and late) fruitrdquo

e (εως λαβη) καρπον τον (προιμον και οψιμον)ldquo(until he receives the early and late) fruitrdquo

f (εως λαβη) και τον (προιμον και οψιμον)ldquo(until he) also (receives) the (early and late one)

local stemma of variants a

b d1 e

c d2 f

Figure 2 Local stemma of variants a more complex case

First of all another example will demonstrate the importance of a close re-lationship between witnesses for the construction of a local stemma Afterthat the two examples will be used to highlight the different textual develop-ments so typical of a contaminated tradition in a given witness and its closestrelatives

In Figure 2 the interpretation of variant a is not entirely clear Is themeaning lsquothe early and late fruitrsquo or something different The vast majority ofwitnesses has variant b resulting in a text of clear meaning Because so manywitnesses have this reading relationships with witnesses of all variants can befound Yet it can be presumed a priori that the oldest layer in the attestationof variant b is to be found in some of its best-known witnesses 02 33 811852 These are more closely related to the witnesses of variant a than to oth-ers The witnesses of variant c are particularly closely interrelated (with 96 to

Gerd Mink

99 agreements) As could be expected further relatives are to be found inthe b attestation Variant c is not represented in an early layer of tradition Forvariant d there is just one single closely related pair of witnesses (996 1661)The remaining witnesses of d (398 1175 a corrector of 01) are not closely re-lated with each other nor is there any special connection with the only witnessof variant e (01 first hand) Variant d seems to have emerged on several oc-casions by introducing the lsquofruitrsquo from the preceding context into the variant1175 has its closest relatives in the a attestation (03 1739) and the b attestation(1243 025 01 of the e attestation is yet more remote) the other d witnesseshave their closest relatives in the b attestation One would therefore trace backtheir variant to variant b (cf d2 in the local stemma) The variant of 1175 (d1)could be considered to derive from either a or b For the time being the ques-tion remains open The witness of e (01) has its closest relative (03) in thea attestation (the relatives in the b attestation are quite remote) And even ifvariant f is not grammatically impossible it is probably an error which couldtheoretically have arisen from variants b d or e The closest relatives of thesingle witness of f (69) are in the b attestation from which accordingly it isbeing derived

It is a prerequisite for constructing a local stemma like the one in Fig-ure 2 that the relationships between the witnesses as based on their degreesof agreement are known In addition onersquos provisional assessments of the ageand quality of a number of well-known witnesses is taken into account Closerelationships between witnesses alone do not say anything about the directionof textual development Yet if in a relationship the ancestor and the descen-dant can be successfully determined further possibilities arise to ascertain alocal stemma Thus in the example the first assumptions were confirmedadditionally the ancestors of 1175 (variant d1) were successfully located inthe a attestation9 and d1 was accordingly derived from it (see Figure 2) Ar-guments for the genealogical connections between variants are based partlyon their content and linguistic form and partly on the relationship betweentheir witnesses The example of Figure 2 makes it clear how these relation-ships can explain why some witnesses share the same variant and why onevariant could arise from another Yet a lack of relationship between witnessescan also reveal coincidental correspondences (cf variant d) Knowledge of thecorrespondences between witnesses provides a first overview which must thenbe examined and supplemented against information about their genealogicalrelationships

A closer look at the two instances under consideration reveals in the dif-ferent attestations of the variants the traces of contamination among related


witnesses Witness P74 has 8 close relatives which contain both instances 0281 03 2344 1735 218 01 1718 (in order of decreasing degree of agreement)10

The distribution of the variants in the two variant places is as follows

James 41210-12 a 01 02 03 81 1735 2344d P74 218 1718

James 5745 a P74 03b 02 81 218 1718 1735 2344e 01

Incidentally it has emerged that all the close relatives mentioned have moreolder variants than P7411 Therefore P74 is probably not the lsquoinventorrsquo of oneof the variants at the two places of variation on the contrary its text is based onvariants which are found in different close relatives with more older variantsand consequently we see the traces of contamination

The Editio Critica Maior of the New Testament

The Institute for New Testament Textual Research in Muumlnster founded in 1959by Kurt Aland has by now accomplished ndash apart from a few exceptions ndash thetask of collecting the basic material of the textual tradition of the New Tes-tament Nearly all known manuscripts of the Greek New Testament are nowavailable on microfilm The next aim was to sift the material in an intelligentway to let new views about important manuscripts find their way into theminor editions of the institute and finally to present an Editio Critica Maiorwhich does justice to present-day standards of knowledge and methodology Afirst stage of work on the ECM has been the research published since 1987 onthe texts of the Greek manuscripts and their relevance (lsquoText und Textwertrsquo)12

This was intended to separate the majority of manuscripts containing the rel-atively uniform text which was standard at the end of the Byzantine traditionfrom the still large number of manuscripts which must be considered relevanton account of their deviations from the majority text The basis was a collationof all available manuscripts according to a system of places of variation Theseso-called test passages (lsquoTeststellenrsquo) were known for the fact that they weredifferent in the newer and older text forms Research into lsquoText und Textwertrsquomade it possible to draw on particularly those manuscripts for the ECM whichdo not contain the uniform text from the end of the textual tradition Never-theless a number of witnesses of this late text are also represented in the ECM

Gerd Mink

According to text tradition the New Testament can be divided into fivesections the Gospels the Acts of the Apostles and the Catholic Letters (whichnearly always follow Acts in the manuscripts making up together with Actsthe corpus of the so-called Apostolos) the Letters of Paul and the Revelation ofJohn The Catholic Letters were the first field of research in the ECM project13

The Letter of James Manuscripts and variants

The Catholic Letters starting with the Letter of James appeared to be especiallysuitable to begin work on the ECM with The number of manuscripts exam-ined in the lsquoText und Textwertrsquo project (552) is of the same order as that forActs which have normally been copied together with the Catholic Letters It issmaller than that for the Letters of Paul (about 750) and a great deal smallerthan that for the Gospels (eg 1787 for the Gospel of Luke) Nevertheless thenumber of witnesses relevant for the textual tradition is in no way smaller Thedegree of contamination made the Catholic Letters very attractive for method-ological investigation As the material for the Letter of James was the first to becompletely available it has been researched most extensively Unless otherwiseindicated the following refers to the Letter of James

As the ECM is mainly interested in the text of the first millennium weexcluded nearly all the uniform witnesses representing the final state of theByzantine tradition from our study of James ie 371 manuscripts of a total of535 In addition 19 lectionaries were selected for the ECM Yet I did not usethem for my genealogical study because they cannot really be compared withthe other manuscripts they do not contain the full continuous text but lessonsselected to be read during services The remaining 164 Greek manuscripts con-tain the continuous text with minor omissions occurring repeatedly Somemanuscripts have also suffered more substantial damage resulting in lacunaeoccurring throughout the text Yet a number of manuscripts is quite badlyfragmented 10 contain less than 150 out of the 761 places of variation in someof them no more than a couple of these places has survived The importantmanuscript 04 at least contains more than half of the text

The text of James contains about 1740 words the exact number dependson textual decisions The selected 164 Greek manuscripts including the frag-mentary ones present 2132 genuine variants at 761 places of variation14 Onlyvariants were counted that appear in the first hands of manuscripts contain-ing the continuous text15 If the variants from Church Fathers and lectionariesand from early versions (Latin Coptic Syriac etc) of the New Testament are


1000

900

800

700

600

500

400

300

200

100

01 2 3 4 5 6 7 8 9 10 11 12 13 14 15

number of witnesses exceeded in an attestation (x10)

nu

mbe

r of

occu

rren

ces

Figure 3 Distribution of size of the attestations

included the total number for these 761 places is 234916 Since many of these761 places comprise more than one word and since the text consists of about1740 words it follows that about half the text is subject to variation

The variants include all types except for type 0 variations17 Type 3 varia-tions hardly play a significant role Typical variations are

A BCDEFGHIJKLMNOPQRSTUVWXYZ (type 1)ABCDEFGH IJKLMNOPQRSTUVWXYZ (type 2)ABC DEFGH IJKLMNO PQRS TUV WXYZ (type 4)

Type 4 variations are the most frequent In Chapter 2 of James spot checksrevealed about 40 type 4 variations 30 type 1 variations and 25 type 2variations The remainder was made up of type 3 variations The distributionwas atypical in some passages A single witness is often responsible for the factthat type 1 variations occur more frequently in a major section of the text Onthe other hand longer units of variation in which several changes are mutuallyinterdependent in all probability lead to type 4 variations

The values in Figures 3 and 4 are based on the variants of the 164 wit-nesses at the 761 places of variation with each witness being represented bythe first hand in a manuscript and by the text (lemma) in a commentary18

In Figure 3 the horizontal axis gives the number of witnesses (to be multipliedby 10) which is exceeded in an attestation The vertical axis tells how oftensuch an attestation occurs (eg about 700 attestations cover more than 110 wit-nesses)19 Figure 3 gives an indication of how the quantity of the attestationsis distributed over their total number The total number of variants and there-

Gerd Mink

nu

mbe

r of

occu

rren

ces

nu

mbe

r of

occu

rren

ces

450

400

350

300

250

200

150

100

50

0

800

700

600

500

400

300

200

100

0

1 2 3 4 5 6 7 8 9

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19 20 21

10 11 12 13 14 15 16 17 18 19 20

number of variants

number of variants exceeded

Figure 4 Distribution of number of variants per place of variation

fore of attestations is 2132 (see above) It is apparent that small attestations (ofup to 10 witnesses) are the most frequent for there are only 965 instances ofmore than 10 witnesses It is not surprising that this result corresponds to ahigh frequency of very large complementary attestations (more than 120 wit-nesses 676 instances) In comparison with that medium and large attestationsmake up only a relatively small part (eg there are only 128 instances of 41ndash120witnesses)

Figure 4 (top) shows how many variants (horizontal axis) occur in howmany places of variation (vertical axis) In Figure 4 (bottom) the horizontalaxis shows the number of variants which is exceeded in one place of variationand the vertical axis gives the number of places of variation with this number ofvariants Thus while at the top it is shown that there are about 125 places with3 variants20 the lower diagram demonstrates that in about 150 places there aremore than 3 variants21 More than half of the places of variation have only 2variants More than 6 variants occur in 25 instances more than 10 in only 4 Acomparatively large part is made up of places of variation with 3 to 6 variants


Letter of James

about 1740 words

535 Greek manuscripts available

164 manuscripts selected761 places of variation2132 variants of the 164 manuscripts

143 cases identical attestation800 occurrences

104 cases identical one witness attestation685 occurrences

39 cases of identical more-than-one-witness-attestation115 ocurrences

Figure 5 The Letter of James some numbers

The existence of 59 places of variation with only one variant is caused by de-viating variants stemming from lectionaries Church Fathers or early versionswhich are assumed to be based on Greek exemplars at these places but cannotbe traced back to the first hands in the 164 manuscripts

The following numbers are based on the same data as Figure 3 and Figure4 the variants of the 164 witnesses in 761 places of variation If correctionsmarginal variants variants in commentaries and the evidence from lectionar-ies Church Fathers and early versions were included in the analysis not onlythe total number of variants would be much larger but the number of occur-rences of identical attestations would also be smaller

The 2132 variants represent 2132 attestations of which 1332 are uniqueie there is no other attestation containing only the same witness or the samecombination of witnesses These unique attestations include those in the 59places mentioned above22 The chances of finding identical attestations amongthese are slight as in each of the 59 places a number of witnesses is lackingfrom the 164 (due to fragmentation unmotivated omission of larger passagesor the like)

There are 143 different cases in which identical attestations occur morethan once The sum of all the occurrences of these 143 cases is 800 Out of these143 cases of identical attestations 104 concern attestations comprising only onewitness The sum of all their occurrences is 685 The 39 remaining cases relateto multiple-witness-attestations The sum of their occurrences is 115 Of these39 cases 12 concern large complementary attestations in two neighbouring

Gerd Mink

places either to only one Greek witness differing from the entire remainder ofthe textual tradition or to witnesses from the field of lectionaries Church Fa-thers or early versions These 12 cases correspond to 24 occurrences When the12 cases are subtracted from the 39 cases of multiple-witness-combinationstwo-witness-combinations are typical among the remaining ones (21 out ofthe 27 cases) they occur up to 6 times typically twice or 3 times Even well-known pairs of manuscripts participate in two-witness-attestations only 5 or6 times There are only 6 cases (apart from the large complementary attesta-tions mentioned above) in which more than two witnesses are combined Therichest combinations are 2 four-witness-cases Each of them occurs 3 times

The conclusion of this is twofold First if we want to explore the ge-nealogical patterns of the textual history of this tradition we cannot base ourresearch on identical combinations of witnesses since they play only a verymodest role Typically attestations do not have duplicates containing the samewitnesses The number of type 2 variants the corner stone in the approachof Salemans (2000) is too small Secondly the James text is highly contami-nated It is therefore unavoidable that we focus our research on the analysis ofcontamination

Contamination as a process

In a dense tradition it is typical of contamination that a witness shares mostof its variants with its closest relative and if it deviates from this relative thevariants concerned can be found in other close relatives In the text of Jamescontamination is the result of small steps That these steps are small is visi-ble only if the number of witnesses of the tradition that have been preservedis large However where the proportion of witnesses that are not preservedis high contamination does not appear to be the result of small steps as somany intermediate witnesses are missing If the density of a tradition is veryhigh (as it was in the middle ages from the 11th century onwards) nearly allthe witnesses have very close relatives The agreement values are typically highbetween 94 and 98 even if the most uniform Byzantine witnesses are ex-cluded This implies that typically only 15ndash45 places of variation saw a changeduring the step from one preserved witness to the most closely related one thathas also survived Contamination in this context occurs in very small steps andthe steps would be even smaller if all the manuscripts had been preserved Thepattern is as follows the two or more manuscripts involved in a copying pro-cess are among those that are most closely related23 and even if they are lost


20

18

16

14

12

10

8

6

4

2

0

nu

mbe

r of

man

usc

ript

s

1 2 3 4 5 6 7 8 9 10

century

P100

P1000103

01802002504404903302

04048

P74

Figure 6 Distribution of manuscripts of the Letter of James in the first millennium

a dense textual tradition may enable us to find very close relatives among thewitnesses that have been preserved

It is rather unlikely that the comparatively few manuscripts to survive fromthe 9th century and before are representative of the totality of the manuscriptsof that time a considerable number must have been lost Figure 6 shows themanuscripts containing James and written in the first millennium that are nottoo fragmented24 Although we do not know how many links are missing be-tween them we can be sure that a part of the variants found only in the latermanuscripts is a reflection of older variants which have not been preserved inthe surviving manuscripts of the first millennium Indeed for this period wemust assume a vast number of missing links Therefore the nearest preservedrelatives of a witness are not as closely related to it as is the case in later timesAs a consequence agreement values are lower than for the later manuscriptswith values of more than 90 being already quite good Contamination in thiscontext looks more radical because of the many missing links All those veryclosely related manuscripts which a given scribe might have consulted havedisappeared At first sight contamination here appears to have had a muchgreater impact than in the later manuscripts but it may still be assumed thatamong the older textual tradition too it has followed the normal pattern themanuscripts of a given copying process must have been among the most closelyrelated ones But in this case they have all disappeared We will however finda lot of traces in the later manuscripts and some of them seem to preserve a

Gerd Mink

very old text only slightly altered in the course of the time For this reason it isimportant to make a distinction between manuscripts and their texts It is im-possible to find the genealogical relationship between manuscripts if most ofthem have not survived It is merely possible to uncover the genealogical struc-ture of the preserved texts In this context therefore the text is the witness andnot the manuscript25

A brief outline of the method Some definitions

It may be helpful to present a brief outline of the method I have developed26

as far as it is relevant for the present subject for it is quite distinct from thegeneral practice of editors Nearly all New Testament textual critics accept thatthe stemma of a tradition contaminated to such a degree as the Letter of Jamescannot be reconstructed However the question of what is the original text fora given passage is widely discussed independent of what the overall stemmamight look like and every editor has to decide what text he will provide Will-ingly or not he must make genealogical decisions concerning the individualplaces of variation as soon as he settles for one of the variants as being theoriginal text at these places

The objective of my method is a comprehensive theory of the structureof the textual tradition with special regard to the problems of contaminationand the coincidental emergence of variants The result will not necessarily bea stemma in the traditional sense of a graphic representation I conceive of thestemma here in a more complex sense It does not only connect witnesses thedata representing the contents of those connections and the data designatingthe quality of those connections predominantly determine the stemmatic hy-pothesis Usually a stemmatic representation only displays information aboutthe direction of development between two witnesses Yet the underlying dataare the real substance of a stemma They are in a more complex sense thenucleus of the stemma27 Graphical representations can show no more thanaspects of this complexity

My method has three elements observations assessments of what wasobserved and structural connections based on these observations and assess-ments according to rules based on a model of manuscript tradition Theserules have to be in harmony with a model of the average situation in the his-tory of transmission of a given text Such a model is necessary in order to havecriteria to assess which results are more probable than others The model de-


pends on what we know about the particular ways of transmission the habitsof the scribes etc in the respective tradition which we are about to explore

In the case of the New Testament the model is as follows ldquoSome essen-tial assumptions are considered more probable than their contrary (withoutexcluding that the contrary might happen at some points) (i) a scribe wantsto copy a manuscript with fidelity primarily the scribe does not want to cre-ate new readings (ii) if the scribe introduces other readings28 they come fromanother source (normally a manuscript) (iii) if the scribe uses more than onesource few rather than many sources will be used and (iv) the source copieshave closely related texts rather than less related onesrdquo29

A comprehensive theory results from the total of observations and assess-ments and allows us to examine the plausibility of each assessment and of eachtextual decision in the light of an overall view This is very important for theeditors of the text for there is a circular argument typical of textual criticismWitnesses are important for reconstructing the initial text and they are impor-tant because of the high number of agreements with the reconstructed initialtext In other words witnesses are good because of their good variants variantsare good because of their good witnesses This circle cannot be avoided but ithas to be controlled We need a method therefore which can provide an over-all view of the consequences of all the decisions we take so that also the overallplausibility of what we are doing can be examined30 In the present methodthis is done through an iterative process especially designed to perform thisexamination

At first I would like to explain some central concepts of the method whichitself will be outlined subsequently

Initial text (A)

The initial text is a hypothetical reconstructed text as it presumably existedaccording to the hypothesis before the beginning of its copying In a hypoth-esis which wants to establish the genealogical relationship between the wit-nesses the initial text corresponds to a hypothetical witness A (lsquoAusgangstextrsquo)The initial text is not identical with the original the text of the author Betweenthe autograph and the initial text considerable changes may have taken placewhich may not have left a single trace in the surviving textual tradition Even ifthis is not the case differences between the original and the initial text must betaken into account

On the other hand the initial text is not simply a reconstruction on the ba-sis of the surviving variants which best explains the emergence of the variants

Gerd Mink

and thus represents the archetype of the tradition Instead several hypothe-ses are possible about the beginnings of the tradition The simplest workinghypothesis must be that there are no differences between the original and theinitial text (except for inevitable scribal slips) In that case the reconstructionof the initial text is not only determined by the subsequent tradition (whichtext form could have been derived from which) but also by the authorrsquos in-tentions as they come to light in the totality of what we know about him (is avariant more likely to come from the pen of the author or from a copyist)31

Another possible hypothesis might involve an editor in-between the author andthe initial text who might possibly have merged several writings of an authorinto one Or there may have been more than one initial text possibly even go-ing back to more than one autograph if for example the author issued severalversions of his work

I should also like to point out in this context that the notion of text needs amore exact definition In textual criticism it is mostly used for what lsquowas writ-ten downrsquo in the manuscript strictly speaking for the sequence of charactersthe copyist wrote down On the other hand the text should convey a mean-ing This aim however is not directly achieved by the written characters Theyrepresent different things in different languages

A copyist may grasp the sense of a text and copy this sense Even if heintends to copy the text true to the letter from the exemplar he may at the levelof the characters produce variants which emerge from the meaning of the textBut a copyist may also particularly in long or difficult sentences copy groupsof words or line by line while losing the overall meaning of the text In the caseof very difficult words he might even copy character by character which maylead to readings which hardly make any sense and even to spectacular errors

Now the problem is that there is no obvious difference between variantsintroduced on purpose by copyist and variants produced accidentally as longas these variants are meaningful in the widest sense The text as carrier ofmeaning is therefore modified by the variants ndash introduced intentionally orunintentionally by the copyist ndash and there is no way of differentiating betweenintentional and unintentional variants

If however a copy contains errors ie readings which clearly do not makesense these do not necessarily modify the text as a carrier of meaning Thecopyist had no intention of changing the text The copy therefore only con-tains an erroneous representation of the same text at the level of the characterswhich nevertheless sometimes renders it unrecognisable as a carrier of mean-ing But not only errors at the level of the characters are possible After allany linguistic element can be affected by mistakes of this sort such as errors of


concord and case also omissions of key words for the context which happenaccidentally and against a scribersquos intention (lapsus)

Even the first exemplar of the entire textual tradition even the very au-tograph may have contained such errors The text as the actual carrier ofmeaning however would not necessarily be affected by these errors it wouldonly be represented defectively

Consequently the ECM records the witnesses of erroneous readings as wit-nesses for the variants which they represent albeit defectively There is even anexample where the best witnesses omit a negation (2 Peter 31048-50) Al-though the preceding passage speaks of the passing away of the heavens andthe dissolution of the elements and the following verses presuppose the dis-solution of heaven and earth (for a new heaven and a new earth are waitedfor) quite superior witnesses here have the reading lsquothe earth and all the worksthat are therein will be found (εὑρθήσονται)rsquo32 when logic demands lsquowill notbe found (οὐχ εὑρεθήσονται)rsquo The meaning as a result is extremely prob-lematic to my mind the reading does not make sense and must therefore beerroneous33 Unquestionably the hyparchetype of all these witnesses did nothave the negation Now there are two variants (ἀφανισθήσονται lsquothey willdisappearrsquo34 and κατακαήσεται lsquothey will be burned uprsquo)35 which presupposeand express more graphically a text containing the negation οὐχ εὑρεθήσονταιlsquothey will not be foundrsquo Although it is not preserved in any Greek manuscript itis probable that the initial text had the negation Even if these variants which in-directly confirm the negation did not exist the assumption should still be thatthe initial text contained the negation required by the sense of the text eventhough the negation is not in the graphemic representation of the archetype36

To my mind this is an almost unavoidable conjecture

Places of variation

Places of variation are places in the text where variants appear At least twodifferent variants occur in a place of variation the maximum in James is 24variants A place of variation may comprise more than one word but it canalso be the space between words Ideally it covers a logical unit of variationThis means that mutually interdependent changes to a text should belong toone unit of variation (eg if a subject and correspondingly the predicate areput in the singular) A unit of variation can also be postulated when a group ofwords presumably belonged together in a copyistrsquos view (eg if a word groupconsisting of articleparticlenoun shows changes in different combinations forthe articlenoun and for the particle) Sometimes very pragmatic considera-

Gerd Mink

tions might be adduced to determine a unit of variation so as to enable thecomparison of all texts at a certain place Places of variation may also overlapIn one place of variation the question may be eg whether a rather large groupof words has been omitted or not yet another instance of variation may resultfrom variants within that group of words whenever it was not left out

The number of variants offered by the first hand of a witness correspondsto the number of places of variation which have been preserved in the witness

Variants vs readings connective variants

ldquoA reading is the generic term for the wording of a textual unit in which amanuscript is distinguished from one or more or from all other manuscriptsA variant refers to one of at least two readings of the same textual unit which isgrammatically correct and logically possible Errors are readings which do notfulfil these criteria rdquo37 Errors are usually deemed as the variant they representincorrectly In some cases if the corresponding variant is no longer availablethe error has to be corrected and incorporated as another variant It is the-oretically possible that new variants originate specifically from an error Inpractice however at least in the New Testament tradition this seems to be rareas the errors are usually corrected into the underlying variants in the furthertradition

ldquoAlternative and orthographically possible forms of the same variants areclassed as orthographicardquo38 Thus in the verb λαμβάνω (lsquoI takersquo) the μ whichis really a formative element of the present tense stem in Koine Greek oftengets infixed in other stems The future form apart from the actually correctλήψομαι is frequently λήμψομαι The two forms are treated as equal and inter-changeable Or the particle ἄν indicating a prospective meaning after relativepronouns and relative adverbs is replaced by ἐάν in later Greek with the samefunction and position ndash a phenomenon caused by the fact that in classicalGreek instead of ἐάν (lsquoif rsquo) people also wrote ἄν Any scribe could adapt thesethings to his usage without affecting the quality of the copy

Certain morphological divergences are considered equivalent to ortho-graphica They have it in common with actual orthographica that in the copyistrsquosview they were also interchangeable without affecting the quality of a goodcopy eg if instead of a strong aorist (eg ἐγενόμην lsquoI becamersquo) he would use amixed aorist (ἐγενάμην)

Often the notion of lsquoreadingrsquo is used in a more general way it is frequentlytaken to mean lsquovariantrsquo It is very common to speak of an original readingfrom which variants were derived in the course of a textrsquos transmission Yet


neutrally speaking even the reading which is held to be part of the originaltext is nothing more than one variant in the textual tradition

For every textual tradition it is necessary to determine ndash in accordance withlanguage historical period and literary genre ndash which readings are to rank asgenuine variants Only variants are the basis of a genealogical relationship ofwitnesses If a variant contributes to genealogical coherencies (see paragraph411) it is called a connective variant of the witnesses concerned Their agree-ment in this variant is assumed to be not coincidental because (i) they usuallyhave variants in common or (ii) the character of the variant argues againstmultiple independent emergence (see paragraph 6)

Witness vs manuscript

A manuscript is the physical carrier of the text A manuscript has proper-ties which can be defined paleographically and codicologically The text in amanuscript may be considerably older than the manuscript itself At the verylatest the text was produced at the same time as the manuscript The witness ofa variant is the text not the manuscript39 A one-witness-attestation is the at-testation of a variant found in the text of one manuscript only A two-witness-attestation is the attestation of a variant found in the text of two manuscriptsWhenever numbers are cited in this study which are generally used to denotemanuscripts they are used here to designate the texts transmitted in them notthe manuscripts as their physical carriers

The hypothetical witness A represents the hypothetical initial text (seeparagraph 41)

Attestation

Attestation is the total of all the witnesses presenting a certain variant at anyone given place of variation Consequently the number of variants equals thenumber of attestations

Local stemma vs global stemma

A local stemma is a stemma representing the presumed genealogical relation-ship between variants at one place of variation It is a key notion in this study Aglobal stemma is a stemma representing the genealogical relationships betweenwitnesses It illustrates the overall genealogical hypothesis A global stemmacan only be true if the relationships it shows between the witnesses are compat-

Gerd Mink

ible with the relationships the witnesses have in every single place of variationaccording to the relationships between their variants as represented in the lo-cal stemmata It must convey the genealogical coherencies (see paragraph 411)of the attestations in any one place of variation correctly It must reflect all thechanges between witnesses and all the non-coincidental correspondences inevery place of variation Meeting these conditions a global stemma is true

A global stemma is the superset of all the optimal substemmata ie itconsists of all their nodes (including any intermediary nodes) and edges It ispossible to have more than one global stemma The number of global stemmatadepends on the number of cases in which there are two or more substem-mata of equal likelihood and on the number of alternative substemmata ineach case For any one set of substemmata containing not more than onesubstemma per witness however only one global stemma is possible40

Intermediary nodes

Normally the nodes in a stemmatic graph stand for witnesses whose genealog-ically relevant contents are made up of all their variants Witness x and witnessy are immediately connected by an arrow (= a directed edge)41 An intermedi-ary node does not represent a witness but signifies only a subset of the variantsof witness x andor witness y which are to be connected with each other Itforms part of the connection between witnesses x and y On the purpose ofintermediary nodes see paragraphs 7 and 8

Optimal substemma

A substemma is part of a global stemma It links one descendant with its hy-pothetical ancestor or ancestors (sources of contamination) A substemma isoptimal if the smallest possible number of ancestors can explain all variants ofthe descendant The ancestors have to exhibit a very high degree of genealog-ical coherence (see paragraph 411) with the descendant in order to excludecoincidental correspondences

Consequently they are to be looked for among the potential ancestors (seeparagraph 410) If at a place of variation a descendant corresponds only toa witness of lesser coherence the nature of the variant must be used to verifywhether the variant in question actually links the two witnesses or whether thecorrespondence of variants is only a matter of coincidence

There are however two cases in which a connection with a non-ancestoris effected by way of an intermediary node (i) A connection between a de-


scendant and a non-ancestor may be compulsory if the latter offers the onlypossible prior variant at some place of variation although it is not a poten-tial ancestor (for the full issue see paragraph 7) If this connection were leftout not all the variants of the descendants would be explained correctly iein accordance with the local stemmata The substemma would not be optimal(ii) The agreements between witnesses with a high degree of undirected ge-nealogical coherence (see paragraph 411) are not covered by stemmata basedon potential ancestry The connection between such witnesses can be effectedby way of intermediary nodes (or undirected edges) Such connections mustalso be integrated into the optimal substemma (for the conditions of this seeparagraph 8)

An optimal substemma is true if it is compatible with the genealogical co-herencies (see paragraph 411) of the attestations in any one place of variationand therefore reflects all the changes between the witnesses concerned and alltheir non-coincidental correspondences

Prior vs posterior

The local stemmata represent a hypothesis about which variant arose fromwhich There each source variant is prior the one developed from it poste-rior In the comparison of pairs of witnesses the proportion of prior variantsand posterior variants plays a decisive role One of the key-questions to beasked is in how many places does witness x have a prior variant from whichthe posterior variant of witness y was derived and in how many places theopposite is true The result reveals witnesses that are predominantly prior orpredominantly posterior

Potential ancestor

Potential ancestors are all those witnesses which show a higher proportionof prior variants than posterior variants in comparison with a given witnessConsequently this also includes witnesses not represented in a stemma or sub-stemma as ancestors because they are not needed to explain the variants of adescendant

Gerd Mink

Coherence and coherencies the different types Pre-genealogical genealogical (directed and undirected) stemmatic

All surviving witnesses are related to each other and there is coherence withinthe entire tradition42 For all the witnesses closest relatives can be found Be-tween a witness and its closest relatives there is the highest degree of coher-ence These closest relatives in turn have their own closest relatives so thatchains of coherencies develop Thus the particular coherency within each pairof witnesses within groups and also within the attestations of variants can beevaluated This way coherence ndash represented by coherencies ndash can be analysedat each place of variation as well as in the entire tradition

Coherencies between witnesses may be qualified as good or poor When-ever a coherency is mentioned without any qualification it refers to a usefulcoherency ie one that is high enough to indicate a closer (genealogical) re-lation The absence of coherence consequently means that there is no usefulcoherency because it is too poor The assessment of the quality of the coherencybetween witness x and witness y depends on the coherencies pertaining to theimmediate genealogical environment of witness x and y The relevant factor inthe assessment of the quality of the coherency are the percentage values of theagreements between each of these witnesses and its closest relatives and not theabsolute values as the number of comparable places of variation changes frompair to pair of the witnesses to be compared A witness can very effectively becharacterised by the percentage of the highest agreement (with its closest rel-ative) and the manner in which the values of agreement (with other relatives)decrease from there A witness may for example have so many individual vari-ants that its text differs very often from the main body of the tradition Insuch a case the rate of agreement with the closest relative may perhaps be only89ndash8743 which under these circumstances would still be useful values Asimilar situation may occur if the text of a witness was compiled from othertexts which are themselves not very closely interrelated The case is differentif none of the close relatives of a witness have survived (especially so if a wit-ness is based on the oldest layer of the textual tradition) In such cases one mayhave to fall back on the range of 92ndash89 agreement to discover traces of thelost relatives If a large number of relatives has agreement rates in the rangeof 96ndash93 however a percentage of 89 in another witness is unlikely toyield more usable coherencies as it is not likely that connective variants will befound there that are not to be found in the large number of closer relatives44

This holds true unless variants which cannot have developed independently ofeach other compel one to take lower coherencies into account


If the assessment of the relationship between witnesses is done only on thebasis of their agreements it is classified as pre-genealogical coherency This typeof coherence is a criterion for determining the probability whether the wit-nesses allow to take into account a genealogical relationship between variantsNormally a lack of pre-genealogical coherence within an attestation implies acoincidental multiple development of conform variants

If it is possible to include genealogical data (from the local stemmata) in anassessment of the relationship it can be referred to as genealogical coherency45

Here two points are essential (i) Based on the number of prior and posteriorvariants (cf the local stemmata) the relative positions of the witnesses in thegeneral textual flow (see paragraph 412) determine a predominant directionof textual flow between the witnesses (directed coherency ndash the normal situa-tion) Directed coherencies determine which witnesses are potential ancestorsof another witness (ii) In addition the rates of correspondence are impor-tant in estimating the probability of stemmatic coherencies between pairs ofwitnesses Genealogical coherence presupposes pre-genealogical coherence

A special case is the undirected genealogical coherency It exists betweenclosely related witnesses which accordingly have a very high pre-genealogicalcoherency but there is no predominant direction of textual flow An undirectedgenealogical coherency is different from a pre-genealogical coherency on ac-count of the fact that in the former case the basis are pre-genealogical andgenealogical data

Stemmatic coherence is found in the global stemma and the optimal sub-stemmata A given stemmatic coherency designates the ultimate stemmaticrelationship in an optimal substemma (see paragraph 48) Stemmatic coher-ence presupposes genealogical coherence or at least one connective variant (cfparagraph 43)

Textual flow Generalparticular globallocal

The genealogical relationships between witnesses are a reflection of the devel-opment of the text There is a textual flow connecting the witnesses The generaltextual flow leads from earlier to later textual states Consequently each witnesscan be assigned a relative position compared to any other witness within thisgeneral textual flow A particular textual flow exists between witnesses in therelative positions of potential ancestor and descendant Here the textual flow isdetermined by the variants the descendant shares with the ancestor and by thevariants of the ancestor from which new variants in the descendant evolvedThey determine the direction of the textual flow The particular textual flow

Gerd Mink

between ancestors and a descendant becomes part of the global textual flowwhen ancestors and descendant in a substemma are part of a global stemmaBy contrast local textual flow between ancestors and a descendant is effected atone place of variation on the basis of the relationships between their variantsin the local stemma and the relationship of the witnesses according to the par-ticular or if possible global textual flow between them Every local textual flowmust be reflected in the global textual flow and vice versa The local textualflow within the attestation of a variant or for an entire place of variation maybe represented in textual flow diagrams These may be based on genealogical orstemmatic coherencies

Procedures

It is of central importance to pay attention to coherencies and to the textualflows which rest upon them The following insight is fundamental for thecoherence-based genealogical method

In a textual tradition where all the copies have survived and where the sourceor (in case of contamination) the sources are also known as well as the ori-gin of every reading in every copy the genealogical interrelationships betweenall the variants at any place of variation must appear in a global stemma ofthe witnesses as genealogical relationship between coherent fields of relationshipsbetween witnesses

Conversely the relationship between each descendant and its ancestor or(in case of contamination) at least a subset of relationships between it andits ancestors should appear at any place of variation namely as the relation-ship between witnesses sharing the same variant there or as the relationshipbetween witnesses between which a change of the variant took place whichsupports the ancestor-descendant relationship Since every descendant may bean ancestor in relationship to other witnesses chains of coherencies are formedwithin attestations These chains help to find out about unique or multipleemergence of variants Chains of coherencies connect attestations of differentvariants where a change of text took place between witnesses46

Consequently the places of variation are the points of departure for discover-ing by way of the relationships between variants something about the relation-ships between witnesses47 The procedures on which my genealogical studiesare based are briefly summarised below48


James 31222-30

o

a

b

c d g

fh1

h2 i k l mj e


Construction of local stemmata

Local stemmata of variants must be created including where possible the vari-ant that is assumed to be the initial text This procedure makes use of all themethods and knowledge provided by textual criticism Two examples have beenset out in the introduction If it is not possible to establish a local stemma theplace of variation does not qualify for further evaluation until a solution hasbeen found It is not necessary to enter all the variants into the local stemma atall costs Further evaluation requires clear statements Either a variant is des-ignated as genealogically dependent on another variant (or on several if theyexplain the variant together eg by merging) or the origin of a variant is desig-nated as uncertain In the latter case the variant is not taken into considerationfor further evaluation unless it should be possible to make a clear statementabout it later All cases in which no definitive statement is possible must remainneutral in the further evaluation

Figure 7 shows the possibilities within a local stemma the letters representthe variants h1 and h2 indicate that the coherency of the attestation of vari-ant h is imperfect (see below and paragraph 53) and therefore the variant isassumed to have emerged twice d is considered a mixture of a and h1 and thequestion mark indicates the questionable source of a variant Variant n doesnot appear in the stemma as its only witness is a lectionary The underlyingtype of variant is ABC DEF G H I J KLM NOP QR S T UV W X The corresponding absolute numbers of the witnesses are (variant letters inbrackets) 7 (a) 8 (b) 1 (c) 1 (d) 1 (e) 1 (f ) 119 (g) 5 (h) 2 (i) 1 (j) 1 (k)

Gerd Mink

Figure 8 Database section with genealogical indications

2 (l) 1 (m) 1 (o) out of the 164 witnesses included 13 witnesses could not beused because the relevant text is missing from them or uncertain

When local stemmata are constructed ndash and this is very important ndash pre-genealogical coherencies of witnesses within an attestation or between attes-tations must be verified Pre-genealogical coherencies are based only on theagreement of the witnesses and do not provide genealogical information If notall the witnesses cohere like a chain or a net the hypothesis is that the varianthas evolved independently more than once (cf variants h1 and h2 in Figure 7)If witnesses in the attestation of a variant do not cohere with witnesses in the at-testation of another variant a genealogical relationship between the respectivevariants is improbable

Analysis of textual flow

The genealogical links between variants become an integral part of the datasets which contain all genealogical information pertaining to each variantsee Figure 8

Figure 8 is a screenshot of a database section the first seven fields areaddress fields the labez field contains the letter addresses of variants and is up-


dated in labneu for genealogical reasons (for instance two variants are treatedas one or one as two)49 the fields Q1 Q2 Q3 contain the letter addresses of thesource variant(s) of the variant in the Labneu field the fields Vq1 Vq1a etcVq2 Vq2a etc contain the letter addresses of the variants that are regarded assource variants of the variant in the fields Q1 Q2 etc

Consequently the witnesses do not only concern variants but variantswhich are genealogically linked to other variants and their attestations Thesedirect links between witnesses give a first idea of the general textual flowthroughout the tradition by providing some basic information on the ge-nealogical relationship between witnesses Above all we can see which wit-nesses may be potential ancestors of a given witness and the particular textualflows connecting them

The general textual flow corresponds to the development of the text (iethe variants) throughout its history This development can be demonstratedat every passage of the text in local stemmata of variants There are differentaspects of this textual flow a general genealogical one and stemma-orientatedaspects cf paragraph 412

The general genealogical aspect concerns two questions What may the po-sition of a witness in the general development be and which role does it playin the general textual flow The position of a witness within the general textualflow compared with the position of another witness can be determined by theratio of two values The first value (cf for instance the xausy fields in Figures21 23 35) is the number of variants that are posterior to those of the witnesscompared the second one (the yausx fields) is the number of variants that areprior These values also determine whether a witness is a potential ancestor ofanother (cf paragraph 410)

The stemma-orientated aspects concern the particular textual flow and theglobal textual flow between witnesses within a stemma or substemma If there isa genealogical hypothesis on the relation between two witnesses it is based onthe agreements and the direction of the textual flow between these witnesses50

A witness is a potential ancestor of another one if the textual flow runspredominantly from that witness to the other one The potential ancestors areclassified according to their degrees of agreement In Figures 9 21 23 and 35we find a witness under the heading of Zeuge1 the potential ancestors of whichare to be determined These can be found under the heading Zeuge2 if the arrowin the Richtg field is pointing to the left51 The more a potential ancestor agreeswith its potential descendant (cf the Proz1 field)52 the more probable is a directgenealogical relationship (= stemmatic coherency) in the global stemma53 Adirect relation is a relation in which no potential ancestor has been preserved

Gerd Mink

Figure 9 List of potential ancestors of 1243 and beginning of list of potential ancestorsof 2412

as an intermediary witness Lists like those in Figures 9 21 23 and 35 allow foran assessment of the probability whether a potential ancestor will become anancestor in a stemma or substemma and whether global textual flow will arisein consequence

Analysis of genealogical coherencies at places of variation

Now an analysis of the genealogical coherencies within one attestation and be-tween attestations is possible

If close genealogical coherencies connect the totality of the witnesses with-out any break the supposed coherency of an attestation is perfect The hypoth-esis is the variant has evolved only once An example of this is to be seen inFigure 10 The local stemma of variants outlines the hypothesis that b derivesfrom a In order to verify the genealogical coherencies within one attestationthe potential ancestors of every witness in the attestation have to be deter-mined Every witness is thus assumed to be a descendant If witnesses are foundwhich offer the same variant among the potential ancestors which possess a suf-ficient degree of agreements the ancestor with the highest degree of agreementis chosen because it makes a direct genealogical relationship more probableWhether a degree of agreement is high enough is to be read from the waythe values decrease (cf paragraph 411)54 In Figure 9 on the left for examplevalues of gt 895 have to be considered as sufficiently high as 1243 offers a rel-atively old text form (which follows from the range of witnesses in the Zeuge2


perfect coherency stemma

variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8

Figure 10 Example of perfect coherence

field) and therefore many more closely related potential ancestors have beenlost In Figure 9 on the right even a value of eg 925 is inadequate as quite anumber of closer related witnesses has been preserved which (not visible here)are not very similar to each other so that among these witnesses all the vari-ants which may indicate a genealogical coherency are likely to be found55 Theobject of the process is to arrive at evaluations of the local stemmata of variantsthrough evaluation of the genealogical coherencies The quality of genealogicalcoherencies actually allows for rather accurate predictions about the probabil-ity that the genealogical coherencies will result in stemmatic coherencies (cfparagraphs 411 and 55)

When genealogical data yield undirected genealogical coherencies of wit-nesses (cf Figure 9 21 23 and 35 where the Richtg field is empty) these co-herencies will also lead to coherencies in the attestations if they are sufficientlyhigh (cf paragraph 8)

If the procedure has been completed for all the witnesses of the variant andthe result is a chain or net of genealogical coherencies ranging over the wholeattestation the coherency of the attestation is perfect The witnesses are allgenealogically interrelated The variant did not emerge repeatedly by chance

In Figure 10 the arrows for variant b in each case point from the poten-tial ancestors towards the descendants For witness 5 no potential ancestor wasfound in the attestation of variant b All the other witnesses are only eligible asdescendants there Witness 5 should therefore be the oldest text to display thisvariant Its potential ancestor must be identified in a different attestation Inthe example it should be witness 3 for variant a This should be the genealog-ical coherency between variants a and b If there is no potential ancestor forwitness 5 with variant a but for example with a variant c this local stemma ofvariants must be false (The other witnesses for a have not been assigned ar-

Gerd Mink

imperfect coherency stemma

variant a

variant a

variant b

witnesses

witnesses

witnesses

a

b2

b1

1 2 3 4

5

9

6

10

7

11

8

12

c

Figure 11 Example of imperfect coherence

rows as the example only concerns the perfect coherency in variant b and theconnection with variant a)

The coherency is imperfect if not all the witnesses of the attestation can beconnected by close genealogical coherencies The hypothesis is the variant hasemerged repeatedly Logically we have two or more variants with coinciden-tally the same text

In Figure 11 the examination of the genealogical coherencies by themethod described resulted into two groups in the attestation of variant b Wit-nesses 5 and 6 have a coherency and so have 7 and 8 There is however nofurther genealogical coherency for witness 5 in the attestation of b The sameis true for witness 7 The coherency of the witnesses of variant b is imperfectWitness 5 for example has its closest potential ancestor with variant a witness7 with variant c Variant b consequently emerged twice It is therefore splitinto b1 and b256

Another case of imperfect coherency different from the one in Figure 11is not crucial for the first analysis In Figure 12 the direct ancestors of witnesses5 and 7 are both to be found with variant a Here the local stemma of variantswould be the one of Figure 10

With the aid of data as found in the examples in Figure 9 the genealogicalcoherencies of the attestations of variants can now be drawn up and assessedin textual flow diagrams The directed genealogical coherencies will form thebasis as they are the only ones giving direction to the textual flow If they arenot of sufficient quality undirected genealogical coherencies must be used butas a surrogate only or fictitiously imperfect coherencies may materialise



variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8


Lists such as the ones in Figure 9 are used to find the potential ancestors(Zeuge2 field if the arrow under Richtg points to the left) of the descendants(Zeuge1 field) and the degrees of agreement (the percentage values underProz1)57 which make it possible to judge whether there is sufficient genealog-ical coherence58 A textual flow diagram now establishes the connections be-tween witnesses and their most closely related potential ancestors within theattestation by means of directed edges (eg Figure 13) It is also possible toshow by which coherencies the attestations of different variants are connected(eg in Figures 14 15)

In Figure 13 such a textual flow diagram is presented It shows an ex-ample of perfect genealogical coherence within an attestation59 The variantconcerned qualified as lectio difficilior The argument lsquolectio difficiliorrsquo is gener-ally used if the variant is regarded as so difficult that a copyist could not haveinvented but only simplified it If a variant is qualified as lectio difficilior it israther uncredible that it could have emerged more than once In the case of1 Peter 1618 there are essentially two variants a and b which may presentthe original text Three further variants derive from them Variant a shows aparticiple (λυπηθέντας) in the accusative The participle correlates to a remoteaccusative (ὑμας) in 1 Peter 1424 although only slightly earlier in 1 Peter166 the persons involved are the subject Variant b relates the participle tothis subject and correctly puts it in the nominative (λυπηθέντες) Without anydoubt variant a constitutes the lectio difficilior and the change to b is naturaland obvious So far variant b has been considered as the original variant asvariant a was not believed to be correct because of its attestation

Figure 13 shows the attestation of variant a As the data necessary for theassignment of genealogical coherencies for 1 Peter are not yet completely avail-able and were largely lacking at the time when the text was established it was

Gerd Mink

genealogical coherency of theattestation of 1 Peter 1618

variant a(accorging to the genea-logical results in James)

A

01

1175

2492

1243 1852

1448

429

17352

2344

307

206 522 1292 1490 2200 453 720 918 1678 24644

4243

01422 0204

1047

18422 254121838

63 884 3264 3983 4422 9154 996515244 21476 25444467 629 642

6234

2818

2805941799 614 1831 630

2412

69

61 1848 218 808 1661 254

621

1359 1127

436 1067 1409 1718 1563 2374

Figure 13 Local textual flow through the attestation of a variant

referred to the genealogical data from James which were related to the pre-genealogical data already known for 1 Peter60 as they are based exclusively onthe agreements The examples in Figures 13 14 and 15 make it very clear thatthereby a good prediction for the genealogical coherencies was also possible for1 Peter For each witness in Figure 13 the most closely related potential ancestorwithin the attestation was looked for A the hypothetical initial text for exam-


ple is the closest potential ancestor of quite a number of witnesses (01 1175etc) In principle it is therefore possible to consider this variant for the originalreading There are also coherent chains (eg A ndash 1852 ndash 1448 ndash 429 ndash 206 ndash 1799)ranging over the complete attestation

If no superscript is given after the number of the witness it means that thepotential ancestor is the one with the highest degree of agreement altogether(not only within the attestation) and therefore has a very high probability ofbeing ancestor in a global stemma (level 1) This is the case for the vast majorityThe superscript numbers indicate lower levels Of these levels of 2 to 5 may ndashgenerally ndash still be considered as good depending on the proximity of the levelsThe minimum here a level of 7 for witness 104 is still acceptable since quite anumber of potential ancestors with almost the same level exists for this witness

It is striking that the superscript numbers predominantly appear with wit-ness 424 and the series below it (0142 020 etc) thereby reflecting an area ofespecially vigorous contamination surrounding the development of the uni-form Byzantine text The main body of the Byzantine witnesses have variant bFor the witnesses followed by a superscript number in Figure 13 the altogetherclosest potential ancestor is to be found in the b attestation Most of themare characterised by the fact that their more closely related potential ancestorsdiffer considerably from one another (which is otherwise atypical)

The verification of the genealogical coherencies does not actually provethat variant a is original as the corresponding test for variant b would re-sult into a comparable picture But the genealogical coherencies in Figure 13demonstrate first of all that derivation of a from A and therefore its original-ity is possible and secondly that no multiple coincidental emergence mustbe assumed

A further example may demonstrate the advantage of observing the textualflow and coherencies in a contaminated textual tradition (Figures 14 15)61

There are two main competing variants in 1 Peter 41624-28 a (τῳ μέρειτούτῳ)62 and b (τῳ ὀνόματι τούτῳ)63 cf Figure 14 So far variant b has alwaysbeen favoured and seen as the original because of its attestation in witnessescounting as outstanding 03 01 02 1739 Variant a in contrast is essentiallyattested by Byzantine witnesses which have generally been judged to be of infe-rior value In addition there are a few more witnesses (025 1448 1735 2298)which are however less highly regarded than some of the most important wit-nesses for variant b Yet variant b which had been considered original in view ofits attestation can hardly have served as the source for a The meaning of vari-ant a is not immediately clear and a represents a far more difficult reading Yetvariant b does not irritate and its phrasing agrees perfectly with a neighbouring

Gerd Mink

14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

642

218

Byz

5 69 88 323 442 915 1832 2718

25412 87621359

1563 4361718 10672374 1409

623623 6212464

2805

12432

813

03

02044 2344 21386142

2492 1505

11752 17392 18522012

57 33 18902412

2464

2805

945 1241 16114

1 Peter 41624-28

variant a

variant b

1359 2541

Figure 14 Multiple genesis of variant b

passage (1 Peter 4146-10)64 There are no grounds however for changing binto a It appears more reasonable to assume that b has developed from a Butis this possible considering the attestations

The genealogical coherence for variant a has proved to be very good anda diagram very much like the one in Figure 13 can be drawn up (symbolisedby the rectangular frame on the upper right in Figure 14 Byz = Byzantine wit-nesses)65 for variant b the genealogical coherence is definitely imperfect Partof the witnesses for b show unmistakable genealogical coherencies with severalwitnesses for a (cf the witnesses on the lower right connected to the frame)66

This means that the witnesses immediately below the frame read variant a butnone of their most closely related potential ancestors does In Figure 14 nextto the large group on the left which derives from A through 03 and presentsperfect coherence within itself a large series of witnesses (to the right underthe frame) happen to read the same variant b If variant b is original variant amust have been derived from b and in the third instance several times variantsmust have been derived from a which happened to have the same phrasing as bThis is quite conceivable because b as indicated presents an obvious readingin the context


14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

1 Peter 41624-28

variant b

3073

14485

180 453 468 720 918 1678 2186 2818 12922

17354

429

206 522 1490 2200

424

025

22982

variant a variant b

A

Figure 15 Differing textual flows with changing initial text hypothesis

This offers an alternative as shown in Figure 15 Following the coherenceexamination both alternatives are possible Figure 15 shows the differenceswhich appear in the textual flow diagram depending on whether variant a(top) or b (bottom) is seen as initial text This naturally results in a changeof position of A In addition there are insignificant changes within the attes-tation of variant a The top diagram shows a number of witnesses deriving theoriginal variant directly from A If the variant is no longer original as in thebottom diagram the textual flow has to explain the association of the witnessesin bold print by way of another witness as A is no longer available In this caseit can easily be done through 025 although that changes the position of 1735considerably

Gerd Mink

That such diverging textual flow diagrams are at all possible is due to thedegree of contamination since it requires more than one ancestor in a sub-stemma To establish such a textual flow diagram the witnesses have to bepresented according to the highest levels of probability that one is ancestor tothe other (cf the witnesses without superscripts) Only very rarely is the sec-ond to fifth highest level of probability needed (cf the superscript numbersfollowing some witnesses)

Process of approximation

After as many local stemmata of variants as possible have been drawn up for theplaces of variation the entire genealogical data and the analysis of genealogicalcoherencies contribute to revising the local stemmata and the cases unsolved sofar Then new genealogical data can be produced which may result in furtherrevisions until a sufficiently stable overall genealogical hypothesis has beenachieved This repetitive procedure is necessary because the knowledge thatwill only be the final result of the entire process would be required for theassessment of the genealogical relationships between the variants An overallhypothesis can therefore only be acquired through approximation and iterativerevision of all the intermediate results I am referring to the circular reasoningwhich cannot entirely be avoided in textual criticism but has to be controlled(cf paragraph 4) Before an overall hypothesis can be achieved the followingpoint is very helpful

Construction of graphs based on predominant textual flows

Since we have now a selection of potential ancestors and the necessary informa-tion to assess the probability that they are ancestors in a global stemma somehypotheses concerning the textual flow through the witnesses can be proposed

The predominant textual flow (first level of probability67 Figure 16)68 thetextual flow with the highest probability and the textual flow which is secondin probability are shown in the top area of the tree (Figure 17)69 In Figures 16and 17 lsquoArsquo is the artificial witness of the initial text containing all the readingsthat are assumed to be initial readings in the local stemmata of variants Theintroduction of A is the logical consequence of making textual decisions70

In the second line of Figure 16 (P74-1852) the witnesses which have thehypothetical initial text as closest potential ancestor appear Comparing the di-agram with usually established stemmata the derivation of such a large numberof witnesses may come as a surprise But the result is based on two arguments


1827

2180

2242

3515

2420

8004

966

518

4518

7418

7516

095

400

322

1390

104

1842

643

2544

8832

132

660

799

612

5113

6717

6526

7413

5911

2761

418

90

254

4318

5062

324

6412

9725

4118

3862

117

5161

1837

1661

876

1832

2494

2652

1563

1718

2374

2412

330

2718

2805

1270

1598

1893

436

1067

1409

2243

378

2147

1595

1066

0142

1853

1848

020

618

3893

181

252

319

323

431

442

456

459

676

915

999

1501

2423

2523

218

808

1799

1831

630

1292

2138

2495

398

018

056

6931

246

761

762

963

164

215

0917

2927

7420

652

214

9022

0015

05

P20

3394

197

424

1840

2718

S42

916

11

0204

417

3523

4418

045

346

872

091

816

7821

8621

9728

1824

9202

4694

512

4122

9814

48

P74

P10

001

0304

025

8130

711

7512

4317

3918

52

A

048

365

Figu

re16

P

redo

min

antt

extu

alfl

owth

rou

ghw

itn

esse

sof

Jam

es

Gerd Mink

A

03

P100 048 01 1175 1852

025 1448

81 1243 307

46804173502

P74 2344 1739

2298

945

1241

33

The two most probableancestors

most probable

second most probable

Figure 17 Predominant textual flows through the top of the tree of Figure 16

Firstly derivation of a predominant textual flow from hyparchetypes is incom-patible with the principle of simplicity which forbids unnecessary assump-tions The problems that conventionally require the usage of hyparchetypes aresolved by a different method that derives the textual flow from several ances-tors in accordance with the contamination (cf Figures 28 36 40) Secondlythe large number of witnesses deriving directly from A is a reflection of the factthat from the first millennium of the textual tradition only a very small percent-age of witnesses has been preserved For in this case the direct predominanttextual flow means nothing else but that of the many witnesses which stoodbetween A and those of the second line not a single one has survived

In the second line many witnesses of undisputed importance appear 03(the best witness) 01 04 1739 and the papyri But also witnesses which sofar have been rated less important like 025 appear or even an almost disre-garded witness like 307 In contrast an illustrious witness like 02 is missing itappears in the third line deriving from 81 Traditionally this relationship hasbeen assumed to be the other way round presumably because the text of 02 ispreserved in an old manuscript (5th century) and that of 81 in a much later one(11th century) But in James 81 has a predominantly older textual state than02 However what is not visible in this case is that 02 nonetheless has readings


deriving from A in 28 instances which do not occur in 81 and therefore can-not derive from 81 (this is reflected in Figure 17 edge from A to 02) As thewitnesses in the second line have A as the potential ancestor with the highestdegree of agreement they are of eminent importance for the reconstruction ofthe textual tradition71

The diagram (Figure 16) demonstrates impressively that there is a con-tinuous flow through 307 468 424 and 617 from the old text into the ex-tensive Byzantine tradition The right hand branch of the diagram shows howthe way leads to the witnesses of another well-know text form (HK) through1852 and 1448

Incidentally it is striking that nodes with many edges emanating fromthem often display numbers belonging to commentary manuscripts (307 inthe second line 424 in line four 617 in line five) Commentary manuscriptscontain apart from the commentaries the continuous New Testament textIt is unlikely that these commentary manuscripts served as exemplars formanuscripts containing the continuous text But it is very plausible that thattext form was chosen as the basis for a commentary which was the most highlyesteemed at that particular time and place Accordingly it would also have beenused as an exemplar in the scriptoria

Figures 16 and 17 are not part of a global stemma they give an overviewof textual flows which in the case of Figure 16 should be found with a veryhigh level of probability in a global stemma as well It is possible that in Figure17 the second most probable textual flow might not explain more variants ofa descendant than the most probable one and therefore becomes superfluousin the global stemma But this is not very probable and so a good overviewof the position of the witness in the textual history has been achieved Figure17 should be compared with the actual results in the optimal substemmata inFigures 28 36 and 40

Typical problems of contamination

Figure 17 already points to some of the problems to be expected in a contami-nated textual tradition Very often we find relationships like the following (cfFigure 18) D originates from C and from B from which C originates Proba-bly the differences between B C and D are not very big Usually the membersof a substemma have a high degree of agreement This is a reflection of thecircumstance that in a region with a dense tradition there exists a stable ideaabout which manuscripts are worth copying

Gerd Mink

B

C

D

Figure 18 Model of a simple contamination

B

C

D

E

Figure 19 Model of a two-stage contamination

B

C

D

E

if altered variants of andif merged and andif altered variants of and

if merged and andif altered -variants included in

according to

C BD B CD B

E D BE C D

B

then may not be potential ancestor ofmay be potential ancestor of

C EE C

Figure 20 Circle possible in a case of multi-stage contamination

The problems arise when there is a chain of relationships like the oneshown in Figure 19 In this case it is clear that B and D are potential ances-tors of E But it is also possible that C is not a potential ancestor of E accordingto the definition of potential ancestry72

If C altered some variants of B and if D merged B and C while alteringsome variants of B and if E merged D and B altering the special C variants in-cluded in D in accordance with the corresponding B variants then C is not


a potential ancestor of D and even a circular relationship becomes proba-ble E looks like a potential ancestor of C The condition for this is that Ewhile it merges D and B retains enough posterior variants compared to DThis problem will be dealt with in more detail below (cf paragraph 9) Theexample demonstrates that edges in a stemma may be misleading if we donot pay attention to the variants whose transmissions and developments arerepresented by them

A situation like that in Figure 20 is possible because usually a numberof the variants of the ancestor in a contaminated tradition are posterior to thecorresponding variants of the descendant and a number of the variants of thedescendant are prior to those of the ancestor (Figure 23 shows the potential an-cestors of witness 1243 in the Zeuge2 field In the yausx field the number ofcases in which 1243 has a prior variant compared with the potential ancestorsis listed In the xausy field we find the number of the posterior variants)

Even if all the substemmata (containing a descendant and its ancestors)are true (as is the case in Figure 19 see below for the meaning of lsquotruersquo)the global stemma may become false if indirect connections are interpretedA graphic representation like Figure 19 can be read as a group of substemmataThe directed edges indicate only the predominant direction of the textual flowbetween two witnesses and what textual flow is necessary for the explanationof the textual state of a descendant A stemma which does not give informationon details of the textual flow (ie which variants it contains) cannot be usedto determine the relation between nodes which are indirectly connected Thedata underlying and qualifying a specific stemmatic connection must thereforebe available for verification and assessment It is also evident that the edges inthe substemmata are of particular importance as carriers of information

A substemma representing a contaminated tradition must relate a descen-dant to an optimal combination of ancestors A combination is optimal if itexplains the totality of the variants of the descendant and if this condition ismet is as small as possible The requirement is that the substemma is true forevery passage of text In the case of Figure 19 each variant of D agrees with Bor agrees with C or has developed from the corresponding variant in B or fromthe corresponding variant in C At least one of these possibilities must apply toevery passage

The following example shows an optimal substemma in which the con-nection is true It is simple and does not contain contamination

It concerns the best witness we have for the Catholic Letters 03 (themanuscript is datable to the 4th century) In Figure 21 the witnesses com-pared are in the Zeuge1 and Zeuge2 fields Further information is to be found

Gerd Mink

Zeuge1 and Zeuge2 ndash the witnesses being compared Richtg ndash the predominant direction ofthe textual flow Proz1 ndash agreement in percentages Kon ndash agreement in absolute numbersProz2 ndash precentage of cases in which Zeuge1 has a variant which represents a further devel-opment of a variant of Zeuge2 Xausy ndash the same in absolute numbers Proz3 ndash percentageof cases in which Zeuge2 has a variant which represents a further development of a variantof Zeuge1 Yausx ndash the same in absolute numbers Proz4 ndash percentage of cases in which thesource of a variant in Zeuge1 is doubtful Qfragl ndash the same in absolute numbers Kv ndash casesin which there is no direct connection between the variants of Zeuge1 and Zeuge2

Figure 21 Potential ancestors of witness 03

A

03

Figure 22 Optimal substemma of witness 03

in the other fields There are three potential ancestors A the initial text whichis hypothetical and fragments 0166 and 0173 They are potential ancestorsbecause compared to 03 more variants of these manuscripts are prior than pos-terior Thus the textual flow predominantly runs towards 03 The textual flowbetween fragment P23 and 03 has no direction

The resulting substemma is simple (Figure 22) A is the ancestor of 03 be-cause in all cases where A and 03 disagree the variants in 03 were created onthe basis of the corresponding readings of A without agreeing with the vari-ants of 0166 0173 or P23 These witnesses do indeed contain prior variantscompared with 03 (cf the xausy field in Figure 21) but in each case they areidentical with the A variants Therefore these witnesses are not needed in asubstemma which explains all the variants of 03 (This conclusion cannot bedrawn from the values in the list we have to know the variants at the placesof variation concerned) It is possible that the text of 03 is also contaminatedalthough no further source of contamination has been preserved73

The substemma in Figure 22 like all substemmata is based on all the placeswhere the local stemmata offer a hypothesis about the relationship betweenthe variants of the descendant and the other variants Places which remain un-solved or where the descendant has uncertain relations with other variants aretherefore not included For 03 this is the case in one place (cf Qfragl in Figure


Figure 23 Potential ancestors of witness 1243 See Figure 21 for field legends

1243 756 places of variation

potential ancesstors

A 1175 025 03 1739 01

686926

694920

648915

670902

677896

648871

XXXXX

XXXXX

XXXX

XX

XX

XX

XX

XX

X

X

SUM

722732732734736

734735

Figure 24 Combinations of potential ancestors of witness 1243

21) I will return to this when I discuss the problem of prior variants foundonly in non-ancestors (cf paragraph 7)

Another example concerning the top area of the global stemma demon-strates contamination We see 1243 (Figure 23 the manuscript is dated to the11th century) and its potential ancestors from A to P100 The first three wit-nesses are fragments and the textual flow between 1243 and each of them hasno direction Their agreements with 1243 are included in the agreements of theother witnesses

Figure 24 shows potential ancestors of 1243 arranged in order of the degreeof agreement (not the absolute number of the agreements) in the next lines thenumbers and percentages of agreeing variants are presented and in the righthand column the total number of variants explained by the agreements with atleast one witness within some combinations (marked with crosses) can be seen

Gerd Mink

1243 has 756 places of variation (ie 756 variants 761 is the maximum possi-ble in James) A is the most closely related witness in this group agreeing at686 places The percentage refers to the totality of places of variation extant in1243 and in the witness compared The agreements with A or 1175 explain 722variants in 1243 If we combine A 1175 and 025 we can explain 732 variantsbut adding 03 does not result in a greater number74 This method allows totest each contribution of potential ancestors to the explanation of the variantsof 1243 The top values are shown in Figure 24 The number of 736 variants isnot exceeded Normally variants not explained by agreements with an ancestormust be secondary altering a variant of at least one of the ancestors75

Most of the variants of 1243 are explained by the combination Andash1175ndash025ndash1739ndash01 (Figure 24 marked line) But are 5 ancestors really necessaryThe overall agreement of 01 (871) is rather poor We have to ask thereforewhether agreements may be coincidental

I summarise the principles of the procedure A substemma links a descen-dant with its hypothetical ancestors which are to be found among the potentialancestors The substemma as defined is optimal if it can explain the textualstate of the descendant from as few ancestors as possible These ancestors willbe very similar to the descendant and consequently its variants must derivefrom the agreements with the ancestors in as many places as possible This isin accordance with the basic model (cf paragraph 4) Their relationship withthe descendant must therefore be based on the highest possible values of agree-ment Lower values raise the question whether the agreements are coincidentalWhat values are to be regarded as high or low depends on the rate at which thepercentage agreement values (Proz1 field) in lists like the one in Figure 23 de-crease76 In cases of doubt the decision whether a variant leads to a stemmaticconnection between two witnesses has to rest on the character of the variant

A fundamental remark is necessary at this point The entire procedure isbased on all the genuine variants excluding errors and orthographica There isno selection of variants assumed to be more important than others In mostcases a scribe copied a manuscript probably using not more than one sourcecopy and he fulfilled his task very well In the view of the scribe there existed novariant He transmitted important and unimportant passages without makingany difference A decision between two variants was only possible if the scribeknew another variant It is a feature of closely related witnesses that they haveimportant as well as unimportant variants in common Whether a variant isconnective (and not accidental) depends on the degree of agreement of two wit-nesses or on the character of the variant If two closely related witnesses agreeon an unimportant variant the variant is connective The coherence within


an attestation is the most important hint in this respect We cannot thereforedecide at the very beginning which variants are connective

If two witnesses not closely related agree the variant is normally not con-nective If these witnesses agree on a significant variant which is unlikely tohave emerged more than once without genealogical dependence the variant isconnective Such a variant is not possible without contamination As a witnesswith relatively few agreements cannot explain enough variants of a descendantit can only be one of several ancestors

All the variants of a descendant which are explained by a new member of acombination of ancestors have to be examined especially if no high degree ofagreement speaks for genealogical relationship Is it necessary to assume thatthe witness is an additional ancestor If we do assume a combination of ances-tors we must analyse the special role of each ancestor and figure out what theconsequences are if we exclude it from the combination In order to make anoptimal decision the database with all the variants of a witness must be used

The database (Figure 25) containing all the variants of 1243 and its po-tential ancestors and showing the genealogical relations of the variants is thestarting point for establishing the optimal substemma The database allows toobserve what variants of a potential ancestor are suitable for explaining thevariants of the descendant In this way the examination of their wording alsobecomes possible

The resulting substemma (Figure 26) only gives the information that oneof the following possibilities is true whatever passage we check (a) at least oneof these 4 ancestors agrees with 1243 ndash this is true for 734 places of variation ndashor (b) if none agrees at least one has a prior variant explaining a posteriorvariant in the descendant ndash this is true for the remaining 22 places of variation

We cannot however see the importance of each ancestor To each edgesome values can be attached (Figure 27 and 28) Of these values those whichexclusively pertain to the agreements have a direct influence on the compo-sition of the substemma The others indicate the respective qualities of thestemmatic connections and make it possible to assess how sensitive the con-nections are to changes in the local stemmata of variants Such changes affectthe ratio of prior and posterior variants which is critical to potential ancestryThe stability of the textual flow from ancestor towards descendants must begt 0 In the case of very low values a change may cause the value to becomelt 0 and the direction of the textual flow to reverse The witness which was anancestor may become a descendant

Figure 28 shows a substemma and values qualifying the edges The valuesof agreement (cf strength1 in Figure 27) and priority (included in strength2

Gerd Mink

Anfadr Endadr ndash beginning and end of the place of variation Hsnr ndash basic witness of thecomparison Labneu ndash code of its variant Rg ndash relation between the variant of the basicwitness and the variant of the following ancestor = ndash agreeing variants lt ndash variant of thebasic witness posterior gt ndash variant of the ancestor posterior ndash relation questionable - ndashno relation possible Mrg ndash mediate relation A Ms1175 Ms025 etc ndash names of potentialancestors

Figure 25 Database section containing the genealogical relationships of the variants ofpotential ancestors of 1243 with the variants of 1243

A 1175 025 1739

1243


in Figure 27) demonstrate the strength of the textual flow It depends on thecharacter of the witnesses and their place in the textual tradition if values ofagreement are considered to be high (cf paragraphs 3 and 411) The priorityvalue need not be high Very high values argue against close relationships asthey lower the number of agreements The stability value gives an idea of thereliability of the ancestor-descendant relationship


The textual flow from a potential ancestor to a descendant

strength 1 = a times 100c

strength 2 =(a + p) times 100

c

stability =(p ndash s) times 100

c

a number of agreementsc number of passages where comparison is possiblep number of prior variants of the potential ancestors number of posterior variants of the potential ancestor

Figure 27 Properties of textual flow

A 1175 025 1739

1243

agreements absolutendash ()

priority variants ()stability of the textual flow

additional contributionmaximal contributionminimal contribution

686926729729

5740

2

694920385013

8723

1

6489154100427ndash8677

3

677896463040

2708

2

Figure 28 Values qualifying the edges in the substemma of 1243

Another value shows the additional contribution of an ancestor within acombination its relevance for it It indicates the number of variants of the de-scendant which could not be explained without this witness77 Witness 1739for example explains 2 additional variants where it does not agree with theother three ancestors

The value of minimal contribution is very interesting Unlike the value ofadditional contribution it concerns cases where coincidental agreement is im-probable on account of the character of the variants In the case of both theadditional variants 1739 is a necessary ancestor of 1243 because no other an-cestor explains the variants of 1243 and the variants are also connective due totheir character78

Gerd Mink

genealogical coherencies in small attestation in James

2142-10

21516

2198-14

2262-4

03 11751243

04

11751243

04

03 1175 1243 2492

variant c

variant b

variant e

variant b

1175 1243

stemma (section)

03

1175

1243

042492

Figure 29 Local representation (in attestations left) of a global stemma (right)

As a result the maximal contribution of 1739 is 708 (agreements plus anyprior variants) and the minimal contribution is 2 In the case of A there are5 additional contributions Only 2 of these can be explained exclusively byA79 The minimal contribution of 1175 is 1 or even 0 for the only readingof 1175 qualifying as minimal contribution could be a simple independentorthographic agreement (as the variant of 1243 could be)80 Nevertheless itis a member of the group of ancestors If we were to remove 1175 from thegroup assuming the agreements to be coincidental (in spite of their number)8 variants of 1243 (cf additional contribution in Figure 28) would remain un-explained by agreements and we would have to search among the potentialancestors for another witness or other witnesses to explain these variants Butshould we find such a witness its general agreement with 1243 would be lowerthan that of 1175 and in case of two or more witnesses being necessary forthe explanation of the 8 variants under discussion the group of ancestors inthe substemma would increase In either case we would have an inferior choiceand it would not be more likely that the agreements were not coincidental thanin the case of 1175

It may be interesting to look at some of the additional contributions whereattestations displaying the assumed relationship of 1175 and 1243 contain only


a small number of witnesses On the right hand side of Figure 29 we see anextract of a stemma On the left hand side there are variants with their attes-tations The edges show their genealogical coherencies which are completelyin harmony with the assumed stemma (cf Figure 43 and the substemmata inFigures 36 and 40) Although 1175 has not been qualified as necessary at theseplaces of variation because there are other witnesses (except for the secondexample) and although the variants are not especially significant81 there is nodoubt that 1175 participates in the textual flow as it usually does The examplessupport the general genealogical relations very well

A last question is why 01 is not needed as a further ancestor although itprovides two additional instances of agreement (cf Figure 24 marked line)The reason is that the agreements are rather coincidental in view of the kind ofvariants and the lower general agreement of 0182

Prior variants found only in non-ancestors

If any two witnesses of a highly contaminated textual tradition are compared atrandom almost every single one has prior and posterior variants as comparedto the other Though prior variants prevail in potential ancestors they alsocontain a (sometimes only slightly) smaller number of posterior variants Theopposite is true for witnesses to which the predominant textual flow is directedThey are therefore not potential ancestors

If the tradition is dense and every witness has a series of closely relatedpotential ancestors it will be easy to find the prior variants among them whichexplain the posterior variants in a descendant Generally speaking this is alsotrue for a less dense tradition but in such a tradition the prior variant whichcorresponds to a posterior one may only occur in one witness which is not apotential ancestor

In the case of two witnesses C and D which both can be traced back toan ancestor B a very simple stemma evolves (Figure 30 left) D is assumed tohave incorporated more changes from B than C has This means that D cannotbe a potential ancestor of C Let us now suppose that there are places where Chas variants which evolved from variants found only in D Yet their number isso small that D is still not a potential ancestor of C C has eg 10 posteriorvariants compared to B D has 30 compared to B C has 5 posterior variantscompared to D D has 25 compared to C For the 5 places where C has posteriorvariants compared to D the left hand stemma of Figure 30 does not apply

Gerd Mink

BB

CC DD

C lt D

Figure 30 Intermediary node in case D has prior variants compared to C withoutbeing its ancestor

A correct stemma should therefore also represent the fact that prior vari-ants are found in a witness which is predominantly not prior and therefore nota potential ancestor The solution is an intermediary node pertaining only toprior variants in a non-ancestor (cf Figure 30 right)

If there were no more surviving witnesses than B C and D it would bepossible to establish the connection between C and D via a hyparchetype Ifthe connection between B C and D is perceived as a substemma within ahighly contaminated tradition however it would not be possible to establish awell founded hyparchetype if the hyparchetype is in turn also the product ofcontamination

The intermediary node does not represent a hyparchetype A hyparchetypewould be a hypothetical witness comparable to other witnesses but only hy-pothetical and would essentially contain variants at all the places of variationThe intermediary node only contains the variants which are relevant in thiscase It is not advisable to merge the intermediary node into the edge point-ing from B to D as it might just as easily be situated on the way from B to Cand as it could in principle also be the product of contamination The inter-mediary node in this case is to be understood as follows There are posteriorvariants in C deriving from prior ones which are found in the non-ancestor Dand which evolved from variants in B C and D are now represented as productsof contamination This may be accurate The intermediary node may also asdescribed be part of the linear development from B to one of its descendantsso that only one of the two witnesses is a result of contamination Also in thisrespect an intermediary node must therefore not be interpreted in the sameway as a main node The stemmatic representation therefore is not a definitehypothesis about the location of contamination in the textual tradition andit raises various possibilities The intermediary node helps to solve a problemwhich is the result of conflicting data (prior and posterior variants in one andthe same witness) and ultimately of contamination and loss of intermediarywitnesses


For an illustration I go back to Figures 21 and 22 The conclusion was thaton the basis of all the places where a certain variant could be assigned to Aand the variant of 03 could be put into a local stemma the ancestor of 03 in anoptimal substemma was A and no other witness Yet according to Figure 21 (cfQfragl) there is a place where the relationship between 03 and A is doubtful inJames 242-4 The variants and witnesses in this place are

a οὐ διεκρίθητε lsquohave you not made distinctionsrsquo (rhetorical question)01 02 03 (additional to the reading of the continuous text) 04 andmany others

b οὐχὶ διεκρίθητε (same meaning as variant a) 2544c διεκρίθητε lsquoyou have made distinctionsrsquo 03 (in the text) 1852d καὶ οὐ διεκρίθητε (same meaning as variant a) 025 and many otherse καὶ διεκρίθητε (same meaning as variant c) 322 323 629

In the ECM variant a was the text established by the editors although it couldwell be a simplification compared to d83 At the time the quality of the attesta-tion of variant d (essentially characterised by 025 and Byzantine witnesses) didnot appear sufficient for assuming it to be the initial text Other editors follow-ing the traditional view of textual history acted likewise So far the secondaryrise of variant d remains unexplained

The first word in variant d is καί normally meaning lsquoand alsorsquo Neitherthese meanings nor any of the other documented meanings make a sense hereThe text preceding the place is a rather long conditional clause interrupted bytwo short phrases in direct speech The following apodosis is then introducedin a completely non-Greek manner by καί a phenomenon which is howeverknown from the realm of semitising Greek as lsquoκαί apodoseosrsquo and was possiblyapplied as well in James 41584 It is rare in the New Testament Later copyistscannot have been familiar with this usage and the word must have appearedsuperfluous to them

Further genealogical investigations have revealed that in the attestation ofvariant d genealogical coherencies are present which could lead by way of wit-nesses 025 and 307 to A the initial text A is potential ancestor of both 025and 307 with the highest degree of agreement (cf Figure 16) Consequentlyvariant d could also qualify as initial text As variant a is easily explained as anadaptation of variant d to common usage and a number of witnesses for varianta are easily traced back to A and as conversely variant d cannot be explainedfrom variant a variant d was subsequently declared the initial text85 Figure 31shows the resulting local stemma

Gerd Mink

d

a e

bc

Figure 31 Local stemma of James 242-4

B

03

04

03 lt 04

Figure 32 Optimal substemma of 03 with intermediary node

Variant c has not been put into the stemma On the one hand it might be asimplification of variant e with the objectionable καί left out in the same way asvariant a probably did on the basis of variant d This however is not very likelyfor the witnesses of e are not potential ancestors and quite remote from 03 aswitness of variant c Within the attestation of c by the way 03 is ancestor of1852 Derivation of variant c from variant d or variant a is more likely The lastword before the place of variation is μου For variant d the resulting context isμου καὶ οὐ διεκρίθητε for variant a μου οὐ διεκρίθητε The reason for variantc appears to be a very common scribal error the copyist eyeskips from onesequence of characters (here ου in μου) to an identical sequence later on in thetext (here οὐ before διεκρίθητε) and continues copying from that point Withboth variant d or variant a as a basis the result would be variant c

If variant c is derived from variant a the problem arises that there is nopotential ancestor of 03 in its attestation Yet its witnesses would have a priorvariant compared to 03 which is not found in any of the potential ancestors of03 For variant a 03rsquos closest relative is 04 A correct stemma should thereforeinclude the fact that there are prior variants in a witness which is predom-inantly not prior The solution is an intermediary node pertaining only tothis one variant (Figure 32) This will ensure that 03 can at the same time bean ancestor of 04 (see Figure 40) If variant c is derived from variant d thesubstemma of Figure 22 is true86


Witnesses which are connected by an undirected genealogical coherencyare also to be classed as non-ancestors No direction of textual flow can bedetected between them for both contain equal number of prior and posteriorvariants Let us assume that this applies to witnesses C and D in the examplein Figure 30 (right) An analogous picture would result if D has a prior variantwhich is essential to explain a variant of C At the same time it is conceivablethat C has a variant which is essential to explain a variant of D In that casethere would not only be an intermediary node D gt C like the one in Figure 30but also an intermediary node C gt D

P23 for example is a fragment However it actually provides text at 49places of variation but there are mostly only one or two witnesses which dif-fer from the initial text at these places Thus P23 has undirected coherencieswith an abundance of witnesses Compared to the initial text P23 differs onlyonce variant ff in James 11740-46 This variant is represented by an erro-neous reading in P23 which is probably based on variant d The witnesses ford are 01 (first hand) and 03 There are also undirected coherencies with theseAs in the previous example 01 and 03 should be linked with P23 by way of anintermediary node in an optimal substemma on account of this prior variant87

P23 in turn has a prior variant compared to 01 and another one comparedto 0388 But in both cases P23 agrees with the initial text A A is ancestor in thesubstemmata of 01 and 03 P23 is therefore in no way necessary to explain avariant in 01 or 0389

Undirected genealogical coherencies

At the beginning the search for the optimal substemmata is based on the po-tential ancestors only because between them and the descendants directedgenealogical coherencies exist and only they offer any information about thedirection of the textual flow When a substemma is found which explains allthe variants of the descendant with as few ancestors as possible the possibilityremains that the substemma does not explain a number of close genealogicalcoherencies resulting from frequent agreements of witnesses within attesta-tions These coherencies may not be indicated by other substemmata eitherand so not be reflected in the global stemma This phenomenon is caused bythe undirected genealogical coherencies

Let us assume a substemma consisting of an ancestor B and a descendantC A further substemma links ancestor B with descendant D The two substem-mata are compatible with the situation at a place of variation as represented in

Gerd Mink

variant b

variant a witness B

witness C witness D

Figure 33 Undirected genealogical coherency between C and D within an attestation

C D

C D

intermediary node

C D

relation without direction

Figure 34 Resolution of an undirected edge through an intermediary node

Figure 33 If there is no link between C and D these witnesses must have devel-oped their variants independently from the variant in B as the diagram seems tosuggest They would coincidentally have the same variant If we further assumea close undirected genealogical link between C and D (cf the link between Cand D in Figure 33) it is impossible that they have developed the same variantcoincidentally This fact has to be reflected in a global stemma

When we find such witnesses with close undirected genealogical relation-ships (cf the undirected edge in Figure 34 left hand graph) intermediarynodes are apparently needed for an exact description of the relationship Thesupposition is that witnesses C and D agree in variants which have not emergedindependently but stem from common ancestors and the condition is that oneor more attestations exist which contain neither a (surviving) common ances-tor of C and D nor a witness which is a descendant of C and an ancestor of Dor the reverse

These intermediary nodes are not hyparchetypes in a traditional sense butshould better be understood as a special kind of connection The content ofsuch a node as in Figure 34 (right hand graph) is all the agreements between Cand D Thus C and D have their ancestors and the intermediary node estab-lishes a link to the ancestor or ancestors (= sources) of the common variantsof C and D If the content of the intermediary node is the result of contamina-tion witness B is only one of the ancestors of the intermediary node (cf Figure



AA 1243 0303 18521739

041175

Figure 36 Connection between two substemmata through an undirected edge

33) The intermediary node has to be incorporated into the substemmata Ifthe substemmata of both witnesses are true the ancestors of the intermediarynode have to be the intersection of the witnesses which are ancestors in the twosubstemmata or a subset of the intersection

For illustration I take two witnesses which are closely related but thereis no predominant direction of textual flow between them 04 and 1175 (cfFigure 35 no arrow in the Richtg field in accordance with the equal values inthe xausy and yausx fields)90 They are shown with their optimal substemmatain Figure 36

04 and 1175 have 12 places where they have secondary variants in com-mon but considering the good coherency between the two witnesses this isno accident Due to the old text form the agreement values (Proz1 field) beginat relatively low percentages According to the optimal substemma 1739 is de-scendant of 04 (Figure 40 top right) and ancestor of 1175 (Figure 36) At 7 outof the 12 places 1739 reads the same variant as 04 and 117591 The hypotheticalroute of the variants at these 7 places could therefore be 04 rarr 1739 rarr 1175At 2 out of the 12 places 03 reads the same variant as 04 and 117592 in whoserespective substemmata 03 is the ancestor (cf Figure 36) The route of the vari-ants could therefore be 04 larr 03 rarr 1175 3 places now remain93 where 04 and1175 agree but in the same attestations no ancestors of at least one of themare available which could connect these witnesses The substemmata in Figure40 are true without a link between 04 and 1175 inasmuch as the variants at

Gerd Mink

04 1175

A

1852 1243

04 1175

1739

03

Figure 37 Resolution of the undirected edge in Figure 36

the three places are further developments of the source variants offered by atleast one of the relevant ancestors found in the substemmata of 04 and 117594

However the relationship between 04 and 1175 would not be expressed by thesubstemmata Rather the fact that there is no stemmatic link between 04 and1175 at the three instances in question would promote the false impressionthat the two witnesses coincidentally share the same variant

Figure 37 shows the result for 04 and 1175 following the model of Figure34 There is no undirected connection between 04 and 1175 Their commonvariants are explained with the aid of the intermediary node connecting thesources of the common variants with 04 and 1175 and connecting 04 and 1175in the same way As a result of the intermediary node one of the stemmaticcoherencies and thereby ancestors may become superfluous as in this case an-cestor 1739 in the substemma of 1175 If there were no instances which make1739 an ancestor in the substemma of 1175 apart from the 7 instances wherethe witness plays the role described above (04 rarr 1739 rarr 1175) this would becorrect Yet there are a large number

The connection of two witnesses by way of an intermediary node and theincorporation of the node into the relevant substemmata (cf the substemmaof 04 in Figure 40) unquestionably calls for renewed examination and possi-bly fresh optimising of the stemmata It is also possible that an intermediarynode is only needed from the perspective of one of the witnesses it is going toconnect This is the case for attestations containing no further witnesses ge-nealogically connecting the witnesses (in contrast to the example 04 rarr 1739rarr 1175) and having no common ancestors (in contrast to the example 04 larr


03 rarr 1175) but an ancestor for one witness only and if no places are left inwhich neither witness has an ancestor in the same attestation Indeed for 04and 1175 such a place remains95 If there were no such place 04 would have hadan ancestor in each place that agrees with 1175 and should not have needed anintermediary node in contrast to 1175 In this case the edge from this node to04 could be marked in order to indicate this fact

An optimal substemma can also be constructed for an intermediary nodeThe ancestors can be found amongst the ancestors common to the substem-mata of the descendants involved The node representing the agreements of 04and 1175 requires A and 03 as ancestors (cf Figure 37)

As a simpler solution but with a certain loss of information instead ofan intermediary node an undirected edge could be used cf paragraph 9 andFigure 43 For intermediary nodes based on undirected coherencies the samerestrictions apply as for nodes based on prior variants in non-ancestors Asfor the latter the stemmatic representation opens up various possibilities forthe location of contamination in the textual tradition (see paragraph 7) Atany rate we must keep in mind that in the case of undirected coherencies thestability of the textual flow has the value 0 (cf Figure 27)

Circular edges

The problems in this field have not been definitely solved and further researchis necessary Figure 20 showed how the problem may emerge If the variantsof an ancestor were mixed with the variants of the ancestorrsquos ancestor overseveral generations of copying thereby reintroducing older variants again andagain this may result in a descendant so rich in older variants that it be-comes a potential ancestor of a mediate ancestor A circle materialises based oncontamination in multi-stage phases But are there more preconditions thanmulti-stage contamination

Figure 38 refers back to Figure 20 The assumed variants of the 4 witnessesat 8 places of variation are shown in a matrix of 8 lines (in a frame) The vari-ants are in accordance with the requirements of Figure 20 On the right arethe local stemmata presupposed at the 8 places of variation This offers the in-formation how often a witness has a prior variant compared to another Thevalues are listed on the left C has prior variants in two cases where D has therespective posterior variants (cf lines 1 and 6 of the matrix) From the valuesthe predominant textual flows follow C gt D D gt E E gt C96 From the tex-

Gerd Mink

witnesses B C D E

directioncases

variants a b c alsquo aldquo

local stemmata for line

C gt DD gt C

D gt EE gt D

E gt CC gt E

21

32

21

aaaaaaaa

abbbbbbb

babbbcalsquoalsquo

bbaabalsquoaldquoaldquo

B C D E 1ndash5 6 7 8aa a

bb balsquo alsquo

c aldquo

C

B

D

E

Figure 38 Circular edges

tual flows follows the potential ancestor of each of the witnesses and a circlematerialises

The local stemmata at the first 5 places allow a statement in each and everycase whether the variant of a witness is prior or posterior to the variant ofanother witness The local stemmata at places 6 7 and 8 however do not allowsuch a deduction for each pair of variants as not all the variants are directlyconnected If there were only direct connections like in the first stemma thecircle would no longer occur97

The question now arises whether the enrichment with older variants in abranch of the textual tradition is the only cause for a circle or whether otherconditions may lead to circles even in small areas of a stemma From Figure 38it already appeared that not all the local stemmata there belonged to type a rarrb and that this made a circle at least possible

Figure 39 represents a circle of 3 witnesses The circle is supposed to bepart of a stemma In the upper half of Figure 39 it is assumed that there isonly one possibility for a local stemma it leads from variant a to variant b Thematrix of variants in the middle section and the information obtained fromit how often each witness has a prior variant compared to another witness(see directioncases) demonstrates the circular relationship of the 3 witnessesThe local stemma of this simple structure does not allow a circular connectionbetween the witnesses (see above) if all the places in the matrix can only befilled by a or by b Three positions (the small boxes in the matrix) remain emptyin order to obtain the values (cf directioncases) needed for a circle If one ofthe boxes were to be filled by a or b there would be no circle In fact in the


witnesses C D E

direction cases

variants a b c (x) (y)

local stemmata

C gt DD gt C

D gt EE gt D

E gt CC gt E

21

21

21

aab

bb

bbaa

abbbaa

C D E

D ED E

C

(x)aaa

abbb c(y)

bcc

a

b

C gt DD gt C

D gt EE gt D

E gt CC gt E

41(1)

21(2)

31

21(1)

21(2)

21

aab

bb

c

bbaacc

abbbaa

C D E

dir cases A cases B

Figure 39 Circular edges in consequence of lacunae (top) or different types of localstemmata

situation where at each place there is just a variant a and a variant b a circle isonly possible if there are appropriate lacunae in the witnesses

It follows that if there are no lacunae the matrix can only be completedwith variants leading to the same circular edges if there are other types of localstemmata The bottom half of Figure 39 shows a matrix of variants in whichthe boxes (ie the lacunae) of the top matrix have all been replaced with variantc We assume that at the first three places (= lines of the matrix) the first type oflocal stemma evidences the genealogical relationship between the variants butat the last three places the second type of local stemma does There c derivesfrom b The a witnesses have no prior variants compared to the variants of thec witnesses A variant after all is only prior compared to a posterior variantwhich developed from it (cf paragraph 49) c however derives from b notfrom a a has mediate priority at most

On the left under lsquodircases Arsquo the relationship between witnesses is shownfrom the perspective of the priority and posteriority of their variants The val-ues are based on the assumption that only the first and the second types of localstemma are true The values without brackets indicate the number of priorvariants Again a circle results C gt D D gt E E gt C The values in brackets

Gerd Mink

show the number of indirect relationships occurring here if one witness readsa and the other c

If we now assume that the last three lines of the matrix do not conformto the second type of local stemma but to the third the resulting values whichare found under lsquodircases Brsquo correspond to the values in the top part of Figure39 The circle remains The changes reflect the fact that c can no longer betraced back to b The mediate priority of a is maintained (values in brackets) Itproceeds by way of a variant y rendered by witnesses not being part of the circle

If the fourth local stemma is assumed to apply to the three last lines of thematrix again the same values as in the top of Figure 39 ensue As a and c nowderive from a common source variant x rendered by witnesses not being partof the circle no mediate priorities are left at all

It would now be possible to consider the use of the mediate relationshipsin the construction of a stemma But there are serious methodical objectionsto this Rightly only direct relationships between variants in the local stem-mata are used to determine potential ancestors and textual flows To determinethe genealogical coherency between two witnesses their degree of agreementis essential first of all and secondly the relation between the number of theirprior and posterior variants ie only data based on identity of variants or di-rect genealogical relationships between them It is only this direct relationshipwhich makes it at all possible to conduct coherence tests at individual places ofvariation98 and to find local chains of coherencies incorporated into a globalstemma and subsets of the global stemma represented in diagrams which showthe textual flow within an attestation99

It is clear that circles can materialise as a result of different circumstancesThe complexity is much higher in reality than in the examples in Figures 38 and39 ndash more witnesses more variants therefore more complex local stemmatathe several types of which (see Figure 39 bottom half) get mixed up and that iswithout taking into account the lacunae In this situation circles may emergebut need not Under the conditions of Figures 38 and 39 a circle does not mate-rialise if variants within the same line of the matrix change places for the samelocal stemmata This is of course due to the very small difference between thenumber of prior and posterior variants of one witness in the circle compared toanother In fact the smallest difference between pairs of witnesses concerneddetermines the stability of a circle and the differences are often very small in-deed The variants have to be distributed among witnesses in a very particularway for a circle to materialise How frequently this may happen also dependson the frequency of non-direct relationships between variants and the degreeof fragmentation of witnesses100


A A1243 1175 025 0403 1739 031852

1243 173904

A 04 1175

04 1175

A

185203

1739

025 1175

1243

04

Figure 40 Top substemmata bottom their combination with a circle

Badly fragmented witnesses certainly present a very special risk The ge-nealogical relationship between a fragmentary witness and another witness isdetermined on the basis of its limited text supply As the distribution of priorand posterior variants in a text can be very different it is possible that an en-tirely different picture would have resulted from the complete witness if it hadbeen preserved

Witness 04 in which roughly the first 3 out of 5 chapters of James havebeen preserved as a continuous text is a good example how the inclusion ofsuch texts causes the emergence of several circles Figure 17 presents the twomost probable genealogical relations in the top area of a global stemma Thistree suggests that in a global stemma 04 will be ancestor of 1739 1243 ancestorof 04 Yet in the substemma of 1243 we find 1739 as an ancestor of 1243 (Fig-ures 28 and 40) Figure 40 represents the optimal substemmata of 04 1243 and1739 At the bottom we see their combination containing the connections inthe substemmata The connection of the intermediary node with 1175 as wellis implied The circle which was to be expected on the basis of Figures 17 and28 does indeed materialise (04ndash1739ndash1243ndash04 cf the bold edges) 04 is onlyavailable at 482 out of 761 places of variation whereas the other witnesses canbe used almost in their entirety If only the first three chapters of James insteadof five are taken as a basis ndash ie roughly the quantity of text preserved in 04 ndash thetextual flow between 1739 and 1243 changes direction101 and the circle disap-pears102 As Figure 36 shows 1739 is also an ancestor in the optimal substemma

Gerd Mink

circular edges

B

D

C

intermediary node

D B

B

C

D

Figure 41 Simplified model for the resolution of circular edges

of 1175 Here again a circle will materialise in which 04 participates It includesthe other circle (04ndash1739ndash1175ndash1243ndash04) The direction of the textual flow be-tween 1175 and 1739 would change likewise for the first three chapters103 Theproblematic side effects of including fragmentary witnesses are evident Theyare aggravated by the fact that the stability of the textual flow between both1175 and 1243 and 1175 and 1739 is very poor104 Of course 04 is too large andtoo important in the textual tradition not to be considered The example waschosen deliberately also in order to illustrate how important it is to have accessto values and facts behind a stemmatic graph

How is such a circular structure to be pictured Figure 41 (left hand dia-gram) shows a simple model of circular edges The circle can be resolved byintroducing an intermediary node containing the variants of D which are thebasis of agreeing and secondary variants in B (Figure 41 right hand diagram)But the problem is that the model of circular edges can be rotated Conse-quently we have to treat all the witnesses in the same way The result is Figure42 (bottom)

The left hand stemma in Figure 43 shows the combination of the sub-stemmata of 04 1243 and 1739 following the model of Figure 42 The circle04-1739-1243-04 is now resolved In addition the undirected coherency be-tween 04 and 1175 is respected (cf Figures 36 and 37) The number of edgesmay be smaller if the edges from A to 04 1243 or 1739 are ldquoemptyrdquo (nottransporting variants for all the variants pass the relevant intermediary node)

In Figure 43 the right hand stemma can be interpreted as representing thesame facts as the left hand one with its intermediary nodes There is a circleconnecting 1739 1243 and 04 1175 and 04 are connected by an undirectededge The logical meaning of the right hand diagram is the same as that ofthe left hand diagram the disadvantage is only that there are no intermediarynodes with their special contents and ancestors On the other hand the cir-


circular edges

intermediary nodes

C D D B B C

B

D

C

B

C

D

Figure 42 Complete model for the resolution of circular edges

04 1175

1243 04 04 1739 1739 1243

03A

18521175

025

04 1243 1739

A

03 025 1175

173912431852

04

Figure 43 Stemmata with resolved undirected and circular edges (left) and unresolvedundirected and circular edges (right)

cles in the left hand diagram are more difficult to discern Yet the right handstemma provides less information However full information can be found inneither of the two stemmata A stemma like the one on the left is only usefulfor specific purposes and even then one has to know the way that each of thevariants follows through the stemma

Further problems arise when more circles materialise in the vicinity of awitness as is indeed the case for 04 (see above) They have not been taken intoaccount in Figure 43 The number of intermediary nodes increases Yet the newnodes and edges may be simplified by combining nodes which are not contra-dictory This will have consequences for the definition of the simplicity of a

Gerd Mink

(sub)stemma When intermediary nodes depending on circles start accumu-lating the limits of depiction are reached fairly soon On the other hand thecircles from the underlying data are often easily made out

In reality circular genealogical relationships are impossible However astemma does not represent historical reality but structures obtained from theavailable data Yet concerning the data the relationships of fragments are es-sentially complicated by the fact that witnesses with different text lengths canonly be compared to a certain extent In any event the result must be in-terpreted with caution both if the character of the whole witness is judgedfrom that of the fragment or if the complete witness is only assessed in thesection which is covered by the fragment Thereby a circle depending on frag-mentation is an artefact Nevertheless according to the rules the substemmatainvolved in the circle remain true In fact in the case of the substemma of 1739(Figure 40) a simpler substemma to explain the variants of 1739 at all placesof variation cannot be found Again the background data are important forestimating to what extent artefacts contribute to the result

Matters are different with circles depending exclusively on the model ofFigure 20 Those circles correspond to a circular development of variants If wedo not know the actual chronological order of the different textual states wecan only detect a circle Rules for the construction of a global stemma whichprevents the emergence of circles will have to result from the respective modelof the textual tradition I do not consider this possible If the data suggest acircular development it should be represented as circular

The construction of a global stemma

The global stemma is the superset comprising all the optimal substemmata assubsets including the intermediary nodes incorporated into them or the cor-responding connections105 The stemmatic coherencies between descendantand ancestors ensue from the optimisation of the substemmata Even if in-termediary nodes are not hyparchetypes and sometimes comprise only onevariant their role is technically comparable to that of the main nodes insofar asthey occupy the same position as ancestors if incorporated into a substemmaThereby the stemmatic connections on the level of the substemmata are es-tablished They contain all the necessary connections and they contain onlythose Each ancestor (apart from A) in a substemma will be a descendantin another substemma Each descendant will either be an ancestor in one or


more substemmata or if not an ancestor represent a terminal node in theglobal stemma

The connections in a global stemma are thus absolutely determined by theconnections in the substemmata Consequently only one single global stemmais possible on the basis of one set of optimised substemmata (with any inter-mediary nodes) Optimising on the level of the global stemma is impossible

Every change in the substemmata leads to a change in the global stemmathe change however being absolutely limited to the range of the substemmatainvolved If there are alternative substemmata of the same quality there arealso alternative global stemmata of equal quality Their numbers may easilymount up If there is another alternative substemma for a given descendantthe number of possible global stemmata increases only by 1 but it doubleswith every alternative substemma for further descendants

In a very dense textual tradition with texts agreeing almost completely thecase will repeatedly occur that an ancestor in a substemma can be replaced byanother without affecting the quality of the substemma The same can happenwith fragments mainly rendering places of variation at which only few wit-nesses depart from the mainstream of the tradition Consequently they willhave numerous close relatives agreeing at the same high degree In addition itmust not be assumed that the character of the lost part of the witnesses cor-responds sufficiently to the preserved part if there is a greatly limited or verysmall number of places of variation Nevertheless it is not necessary to refrainfrom the use of fragments in stemmata at all but it is essential to know theirprecise data It is not possible to refrain from using a witness as important as04 (see above) in the construction of the stemma although two chapters ofJames are missing Yet 04 belongs to an area of the textual tradition in whichnearly all witnesses are lost The smaller fragments are very unlikely to appearas ancestors anyway as they explain too few variants in a descendant

It would not be very worthwhile to construct all possible stemmata Asoptimising takes place at the level of the substemmata and not in the course ofthe creation of the global stemma connections in the global stemma would notchange if it is incomplete It is therefore possible to construct parts of a globalstemma and leave out those areas where there are alternatives The remainingconnections will still remain true The informational value of an area in thestemma with many equally good alternatives is meagre at any rate Yet on theother hand it may be desirable to see what the development in a specific areais like cum grano salis This is possible if all the alternatives lead to very similarresults in any case It would then be sufficient to mark the connections for

Gerd Mink

AA 1175 02503A

A

1739

17391739

1243117503

1175

1243

5

8

1

025

03

4

6

7

23

1 973 19 02 925 55 03 926 54 04 908 47 125 904 31 306 920 29 287 896 35 328 915 29 26

Figure 44 Part of a global stemma compiled from substemmata Left qualification ofedges agreements in number of prior and posterior variants of the ancestor

which there are alternatives This is particularly useful where in a stemma areaswithout alternatives are separated from each other by an area with alternatives

As the completeness of a stemma is no prerequisite for the accuracy of theconnections it is of course possible to extract any section of the global stemmaeg the top of the stemma the stemma of a group without constructing thecomplete stemma Figure 44 demonstrates that there is no other possibility forconstructing this section of the global stemma on the basis of the substemmataIf any ancestors are available here then all of them are available If there are noancestors available as with 025 and 1739 they are situated outside the sectionof the global stemma If the global stemma were complete the section of thestemma in Figure 44 would not change106

Conclusion

It is very clear that the genealogical structure of a contaminated tradition can-not be revealed if evaluations of variants and automatic procedures are notalternated permanently Accordingly three levels of falsifiability have to be ac-counted for the level of facts the level of evaluation of readings and the level ofprocedures and their rules Furthermore we have to be aware of the undesirableartefacts caused by method point of view and way of visualisation107


The user of a stemma has some very simple questions Why is there anedge connecting x and y and what is the reason for its direction How reliableis the connection He needs more information than stemmata usually provideThe stemmatic graph of a contaminated tradition is only one possible surfaceof a very complex situation The facts behind the stemmatic connections areextremely important and must be accessible for assessment The genealogicalevaluation of the facts is the basis for their graphic visualisation if the con-nections rendered in it are to represent genealogical directions If a stemmaticgraph is the visualisation of a hypothesis and if it is a surface showing only cer-tain aspects of the material different methodological approaches will be usefulin order to recover the patterns within the preserved material It is all the moreimportant then that the user of stemmatic representations should know howto read them and understand which interpretations are justified and which arenot with respect to the theoretical background and all the procedures whichcontribute to finding the stemma

Notes

This leads to a more complex concept of a stemma cf paragraphs 4 10 and 11

The word address in even numbers marks the word in the text established in the ECMwhere the variation starts and if not identical the word with which the variation ends theaddress uses odd numbers to mark spaces between words which is important for additions

The last word before the place of variation ends in the same letter or letters as the omittedvariant

The witnesses are 467 643 and 1848

631

The degree of relationship is determined by the degree of agreement Cf for this para-graph 411 (pre-genealogical coherence)

According to the pre-genealogical coherencies Nearest relative here is 424 with only859 correspondence Also when the genealogical coherence (cf paragraph 411) is takeninto consideration the closest potential ancestor (again 424) falls within the d attestation

If variant c is assumed to stem from variant a only the link between d and c would dis-appear from the stemma The relationship between the a witnesses and the c witness wouldnot change the relationship between the d witnesses and the c witness would become onlyslightly remoter (some 02 percentage points) Actually a choice has to be made between thetwo possibilities as further processing of alternative local stemmata is not possible

Cf Figure 36

Gerd Mink

For P74 the degrees of agreement start off very modestly 896 with 02 The relativeremoteness of the next relatives is caused by a larger number of peculiar variants attestedonly in P74 or shared only coincidentally with more remote witnesses

They are potential ancestors cf paragraph 410 In this context the dating of themanuscripts is not important cf paragraph 44

Cf Aland (1987 1991 1993 1998ndash1999)

Cf Aland Aland Mink and Wachtel (1997 2000 2003)

The numerous orthographic modifications and 592 faulty readings which almost al-ways occur in conjunction with a correct variant were not counted as genuine variants cfparagraph 43 The places of variation include also 59 places where only variants attested bylectionaries Church Fathers or (mostly) early versions deviate from the mainstream of thetradition In the case of Church Fathers and versions variants as a matter of fact were onlyrecorded if they certainly or presumably rest upon a Greek exemplar in a now lost Greektext witness

So not counting variants by correctors in marginal text and commentaries

Not including versional variants which are not assumed to rest upon a Greek exemplar

For the types cf Salemans (200024ndash25)

Again not including corrections marginal text and commentaries as they do not resp-resent continuous text

To be exact 704 attestations The dots in the diagram correspond with the followingvalues (from left to right) 969 875 832 804 794 782 766 753 734 722 704 676 629533 154

To be exact 124 places The dots in the diagram correspond with following values (fromleft to right) 59 418 124 71 37 25 6 6 6 3 1 1 1

To be exact 158 places The dots in the diagram correspond with following values (fromleft to right) 702 283 158 87 50 25 19 13 7 4 3 3 3 2 2 2 2 2 2 1

Although the 59 places do not give any information about the genealogical relation-ships between the 164 Greek manuscripts involved they are nevertheless included in thegenealogically relevant data as they are needed for the genealogical assessment of the textsof the lectonaries Church Fathers and in particular the Greek exemplars of the early ver-sions The versions can then be subjected to the same procedures as the Greek witnesses CfSpencer Wachtel and Howe (2002)

Cf the model in paragraph 4

The dark part of the bars in Figure 6 indicates manuscripts of uncertain date

Cf witness vs manuscript paragraph 44

Cf also the preliminary study Mink (1993) and Mink (2000) Mink (1993) representsthe state of that time and could only be based on the material in the test passages (seeparagraph 1 25 instead of 761 places)

Cf eg the case of 1243 paragraph 6

Readings here meaning variants


Mink (200052)

Cf Mink (1993489)

An author does not necessarily always do justice to his intentions He may formulatea text which is barely understandable or which requires a great intellectual effort on thepart of the reader while to the author himself its meaning is absolutely clear A copyistis more likely to produce variants here in order to replace a lectio difficilior according tohis understanding at any rate He might also as he does not understand the text himselfproduce a variant which makes no sense and thus generate a lectio difficilior as compared tothe original

Amongst others 01 03 025 1175 1739T and 1852

If εὑρεθήσονται is accepted without the negation exegetists and translators attempt toassign meanings to the word which are not recorded anywhere else Many also have re-alised the absurdity of this variant and proposed a multitude of imaginative conjectures cfMetzger (1971706)

Given by 04 and therefore certainly an old variant

This variant is given by the majority of the witnesses Some of these also point to the ageof the text form 02 33 81 307 2298 2344 Further variants (κατακαήσονται καήσονται)have the same meaning

lsquoThey will not be foundrsquo is actually attested in non-Greek text traditions It is regularlyfound in the Sahidic version in the witness of the Coptic dialect V and also in some of thewitnesses of a Syriac version (Philoxeniana) Either the original variant truly survived here ndashwhich is at any rate quite possible for the Sahidic ndash or this variant is also already a conjectureAn old conjecture is already to be found in P72 εὑρθήσονται λυόμενα lsquothey will be found asdissolved onesrsquo

Aland Aland Mink and Wachtel (199716)

Ibid

The distinction between manuscript and text is essential Cf Mink (200052) ldquoThemanuscript having a palaeographical date gives us only the terminus ante quem non of thetextrdquo there is no earlier attestation of the text It is also possible to speak of the terminus postquem non the text in this form came into being at the very latest at this point in time

For the complete issue cf paragraph 10

For the special case of an undirected connection cf paragraph 8

Cf Mink (200053) on genealogical coherence ldquoIn a system whose constituents are notindependent of one another those constituents must cohere in a definable manner if we areto understand that system Coherence within a group of witnesses means that the membersof the group are connected by immediate genealogical relationsrdquo The latter is true if we knowthe ways of transmission Normally they are not known and only part of the witnesses havebeen preserved In such a case coherence is represented by hypothetical coherencies withingroups and the genealogical relations are immediate only because a part of the tradition islost

Cf P74 see note 10

Gerd Mink

See for this issue also paragraph 6

For perfect and imperfect coherence at places of variation see paragraph 53

Translated from Mink (2002) Expansive treatment in Mink (200341ndash46)

For places of variation see 42

The present study uses the data available in November 2003 For limitations see Note56

If very similar variants obviously emerge repeatedly from each other or respectively ifthere is an imperfect coherence within the attestation

For the special problem of undirected genealogical coherencies cf paragraphs 411 and8

If the field is empty no predominant direction can be determined

The percentage value is the relevant one cf paragraph 4

This is in keeping with the model of the textual tradition cf paragraph 411 The degreeof agreement does not conclusively determine the probability it only influences it for itdoes not allow to discern whether the textual flow between a potential ancestor x and adescendant z is included in a textual flow between a potential ancestor y and the descendantz In such a case x would not be ancestor in an optimal substemma Cf paragraphs 48 and 6

Smaller fragments must be ignored Their agreement percentages are often extremelyhigh as they preserve areas in which only few witnesses deviate from the mainstream This istrue for the major part of the text of James (cf paragraph 2) It is impossible to say how sucha text would look like if it had been preserved in complete form Even a partially fragmentedtext like 04 which nevertheless renders more than half of the text in contiuous form andcannot be left out of consideration because of its genealogical importance presents somedifficulties see paragraph 9

Further examples of the evaluation of similar lists in Mink (200353ndash56)

Cf h1 and h2 in Figure 7 In the most cases a certain number of erroneous assessmentsof coherencies in the first step does not affect the overall results very much If we do not splita variant logically into two or more variants despite the imperfect coherency the numberof agreements of each of the pairs of witnesses in the attestation will increase by 1 due tothis variant If we do split the variant only the number of agreements of pairs in the samecoherency chain will increase by 1 But if the number of agreeing variants of a pair is largethere is a high coherency of those witnesses even if some local coherencies are not realisedand if the number is small coherence remains low even if some local coherencies are as-cribed erroneously ndash As regards James the coherencies within attestations will need a finalrevision based on the last genealogical coherencies and particular textual flows Until nowattestations have been split only in cases of obvious imperfect coherencies A revision is ex-pected to result in splitting more attestations and strengthen the tendencies which can beobserved now In some cases of poor stability of textual flow ancestor-descendant relationsmay change

Under kon the numbers of agreements are listed


The lists for two quite differing witnesses have been taken as an example here 1243 hasfew potential ancestors which indicates a position in the upper regions of a global stemmaThe list for 2412 has been shortened and in reality counted 138 lines From the top of the listit is obvious that the well-known witness 614 is the most closely related potential ancestorof 2412 The next lines document that the well-known HK group is to be counted amongthe potential ancestors

Witness A the root of the tree is hypothetical cf paragraph 41

Cf Aland Aland Mink and Wachtel (200021ndash22) The same procedure was notsuitable for 1 John because the witnesses are on the average closer to the initial text A and toeach other than in James or the Letters of Peter Therefore it was necessary to start at thoseplaces of variation that can be assessed easily In this way determining potential ancestorsand predominant textual flows could be based on 621 out of 761 places of variation Theresults are preliminary of course Yet they proved to be plausible and sufficient for a firstevaluation of genealogical coherence and construction of textual flow diagrams Cf AlandAland Mink and Wachtel (200329)

Cf for Figure 14 Mink (200360ndash62) Figure 11 is a pre-form of Figure 14 here

ldquoYet if he suffers as a Christian he should not be ashamed but he should glorify God inthis part() in this case() on this behalf ()rdquo The word μέρος has many meanings

ldquoYet if he suffers as a Christian he should not be ashamed but he should glorify God inthis namerdquo

ldquoIf you are reviled in the name of Christ you are blessedrdquo

Cf the textual flow diagram representing perfect coherence in Mink ibid Figure 10

A question mark at an edge means that the connection is doubtful and appears else-where in the diagram with another question mark

The level of probability depends on the position of a potential ancestor in a list withdescending rates of agreement cf Figures 9 21 23 35 and paragraph 53

A few witnesses are lacking in Figure 16 Fragments P23 and 0166 contain very littletext and present few places of variation in these passages Thus no directions of predomi-nant textual flow emerge Fragments P54 0173 and 1846 contain very little text too Herehowever there are too many witnesses with equally strong textual flows directed towardsthese fragments With 2718S the predominant textual flow from two directions is equallystrong and has been drawn with interrupted lines A pre-form of this graph was earlier madeavailable for a lecture of Barbara Aland which was printed in the Korean Journal of BiblicalResearch (Aland 2000)

A section of Figure 17 including the textual flows which are third in probability is tobe found in Mink (200357 Note 17) Figures 16 and 17 do not include intermediary nodes(cf paragraph 47) Usually the first 3 to 5 levels of probability are important when we arelooking for the sources of the textual state of a descendant In cases of variants which areconnective because of their character and if the difference between the levels of probability israther small lower levels may be used to give a better choice Cf the method for constructingsubstemmata paragraph 6 If the differences between the levels are small the number ofwitnesses which do not contribute new variants in a substemma increases

Gerd Mink

Where no initial reading could be hypothesized A is treated as if it has a lacuna

Those witnesses with good genealogical coherencies with A are of course of prime im-portance since the variants in those witnesses could be expected to be part of the initial textsee Aland Aland Mink and Wachtel (200021ndash22)

Cf paragraph 410

Otherwise one would have to assume that 19 variants (cf Figure 21 xausy field) emergedon the way from A to 03 without any intermediate stages or without any contamination inany of the non-preserved intermediary witnesses

03 actually agrees with 1243 in more places than 025 does but the degree of kinshipis smaller This degree is however decisive because of the respective number of places ofvariation where a pair of witnesses can be compared A certain degree of fragmentation isnormal 1243 can be compared with 03 in 743 places with 025 in 708 places Only strongerfragmentation leads to less significant values resulting from the comparison and moreoverlessens the chance that the witness will become ancestor in substemmata as it explains toofew places of the descendants

If the tradition is wide-spread it may happen in a small number of cases that a descen-dant has a prior variant compared to all potential ancestors Cf paragraph 7

For the conditions cf paragraph 411

For 025 an additional contribution of 7ndash8 is indicated This is based on the fact thatat one place it is not clear which variant should be the reading of the initial text A and025 makes an additional contribution only if it does not read the same variant as A (James51026-32)

The common omission of the sentencersquos initial conjunction ὅτι in James 1232 wasregarded as not coincidental The same applies to the addition of ἔργων in James 31733

James 2350-56 (ὑπό not ἐπί) 21318 (κατακαυχαται not ndashσθε or ndashσθω)

James 21516 variant b aorist participle λιπόμενοι vs present participle λειπόμενοιthe pronunciation of the two words was the same There is another place where a minimalcontribution could be considered James 52020-28 1243 has σώσει τὴν ψυχὴν ἐκ θανάτουI derive this from σώσει ψυχὴν ἐκ θανάτου the variant of 1175 and of no other ancestorBut although evolvement of the variant of 1243 from σώσει ψυχὴν αὐτου ἐκ θανάτου is lessprobable it is not completely impossible There are two more witnesses for the variant of1243 049 and 2492 1243 is the closest potential ancestor of 2492 049 however is not partof the environment of 1243 at all (only 87 agreement) and only coincidentally reads thesame variant which goes back to the same source variant

James 2142-10 variant c omission of an article which is not obligatory 21516 variantb aorist participle λιπόμενοι vs present participle λειπόμενοι ndash the difference is just onecharacter the pronunciation of the two words was the same 2198-14 variant e reversal ofwords 2262-4 variant b omission of γάρ lsquoforrsquo

In two additional places 01 could explain the variants of 1243 (cf Figure 24) In James596-10 1243 reads variant c κατrsquo ἀλλήλων ἀδελφοί 01 has the same with slightly differ-ent orthography κατὰ ἀλλήλων ἀδελφοί Compared to variant a of the ancestors A 03 025(ἀδελφοί κατrsquo ἀλλήλων) this is a reversal of word order which could have independently


emerged several times In James 51240 (variant d) 01 (hand 1) and 1243 like several wit-nesses change the James text (variant a the reading of the ancestors of 1243) into the muchbetter known parallel in Matthew 537 a variant which is not connective

Cf Aland Aland Mink and Wachtel (199728)

Cf Blass and Debrunner (2001) sect442 especially Note 14

Cf Aland Aland Mink and Wachtel (200022 Note 4)

The condition is that d has the initial text If a is regarded as the initial text the problemis the same there would be no potential ancestor of 03 with variant d if one derives c from d

However an optimal substemma in the case of P23 cannot be established as there is alarge number of equally good hypotheses

01 has variant b in James 11738 03 has variant b in James 11144 P23 in both caseshas variant a

The graph might be simplified as a result by attempting to link 01 or 03 directly toP23 even though there is no direction of textual flow Such a decision might however havefar-reaching consequences In this case it would not only be decisive for an immediate anddirected stemmatic connection whether a directed textual flow can be detected between twowitnesses but also whether the variants which are not in accordance with the direction of thetextual flow are still needed for the construction of a stemma It is better to postpone suchsimplifications until the consequences for the basics of the method and the construction ofthe global stemma are clear

04 is rather badly fragmented (482 out of 761 places of variantion are extant) cf para-graph 9 so it is impossible to conjecture what the result would be if the complete textwould have survived In the context of the principles of this procedure however this is notimportant

James 11826-28 (variant b) 12012-14 (variant b) 12214-16 (variant b) 1262-4(variant b) 21320 (variant b) 22627 (variant b) 3432-42 (variant b)

James 2142-10 (variant c) 21642-46 (variant b)

James 2153 (variant c) 21824-34 (variant b) 2198-14 (variant e)

At 2 out of these 3 places 1243 an ancestor in the substemma of 04 even reads the samevariant as 04 James 21824-34 (variant b) 2198-14 (variant e) But 1243 is not an ancestorof 1175

James 2153 variant c

Witness B is not presented here as it is not part of the circle B has only prior variants

In James I have not yet found a case in point where purely as the result of the enrich-ment of a branch of the tradition with older variants such a circle actually materialises inthe construction of a group of substemmata I am still expecting this to happen as the prob-lem arises in diagrams representing textual flows like in Figure 17 but with a larger rangeand taking into account potential ancestors with lower levels of probability

Cf paragraph 53

If the local and the global textual flows do not correspond in this way the global stemmais falsified Cf paragraph 46

Gerd Mink

In James there are 1723 genealogical relationships between variants 209 of these per-mit indirect relationships as they determine source variants of source variants The numberof witnesses concerned by this is relatively small however A pair of witnesses can on aver-age be compared at 641 places In only 45 instances on average this does not result in directgenealogical relationships between variants of a pair 697 places on average are comparableif fragments are excluded which are only comparable at 50 places at most In that case 49places on average do not produce direct genealogical relationships

The direction is determined by the values found in Figure 23 in the xausy and yausxfields For 1739 the ratio of the values is 3532 This implies a predominant textual flowfrom 1739 towards 1243 If only Chapters 1ndash3 are taken as a basis the ratio is 1419 so thetextual flow changes direction

This does not mean that 1243 becomes an ancestor in an optimal substemma for Chap-ters 1ndash3 as the textual flow from 1243 towards 1739 is included in the textual flow from Atowards 1739 and from 03 towards 1739

Instead of 3130 (in Figure 35) the ratio is now 1821

Cf for the concept lsquostability of textual flowrsquo also paragraph 6

Cf paragraphs 8 and 9 and Figure 43

A section in a stemmatic graph rather creates the possibility of adding information tothe edges The values displayed are typical for the top of the stemma the percentage of agree-ments is not particularly high The unusual role of 03 becomes clearer through the figuresThe number of posterior variants of the ancestors compared to the prior ones indicates therelative quality of the edges

For instance in the method discussed the fact that there is one initial text may bean undesired artefact in some other traditions Another (negative) artefact is the fact thata relationship in which as a result of contamination an immediate descendant acquiredmore prior variants than one of the immediate ancestors would not be detected In this casedescendant and ancestor may even change places

References

Aland B (2000) ldquoDie editio critica maior des Neuen Testaments Ihre Anlage ihre Aufgabedie neu entwickelten Methoden der Textkritikrdquo Journal of Biblical Text Research 7 7ndash23

Aland B K Aland G Mink amp K Wachtel (Eds) (1997) Novum Testamentum GraecumEditio Critica Maior ed by Institute for New Testament Textual Research Vol IVCatholic Letters Installment 1 James Stuttgart Deutsche Bibelgesellschaft

mdashmdash (2000) Novum Testamentum Graecum Editio Critica Maior ed by Institute for NewTestament Textual Research Vol IV Catholic Letters Installment 2 The Letters of PeterStuttgart Deutsche Bibelgesellschaft

mdashmdash (2003) Novum Testamentum Graecum Editio Critica Maior ed by Institute for NewTestament Textual Research Vol IV Catholic Letters Installment 3 The First Letter ofJohn Stuttgart Deutsche Bibelgesellschaft


Aland K et al (Eds) (1987) Text und Textwert der griechischen Handschriften des NeuenTestaments Bd 1 Die Katholischen Briefe ANTT 9ndash11 Berlin New York De Gruyter

mdashmdash (1991) Text und Textwert der griechischen Handschriften des Neuen Testaments Bd 2Die Paulinischen Briefe ANTT 16ndash19 Berlin New York De Gruyter

mdashmdash (1993) Text und Textwert der griechischen Handschriften des Neuen Testaments Bd 3Die Apostelgeschichte ANTT 20ndash21 Berlin New York De Gruyter

mdashmdash (1998ndash1999) Text und Textwert der griechischen Handschriften des Neuen TestamentsBd 4 Die synoptischen Evangelien ANTT 26 28 30 Berlin New York De Gruyter

Blass F amp A Debrunner (2001) Grammatik des neutestamentlichen Griechisch Bearbeitetvon Friedrich Rehkopf (18th edition) Goumlttingen Vandenhoek amp Ruprecht

Metzger B M (1971) A Textual Commentary on the Greek New Testament New YorkUnited Bible Societies

Mink G (1993) ldquoEine umfassende Genealogie der neutestamentlichen UumlberlieferungldquoNew Testament Studies 39 481ndash499

mdashmdash (2000) ldquoEditing and Genealogical Studies the New Testamentrdquo Literary and LinguisticComputing 15 51ndash56

mdashmdash (2002) ldquoKohaumlrenzbasierte Genealogische Methode ndash Worum geht esrdquo lthttpwwwuni-muensterdeNTTextforschungGenealogische_Methodehtmlgt

mdashmdash (2003) ldquoWas veraumlndert sich in der Textkritik durch die Beachtung genealogischerKohaumlrenzrdquo In W Weren amp D-A Koch (Eds) New Developments in Textual CriticismNew Testament Early-Christian and Jewish Literature (pp 39ndash68) STAR 8 Assen Royalvan Gorcum

Salemans B J P (2000) Building Stemmata with the Computer in a Cladistic Neo-Lachmannian Way Nijmegen Nijmegen University Press

Spencer M K Wachtel amp C J Howe (2002) ldquoThe Greek Vorlage of the Syra HarclensisA Comparative Study on Method in Exploring Textual Genealogyrdquo TC A Journal ofBiblical Textual Criticism 7 lthttppurlorgTCvol07SWH2002gt

Kinds of variants in the manuscript traditionof the Greek New Testament

Klaus WachtelWestfaumllische Wilhelms-Universitaumlt Muumlnster

Introduction

One problem which editors of the Greek New Testament have to face is the largenumber of manuscripts which transmitted the text from antiquity to the age ofprinting For the first installment of the Muumlnster Editio Critica Maior (ECM)1

we had to deal with a total of 553 complete or fragmentary manuscripts Forthe Gospels this number is even three to four times larger A solution to thisproblem is offered by an outstanding feature of rich manuscript traditions ofantique texts in general most of the manuscripts are very much alike

Before we decided which manuscripts to collate in full for the ECM all ofthem had been tested by probe collations in 98 units of variation in the sevenCatholic Letters2 We could exclude more than two thirds of the manuscriptsavailable because with few exceptions they witness to the late medieval Byzan-tine text the Greek Vulgate or Koine The manuscript basis of the ECM as wellas that of Gerd Minkrsquos study in the present volume is a selection of 164 com-plete or fragmentary manuscripts The Byzantine text itself is represented in theedition by seven manuscripts which share the majority readings at the 98 testpassages with a rate of nearly or actually 100 percent3 I am going to deal firstwith the variants which distinguish the late Byzantine from earlier textformsand secondly with variations which occur within the Byzantine tradition Thefirst part aims at a revised assessment of the late Byzantine textform the secondat a better documentation of the Byzantine textform itself one which wouldnot simply have to rely on an admittedly arbitrary rate of agreements with themajority text which we set at 90 for the Catholic Letters

Klaus Wachtel

It may be useful to define the terms lsquoMajority Textrsquo and lsquoByzantine Textrsquo atthis point lsquoMajority Textrsquo is a merely quantitative term while lsquoByzantine Textrsquois a historical and text-critical term The Majority Text consists of the majorityreadings and passages that have been transmitted without substantial varia-tion A majority reading is a variant attested to by the majority of extant Greekmanuscripts As a rule such a reading is attested to by the oldest and best aswell as by the largest number of medieval manuscripts Normally there is lit-tle reason to doubt that it is also the original reading Yet the Byzantine Textis characterised by those majority readings which differ from the supposedlyoriginal text and additionally by those readings which result from a division ofthe main stream into two and sometimes three branches The branching inthe main stream of the manuscript tradition allows the Byzantine and Major-ity texts to be clearly distinguished Moreover it allows us to understand whyfor example two manuscripts that agree with the Majority Text at a rate ofabout 90 can still differ from each other at 20 of the variants in questionIt should be possible to make use of these differences to come to grips with theproblem of an adequate representation of the so-called Byzantine subgroups Ishall return to this point later But for now some remarks about the features ofthe main stream are in order

The late Byzantine text of the New Testament4

Certain kinds of variants are regarded as typical of the Byzantine text In thefirst place there are those variants which enhance the pragmatism of the textInsertions of particles and pronouns which join sentences or improve their in-ner structure occur very often They make it easier to read the text and appearto be meant to avoid any misunderstanding Forms word-order and the word-ing itself are changed to the same effect Although there are cases of Byzantineomissions this textform is considerably longer than older textforms In manyplaces it seems that a striving for clarity and completeness motivated the riseand dissemination of Byzantine readings Moreover we find many harmonisa-tions with the closer or wider context The wording of the synoptic Gospels isassimilated and quotations from the Old or New Testaments are rendered moreprecisely Nevertheless although there is abundant variation it is rare for suchinterpolations to change the meaning of the text considerably As a rule theyonly emphasise what the text says in older text forms as well

Are these variants intentional changes Might they even be regarded astraces of an overall revision of the text in the third or the beginning of the

Kinds of variants in the Greek New Testament

fourth centuries This stance was taken by eminent New Testament scholars inthe past and is still held by some5 I think we can derive persuasive argumentsagainst this theory from the nature of the variants found in the Byzantine textas compared with older text-forms

First the changes are inconsistent There are many places where there is nosmoothing to improve the coherence of the text I am thinking of some difficultpassages in Paul for example On the other hand one sometimes finds an ob-viously intentional change close to a very similar instance which has been leftuntouched This may be demonstrated by the following example In the secondchapter of Luke the infant Jesus is hailed as the coming salvation by Simeon aman inspired by the Spirit According to the oldest and best manuscripts ldquohisfather and mother marvelled at what was said about himrdquo (Lk 233) The major-ity of the Greek as well as Latin Syriac and Coptic manuscripts have ldquoJosephrdquoinstead of ldquohis fatherrdquo For obvious reasons one may think ndash wasnrsquot Jesus bornof a virgin Did he not say that his father is in heaven Might this variant not beclassified as an ldquoorthodox corruption of scripturerdquo a term used by Bart Ehrmanin his celebrated study of the same title (Ehrman 1993) Similarly it is said inLk 243 that Jesusrsquo parents did not know that he was with the teachers in thetemple whereas the majority of witnesses read that ldquoJoseph and his motherrdquodid not know where he was But between these passages there is another whereJoseph and Mary are called ldquohis parentsrdquo (Lk 241) almost unanimously andat 248 Mary says to Jesus according to all but a few witnesses ldquoYour fatherand I have been looking for you anxiouslyrdquo This inconsistency which can beobserved in many more instances excludes orthodoxy as the conscious motivebehind the variants that distinguish the Byzantine from older text-forms

If we study the kinds of variants found in our oldest and best manuscriptswe will observe that the characteristics of supposedly typically Byzantinereadings occur in these too and moreover in each and every manuscriptThis means that such readings are not typical of the Byzantine text but ofmanuscript traditions in general A scribe occasionally does not write whathe reads in the exemplar but what he understands the text to mean insteadHence we frequently find what may be called scilicet variants They normallyresult in an interchange of synonyms The replacement of ldquohis fatherrdquo withldquoJosephrdquo in the above example is a variant of this kind too The scribe may verywell have been subliminally puzzled by the wording of the text he was copy-ing and his interpretation crept into the text The oldest Greek manuscriptwitnessing to this variant is from the 5th century but there are many clearlysecondary majority readings found in the earliest witnesses Defenders of thegeneral originality of the Majority Text have tried to strengthen their case by

Klaus Wachtel

listing single majority readings attested by early papyri However these read-ings demonstrate neither the early provenance of the Majority Text nor itsoriginality Rather they demonstrate only the early genesis of many of the ma-jority readings and the susceptibility of scribes to what Eduard Schwartz calledldquohalf-conscious trivialisationrdquo6

Yet on the other hand there are many atypical Byzantine readings ie ma-jority readings which by no means smooth out the text but instead renderit more difficult or seemingly distorted by obvious scribal errors A majorityreading from the letter of James serves as an example Near the end of the letter(413f) the author exhorts those striving for worldly gain ldquoCome now youwho say today or tomorrow we will go to that town and spend a year there andtrade and get gain you who do not know what will be tomorrow what yourlife will be like for you are [ἔστε] a vapour appearing for a little while rdquo7 Themajority reading here is ldquofor there will be [ἔσται] a vapour for a little whilerdquo orreferring to life ldquoit will be [ἔσται] a vapour for a little whilerdquo This is an obviousscribal slip with an equally obvious reason interchange of the diphthong -αι-and the vowel -7- both of which have the sound [e] in later Greek

Finally it has to be taken into account that there is evidence to show thatthe distinctive readings of the Koine text accumulated in stages (cp Wachtel1995180ndash198) It is probable that the last decisive step to standardisation wasthe change to minuscule production in the ninth century It was for this pur-pose that archetypes of many manuscript traditions of Greek literature wereproduced For the Greek New Testament this was the text used and to be usedin liturgy and theological study

There can be little doubt that the Byzantine text is distinguished from theoriginal text by many readings There seems to be a large gap between the so-called Byzantine texttype and textforms preserved in manuscripts from the 3rdto 4th centuries if we focus on the variants attested to by the mass of youngermanuscripts against the noble few In recent times editors of the Greek NewTestament have striven for a reconstruction of the original text according tothose old and trustworthy witnesses and have shunned Erasmusrsquo Textus Recep-tus which made the Byzantine text dominate printed editions of the Greek NewTestament for three centuries But if we take the text of the Nestle-Aland edi-tion (NA 27) which is most commonly used for scholarly purposes today as areasonably good approximation to the original text and calculate the distanceof each manuscript from this text by counting agreements with and differencesfrom it we find that pure Byzantine manuscripts come closer to the lsquooriginaltextrsquo than many older documents and their descendants This has motivated anew interest in the Byzantine text As editors of the Greek NT we take it more


seriously in all cases that do not show clear signs of scribal error It is indeedworthwhile to ask which ldquolines of textual tradition fed into the tenth andeleventh centuriesrdquo as did Frederik Wisse in developing the Claremont ProfileMethod for classifying and evaluating manuscript evidence (Wisse 19825)

Group readings of Koine manuscripts

Due to the large number of extant manuscripts and the fact that most ofthem witness to the Byzantine text a New Testament editorrsquos first task is thedetermination of three classes of manuscripts

1 those witnessing to non-Byzantine texts which probably represent oldertext-forms

2 those witnessing to the Byzantine text of which as many as possible willbe eliminated from the list of witnesses that should be included in theedition and

3 those suitable for representing the Byzantine Codices in the edition

In the following I am going to deal with the last of these points The questionis which manuscripts are suited to serve as representatives of the Byzantinetext in a more discriminating and not merely quantitative manner I shall usetest collations of 1785 manuscripts for this These collations were done at 153short passages in the Gospel of John as a first step towards a future criticaledition of this writing The results were entered into a database I shall usethese results to mark out Byzantine groups for the purpose of finding whichmanuscripts are typical of the main varieties of the Byzantine text Such groupsmay provisionally be defined as sharing more readings with each other thanwith the majority text

Not all kinds of readings are suited for this evaluation Above all it is neces-sary to distinguish between variants and mere readings A reading is the genericterm for the wording of a passage by which a manuscript is distinguished fromone or more or from all other manuscripts A variant is defined as one ofat least two readings of the same textual unit which is grammatically correctand logically possible A reading which does not fulfil these criteria is an errorThis means that the manuscript will be treated as witnessing defectively to thevariant rendered Like errors orthographically or morphologically equivalentforms of the same variant are not classed as textual differences

Klaus Wachtel

Groups of manuscripts as shown below in the extract from the table ofmanuscript groupings were determined by the following procedures and qual-ifications

1 For each pair of manuscripts which share at least one reading apart fromthe majority text the total numbers of agreements including and excludingthe majority readings were calculated This was done by a program whichcompared their patterns of variants at the 153 test passages

2 Two manuscripts qualified as members of a group if the percentage of theirmutual agreements was greater than the percentage of the readings that oneor both of them shared with the majority

3 The degrees of agreement reached by a manuscript as compared to otherswere classified Only those manuscripts which showed at least a third-ratedegree of agreement with the base manuscript were included in a group8

4 If the condition under point 2 was met but the percentage of agreementsapart from the majority text was less than 50 the respective manuscriptswere regarded as weakly related Manuscripts with such rates of agreementare found in many groups defined by the condition under point 2

This sifting procedure resulted in 1125 groupings like those shown in the ex-tract Most of them consist of between two and ten manuscripts and in therealm of the Koine it is not uncommon for some of the manuscripts to agree ata rate of 100 percent Such manuscripts are very likely to have been copied froma common exemplar which may have survived as one of the group members

The biggest clear-cut Byzantine group was named Kr by Hermann von So-den who discovered it (Soden 1911 (I2) 757ndash765) The ldquoKrdquo means Koinethe ldquorrdquo recension It is a diligently made manuscript edition which shows thedegree of perfection that could be achieved by scribes There are 40 codicescontaining the gospel of John which witness to this edition without any de-viation at the 153 test passages The group is distinguished from the majorityby four readings that while supported by Kr are usually supported by consid-erably more manuscripts besides The lowest number of witnesses for one ofthe Kr readings is 272 the highest 367 This indicates that the editor(s) madedeliberate decisions where they knew of variants

There are four group readings of Kr at the test passages but they arenot distinctive readings in the proper sense of the word as they are sharedby manuscripts belonging to other groups This is the rule with manuscriptswhich have such a large share of majority readings

Thus it is clear that the starting point for an inquiry into the genealogicalbackground of Kr must be the pattern of variants distinguishing the group from


Table 1 Extract from the table of manuscript groupings

ms MT cms dg n agr1 agr2

4 908 973 2 1 92 (129141) 67 (23)8 974 65 2 8 98 (150153) 100 (11)

774 1 2 98 (150153) 100 (11)1168 3 1 98 (150153) 67 (23)1514 1 3 98 (150153) 100 (22)

10 908 342 1 48 100 (1212) 100 (11)895 2 1 99 (151153) 100 (1414)1091 2 1 98 (150153) 92 (1213)1194 2 1 98 (150153) 92 (1213)1517 2 1 96 (145151) 77 (1013)2676 3 1 96 (142148) 91 (1011)248 2 1 93 (142152) 100 (99)

11 961 1212 2 1 99 (152153) 100 (55)570 1 3 99 (151152) 100 (55)1207 1 3 99 (150151) 100 (66)200 2 2 99 (150152) 100 (66)944 2 3 99 (150152) 100 (55)905 2 1 99 (149151) 100 (66)1444 2 3 99 (148150) 100 (44)188 2 1 98 (150153) 100 (55)1179 3 1 98 (150153) 100 (44)1351 2 14 98 (4748) 100 (11)2682 2 131 97 (3132) 100 (11)29 3 3 97 (147152) 100 (44)148 3 2 96 (129134) 100 (55)

13 757 543 2 1 93 (119128) 80 (2025)828 3 1 91 (138152) 79 (2633)788 3 1 89 (125140) 89 (2326)346 2 1 89 (7382) 87 (1315)

14 987 140 1 6 100 (153153) 100 (22)1343 1 6 100 (153153) 100 (22)2224 1 6 100 (153153) 100 (22)2522 1 6 100 (153153) 100 (22)95 1 6 100 (150150) 100 (22)405 1 20 100 (106106) 100 (11)123 2 10 99 (152153) 100 (22)208 2 6 99 (152153) 100 (22)1080 2 10 99 (152153) 100 (22)1191 2 10 99 (152153) 100 (11)1225 1 11 99 (152153) 100 (22)1585 1 9 99 (152153) 100 (11)

ms manuscriptMT percentage of agreementswith the majority textcms compared manuscriptdg degree of agreement reachedby the compared manuscriptn number of manuscripts forwhich the same degree of agree-ment is reached by the com-pared manuscriptagr1 percentage and number ofagreements proportional to thenumber of shared test passagesincluding majority readingsagr2 percentage and number ofagreements proportional to thenumber of shared test passagesexcluding majority readings

ExamplesFor manuscript 4 (ms first line)there is only one manuscriptwhich agrees more frequentlywith it than with the majoritytext manuscript 973 (cms)The figures in agr1 show thatthe two manuscripts agree ata rate of 92 or in 129 of the141 test passages they shareincluding the majority readingsAccording to agr2 4 and 973agree at a rate of 67 or in2 of the 3 test passages theyshare apart from the majorityreadings According to the entryin dg this is only the second bestdegree of agreement reached by973 The entry in n shows that4 is the only manuscript which973 agrees with at the samedegree

Klaus Wachtel

Table 1 (continued)


2509 2 10 99 (152153) 100 (11)584 2 10 99 (151152) 100 (22)1058 2 15 99 (104105) 100 (22)

15 935 2562 1 1 99 (151153) 100 (88)1439 2 1 98 (149152) 90 (910)1163 1 1 98 (145148) 90 (910)53 3 1 95 (146153) 75 (68)902 3 1 95 (146153) 86 (67)2374 2 1 95 (143151) 100 (44)2502 2 2 94 (141150) 100 (22)

18 974 35 1 40 100 (153153) 100 (44)55 1 40 100 (153153) 100 (44)128 1 40 100 (153153) 100 (44)201 1 40 100 (153153) 100 (44)479 1 40 100 (153153) 100 (44)480 1 40 100 (153153) 100 (44)645 1 40 100 (153153) 100 (44)696 1 40 100 (153153) 100 (44)769 1 40 100 (153153) 100 (44)789 1 40 100 (153153) 100 (44)867 1 40 100 (153153) 100 (44)928 1 40 100 (153153) 100 (44)955 1 40 100 (153153) 100 (44)1023 1 40 100 (153153) 100 (44)1072 1 40 100 (153153) 100 (44)1117 1 40 100 (153153) 100 (44)1147 1 40 100 (153153) 100 (44)1339 1 40 100 (153153) 100 (44)1401 1 40 100 (153153) 100 (44)1493 1 40 100 (153153) 100 (44)1496 1 40 100 (153153) 100 (44)1550 1 40 100 (153153) 100 (44)1560 1 40 100 (153153) 100 (44)1584 1 40 100 (153153) 100 (44)

This means that 4 and 973are only weakly related But asthere is no other manuscriptwhich comes closer to 4 wemight still start investigatingwhich non-majority readingsconnect the two manuscripts ifwe were interested in the rela-tions of ms 4 with the varietiesof the Byzantine text We mightthen look at the manuscripts towhich 4 is related by 973 (ie thecloser relatives of 973 not shownin the present extract)

The last manuscript in thisextract 18 which agrees withthe majority text at a rate of974 is shown to have manyrelatives which share morereadings with it than with themajority The first comparedmanuscript 35 is one out of 40which agree at the same rate of100 with ms 18 including orexcluding the majority readings

Manuscripts 4 and 18 rep-resent extremes Most groupingsconsist of between two and tenmanuscripts

the majority text (ldquoMTrdquo in Table 1) The total number of Byzantine groupssharing at least one reading with Kr apart from the majority is 391 57 of themare connected by patterns of variants which include the complete Kr patternThe total number of manuscripts involved is 127 The following table showsthe patterns coming closest to that of Kr being represented by minuscule 189


Table 2

Ms MT Group Pattern

18 974 322 555 1004 150483 967 322 555 1004

1279 1504141 967 1612B 322 555 1004 1504147 961 263 322 555 1004B 1504

693155 961 322 555 1004 1504

992 1513167 954 273 322 555 1004B 1504

603664

189 935 123 322 555 1004B 1504394 693 1323

894 1343246 974 322 555 1004B 1504386 967 322 555 1004 1504

963394 967 322 555 1004B 1504

1183547 966 322 555 1004 1504

693553 967 322 555 1004 1504

762685 967 322 555 1004B 1504

1525689 961 92 322 555 1004B 1504

123763 967 322 555 1004B 1504

762781 967 322 555 1004B 1504

548797 954 182 322 555 1004 1504

322B 6912D547

The first line shows that ms 18 a minuscule from the 14th century whichrepresents von Sodenrsquos Kr in a particularly pure form agrees with the major-ity text at a rate of 974 of the test passages The group pattern consists ofvariants at the test passages that distinguish the group from the majority (eg322 is test passage 32 variant 2) Exactly the same pattern is shared by 41manuscripts 40 of which are not shown in the figure The second manuscript

Klaus Wachtel

in the figure ms 83 has all the readings of the Kr pattern but differs from themajority at one more test passage (1279) The mss which follow in the tableall include the Kr pattern but all differ from the majority at one or more addi-tional test passages (Note that in ms 246 the difference is but a subvariant of1004 indicated by the ldquoBrdquo)

It can hardly be due to chance that all of the manuscripts attesting the Kr

pattern as a part of their own are found to agree with the majority at more than92 of the test passages We have witnesses more remote from the majority textfor each single reading of the pattern but not for the pattern itself This againindicates that the text of Kr was carefully edited The editors wanted the newtext as much as possible to be in concordance with those in official use thusintroducing a new standard

Our chance to trace the genealogy of Kr and the groups containing its char-acteristic pattern is not very good As a rule the readings of the patterns includ-ing the one of Kr are attested to by relatively strong minorities of manuscriptsand were occasionally picked up by scribes But our test collations are not asuitable basis for genealogical analysis anyway This is because the system oftest passages cannot serve as a model for the entire text They were selectedin the first place for the purpose of distinguishing witnesses of the Byzantinetext from those of other textforms Yet the present study shows that we cango one step further The patterns of variants at the test passages allow a morediscriminating selection of Koine manuscripts which should be subjected tofull collations and then to genealogical methods as shown by Gerd Mink in thepresent volume

Conclusions

1 There are 1484 manuscripts containing the Gospel of John which agreewith the majority at more than 90 percent of the test passages The remain-ing 301 manuscripts are certainly not all candidates for the apparatus of afuture Editio Maior of the Gospel of John because many of the differencesfrom the majority may turn out to be distortions of majority readings Thishas to be worked out

2 The Koine group Kr was identified by means of a distinctive pattern of vari-ants at four test passages This means that the large number of manuscriptswhich deviate from the majority in between one to all four of the Kr

variants will be well represented by a single Kr manuscript


3 As a clearly defined group Kr can serve as a starting point for further clari-fication of the group structure of the Koine manuscripts One question willbe whether manuscripts which include the full Kr pattern are more likelyto be predecessors or descendants of Kr Another promising investigationwill deal with those Koine manuscripts which share no reading with Kr atthe test passages

4 One basic conclusion may be drawn as to the usefulness of test collationfor determining manuscripts for further study it is demonstrated by thefact that the outlines of the structure of the tradition can be drawn on thatbasis

Notes

ECM IV 19972003

Text und Textwert 1989ndash1999

Cp ECM IV 1 11ndash12 ECM IV 2 B8ndashB9

This chapter contains statements and conclusions which are explained on a larger scalein Wachtel (1995)

Cp eg WestcottHort (1882132ndash139) Soden (1911(I2) 707ndash713) Metzger (1992212)

Schwartz (1909CXLVI) (ldquohalbbewusste Trivialisierungrdquo)

Translation quoted from the Revised Standard Version (21971)

A first-rate degree of agreement means that the manuscript being compared does notreach a higher percentage of agreements with any other manuscript If it has a second-ratedegree of agreement there is at least one manuscript which comes closer to it than the onewhich the second-rate degree applies to

[322 ndash 41 ndash lsquoJesusrsquo instead of lsquothe Lordrsquo Context is ldquoNow when the Lord knew that thePharisees had heard that Jesus was making and baptizing more disciples than John althoughJesus himself did not baptizerdquo It is quite likely that original lsquoJesusrsquo was replaced with lsquotheLordrsquo here to avoid the clumsy repetition of lsquoJesusrsquo the more so because lsquothe Lordrsquo was in-creasingly used in reference to Jesus There is a relatively strong fraction of 367 witnesseswhich share the reading of Kr some of them ranging among the oldest and best [555 ndash544 ndash replacement of lsquofrom one anotherrsquo by lsquofrom menrsquo in the sentence ldquoHow can you be-lieve who receive glory from one another and do not seek the glory that comes from theonly Godrdquo The reading of Kr makes for a more obvious opposition to lsquothe only Godrsquo Thetotal number of witnesses is 381 here [1004 ndash 753ndash811 ndash This passage refers to the storyof the adulteress who is brought to Jesus by the scribes and Pharisees and who finds for-giveness with him There is overwhelming evidence for the non-Johannine origin of thepericope (cp Metzger 1994187ndash189) which is found in the majority of witnesses Kr has itbut obelises it which is an obvious trace of editorial work The total number of witnessesfor this obelization is 272 [1504 ndash 1039 ndash ldquoagain they tried to arrest himrdquo The variant of

Klaus Wachtel

Kr is a transposition which does not affect the meaning at all Here we have branches of 302witnesses for the reading of Kr and 439 for still another word-order

References

ECM IV (19972000) Novum Testamentum Graecum Editio critica maior IV DieKatholischen Briefe hg von B Aland K Aland dagger G Mink K Wachtel Part 1 TextPart 2 Begleitende Materialien 1 Lieferung 1997 Der Jakobusbrief 2 Lieferung 2000Die Petrusbriefe 3 Lieferung 2003 Der Erste Johannesbrief Stuttgart

Ehrman B D (1993) The Orthodox Corruption of Scripture The Effect of Early ChristologicalControversies on the Text of the New Testament New YorkOxford

Metzger B M (1992) The Text of the New Testament its Transmission Corruption andRestoration Oxford 1964 (31992)

Metzger B M (1994) A Textual Commentary on the Greek Text of the New TestamentStuttgart 1971 (21994)

NA27 Novum Testamentum Graece post Eberhard et Erwin Nestle editione vicesimaseptima revisa communiter ediderunt B et K Aland J Karavidopoulos C M MartiniB M Metzger A Wikgren Stuttgart 1993

Schwartz E (1909) ldquoProlegomena zu Eusebs Kirchengeschichterdquo Eusebius Werke II 3 TeilEinleitungen Uumlbersichten und Register (GCS 93) Leipzig

Soden H Frhr v (1911) Die Schriften des Neuen Testaments in ihrer aumlltesten erreichbarenTextgestalt Bd I 1ndash3 Untersuchungen Goumlttingen 21911 Bd II Text und ApparatGoumlttingen 1913

Text und Textwert (1989ndash1999) Text und Textwert der griechischen Handschriften des NeuenTestaments ed by Kurt Aland et al Vol 1 Die Katholischen Briefe (1987) vol2 Die Paulinischen Briefe (1991) vol 3 Die Apostelgeschichte (1993) vol 4 Diesynoptischen Evangelien (19981999) Berlin

Wachtel K (1995) Der Byzantinische Text der Katholischen Briefe Eine Untersuchung zurEntstehung der Koine des Neuen Testaments Berlin

Westcott B F amp F J A Hort (1882) The New Testament in the Original Greek Bd I TextBd II Introduction CambridgeLondon 1881 and 1882

Wisse F (1982) The Profile Method for the Classification and Evaluation of ManuscriptEvidence as Applied to the Continuous Greek Text of the Gospel of Luke Studies andDocuments 44 Grand Rapids Michigan

How shock waves revealedsuccessive contaminationA cardiogram of early sixteenth-centuryprinted Dutch Bibles

A A den HollanderUniversiteit van Amsterdam Vrije Universiteit Amsterdam

Introduction1

One of the complex problems a philologist has to deal with is a text traditionthat is so entangled that its transmission cannot be assessed in a straightfor-ward way Such a text tradition is generally considered to be lsquocontaminatedrsquosome or all of its members present a text not derived from one single exem-plar but composed from several predecessors Until recently scholars had noaids in trying to handle this problem2 In their article in the first volume ofStudies of Stemmatology however E Wattel and M J P van Mulken offeredan instrument which could deal at least partially with the phenomenon ofcontamination (Wattel amp Van Mulken 1996)

In their contribution they distinguished three types of contamination

1 simultaneous several exemplars being used at the same time2 successive one exemplar being used for one part of a text another for a

second part and another (or again the first) for yet another part of thetext etc

3 incidental one exemplar being used for the text and other exemplars usedto verify or improve the text through local interventions

The instrument Wattel and Van Mulken offered helps to trace the second typeof contamination successive contamination In addition they suggest a proce-dure for handling stemmatological problems caused by this phenomenon Suc-

A A den Hollander

cessive contamination makes it impossible to draw one stemma which correctlyrepresents the genealogical relationships throughout the entire text Stemmascan only be drawn for parts of the text and will only represent the relationshipsin those particular parts It is therefore advisable to split up the text into sec-tions for each of which one valid stemma can be drawn For making sound textdivisions in a (successively) contaminated text Wattel and Van Mulken suggestusing a so-called cardiogram of the text tradition

In this article I intend to show how a cardiogram has actually been of greathelp in tracing successive contamination in the text tradition of the Dutchbibles printed between 1522 and 1545 On the basis of a cardiogram the texttradition could be split up into parts each of which corresponded with logi-cal units one or more bible books It turned out that the bible text in thesesuccessive units had indeed been derived from various sources

The cardiogram of a text tradition

A cardiogram is a graphic presentation of the distances between text witnesses(manuscriptseditions) In a text tradition with unambiguous relationshipsthe similarity (or distance) between two witnesses will remain more or lessstable In case of successive contamination the similarity (or distance) betweentwo witnesses will change at the point of change in relationship3 ndash probablydramatically In order to find out whether and at what points in a text suchchanges in distances between witnesses occur Wattel developed a distance dis-tribution function which records the distances between each pair of witnessesat every instance (marked by a variant) of a text tradition

The starting point in the production process of a cardiogram is the listof variants This list of variants should have a clear data structure and shouldcontain at least the following elements

1 heading line ndash presenting the total number of lines and all the wit-nesses (manuscripts or editions) involved Each witness is represented by aunique siglum

and each next line stating the following elements in a fixed order

2 location ndash of the variant eg number of chapter verse or line etc3 formula ndash mathematical representation of the relation between witnesses4 readings ndash the various readings at a certain variation place

The following fictitious list of variants may serve as an example

How shock waves revealed successive contamination

6 abcdefg (heading line)1 abcdefg reading 1 reading 221 abcdefg reading 1 reading 2 reading 322 abcdefg reading 1 reading 24 abcdefg reading 1 reading 2

In this example the text has 6 lines with one variant in line 1 two variants inline 2 and again one variant in line 4

Additional information may be added in other columns in the list of vari-ants such as a typology of variants and weight factors A typology of variantsassigns each variant to a certain type eg the type ldquoinversion of wordsrdquo Such atypology makes it possible to test which types of variants are kinship-revealingin a given text tradition and which are not and should therefore be left outTherefore no decision concerning possible genealogical relevance of a certainkind of variant must be taken before the entire list of variants is completed

A weight factor may also be added to each variant A different weight factormay be assigned to the various types of formulas in this case concerning ty-pology of variations not of variants Type 2 variations for example show ldquotwocompetitive variant readings which are present in precisely two true groups oftext versionsrdquo4 a true group being a group containing two or more membersType 2 variations which are considered to be the most important materials forthe stemma can be given a much higher weight factor than the other variationtypes Again the whole process of weighing may be changed at any time and isreversible as well No final decision concerning possible genealogical relevanceof a certain type of variation must be taken during the process of building thelist of variants

After the list of variants has been established the similarity between eachpair of witnesses will be measured throughout the entire list of variants Thisstep will result in a list for each pair stating whether the two do or do not sharea reading Two witnesses sharing the same reading will have a positive score(similar) two witnesses not sharing the same reading a negative score (notsimilar)5 When adding up all the scores of a given pair over a large numberof formulas results in a positive score it can be concluded that these witnessescorrespond in most cases When the result is a negative score it means that thetwo witnesses have different readings in most formulas

Wattel defined a distribution formula which makes it possible to visual-ize the lsquosimilarity-scorersquo of a pair of witnesses as a function over the entire textThe graph produced by this function fluctuates somewhere between its extremevalues +100 (complete similarity) and ndash100 (complete dissimilarity)6 This

A A den Hollander

F

D

C

B2

B1

A4

A3

A2

total

A1

47ndash2

ndash44ndash99

37ndash88

10078

89ndash39

67ndash39

89ndash39

89ndash39

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500

25

ndash92

ndash39

97

26

23

27

24

Figure 1 Similarity of all mss of Charroi de Nicircmes with ms B1 Horizontally line num-bers of the text vertically sigla of the mss on the right maximum and minimum valuesof the lsquosimilarity-scorersquo on the left the average values

lsquosimilarity-graphrsquo clearly shows how each of the other witnesses relates to agiven witness and whether any dramatic changes in relationship occur A rela-tionship is stable when the similarity remains stable through the entire text Astable relationship will be visualised by a flat graph without many fluctuationsThe graphs in Figure 1 show the relationships of eight manuscripts (A1ndash4 B2C D and F) of the manuscript tradition of Charroi de Nicircmes with manuscriptB1 of the same tradition7

The graphs clearly show that the relationships between B1 and B2 as well asbetween B1 and D are stable B1 and B2 are constantly very similar (maximum+100 minimum +78 average +97 without many fluctuations) and B1 andD very dissimilar (maximum ndash44 minimum ndash99 average ndash92 without manyfluctuations)

In this article however we are not interested in stable relationships Wewant to know for each pair of witnesses where major changes in their rela-tionships occur We are in other words not interested in the flat parts of thelsquosimilarity-graphsrsquo indicating stable relationships but in those parts where thegraph rises or falls sharply8 These rises and falls can be visualised in anothergraph ndash a so-called shock wave This shock wave shows for each pair of wit-


F

D

C

B2

B1

A4

A3

A2

total

A1

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500

Figure 2 Shock waves of all the manuscripts of Charroi de Nicircmes Horizontally the linenumbers of the text vertically the sigla of the manuscripts

nesses where the lsquosimilarity-graphrsquo is flat and where it fluctuates In the caseof a sharp fall or rise the lsquoshock waversquo has a high value in the case of a sta-ble relationship it has a low value The minimum value of the shock wave iszero (completely flat lsquosimilarity-graphrsquo) and its maximum is one (maximumfall or rise)

Counting up the values from the shock waves of a given witness with allthe other witnesses of a text tradition results in the ultimate shock wave of awitness Figure 1 eg shows the shock wave of manuscript B1 It representsthe average value of all the shock waves of pairs of manuscripts of the traditionof Charroi de Nicircmes containing manuscript B1 The points where this averagevalue graph sharply rises or falls still indicate important changes in relation-ships of manuscript B They seem to occur in the vicinity of line 500 950and 1050

It is possible to compute shock waves of all the members of a text tradi-tion Figure 2 shows the shock waves of all the manuscripts of the tradition ofCharroi de Nicircmes

The graph at the base line gives the average value of all the shock waves9

This graph is the shock wave of the entire text tradition of Charroi de Nicircmes

A A den Hollander

Figure 2 gives the heart beat ndash the cardiogram ndash of the text tradition The pointswhere the shock wave of the entire text tradition peaks indicate changes in rela-tionship in the text tradition When all the manuscripts or at least a substantialnumber peak at the same point this could well be an indication of successivecontamination10

Early sixteenth-century printed Dutch bibles

During the period 1522-1545 some eighty Dutch bible editions were pub-lished11 Six of them had a reissue during that period one as many as threereissues Seventeen of these eighty-nine editions were complete bibles Fiveeditions contained the text of the Old Testament or parts of it only andseventy-six editions the text of the New Testament or parts of it only

On the basis of an extended random sampling a comparison was made ofthe bible text in the various editions The similarities and differences in read-ing between the editions were taken up into a list of variants Separate listswere made for the Old and the New Testament The list of variants for the OldTestament counted 5099 variants and for the New Testament 4573 variants

With the help of the instruments Wattel had developed two separate car-diograms were made for the Old Testament and for the New Testament Figure3 gives the cardiogram of the Old Testament consisting of the shock waves ofthirteen editions of (parts of) the Old Testament leaving out the reissues atthe base line is given the shock wave of the text tradition of the Old Testamentin early sixteenth-century printed Dutch bibles12

Several observations can be made after a first glance at the cardiogramFirst observing the shock waves of the thirteen editions it is obvious that thegraphs are quite flat and do not show a lot of fluctuation There are a number oflocal symptoms such as JvL1526 en DP1527 peaking sharply at line 780 ndash andsome others peaking less at the same spot The graphs also show a few moregeneral phenomena such as the fluctuations at line 213 line 1145 and line1300 Also the shock wave of the entire tradition at the base line is quite flat andwithout a great deal of fluctuation let alone any number of significant peaksExcept for the peak at line 1145 all fluctuations could be considered as noise

However looking into the graphs in more detail revealed that at every pointwhere the heart beat of the text tradition at the base line fluctuated visibly achange of relationship did indeed take place Even points with minimal fluc-tuation still marked an actual change in relationship somewhere in the texttradition13 With the aid of the heart beat the text tradition could be split up


III II II VI VI IV

III IV V II VII VIII

CvR

SM

JvL

WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 180 360 540 720 900 1080 1260

1525

1544

1527

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525

Figure 3 The heart beat of the Old Testament Horizontally the line numbers (basedon the list of variants) vertically the abbreviations of the various editions14

into thirteen separate parts15 Within these thirteen parts all the relationshipsremained stable The thirteen parts all correspond with coherent text unitsfrom the Old Testament ndash one or more bible books16 ndash and for all thirteenparts separate stemmas were drawn

A A den Hollander

1 Genesis Exodus Leviticus Numbers Deuteronomy2 Joshua Judges Ruth 1+2 Samuel 1+2 Kings 1+2 Chronicles Ezra Ne-

hemiah Esther (canonical part)3 Esther (apocryphal part)4 Job5 Psalms6 Proverbs Ecclesiastes Song of Solomon7 Isaiah8 Jeremiah Lamentations Baruch Ezekiel Daniel Hosea Joel Amos Oba-

diah9 Jonah10 Micah Nahum Habakkuk Zephaniah Haggai Zechariah Malachi11 3+4 Esdras12 Tobith Judith Wisdom of Solomon Ecclesiasticus Daniel (apocr)13 1+2 Machabees

Passing from one of these thirteen parts of the Old Testament to the next alwaysresulted in change of relationship somewhere in the text tradition Thereforewith each transition a new stemma was required to represent the genealogicalrelationships in the next part Not all thirteen stemmas however were entirelydifferent since two or more non-successive stemmas were sometimes identi-cal A total number of eight different stemmas was required to represent thegenealogical relationships in the entire text tradition of the Old Testament inthe early sixteenth-century printed Dutch bibles17

The process of dividing the text tradition into sections on the basis of thecardiogram drew our attention to some striking features of the text tradition ofthe Old Testament Why for example was there a change in relationships be-tween the books of Isaiah and Jonah These books belong to a coherent part ofthe Old Testament major and minor Prophets respectively in which changesin relationship were not be expected A closer look into the part of the Prophetsthrough study of the list of variants and the lsquosimilarity graphsrsquo in the relevantsection of the text tradition revealed the cause of the change in relationshipvarious sources were used for the books of the Prophets in the 1526 Bible of theAntwerp printer Liesvelt

Jacob van Liesvelt was the first one to publish a complete bible in the Dutchlanguage18 His aim was to provide the Netherlands with a Dutch translation ofthe Luther Bible In 1526 however Luther had not completed his bible trans-lation in full He had translated the entire New Testament a few years earlier


but at that time substantial parts of the Old Testament remained to be doneLuthers translation of the Prophets was not finished until the year 1534 In 1526Luthers translations of two of the minor Prophets had been issued separatelyJonah and Habakkuk19 Jacob van Liesvelt must have been able to lay hands ona copy of the edition of Jonah almost at once Only a few months later thisLuther edition appeared in a Dutch translation in the Liesvelt Bible For theother books of the Prophets except for Isaiah Liesvelt simply reproduced theexisting Dutch translation from the 1525 edition of the Old Testament of theAntwerp printer Hans van Ruremunde (HvR1525) which explains the suddenchange in relationship20

A cardiogram was also computed for the New Testament see Figure 4On the basis of this cardiogram the text of the New Testament was split up

into nine parts Again all parts were coherent text units consisting of one ormore bible books21

1 Matthew Mark Luke John2 Acts3 Romans4 1+2 Corinthians Galatians Ephesians Philippians Colossians 1+2 Thes-

salonians5 1+2 Timothy Titus Philemon6 Hebrews7 James 1+2 Peter 1+2+3 John8 Jude9 Revelation

A total number of eight different stemmas was necessary to represent all thevarious relationships for the nine sections22 The same stemma was valid forparts (2) and (9) In this case it was not very surprising that the same stemmawas valid for these two parts The text of most of the early printed sixteenth-century editions of the New Testament had been derived from other existingDutch translations Some of the editions of the New Testament were printedand published in parts usually one part containing the Gospels another partall the Letters and a third part the two remaining books Acts and RevelationThese parts were independently distributed and were therefore separatedlyused as sources for new editions

Suppose that the text of an edition of the New Testament A went backto another existing Dutch translation Suppose also that for the text of Actsand Revelation one of the above-mentioned separate editions was used whichgave a completely different text from the text in the existing edition In that

A A den Hollander

I III IV V VI VII VII

0 120 160 200 240 280 320 360 40040 80

IX IIII IX(B)

MC

AvB

JvL

JvL

CL

JvG

JvL

MHvH

HPvM

JvG

JvL

JvL

JvG

AvB

AP

AvB

SM

AvB

WV

CvR

HvR

CvR

WV

JvG

CvR

MKGvdH

DP

HF

total

1539

1525

15232

15231

1524

1525

1526

1527

1543

1524

1522

1523

15262

1523

1525

1524

1545

1524

1529

1525

1525

1528

1528

15261

1526

1525

1523

1525

Figure 4 The heart beat of the New Testament Horizontally the line numbers (basedon the list of variants) vertically the abbreviations of the various editions23

case the shock wave of this New Testament A would peak between just beforeand directly after Acts as well as just before Revelation indicating changes inrelationship24 On the basis of this shock wave the text would have to be split


up into four parts The final conclusion would be that parts (1) and (3) wouldshare one stemma and so would parts (2) and (4)

Conclusions

1 The cardiogram has proved to be a useful instrument for getting at oneglance an overall picture of the genealogical relationships within the texttradition of the early sixteenth-century printed Dutch bibles It should bestressed that this overview could be obtained on the basis of just the listof variants

2 The cardiogram made it visible at a glance how stable the genealogicalrelationships within the entire text tradition were The shock waves alsoclearly revealed these relationships for each of the witnesses It shouldbe noted however that the text tradition of the early sixteenth-centuryprinted Dutch bibles is a very lsquocleanrsquo one Incidental contamination forexample can obscure a clear view25

3 By means of the cardiogram it was also possible to point out those placesin the text where changes in relationship might have taken place Furtherresearch with the benefit of the list of variants and the lsquosimilarity-graphsrsquodid in fact reveal successive contamination As a result the text traditionwas split up into coherent parts and for each of these a separate stemmawas drawn

4 It turned out that the coherent parts were not only genealogically coher-ent text units but that these parts also yielded a profile of the actual usedsources The fact for example that the single book of Jonah was separatedfrom the other books of the minor Prophets as a non-coherent part raisedthe question what actual source had been used for the book of Jonah Thesearch for a potential source of Jonah led to the conclusion that its sourcehad been a separately issued book of Jonah This meant that other coherenttext parts established after division on the basis of the cardiogram mightalso reflect (the size of) the actual used sources This turned out to be trueas was shown above for separately issued parts of the New Testament

The same conclusions may hold for handwritten texts Suppose that the car-diogram of a manuscript tradition divides up a text into parts Since we knowthat manuscripts were sometimes split up into quires for reproduction ourtext division on the basis of the cardiogram might very well result in text partswhich relate directly to the separate quires of the exemplars In that case the

A A den Hollander

cardiogram could also function as a quire separator indicating the actual sizeof the exemplar

Notes

This article uses results of my earlier study (Den Hollander 1997 esp Chapter 3 ldquoDeteksttraditierdquo 127ndash242)

In 1957 Paul Maas uttered ldquoGegen die Kontamination ist kein Kraut gewachsenrdquo (Maas195731)

See for an example Wattel and Van Mulken (1996107)

See for this definition and a definition of other types of variations Salemans (200023ndash25) See also the contributions of Salemans and Wattel in this book

See Wattel and Van Mulken (1996110ndash111) to learn how the scores could be computed

The shock waves show the actual extreme values of each graph as well as the averagescore

This example was derived from Wattel and Van Mulken (1996116)

In order to level out the graph the maximum rise or fall between two successive measur-ing points is about 30 Increase of similarity for example from 5 to 95 requires a pathof four successive measuring points minimal The horizontal line consists of 300 measuringpoints each measuring point corresponding with 5 textlines In this way local informationis suppressed in order to correct incidental influences

That the shock wave of manuscript D hardly fluctuates is due to the fact that D hardlyever agrees with any of the other manuscripts in the list of variants

It is of course always necessary to relate these findings to the text tradition itself (suchas the list of variants)

Bible editions in this context are complete bibles individual issues of the Old and NewTestaments individually issued books of the bible and fragments I exclude individual edi-tions of the Psalms and editions of the lessons from the lsquoepistles and gospelsrsquo since thesetwo genres constitute separate text traditions The bibles printed in this period form awell-defined group and can be considered as an indepedent text tradition

This cardiogram already gives an impression of which editions are related and which arenot Related editions will have a similar shock wave

The small peak at line 175 (between Numbers and Deuteronomy) however did not referto a change in relationship The peak in WV1528 indicated a local phenomenon

The abbreviations consist of the abbreviated name of the printer and the year of issueHvR = Hans van Ruremunde JvL = Jacob van Liesvelt DP = Doen Pietersoen WV = WillemVorsterman HPvM = Henrick Peetersen van Middelburch HvL = Hansken van LiesveltSM = Steven Mierdmans CvR = Christoffel van Ruremunde


The parts covered lines (1) 0ndash123 (2) 219ndash509 (3) 515ndash519 (4) 525ndash566 (5) 572ndash717(6) 723ndash781 (7) 787ndash851 (8) 857ndash1044 (9) 1050ndash1053 (10) 1059ndash1124 (11) 1130ndash1159(12) 1165ndash1295 (13) 1301ndash1335

Separate cardiograms can be made for each of the thirteen parts which give a moredetailed look into the interrelations see Appendix A for the heart beat of part one of the OldTestament

Eight different stemmas were valid for various parts no 1 for part (1) no 2 for parts(2) (4) (6) and (9) no 3 for part (3) no 4 for parts (5) and (12) no 5 for part (7) no 6for part (8) no 7 for part (11) no 8 for part (13)

More extensively in Den Hollander (1999)

Luthers Werke Die Deutsche Bibel vol 2 392ndash395

For the book of Isaiah Liesvelt once more made use of a different source This time hepublished a Dutch translation of the Latin text of Isaiah from Oecolampadius (1525) Com-mentary on the book of Isaiah This Latin text was not a Vulgate text but Oecolampadiusrsquoown Latin translation from the Hebrew See Den Hollander (1997185)

The parts covered lines (1) 0ndash109 (2) 115ndash142 (3) 148ndash163 (4) 169ndash260 (5) 266ndash294(6) 300ndash312 (7) 318ndash362 (8) 368 (9) 374ndash395

See for the stemmas Den Hollander (199799ndash211)

The total number of editions had been reduced through clustering see Den Hollander(1997155ndash157) Only the representatives of these clusters have been presented in the cardio-gram The abbreviations consist of the abbreviated name of the printer and the year of issueHF = Hiero Fuchs DP = Doen Pietersoen MKGvdH = Merten de KeyserGovaert van derHaeghen CvR = Christoffel van Ruremunde JvG = Jan van Ghelen WV = Willem Vorster-man HvR = Hans van Ruremunde AvB = Adriaen van Berghen SM = Steven MierdmansAP = Albert Pafraet JvL = Jacob van Liesvelt HPvM = Henrick Peetersen van MiddelburchMHvH = Michiel Hillen van Hoochstraten CL = Cornelis Lettersnijder MC = MatthiasCrom

Presuming that Revelation is the final book Otherwise the shock wave would also peakdirectly after Revelation dividing the text in five parts

See for example the cardiogram of the text tradition of Perceval (Figure 5) in Watteland Van Mulken (1996119)

References

Hollander A A den (1997) De Nederlandse BijbelvertalingenDutch Translations of the Bible1522ndash1545 Nieuwkoop De Graaf

Hollander A A den (1999) ldquoDat Oude ende dat Nieuwe Testament Jacob van Liesvelten de nieuwe markt voor bijbels in de zestiende eeuwrdquo Jaarboek voor NederlandseBoekgeschiedenis 105ndash122 Leiden Nederlandse Boekhistorische Vereniging

Luthers (D Martin) Werke Kritische Gesamtausgabe Die Deutsche Bibel 12 vols (1906ndash1961) Weimar

A A den Hollander

Maas P (1957) Textkritik (3rd rev edition) Leipzig TeubnerSalemans B (2000) Building Stemmas with the Computer in a Cladistic neo-Lachmannian

way The Case of Fourteen Text Versions of Lanseloet van Denemarken (doctoral thesis)Nijmegen Nijmegen University Press

Wattel E amp M J P van Mulken (1996) ldquoShock Waves in Text Traditions Cardiogramsof the Medieval Literaturerdquo In P van Reenen amp M J P van Mulken (Eds) Studies inStemmatology (pp 105ndash121) Amsterdam John Benjamins

Appendix A The heart beat of part one of the Old Testament

WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 20 40 60 80 100 120 140 160 180 200 220

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525

The manuscript tradition of the Cligeacutesof Chreacutetien de Troyes

A stemmatological approach

Margot van MulkenKatholieke Universiteit Nijmegen

Introduction

In order to verify whether the proliferation of intricate relationships betweenmanuscripts is not typical of the Perceval of Chreacutetien de Troyes but at leasttypical of other works of the same author I decided to examine the relation-ships between de manuscripts of the Cligeacutes and to subject the manuscripts toa similar treatment as the procedure for the manuscripts of the Perceval (VanMulken 1993) Like the Perceval the Cligeacutes is a complex tradition To quoteFoerster

So erklaumlrt sich denn nur durch diesen merkwuumlrdigen Zustand der Uumlberliefer-ung die dann nicht mehr auffaumlllige Tatsache daszlig der Cligeacutestext jedesmal einesolch groszlige Zahl von Kritikern gefunden hat waumlhrend andere Kristiantexte []so gut wie ganz von denselben Herren Kritikern gemieden worden sind dennhier handelt es sich um wirkliche Besserungen und nicht um das Vorziehen einerunter zwei an sich passablen Lesarten und es ist dazu nicht nur eine wirklichgruumlndliche Kenntnis der Sprache sondern auch wirklicher Scharfsinn unbedingtnoumltig (Foerster 1910 lxxv)

The MS tradition of the Cligeacutes consists of 7 more or less complete manuscripts(the fragments will be left out of consideration) In Table 1 all the manuscriptsare listed together with an approximate date attribution (based on Nixon1993) and localization (based on Gregory amp Luttrell 1993 Van Mulken1993 1999)

Margot van Mulken

Table 1 MS tradition Cligeacutes

Manuscript Siglum Date Attribution Localisation

Tours BM 942 T 1213 AnjouParis BN fr 794 A 13 2nd q ChampagneParis BN fr 1450 B 13 2nd q North EastParis BN fr 12560 C 13 mid Hand 1 East of Ile de France

Hand 2 West Anglo-NormanParis BN fr 1420 R 13 mid Ile de FranceParis BN fr 1374 S 13 3rd q YonneParis BN fr 375 P 1314 Arras

q = quarter mid = middle

On the basis of textual differences and similarities a list of 370 variants hasbeen established Criteria for admittance into the list were derived from DenHollander (1997) However the following variants have not been included

1 Variants concerning a change in word order (see Van Mulken 1993)2 Lacunae or discrepancies in metre or prosody3 Variants concerning pronouns prepositions particles (Salemans 2000)4 Variants concerning prefixes (Salemans 2000)5 Variants concerning tense or mood (Salemans 2000)

It is important to first determine the relationships within the tradition withoutconsidering the direction of the variants (ie the origin of the readings) sincein a first attempt the number of subjective interventions should be reduced asmuch as possible (in accordance with the Three Level Method also known asDees Method Van Mulken 199345ndash71) The list of variants adapted to theguidelines published in Wattel and Van Mulken (1995) served as input to gen-erate a table of quadruples1 and a first draft of a stemma (Figure 1) Of coursethe reliability of this stemma depends on the evaluation of the philologist adraft of a tree can always be drawn on the basis of quadruples but the relia-bility rate will not always be high enough All depends on the extent of noisethe philologist is willing to accept and the degree of inconsistency heshe isready to allow in order to build a stemma Some philologists insist upon thereduction of all noise and are therefore willing to reduce the number of vari-ants other philologists are ready to allow for a reasonable amount of noisesince some variants may be due to parallelism (Salemans 2000 Van Mulken199329ndash32) In this particular case I estimated that the first draft is far fromreliable only 103 out of the 370 variants (28) are in concordance with the

The manuscript tradition of the Cligeacutes of Chreacutetien de Troyes

C A

R B T P S

Figure 1 Draft structure Cligeacutes 72 of variants to be rejected

tree a rather too large degree of inconsistency We cannot but conclude thatthe Cligeacutes is not a closed but an open manuscript tradition

Some reputable philologists working on the Cligeacutes were also disappointedwith their trees Compare the following laments

So ist denn schlieszliglich der Stammbaum stellenweise recht unsicher ndash es muumlszligteneigentlich deren mehrere aufgestellt werden je nach den verschiedenen Teilen

(Foerster 1910 lxxiii)

Dieses schwankende Handschriftenverhaumlltnis [] fuumlhrt dazu daszlig man an vielenStellen schwankend wird und mit der gewoumlhnlichen Formel des Stammbaumesnicht auskommen zu koumlnnen glaubt (Foerster 1910 lxxv)

[] Il (le copiste du groupe y) a copieacute un ms contamineacute qui offre dans son textedes leccedilons tantocirct de S tantocirct de P tantocirct de B Crsquoest lrsquoexplication agrave adopter si lrsquoonimagine difficilement lrsquoarcheacutetype y allant consulter trois mss (Micha 1966114)

Il ressort deacutejagrave que le ms B nrsquoest pas le seul oscillant puisque S a des rapports assezsuivis avec y (Micha 1966112)

In such a case a philologist will conclude that the tradition is contaminated Sodid these philologists

[Le ms drsquoAnnonay] confirme lrsquoexistence de ces groupes intermeacutediaires imputablesagrave une tradition neacutecessairement contamineacutee (Micha 1966121)

Einzelne Schreiber von ihren Auftraggebern mit den Vorlagen versehn [hatten]zufaumlllig beide Auflagen vor sich und [folgten] ihnen nach Gutduumlnken falls nichtgroumlszligere Luumlcken der Vorlage aus einer anderen der 2 Redakzion angehoumlrigen Hsausgefuumlllt werden muszligten [] (Foerster 1910 lxxiv)

Contamination

If there is question of contamination the premise is always that more thanone version of a text was present in a scriptorium or library Of course this

Margot van Mulken

is always the case when a new text is produced after transcription there arealways two copies of the same text in one location However there are twopossible concepts of contamination

1 Simultaneous contamination presupposes that the copyist used severalsources simultaneously to produce a new text When this is the case it isonly logical that the copyist did so intentionally he compared readingsand made choices The role of the copyist in the reproduction processtherefore was very important the choice of a particular reading is a well-considered and deliberate choice since the copyist judged that determiningthe lsquocorrectrsquo reading was hyper-important

2 With successive contamination the copyist refers to different sources oneafter another and in this case it is improbable that he does so consciouslyThe copyist is unaware of the fact that he is using more than one text asexemplar he is unconscious of the differences between the versions Theexplanation for the fact that he could have had several exemplars at hisdisposal lies in the transcription process group production was not unfa-miliar in the Middle Ages when several copyists collaborated to produce anew text and when copies were produced by quire After transcription allthe quires were gathered and stitched Since copyists did not distinguishbetween old and new quires it became possible that newly transcribedquires were gathered together with quires of the original exemplar

In addition it was possible that several exemplars of a text were present in alibrary already at the beginning of the transcription In such a case successivecontamination could occur when a copyist unaware of the differences betweenthe exemplars switched exemplars in the transcription phase

In three manuscripts of the Cligeacutes tradition (P C and S) paleographers havedistinguished the presence of more than one hand (Stewart amp Luttrell 1993)They are all manuscripts dating from the Later Middle Ages So at least forthese manuscripts collaboration of copyists has been established

A combination of both types of contamination is of course possible butit remains difficult to prove the existence of simultaneous contamination Theproblem with successive contamination is only slightly smaller the shift in re-lationships often occurred at a stage prior to the extant manuscript when (oneof) the ancestors of the extant manuscript may have been subject to successivecontamination In such a case contamination can only be proved indirectlyWhat remains evident is that successive contamination always betrays itself bythe occurrence of a sudden change in relationships


The concept of contamination (and the type the philologist decides it be-longs to) depends heavily on the view of text one ascribes to a copyist whowrites in the vernacular In the case of the Cligeacutes I find it highly improba-ble also in view of the large number of singular readings in every extant copyin this tradition that the copyists were so preoccupied by the exact lsquocorrectrsquoreading of a theme that they would have wanted to refer to several exemplars2

The phenomenon of lsquomouvancersquo and the liberal conception of text were gener-ally accepted for the production of vernacular texts in contrast to other genres(Zumthor 1972)

The consequence of this presupposition as Foerster himself suggests isthat several passages in the Cligeacutes require the drawing of different trees Suc-cessive contamination is an option that needs to be taken seriously in order tosolve the contamination in the Cligeacutes

The quire separator

Thanks to the work of Evert Wattel (Free University Amsterdam) it is nowpossible to determine quite exactly at what places in the tradition a shift inrelationships occurs since a lsquosuddenrsquo change in relationship can be detected bythe presence of contradicting quadruples (see Note 1) (Wattel amp Van Mulken1993 Wattel this volume Den Hollander this volume) In the case of a newaffinity manuscripts will regroup into new mutual relations Between shiftsin relationships steady relations can be detected by the presence of consistentquadruples

According to this quire detector the Cligeacutes presents two important shiftsin relationships

1 The first consistent part covers the beginning of the tradition up to verse1250 (Structure I)3

2 The second part covers verse 1250 up to verse 4800 (Structure II) 4

3 The third part covers verse 4800 to the end (Structure III) 5

The relation shifts occur around verses 1250 and 4800 It is mss B and S whichare responsible for the first shift they change places in the region of verse 1250It is striking that this shift actually occurs at a stage where a change of quirescould have taken place the length of the first part coincides with the numberof verses in a quire consisting of 16 pages (ie 8 folios) with two columns of3840 verses each (ie 1216ndash1280 verses) This unit is among the most common

Margot van Mulken

C A

R S P B

Figure 2 Structure I Cligeacutes vv 1-1250

C A

R B T P S

Figure 3 Structure II Cligeacutes vv 1250-4800

C A

R

B

T

P

S

Figure 4 Structure III Cligeacutes vv 4800-end

quire partitions in the Middle Ages and the extant manuscript S has the samedistribution

In the region of verse 4800 manuscript P changes places with a non-transmitted intermediary This second change corresponds more or less witha quire of 24 pages (ie 12 folios) with 3 columns of 5960 verses each (ie4248ndash4320 verses)6 This is a less current format but it is precisely the formatof manuscript P in the Cligeacutes tradition

It is therefore conceivable that in the production phase of (ancestors of)manuscripts S and P quire shifts occurred and that these shifts are responsiblefor the multiple stemmata This can only be established by careful codicological


Table 2

1567Einz li est touz li sens foiz Einccedilois li est li sens faillizSi que pres an est amuiz Si que pres qu il nrsquoest amuizMss ap Mss bcrsT

research into the extant manuscripts ndash and it remains possible that it was theancestors of these manuscripts which were subject to quire shifts and not themanuscripts themselves

After having determined these consecutive structures I recalculated theconsistency within the trees and it appeared that now in stead of 28 51of the 370 variants could be preserved This still implies a refutation of 49of the variants which is of course quite considerable Every variant that hasto be dismissed should therefore be carefully examined and evaluated in thelight of parallelism We shall not discuss all the contradictory readings butmany of the important conflicting variants concern cases of lsquorich rhymersquo Forexample Table 2

We see here that manuscripts A and P have a rich rhyme lsquofoiz amuizrsquo Ac-cording to Meacutela this must be an original reading (ldquo[L]rsquoaccord AP justifie lacorrection [de C] pour une rime plus richerdquo) (Meacutela 1994136) However theproliferation of rich rhymes is generally thought to be a posterior trend In thisrespect a remark of Foerster is relevant ldquo[] daher auch die reichen Reime derspaumlteren schlechteren Hssrdquo (Foerster 1910 lxxiii) What is clear is that cases ofrich rhyme are likely to have been subject to parallelism and that they shouldtherefore not be included in the variant list

In spite of the large number of variants that have to be dismissed I con-sider the three structures as relevant and corresponding to actual underlyingrelation shifts

Orientation

The consequence of the assessment of three underlying structures is that theother phases in the stemmatological process according to the Three LevelMethod are also to be multiplied by three And this implies that for eachstructure the orientation and intermediate nodes must be determined

In earlier days the orientation phase served as a basis to reconstruct theoriginal text of the author Nowadays such an attempt is generally consideredunrealistic and too far-fetched The orientation phase is however still relevantin the study of the text production process by orienting the structures we ob-

Margot van Mulken

Table 3

0313tuit li baron les esgardoient et li baron forment se taisentpor ce que biax et genz les voient que li vallet trestuit lor plesentcar li vaslet molt lor pleisoient por ce que biaus et genz les voientne cuident pas que il ne soient ne cuident pas que il ne soientmss ab mss cprs

Table 4

1027et srsquoil nrsquoaimme ne nrsquoa ameacute et srsquoil nrsquoaime ne nrsquoa ameacutedonc ai ge en la mer semeacute donc ai ge en lrsquoareine semeacutemss ab mss crT (p s)

tain a better insight into the direction of the pedigree In order to do this Itried to determine the original reading of some of the more important vari-ants Here I will discuss some of the variants that constitute the arguments forthe final orientation

In Table 3 we see that mss A and B share a reading that does not reflect apossible original reading (the intersection of verses 0313 and 0315 producinga repeated rhyme on -oient) whereas the other manuscripts present a morepoignant rhyme The reverse is true for the next variant in Table 4 where Aand B share a probably original rhyme whereas the other manuscripts have areading that does not refer to the preceding passage of the lsquosecret of the searsquo inverses 548-564 (cf Meacutela 1994104)

The orientation of the first structure must therefore not be looked for inthe group AB nor in the group CPRST and this implies that I located theorientation on the edge between manuscripts AB and CPRS(T) Since T onlycommences a few verses before the start of Structure II because its first quire islacking it is difficult to determine the position of T in this structure I decidedto leave T out of the first Structure

In the second part we find the readings as presented in Table 5The versions in manuscripts APT though not entirely identical are similar

enough to group them together confronting the other group BCR The ver-sion in the first group is far more lively and detailed and can be consideredarchetypal

The modification of lsquoiert toz jorz establersquo to lsquoest si veritablersquo (see Table 6)is perhaps semantically cognate but the fact that four words were modifiedindicates that a case of parallelism is improbable


Table 5

1744car de bien ferir se travaille va un chevalier envairel plus espeacutes de la bataille si le fiert par si grant airvet ensi ferir un gloton que mort jus des arccedilons lrsquoabatque ne li valut un boton si qursquoil ne se plaint ne debatne li escuz ne li haubers (br) (c)ne li valut un cendal persms a (ms s pareisin)(mss pT krsquoen terre ne lrsquoemporteenvers)

Table 6

3118cest amors ne fu pas resnable ceste amor ne fu pas renablemes la moie iert toz jorz estable mais la moie est si veritablemss ab mss bprsT

Table 7

5725onques meis si male goleacutee onques meis si male goleacuteene pois tu doner au monde ne pot la mors haper au mondemss acrs mss bp

In part III in verses 5725-7 (see Table 7) we see that manuscripts BPpresent a reading that cannot possibly be original since it occurs in a pas-sage where Death is directly addressed and the third person singular (lsquopot lamorsrsquo) is a sudden stylistic change of perspective The wordplay with lsquomorsrsquo inthe sense of lsquodeathrsquo and lsquobitersquo in the context of lsquogoleacuteersquo and lsquohaperrsquo must be arefection

In the second and third parts of the MS tradition in Structures II andIII the orientation must be located on the edge between AS and the othermanuscripts

Conclusion

We see then that the orientation for each structure must be looked for in theneighbourhood of manuscript A the Guiot copy This is not really surprisingsince Micha and Foerster came to the same conclusion with regard to the gen-eral quality of the readings and in view of the relative age of the manuscript

Margot van Mulken

(second quarter of the 13th century) However Foerster prefers manuscript Sin the case of the Cligeacutes He motivates his choice as follows

Einmal ist es die Hs S die Abschrift eines recht (auch des Franzoumlsischen) un-kundigen Schreibers nach einer guten Vorlage die an manchen Stellen wiederso schlechte und wertlose Lesarten bietet daszlig man nicht begreift wie sie in densonst so guten Text kommen konnten [ ] Hier wird (meiner Wertschaumltzungder Hs entsprechend) meist an S unter allen Umstaumlnden gegen die andern Hssfestgehalten (Foerster 1910 lxxvndashvi)

Micha on the contrary continued to consider the Guiot copy as the bestmanuscript in the tradition of Beacutedier and used it as manuscrit de base forhis edition of 1957 Similar arguments led Lutrell and Gregory to edit theGuiot copy in 1993 just as Poirion decided to edit Guiot in 1993 It is truethat manuscript A is a remarkable manuscript but in Van Mulken (2002) Idemonstrated that a picardian predecessor must be inferred in the case of theCligeacutes The version of the Cligeacutes as laid down in Guiotrsquos copy is coloured bydistinct picardisms although the champenois provenance of the manuscriptremains undisputed (one of) the ancestor(s) of Guiot must have been a pi-cardian manuscript But we also know that Chreacutetienrsquos dialect was originally achampenois dialect This implies that the version of Cligeacutes in manuscript Afirst travelled to Picardy and then returned to Champagne to be transcribedin the Guiot copy In other words even the best manuscript in a tradition onthe grounds of quality and age may have had a considerable history before itbecame the best manuscript

The dialectological argument confirms the fact that manuscripts in thevernacular had a complicate production history as demonstrated by the stem-matological analysis of the Cligeacutes where the use of the quire separator throwsan important light on the contemporary view on vernacular texts in the MiddleAges Not only the Perceval but also the Cligeacutes proves to be a lsquorichrsquo tradition

Notes

A quadruple is the minimal unit in a Type-2 variant it covers two similarities and a differ-ence A variant such as lsquoabrscprsquo can be broken down into abcp arcp ascp brcp bscprscp The advantage of quadruples is that no more than two Type-2 variations exist thatcan be in contradiction with the quadruple lsquoabcprsquo can only be in contradiction with lsquoacbprsquoand lsquoapbcrsquo If a manuscript tradition is closed and linear this implies that for each quadru-ple the score for the competing quadruples will be zero or next to zero (depending on the


degree of inconsistency admitted by the philologist) Thanks to the table of quadruples thephilologists can see at a glance whether a tradition is closed or not

Foerster even goes so far as to presuppose an authorized review of the text by Chreacutetien(Foerster 1910 lxxiv) Micha thinks several copyists had more than one exemplar at theirdisposal (Micha 1966114)

All verse references follow CFMA Micha 1957 The following important variants are inaccordance with the stemma structure of the first part

abcprs 0022 0176 0290 0297 0313 0331 0430 1028abpcrs 0044 0054 0327 0432 0631 0905 0973abpscr 0527 0724 0806 1083

The following variants are in accordance with Structure II

asbcprT 1190 1380 1386 1409 1487 1535 1572 1612 1640 2368 26282 2666 28002987 3063 3086 3118 3162 3362 3555 3602 3743 3948 4133 4380 44464695

apsbcrT 1344 1590 1613 1633 1638 2360 2512 2628 2945 3105 3420 3478 35453631 3651 3657 3776 3824 4400 4476 4607

apsTbcr 1151 1201 1244 1250 1285 1330 1348 1403 1521 1544 1666 1719 17221741 1744 1745 1754 1797 2247 2394 2405 2428 2588 2601 2679 26812703 2726 2836 2919 2921 3026 3253 3430

abpsTcr 1296 2402 3106 3237 40822 1232 1244 1322 1544 1602 1719 2375 23942428 30862 3577 37072

The following important variants are in concordance with Structure III

asbcprT 5309 5560 5773 6040asTbcpr ndashcrabpsT 5693 5926 6084bpcprsT 4780 4769 4933 5021 5497 5699 5700 5715 5725 6170

There still remains a difference of some 480 verses to be explained Since the exact tran-sition of an underlying structure into another cannot be detected due to lack of variantsat the precise border of a structure it is very well possible that a different combination ofunderlying quires must be presupposed

References

Dees A (1988) ldquoAnalyse par lrsquoordinateur de la tradition manuscrite du Cligeacutes de Chreacutetiende Troyesrdquo In D Kremer (Ed) Actes du XVIIIe Congregraves International de Linguistique etde Philologie Romanes (pp 62ndash75) Tuumlbingen Niemeyer

Gregory S amp C Luttrell (1993) ldquoThe Manuscripts of Cligeacutesrdquo In Keith Busby TerryNixon Alison Stones amp Lori Walters (Eds) Les Manuscrits de Chreacutetien de Troyes TheManuscripts of Chreacutetien de Troyes (pp 41ndash48) Amsterdam Rodopi

Hollander A A den (1997) De Nederlandse Bijbelvertalingen 1522ndash1545 Nieuwkoop DeGraaf

Micha A (1966) La tradition manuscrite des romans de Chreacutetien de Troyes Genegraveve Droz

Margot van Mulken

Mulken M van (1993) The Manuscript Tradition of the Perceval of Chreacutetien de Troyes astemmatological and dialectological approach PhD Dissertation Amsterdam

Mulken M van (1999) ldquoLes changements de parenteacute dans le Cligeacutes de Chreacutetien de TroyesrdquoAtti dei Convegni Lincei 151 105ndash114 Roma Accademia Nazionale dei Lincei

Mulken M van (2002) ldquoLe manuscript de Guiot et les influences picardes dans le Cligeacutesde Chreacutetien de Troyesrdquo In H Jacobs amp L Wetzels (Eds) Liber Amicorum BernardBichakjian Nijmegen University Press

Nixon T (1993) ldquoRomance Connections and the Manuscripts of Chreacutetien de Troyesrdquo InKeith Busby Terry Nixon Alison Stones amp Lori Walters (Eds) Les Manuscrits deChreacutetien de Troyes The Manuscripts of Chreacutetien de Troyes (pp 17ndash26) AmsterdamRodopi

Salemans B (2000) Building Stemmas with the Computer in a Cladistic Neo-LachmannianWay PhD Thesis Katholieke Universiteit Nijmegen Nijmegen University Press

Wattel E amp M van Mulken (1996) Shock Waves in Text Traditions ndash Cardiograms of theMedieval Literature In P van Reenen amp M van Mulken (Eds) Studies in Stemmatology(pp 105ndash121) Amsterdam Benjamins

Zumthor P (1972) Essai de poeacutetique meacutedieacutevale Paris Seuil

Editions

Foerster W (1910) Kristian von Troyes Cligeacutes Textausgabe mit Variantenauswahl Einlei-tung Anmerkungen und vollstaumlndigem Glossar Halle Niemeyer3

Luttrell C amp S Gregory (1993) Cligeacutes ndash Chreacutetien de Troyes Bury St Edmunds StEdmundsbury Press

Meacutela Ch (1994) Chreacutetien de Troyes Cligeacutes Edition critique du manuscrit BN fr 12560Lettres Gothiques Livre de Poche

Micha A (1957) Les Romans de Chreacutetien de Troyes Cligeacutes eacutediteacutes drsquoapregraves la copie de Guiot(Bibl nat fr 794) Les Classiques Franccedilais du Moyen Age Paris Champion

Poirion D (1993) Œuvres Complegravetes de Chreacutetien de Troyes Paris Gallimard

P II

Textual variation

Genealogy by chance

On the significance of accidentalvariation (parallelisms)

Ulrich SchmidKirchliche Hochschule Bethel Bielefeld Vrije Universiteit Amsterdam

Introduction

Relationships between witnesses of a given text tradition are usually defined bythe variant readings they share Structures like lsquogroupsrsquo or lsquoclustersrsquo or lsquofami-liesrsquo appear and ideally an overall stemma of the tradition can be establishedby means of the kinship-revealing process of comparing the variant readingsthat are displayed by the tradition Such a process can be executed in dif-ferent ways No matter which way is chosen usually one presupposition isaccepted by all Agreement in reading is agreement in ancestry Although wemay consider some variant readings or even some types of variant readings asnot very impressive with regard to their kinship-revealing nature we usuallydo not exclude them from the body of variant readings which is registeredcounted and processed This is all the more true for variant readings thatare not mere lsquoorthographicalsrsquo or lsquononsense readingsrsquo1 but sensible readingsproviding a readable and intelligible text If variant readings of the latter typeare to be noted and found in more than one witness we must not excludethem from our data processing Quite to the contrary This type of evidenceis the backbone of every genealogical or stemmatological work Agreement inreading is agreement in ancestry Yet the problem of accidental variation ieidentical variant readings created purely by chance has not escaped the eye oftextual scholars

Most recently B J P Salemans using the designation parallelism to de-scribe the phenomenon of accidental variation has treated the issue (Salemans200064ndash71) He formulates the ldquotext-genealogical basic rule A genealogical

Ulrich Schmid

(or relationship-revealing) variant is a textual difference that fits well and in-conspicuously in a text versionrdquo (Salemans 200064) The rule is intended toleave the genealogically relevant information exclusively with readings that donot attract the attention of a copyist nor hint at his or her orthographical inter-punctional dialectal etc proclivities Additionally Salemans describes variousforms of parallelisms which include the exchange of synonyms on the level ofindividual words exchange of tenses numbers or cases on the level of indi-vidual lexems and the ldquoabsence of small (highly) frequently used wordsrdquo2 Inhis view textual phenomena of those types should not be used for genealogicalpurposes Of course it is one thing to theoretically assess the liability of certaintypes of variant readings to accidental variation and it is quite another thing toactually observe real parallelisms Therefore it is very interesting to note thatSalemans from within his own deductive framework is even able to observethese kind of phenomena after processing the textual information collectedfrom a body of fourteen versions of the late medieval vernacular text Lanceloetvan Denemerken The way he proceeds is based upon complete (electronic)transcriptions of all fourteen versions Then a computer program designed toapply the text-genealogical rules and criteria as developed deductively filtersout all the information that is considered to be genealogically unfit accord-ing to these criteria The remaining body of information is then used to build astemma A large part of Salemansrsquo study is devoted to the evaluation of his text-genealogical rules and criteria by comparing the results based upon the varioustypes of unfiltered variant readings with the lsquotruersquo stemma based upon the fil-tered set of data From within his own framework Salemans is able to vindicatemost of his lsquoprejudicesrsquo against certain types of variant readings that he has dis-carded from the onset Regarding the omissioninterpolation of small wordseg Salemans is now even more confident in defining that type of words ldquocon-taining four or fewer phonemes and belonging to other word categories thannouns and verbsrdquo (Salemans 2000294) Nevertheless he does not consider hisfilters and their evaluation to be lsquouniversalrsquo Other text traditions may requireadaptations and refinements (Salemans 2000297)

Undoubtedly Salemans has made a good case for critically and systemati-cally reviewing the types of variant readings that are being used in genealogicalstudies Moreover he certainly pinned down the ones that are predominantlysuspect of accidental variation However his case is built by means of data de-liberately manipulated according to certain lsquoprejudicesrsquo set out before From ascientific viewpoint this is of course perfectly legitimate even necessary But itshould be interesting to have a look at the issue of parallelism based upon datathat have not been selected and manipulated for the purpose Ideally we should

Genealogy by chance

want to find a set of data that had been selected and already used for specificgenealogical purposes because it was considered as allowing genealogically safeconclusions Upon reviewing those data part of them should be unmistakinglyidentifiable as purely accidental Then we should be able to (a) assess the typesof variant readings that happened to be accidental in a certain constellation ofwitnesses and (b) relate the subset of parallelistic readings to the entire body ofdata that had been selected for genealogical purposes among those witnessesIf it turned out that quantity and quality of the demonstrably parallelistic read-ings does actually outweigh the testimony of the non-parallelistic readings weshould have an empirically based case for Genealogy by Chance Although itseemed extremely unlikely that anyone should come across such a case thanksto the infallible fallibility of human nature I am able to provide one The caseis built upon the fact that in 1919 a scholar published a list of readings drawnfrom a 9th-century Gospel Harmony manuscript which was intended to linkthe basically Vulgate Harmony text under scrutiny to a particular allegedly con-siderably older textual tradition The text that was used however was not thatof the 9th century historical artefact but that of its 19th-century edition Uponcomparison of the edition with the real manuscript a considerable number oferrors showed up which were partly incorporated in the list of readings men-tioned thus linking the 19th-century counterfeit but not the real thing to theolder textual tradition

In what follows I will elaborate the case in three steps First of all the schol-arly debate in which the mentioned Harmony text is situated will be recon-structed Secondly the parallelistic readings will be identified and discussedFinally some of the implications of this case will be pointed out

Medieval Latin Gospel Harmony manuscripts ndash Reconstructing thescholarly debate

Gospel Harmonies are texts which present the materials of the four separateGospels of the New Testament as one single coherent narrative Gospel Har-monies are as old as the second half of the 2nd century when a certain Tatiancomposed his famous Diatessaron3 of which not a single direct witness sur-vives The oldest extant Gospel Harmony is in fact preserved in a famous LatinVulgate manuscript the so-called Codex Fuldensis4 Codex Fuldensis containsthe entire New Testament with a Gospel Harmony replacing the usual fourcanonical Gospels Victor Bishop of Capua5 commissioned the manuscriptfinally approving it on April 12 547 CE6 In the 8th century the famous Anglo-

Ulrich Schmid

Saxon missionary St Boniface acquired the manuscript and donated it to thethen newly founded monastery of Fulda whence it derives its name TodayCodex Fuldensis still resides in Fulda now in the Landesbibliothek The old-est copies of the Fuldensis Harmony date from the 9th century Today we havegot a total of four of these 9th century manuscripts and the question of howclosely they are related to Codex Fuldensis is a matter of scholarly debate

Apart from Codex Fuldensis two of the four Harmony manuscripts fromthe 9th century have been edited The editors were Germanists who stud-ied two of the oldest texts in vernacular German One of the two Harmonymanuscripts the 9th century Codex Sangallensis7 actually contained one of theoldest texts in Old High German namely a translation of the Gospel HarmonyTechnically speaking St Gallen Stiftsbibliothek Cod 56 is a Latin-Old HighGerman bilingual manuscript presenting on every page a column with theLatin Harmony text on the left and an Old High German version on the rightThe second manuscript the so-called Codex Cassellanus8 was edited becauseof 19th century research into the sources of the Heliand9 The Heliand is an OldSaxon poem again one of the oldest texts in that Germanic idiom retelling theGospel stories in the form of a Harmony Although the outline of the Heliandis basically the same as found in Codex Fuldensis yet for some of the textualdetails C W M Grein the editor of Codex Cassellanus felt that the latter wascloser to the Heliand than the former From the Germanistrsquos point of view itis absolutely crucial to identify as precisely as possible the sources of the twovernacular texts because this may lead to a better understanding of the trans-latorrsquos choices Moreover should it be possible to locate these sources in timeand space it will help to allocate the specific blend of the vernacular dialectas displayed by the mentioned texts to a certain area This of course is pri-mary evidence for compiling a map of the historic development of a languageIt comes as no surprise therefore that Germanists were very pleased to viewCodex Sangallensis as a direct copy of Codex Fuldensis as it attributed a placeof origin to the Sangallensisrsquo Old High German blend On the other hand withrespect to the intermediary Codex Cassellanus which separates the old CodexFuldensis from the Heliand the link between the Heliand and Fulda as its placeof origin seems not so strong Anyway most Germanists believed that CodexFuldensis was the ultimate source of all of their vernacular harmonised textseither directly ndash in the case of the bilingual Codex Sangallensis ndash or perhapsindirectly as with respect to the Heliand

These same three Harmony manuscripts Codex Fuldensis Codex Sangal-lensis and Codex Cassellanus are now studied from a different angle As alreadynoted the oldest and most famous Gospel Harmony Tatianrsquos Diatessaron has

Genealogy by chance

not been preserved in the original Scholars who were interested in reconstruct-ing the lost text studied Gospel Harmonies that had been preserved for possibledistant echoes from this text which must once have been very popular Dia-tessaronic scholarship developed the theory of a very ancient Old Latin iepre-Vulgate translation of Tatianrsquos Harmony which was composed in eitherSyriac or Greek This Old Latin Harmony was then textually adapted to theVulgate Textform which is exactly what Codex Fuldensis represents Apart fromthis however the Old Latin or less vulgatized Latin version of Tatianrsquos Diates-saron is thought to have exerted its influence on other parts of the Harmonytradition in Western Europe Contrary to what most Germanists believe Dia-tessaronic scholars suspected not Codex Fuldensis to be the ultimate source ofthe Western Harmony tradition but the Old Latin Harmony10 Consequentlythey looked for evidence for their guess by screening bilingual Codex Sangal-lensis Latin Codex Cassellanus and even the Heliand against Codex FuldensisThe idea was that every difference between Codex Fuldensis and the later textsweakens the position of Codex Fuldensis as the ultimate source of the WesternHarmony tradition Their idea was corroborated by the deviations from CodexFuldensis which could often be paralleled from known Old Latin text tradi-tions of the separate Gospels andor remote witnesses such as Syriac or Arabictexts or Church writersrsquo testimonies of the first centuries These deviationsthen suggested that some source other than the vulgatized Codex Fuldensismust have been operative namely the ancient Old Latin translation of TatianrsquosDiatessaron

After a time of reorientation on the Germanistsrsquo side when they tried tocome to terms with the new situation of a dethroned Codex Fuldensis the pen-dulum swung back The Germanist J Rathofer devoted detailed studies to therelation between Codex Sangallensis and Codex Fuldensis and concluded thatthe former was a direct copy of the latter11 Once again the Germanists arehappy to have their most interesting Old High German text located in timeand space to the mid 9th-century monastery of Fulda Rathoferrsquos successfuldemonstration was based on reconsidering the manuscripts anew with spe-cial focus on features that had not been readily available in the 19th-centuryeditions of Codex Fuldensis Codex Sangallensis and Codex Cassellanus namelypalaeographical data layout phenomena and marginal addenda Incidentallythis demonstrates that kinship-revealing features are also even decisively to befound outside the narrowly defined textual variants of a manuscript traditionAnyway Rathoferrsquos move away from the editions and back to the manuscriptsclears the stage for my demonstration of accidental variation resulting in iden-tical readings yet without any genealogical link whatsoever

Ulrich Schmid

An empirical case for accidental variation

The evidence for accidental variation

In 1919 H J Vogels published a study on a variety of Latin Harmony manu-scripts from different times (Vogels 1919) Part of his study was devoted tothe presenting of examples of textual differences between the oldest still al-most purely Vulgate manuscript Codex Fuldensis and the younger manuscriptspartly or considerably differing from Codex Fuldensis For his study Vogels usedthe editions of the manuscripts when available among them Greinrsquos edition ofCodex Cassellanus and of course Rankersquos famous edition of Codex FuldensisOn two pages Vogels presented a list of 44 readings where 9th-century CodexCassellanus deviates from 6th-century Codex Fuldensis (Vogels 1919128ndash129)The significance of this list for Vogels and other Diatessaron scholars lay in thefact that all of these deviations can be paralleled from either Old Latin wit-nesses andor remote witnesses such as Greek Syriac Arabic and Armeniantexts It was exactly this type of evidence that caused Diatessaronic scholarshipto postulate influence even descent from an ancient source representing a lessvulgatized text than Codex Fuldensis

Now when Rathofer as explained above made his move back to themanuscripts he noted in passing that especially Greinrsquos edition of Codex Cassel-lanus contains a large number of mistakes Rathofer counted 360 mistakes forthe Harmony text some of which even entered Vogelsrsquo list (Rathofer 1972348)This goes right to the heart of the problem After checking Vogelsrsquo 44 devia-tions of Codex Cassellanus from Codex Fuldensis against Greinrsquos edition and amicrofilm of Codex Cassellanus I detected that 24 of them are based on er-rors in Greinrsquos edition Vogels didnot make mistakes he simply recorded thedifferences between two 19th-century editions and assumed that they accu-rately represented the 6th and 9th-century manuscripts This highlights first ofall the usually neglected fact that modern editors of ancient manuscripts arebasically less or more accurate copyists of the manuscripts they edit In thatsense they not only contribute to the study of a manuscript tradition but withtheir errors they are part of the manuscript tradition itself This simple insightis usually obscured by the fact that modern editorsrsquo lsquomanuscriptsrsquo are extantin several hundred identical copies In a broader perspective however Greinrsquoserrors as unsuspectingly reproduced by Vogels severely question the receivedperception of manuscript genealogies First of all there can be little doubt thatVogelsrsquo conclusions based on the 44 readings are highly questionable for hethought they ruled out the possibility that Codex Cassellanus was a direct copy

Genealogy by chance

of Codex Fuldensis (Vogels 1919126) Then by consequence the fact that morethan 50 (24 out of 44) of those readings are not to be found in Codex Cassel-lanus but only in its 19th-century edition should cast severe doubts on whatthe edition actually pretends to be namely an edition ndash ie a direct copy ndash ofCodex Cassellanus Secondly and even more devastatingly the 44 readings werenot chosen randomly just because they were supposed deviations from CodexFuldensis Quite to the contrary they were carefully selected as genealogicallysignificant deviations from Codex Fuldensis because they could be paralleledfrom Old Latin and even more remote Syriac Arabic or Armenian witnessesThey should serve to make the point that some influence other than the stan-dard Vulgate text as displayed by Codex Fuldensis was operative within the LatinHarmony tradition The repercussion of this assumption is devastating sincewhatever may have influenced the copying process of Gospel harmonies in theearly Middle Ages it left exactly the same traces in a 19th-century copy of aGospel Harmony ie in Greinrsquos edition of Codex Cassellanus It is essential torealise that Vogelsrsquo findings remain valid because the 44 readings are obviouslyshared by the remote witnesses he adduced even though only 20 of them ac-tually belong to the actual manuscript he was aiming at But what to thinkof the other 24 readings that were produced within the 19th-century copy ofCodex Cassellanus All of them can also be paralleled from the same type ofremote witnesses If it is assumed that the 20 readings belonging to the ac-tual manuscript are due to the influence of a remote Old Latin text by thesame token this has to be assumed for the additional 24 readings of the 19thcentury edition as well This second assumption however is simply beyondimagination I can think of no condition under which a modern scholarlyeditor would consciously modify the text he or she is editing in such a waywithout mentioning anything about it If the idea of external influence is to beupheld unconscious or subconscious ways of infiltration must be looked forShould we suppose that Grein was influenced by some sort of rsquolocalrsquo text hegrew up with or was familiar with What could such a text have looked likein mid-nineteenth century Germany If he were a good Protestant Grein cer-tainly would have been exposed to Lutherrsquos translation of the Bible in Sundayservices and confirmation classes How likely is it that such a vernacular up-bringing would influence the editing of a Latin Gospel Harmony Even if onewere a Roman Catholic in those days which was probably the only way to growup with or be surrounded by a Latin Bible text the text should have been thatof the Clementine Vulgate not any Old Latin type of text In short it is highlyunlikely that unconscious or subconscious influence from a text contemporaryto the editor was responsible for the deviations from his exemplar But even

Ulrich Schmid

if such a process did take place the resulting contamination would have beencompletely coincidental and would thus not allow any scientific conclusion Imay therefore conclude that the deviations between the text of the 9th-centurymanuscript and its 19th century edition were not purposely created nor in anylikelihood due to influence from a lsquolocalrsquo text the editor was familiar with In-stead they were created ad hoc by the 19th-century editor while transcribingthe manuscript he edited by committing the usual blunders that every copyistof substantial portions of text is guilty of Then by consequence these de-viations between the manuscript and its edition hit any of their lsquoparallelsrsquo asadduced by Vogels purely by chance Since the proportion of accidental hits ismore than 50 (24 out of 44) there is nothing significant left with respect tothe rest In other words The 20 readings which are not based on errors in theedition might have hit their parallels by chance too

This brings the whole case full circle It must be kept in mind that thisdemonstration was performed on material not intentionally selected for thepurpose Quite to the contrary another scholar made the selection in full con-fidence of the kinship-revealing quality of the chosen variant readings Withall the parallels that he adduced he aimed at a genealogically lsquosafersquo conclusionwith respect to a certain constellation of witnesses Yet unexpectedly and un-intentionally he not only demonstrated that his conclusion was genealogicallyunsafe but once and for all that the genealogical method in itself is intrinsicallyvulnerable Not only is there empirical evidence that genealogically relevantinformation could have been created by chance In the case under discussionthe factual proportion of accidental variation did actually blow up the entiregenealogical conclusion based on the evidence presented Thus Vogelsrsquo geneal-ogy regarding Codex Cassellanus is completely arbitrary So let us now considerthe type of evidence the study of Codex Cassellanus and its 19th-century editiondoes provide us with

A short empirical survey of scribal errors and accidental variation

The following discussion is based on the evidence gathered from comparing amicrofilm of a 9th-century Gospel Harmony manuscript (Codex Cassellanus)with its 19th-century edition I do not aim to present an exhaustive collectionof scribal errors or of accidental variation (parallelisms) I would not even dareto think of compiling such lists I am interested in a rough grouping of thephenomena I encountered according to the criteria as outlined by Salemans(see above) by highlighting those phenomena he does not mention

Genealogy by chance

The selected evidence is of two kinds First of all I collated a sample of ninerandomly chosen pages from Greinrsquos edition of Codex Cassellanus12 against themicrofilm of the manuscript This was done in order to gain a general insightinto the overall distribution of errors as well as the types of errors evidencedSecondly I checked the 44 variant readings given by Vogels in his list of sup-posed deviations between Codex Fuldensis and Codex Cassellanus against themicrofilm of the manuscript

Spot checks on Greinrsquos overall error rateThe lemma (italics) gives Greinrsquos text The reading of the manuscript is in boldtype

a Grein p 130

Lk 14 visum est mihi] visum est et mihiJo 134 factum est nihil quod factum est In ipso vita erat]

factum est nihil bull Quod factum est in ipso vita erat

Although the manuscriptrsquos system of punctuation is different from modernsystems in this very prominent case Grein choose to ignore the capitalisedldquoQ(uod)rdquoand the lower case ldquoi(n)rdquo given by the manuscript as well as the raisedstop after ldquonihilrdquo and the normal stop after lsquoestrsquo I will not go into other punc-tuation questions But in this particular case Greinrsquos edition definitely alters themeaning of the manuscriptrsquos reading

Jo 14 et vita erat et lux hominum] et vita erat lux hominumLk 15 Herodis regis] Herodis recis lsquoregisrsquo corrected (scribe)

16 sine querela] sine quaeligrela

b Grein p 131 No error detected

c Grein p 132

Lk 145 dicta sunt ei] dicta sunt lsquoeirsquo supplied supralinear by scribe160 sed vocabitur] sed vobitur (sic)162 patri eis] patri eius163 pugilarem] pugillarem163 et nomen] est nomen164 dominum] deum

d Grein p 190 No error detected

(Grein choose to represent the numerals XCVIIII [Mt 1812ndash13] and X [Lk 158]given by the manuscript with lsquononaginta novemrsquo and lsquodecemrsquo)

Ulrich Schmid

e Grein p 191

Lk 1517 mercenarii] mercennarii13

1519 mercenariis] mercennariis1520 in misericordia] misericordia1521 mercenariis] mercennariis1522 in manum] in manu173 si peccaverit] attendite vobis si peccaverit173 lucratis eris] lucratis es (e)ri(s) supplied supralinear

(scribe)Mt 1817 si autem aecclesiam] si autem et aecclesiam

f Grein p 192

Mt 1821 quotiens peccavit] quotiens peccavit in me

g Grein p 259

Jo 2019 dum esset sero] cum esset sero

h Grein p 260

(Jo 2026 octo] VIII)Jo 2029 quia audisti me] quia vidisti meJo 2111 et cum tanta] et cum tanti

i Grein p 261

Jo 2117 Simon Johannis amas me] + contristatus est petrus quiadixit ei tertio amas me

2122 quid ad te] + tu me sequere

Our spot check reveals a total of 25 errors in nine pages (excluding the differ-ent representation of the numerals) involving about 40 words Extrapolatingfrom that figure to the 1315 pages of Greinrsquos edition adds up to a total of c365 errors involving about 584 words (c 28 errors per page) This figure isvirtually identical with the number of errors recorded by Rathofer All sortsof errors occur apart from (presumingly) printing errors (eg Lk 162632)and unrecorded corrections (Lk 1545 173) there are orthographicals (egLk 16631 Lk 15171921) punctuation (Jo 134) and a homoioteleuton er-ror (Jo 2117) Moreover we find two omissions of the highly frequently usedconjunction et (Lk 14 Mt 1817) two additions of small words (Jo 14 Lk1520) three substitutions of words (Lk 164 Jo 201929)14 and three changesof word forms (Lk 1522 Jo 2019 2111) This is in perfect agreement withthe types of readings that Salemans records as especially liable to parallelismHowever we find another three readings (out of a total of 25 ie 12) which

Genealogy by chance

are as far as I can judge not a type recorded by Salemans namely the omissionor addition of more than one word (Lk 173 Mt 1821 Jo 2122)15 Mt 1821 in-cludes the omission of a prepositional object (two words) Lk 173 (two words)and Jo 2122 (three words) include the addition of entire sentences Within thatcontext we may even add the homoioteleuton error (Jo 2117) which causedthe omission of nine words16

Greinrsquos errors spotted from Vogelsrsquo list (Vogels 1919128ndash129)a It must be remembered that Vogelrsquos list was intended to register differencesbetween Codex Cassellanus and Codex Fuldensis which could be paralleled fromother remote witnesses that is to say that the text of Codex Fuldensis is thepoint of reference In order to simplify comparison I reproduce Vogelsrsquo list-ings according to his conventions eg infin aut l neque (Grein p 151) meansCodex Cassellanus (according to Greinrsquos edition p 151) has aut where CodexFuldensis has neque However in every single case listed below the 9th-centuryartefact Codex Cassellanus does not deviate from Codex Fuldensis only the er-rors in Greinrsquos edition make it deviate Consequently all these readings arepurely parallelistic

Mt 44 dixit] + ei (Grein p 141)625 infin aut l neque (Grein p 151)156 infin et l aut (Grein p 179)175 om ipsum audite (Grein p 187)2231 infin domino l deo (Grein p 210)2414 om hoc (Grein p 227)2421 infin saeculi l mundi (Grein p 227)287 om ecce 1o (Grein p 256)

Mc 317 infin nomen l nomina (Grein p 146)819 om plenos (Grein p 185)

Lk 226 infin christum dominum l chr domini (Grein p 135)1318 infin regn caeli l regn dei (Grein p 171)143 om dicens (Grein p 199)1412 om neque fratres tuos (Grein p 200)1627 om enim (Grein p 197)2237 om hoc (Grein p 238)

Jo 119 om ad eum (Grein p 140)46 om sic (Grein p 181)430 infin et exierunt l exierunt (Grein p 182)1150 om homo (Grein p 220)

Ulrich Schmid

1228 infin glorifica glorificavi glorificabo l clarif (Grein p 222)146 et nemo l nemo (Grein p 237)

b The following readings need to be specified

Mt 201 om enim (Grein p 198)

Grein failed to note that rsquoenimrsquo was supplied supralinear by the scribe himself

Jo 141 inc et ait discipulis suis (Grein p 236)

Grein failed to note that the words were added by a different post 9th-centuryhand at the bottom of the page partly exceeding the usual layout of 28lines per page

When we now consider the character of the 24 errors from Greinrsquos edi-tion that Vogels paraleled in his list we find that orthographicals punctuationand nonsense readings are not included In agreement with Salemansrsquo criteriaof what is likely to be due to parallelism we find nine omissions of one word(Mt 201 2414 287 Mc 819 Lk 143 1627 2237 Jo 46 1150) thoughone omission involves a noun (Jo 1150) one omission of three words is (mostlikely) due to homoioteleuton (Lk 1412) Moreover we have got three addi-tions of one word (Mt 44 Jo 430 146) five substitutions of one word (Mt625 156 2231 2421 Jo 1228) of which in one case three different forms ofthe same word have been substituted by the pertinent forms of a potential syn-onym (cf Jo 1228) and three changes of word forms (Mc 317 Lk 226 1318)And yet three of the real parallelisms ie 125 of the total of 24 variant read-ings are not accounted for by Salemans Two of them involve the omissions oftwo words (Mt 175 Jo 119) one of which is actually the omission of a com-plete sentence This is in perfect line with what we already discovered in ourfirst sample (see above) In addition a new type of variant reading that is liableto parallelism occurred ie the addition of a whole sentence consisting of fourwords (Joh 141)

To sum up It is true that almost 90 of the real parallelisms presentedhere are more or less lsquotypicallyrsquo covered by Salemans Nevertheless the re-maining more than 10 send most disturbing signals Not only comparativelysimple omissions and additions of single words but omissions of compoundexpressions and sentences even the addition of an entire sentence can hit arsquoparallelrsquo purely by chance Thus the issue of parallelism is considerably morecomplicated than hitherto thought of

Genealogy by chance

Concluding observations

I should like to conclude this paper by adding two observations one on whatmight have contributed to the devastating results of Vogelsrsquo genealogy and oneon using typologies of variant readings with respect to issues of parallelism

1 To my mind it is a most remarkable fact that Vogelsrsquo list of 44 variantreadings from Codex Cassellanus for which he gave parallels from remotewitnesses provides us with a full empirical case of Genealogy by Chance Itis important to keep in mind that our findings did not simply raise doubtsas to whether or not Vogelsrsquo overall genealogical conclusions might still bevalid because some of his readings could be proven to be accidental hitsThe case is far more serious Since the proportion of accidental hits is morethan 50 Vogelsrsquo entire case crumbles to pieces This proves that the com-parison he performed must have been completely inadequate What canhave contributed to this devastating result

1a First of all the entire Harmony text consists of about 40000 words17 Uponcomparison of two Harmony manuscripts Vogels produced out of thema list of 44 variant readings involving 63 words that he paralleled fromremote witnesses These proportions could simply be too low since theyinvolve only 016 of the actual text

1b Secondly Vogels comparison base consists of witnesses of different genresand languages and varying degrees of preservation18 This alone shouldqualify as a big question mark as to whether Vogels comparison base canindeed be accepted as forming one single coherent text tradition at all Acloser look at the witnesses which were used to form the comparison basereveals that they are taken individually not consistent in providing paral-lels for all of the 44 variant readings Only as a group consisting of about30 members do they testify to all 44 of them However if we split up thegroup we find that the highest number for a single witness is 13 out of44 this witness is Old Latin Gospel manuscript b If we combine the tes-timony of the four individual Syriac witnesses (sycscpalp) we reach 14 outof 44 However the usual frequency for individual witnesses is 6 to 10 outof 44 variant readings We may conclude then that there is a considerablelack of coherence among the witnesses that form the comparison base

1c Thirdly the witnesses that were used to form the comparison base were se-lected because of their historical remoteness with respect to the two VulgateGospel Harmony manuscripts under review Moreover their remotenesswas also taken to indicate the overall rareness of their shared readings when

Ulrich Schmid

they deviate from what is considered to be the standard Vulgate text Al-though not explicitly stated by Vogels this combination of remoteness ofwitnesses and alleged rareness of readings could be understood as somesort of antidote against the problems of (a) low numbers and (b) incoher-ent comparison base The implicit reasoning could have been readings thatare only rarely testified to by scattered testimonies must be lsquosurvivorsrsquo of anolder Textform which successfully resisted the assimilation process to thenow dominant Textform Thus they are safe even if they are few in num-ber and incoherently present in a rather diverse array of witnesses Vogelsrsquoactual case of the 44 readings from Codex Cassellanus suggests a differentinterpretation shared readings which are only few in number and testifiedto by historically remote witnesses appear to be more likely than not ac-cidental hits or lsquofalse positivesrsquo The reasons for such an impression needto be further explored which is beyond the scope of this paper Howevermy guess is that it has to do with the sheer quantity of the text traditioninvolved The New Testament Gospels display probably the largest text tra-dition ever produced in the Western Hemisphere with tens of thousandsof mostly unexplored manuscripts in Latin alone Some amount of co-incidental variation (accidental hits parallelism) seems to be inevitableHowever one may wonder whether a comparison base which is selected ina different way would provide better results It seems reasonable to assumethat with witnesses of which the historical relations (in time and place)other than shared variant readings provide a link with the two Harmonieseven a small fraction of such readings might prove to be significant

2 There can be little doubt that the vast majority of the parallelisms discov-ered in our sample are accounted for by Salemans There are howeversome that are not accounted for ie omissionaddition of more than oneword and most notably of entire sentences Should we now simply banthose types of variant readings like the other ones Salemans already ex-cluded in order to stay on the safe side in genealogical studies The problemwith such an approach is threefold

2a First of all the fact that variant readings which are assigned to certain typescan be shown to be parallelistic does not mean that the reverse is true aswell namely that all or even most of the variant readings belonging tothose types are indeed parallelistic Quite to the contrary more often thanexpected even in Salemansrsquo own analysis some highly suspect types of vari-ant readings seem to fall in neatly with the lsquotruersquo stemma19 In other wordsTheir assembled testimony may not effectively obstruct the stemma basedon the filtered set of data

Genealogy by chance

2b Secondly to the extent that the present study adds new types of readingsto the pool of possibly parallelistic readings severe doubt is cast upon theremaining set of data With only 24 readings we were able to identify ad-ditional cases of parallelistic readings which Salemans did not account forA larger sample might have provided even more evidence and could haveadded different types This in turn enhances the possibility that we mightnot be able to find a single type of variant readings that is entirely free fromliability to parallelism

2c Thirdly let us assume that there is indeed no completely safe line to drawbetween those types of variant readings that are prone to parallelism andthose that are never exposed to the threat The challenge we are facing isthat genealogical study must either surrender or intelligently use data thatin all likelihood will never be completely weeded of potentially parallelis-tic inclinations One solution could be to determine statistical correlationsbetween individual subsets of variant readings and their liability to par-allelism as observed in a given text tradition in order to weigh them ac-cordingly A typology of variants like the one proposed by Salemans iscertainly helpful to identify appropriate subsets as well as the usual sus-pects ie those types of variants that seem to be more likely exposed toparallelism than others But it should not be used to eliminate data with-out proper evaluation of their statistical inclination towards parallelism ina given text tradition Otherwise we should be left with individual sub-sets of data in which the differing parallelistic inclinations are completelyunaccounted for

Notes

lsquoNonsense readingsrsquo are readings that leave the text unintelligible at a certain passage

Salemans (200070) see also the list on p 67ndash68 n 44

Cf Petersen (1994)

Ranke (1848) Descriptions of the manuscript can be found in Scherer (19056ndash12) andFischer (1963519ndash600 esp 545ndash557)

A town in Northern Italy in the region of Campania (medieval liburia)

Fischer (1963546)

Cf the recent edition Masser (1994) Descriptions of the manuscript are given inMasserrsquos edition as well as in an older edition by Sievers (1892) An extensive study of themanuscript (including photographs) was presented by Masser (1991)

A recent description is given in Kahlfuss (199437ndash39)

Ulrich Schmid

The manuscript was edited as an appendix to Grein (1869127ndash262)

The first scholar to issue that theory was Th Zahn see Zahn (1881300ndash303) and Zahn(189487ndash107)

Rathofer (1971 1972 1973)

These include the first and the last three pages and three pages from the center ie pp130ndash132 190ndash192 259ndash261 out of a total of 1315 pages

On p 217 (Jo 1012) Grein records the different spelling of mercen(n)arius in a note

The substitution of vidisti with audisti is certainly supported by the very similar graphicrepresentations of the two words in Codex Cassellanus The ms like most early medieval mssdoes not distinguish between the vowel lsquoursquoand the semi-vowel lsquoursquo = lsquovrsquo the back stroke of thelsquoarsquo is very upright with a rounded foot towards the right thus closing the gap to the nextletter Therefore the two letter-combinations lsquoau-rsquo and lsquouirsquo are basically both dominatedby three vertical strokes As far as I can see Salemans does not place much emphasise ongraphic similarities between words as likely causes for accidental variation

Salemans discusses the ldquoaddition (or interpolation) and omission of complete versesrdquo(Salemans 2000101ndash102 294ndash295) This is very likely due to the fact that the text heresearched is a poetic piece of literature

As far as I can see Salemans does not discuss issues of homoioteleuton (saute des yeux)

In what follows I make use of results that will be published in my book Unum ExQuatuor Genealogie und Rezeption einer Lateinischen Evangelienharmonieuumlberlieferung TheHarmony text in Patrologia Latina 68 contains 40770 words The edition is based on amanuscript from the 13th century Spot checks revealed that the later versions are slightlyexpanded (400ndash1000 words) compared to the old version from the 6th to 9th centuries

The witnesses adduced include Gospel Harmonies a Harmony commentary a fewChurch father testimonies and Gospel manuscripts out of which some are fragmentaryin at least five different languages

See eg Salemans (2000257) ldquoEvaluation of the Word Category Conjunctionsrdquo Al-though there are examples of parallelisms with the medieval Dutch conjunction ldquoenderdquo thisdoes obviously not hold true for other conjunctions ldquoWe can offer many examples in whicha conjunction is added at the beginning of a verse However these interpolations all agreewith our stemma The conclusion at the end of this section must be stated cautiously be-cause we did not find clear convincing examples to show that conjunctions are parallelisticYet I am still convinced that they are parallelistic in view of the ease with which they havebeen interpolated in the Lanseloet textsrdquo (pp 257ndash258)

References

Fischer B (1963) ldquoBibelausgaben des fruumlhen Mittelaltersrdquo [La Bibbia nellrsquoalto Medioevo]Settimane di studio del centro italiano di studi sullrsquoalto medioevo 10 Spoleto

Grein C W M (1869) Die Quellen des Heliand Cassel

Genealogy by chance

Kahlfuss H-J (1994) Die Handschriften der Gesamthochschulbibliothek Kassel Landes-bibliothek und Murrhardsche Bibliothek der Stadt Kassel H-J Kahlfuss (Ed) Bd 11Manuscripta Theologica Die Handschriften in Folio (bearbeitet von K Wiedemann)Wiesbaden

Masser A (1991) In Nachrichten der Akademie der Wissenschaften in Goumlttingen IPhilologisch-Historische Klasse Jahrgang 1991 Nr 3 Goumlttingen

Masser A (1994) Die lateinisch-althochdeutsche Tatianbiblingue Stiftsbibliothek St GallenCod 56 SAHD 25 Goumlttingen

Petersen W L (1994) Tatianrsquos Diatessaron Itrsquos Creation Dissemination Significance andHistory in Scholarship VigChrS 25 Leiden

Ranke E (1848) Codex Fuldensis ndash Novum Testamentum Latine Interprete Hieronymo exmanuscripto Victoris Capuani edidit E Ranke (Ed) Marburgi et Lipsiae

Rathofer J (1971) ldquoZur Heimatfrage des althochdeutschen Tatian Das Votum derHandschriftenrdquo Annali (instituto universitario orientale sezione germanica) 14 7ndash104

Rathofer J (1972) ldquolsquoTatianrsquo und Fulda Die St Galler Handschrift und der Victor-CodexrdquoIn K-H Schirmer amp B Sowinski (Eds) Zeiten und Formen in Sprache und DichtungFestschrift fuumlr Fritz Tschirch zum 70 Geburtstag (pp 337ndash356) KoumllnWien

Rathofer J (1973) ldquoDie Einwirkung des Fuldischen Evangelientextes auf den althoch-deutschen lsquoTatianrsquo Abkehr von der Methode der Diatessaronforschungrdquo In AOumlnnerfors et al (Eds) Literatur und Sprache im europaumlischen Mittelalter Festschrift fuumlrKarl Langosch zum 70 Geburtstag (pp 256ndash308) Darmstadt

Salemans B J P (2000) Building Stemmas with the Computer in a Cladistic Neo-Lachmannian Way The Case of Fourteen Text versions of Lanselot van Denemerken(Doct Thesis) Nijmegen

Scherer C (1905) ldquoDie Codices Bonifatiani in der Landesbibliothek zu Fuldardquo Festgabe zumBonifatius-Jubilaumlum 1905 Fulda

Sievers E (1892) Tatian lateinisch und altdeutsch BADLD 5 Paderborn 18922Vogels H J (1919) Beitraumlge zur Geschichte des Diatessaron im Abendland NTA 81 MuumlnsterZahn Th (1881) Tatianrsquos Diatessaron FGNK 1 ErlangenZahn Th (1894) ldquoZur Geschichte von Tatians Diatessaron im Abendlandrdquo NKZ 5 85ndash120

Constructing initial binarytrees in stemmatology

E WattelVrije Universiteit Amsterdam

Introduction

The main purpose of stemmatology is to construct a pedigree of actual andputative texts in order to describe the relations between the manuscripts ofa historical text tradition The expectation is that such a pedigree which isusually called a stemma will help to obtain insight into the transmission pro-cess which led to this text tradition Questions like lsquoIs one manuscript a directancestor of another onersquo and lsquoWhich reading in case of variants should beconsidered originalrsquo can sometimes be solved if a reliable stemma is available

Since the only information available is usually in the texts themselves weneed a method to extract the genealogical information from the comparison oftexts To enable the use of modern computers the coding of this informationshould be well defined and easily available for automatic processing Thereforewe will use the database conventions of eg Wattel and Mulken (1996a) whichis described in the next section

The genealogical information which is coded in version formulas is usuallyinternally inconsistent and so many investigators direct their efforts towardsexplaining away version formulas which disturb favoured parts of a stemmaunder construction Most of the arguments used to delete a version formulaare ad hoc and a posteriori and are not suited for an objective constructionmethod A much better approach to the problem was investigated by Salemans(2000) In this work a version formula has to meet eleven admission criteriabased upon seven basic principles to be accepted for stemma constructionIf this leads to a complete stemma we have certainly found a reliable objec-tive solution but it is possible that some inconsistencies remain or that the

E Wattel

number of version formulas which meet the requirements is insufficient toconstruct a stemma

An alternative approach is to keep in mind that no information in the ver-sion formulas is completely and absolutely reliable So the best we can hope foris a stemma with only a small amount of inconsistency

In order to estimate these inconsistencies we assign to each version for-mula a positive formula weight After the construction of a tree we can add theweights of all the formulas which contradict the tree and so we can estimatethe performance of a tree A tree is better when the estimated inconsistency is asmaller number Now a stemma can be evaluated with respect to the values ofthe version formulas and the stemma with optimal value can be chosen Sincethe Salemans (2000) admission criteria are suitable for assessing the structureand the contents of variation formulas meeting those criteria will get a highformula weight but formulas previously rejected can still be kept albeit witha much lower weight In this way we hope that good information will not bechoked by bad information but we can still use downgraded indications incases where nothing else is available

At this point it is necessary to assess variation formulas We can distinguishtwo types of observations which influence the value of the formula

i Linguistic These observations consider the contents of the versions Herewe take grammatical semantical logical and aesthetic aspects of the ver-sions in account

ii Computational Here we judge how suitable the structure of a version for-mula is for the construction of a stemma The most prominent of thosecriteria is the type 2 which indicates that a formula contains precisely twoconcurrent readings both shared by at least two manuscripts

In order to separate the data collection phase from the data processing it isstrongly advised to discard the information mentioned in (ii) when the valueof a version formula is determined in the data base During the constructionof a stemma from the data base the information in (i) should have no effectswhich are not coded into the value of the formula In this way one investigatorcan supervise the data acquisition and a different expert can monitor the dataprocessing

In this contribution we will concentrate on the computational problemsunder the assumption that the linguistic considerations have already deter-mined the value fields in the version formulas In Wattel and Mulken (1996b)the main subject is how an arbitrary tree could be evaluated with respect to aset of version formulas with valuation fields If this tree evaluation procedure is

Constructing initial binary trees in stemmatology

well defined it seems reasonable to evaluate all the possible trees and declarethe tree with the optimal value to be the stemma However such a procedure isnot feasible

The reason for this failure is called computational complexity which meansthat the number of possible trees is very large even for a small number ofavailable manuscripts

In order to get a credible final solution we start out from a reasonable treeand evaluate the sum of the version formulas which contradict it We try to im-prove this tree by a search algorithm which proceeds along the following linesWe construct a small set of related trees and evaluate the sum of the contradict-ing formulas for each of these trees If the evaluated sum is smaller we obtain abetter tree If a better tree is found we discard the first tree and continue withthe best tree encountered This process is repeated until no further improve-ment is obtained An algorithm of this kind is called a local search method Instemma construction this local search is quite applicable but again the theoryof computational complexity implies that success cannot be guaranteed Onepossibility is that the search goes on for ever effecting smaller and smaller im-provements all the time Another possibility is that the search never as muchas approaches the optimum over all possible trees and that the result is a fakewhich is far from optimal We should especially avoid the second possibility

Since expectations for the search algorithms cannot be very high a greatdeal of effort should be put into the construction of the initial tree Since ourmaterial is not collected in order to supply just any arbitrary tree but specif-ically to yield a stemma reflecting the transmission we hope that the truestemma will be clearly distinguishable from the background noise of incon-sistencies The main aim of this paper is establishing a method to constructan initial stemma which is hopefully close enough to the optimal stemma tofunction as a starting point for any type of local search

From the above it should be clear that the program described here is part ofa larger set Obviously other programs should deal with stepwise refinementand edge contraction (see for those programs Wattel amp Mulken 1996b) Oneof the set is a program which can spot version formulas which are in contra-diction with their context and therefore must be erroneous or contaminatedThe method used is an implementation of the ideas of Dees on this subject Itis also possible to look for a stable section of the text by means of the shockwave method described in Wattel and Mulken (1996a) Although the programdescribed here is meant to yield an initial tree which should be refined byother programs with the algorithm of this present contribution it is possible

E Wattel

to construct an unordered tree structure straight from a collection of versionformulas

In order to get a complete stemma we have to define an ordering for themanuscripts and this can be done by adding an extra manuscript in the versionformulas which is only mentioned when the original version for the formulacan be determined The position of this manuscript in the final result will be atthe top of the tree which then defines the direction of all the edges

In the present contribution we have tried to implement the experienceswhich have been obtained in the last ten years In this respect the work ofSalemans (2000) was essential The main difference between Salemansrsquo ap-proach and ours is that he refuses to use version formulas which are notcompletely reliable while in our approach the reliability of the formulas is ac-counted for in the weighing schemes In the near future we hope to redesignthe stepwise refinement and contraction programs in order to improve theirperformance and to get statistical information on the probability of error ofthe final stemma

In this paper Section 2 will be devoted to the structure of the data baseThe subject of Section 3 are lacunary version formulas ie formulas in whichsome manuscripts of the text tradition are missing In Section 4 we will give asmall introduction to the theory of complexity and explain the computationaldifficulties of exhaustive search methods This section is the main motivationfor the binary tree construction but can be skipped if the reader is not inter-ested in the theory In Section 5 we will give a construction algorithm for initialtrees in an iterative way in which formula weights are adapted in the processif the corresponding version differences turn out to be either unreliable or notsuited for tree construction

Material

Following eg Wattel and Mulken (1996a) version differences can be coded in adata base consisting of a single header line and a collection of version formulasas in Table 1

In our convention the heading line consists of two fields firstly an integerindicating the maximum number of verses in the text and secondly a list ofcodes indicating the available manuscripts

In this example the total number of lines is 5 and there are seven availablemanuscripts lsquoa lsquob lsquoc lsquod lsquoe lsquof and lsquog The verse numbers should not be taken toostrictly they are used to find corresponding verses in different manuscripts


Table 1 A simple data base

5 lsquoa lsquob lsquoc lsquod lsquoe lsquof lsquog (heading line)11 lsquoa lsquoc lsquod lsquob lsquoe 75 (first formula)16 lsquoa lsquob lsquod lsquog lsquoc lsquoe 05 (second formula)3 lsquoa lsquod lsquob lsquoc lsquof lsquog (third formula)3 lsquoa lsquob lsquoc lsquoe lsquof lsquog 5 (fourth formula)5 lsquoa lsquof lsquob lsquoc lsquod lsquoe lsquog 075 (fifth formula)

and in eg Wattel and Mulken (1996a) verse numbers are used to indicate thenatural order of the episodes and verses in the texts

All the lines apart from the heading line in the data base contain codedinformation from variant readings in the manuscripts The version formulasin the data base consist of two or three fields The first field indicates the po-sition of the variation usually a line number or a verse number where a set ofcomparable variant readings can be found The important part in such a vari-ant formula is the second field which shows which manuscripts share each ofthe various readings Manuscripts sharing a reading are listed in groups andthese groups are separated by slashes ( ) The third field may contain a numberindicating the relative importance of the formula If a version formula revealsmore about the relationship of the manuscripts than usual this field shouldcontain a number which is larger than one if the version formula indicatesdifferences which can easily arise from other sources than the used examplarthis field should contain a small number If the third field does not exist theimportance is supposed to be 1 which means that the information in the for-mula is moderately revealing Fields are separated by tabs In order to keep thedata base readable it is possible to add comments and version indications asfourth fifth etc fields in a database line Usually the fourth field contains thevarious readings themselves The program uses the coded information of thesecond field

Verse numbers in the version formulas need not be unique One verse maycontain several version differences In the example the third and fourth versiondifferences both occur in verse number 3 Also shown in the example is that wecan use fractional verse numbers in stead of integer ones to indicate wherein the verse the difference occurs When one version difference occurs at thebeginning of a verse and another at the end we use eg 11 and 16 to indicatethat both formulas were extracted from verse 1 but one from the beginning andone from the end

The lines with verse numbers 11 and 16 indicate precisely two differentreadings and each reading is shared by more than one manuscript Version

E Wattel

formulas of this kind are called type 2 The version formulas of this type 2 areby far the most suitable for constructingstemmas

Especially when all the manuscripts are represented in an unquestionabletype 2 formula the group of manuscripts is divided into two subgroups Wecall such a partition of all the material into exactly two groups a dichotomy Allthe lines in a tree represent a dichotomy of such a kind If we have a numbern original manuscripts then we need n ndash 3 of those dichotomies to constructa tree The requirements for this set of n ndash 3 variants are that they are differ-ent unquestionable complete and compatible The ideal case in which sucha collection can immediately be selected from the data base is like hitting thejackpot Types one three and four are also defined (see also Salemans 2000)A formula is type 1 when precisely two variant readings exist but one of thesereadings is found in only one manuscript A formula is type 3 when there aremore than two different readings but all except one of those readings are foundin only one manuscript The formula for verse 5 of the example is type 3 A ver-sion formula is type 4 when there are more than two different readings andat least two of those readings are shared by more than one manuscript Theformula with verse number 16 and both formulas for verse number 3 of theexample are version formulas of type 4 It must be observed that formulas inwhich no two manuscripts share a reading are useless for tree construction andare therefore omitted from the data base

Lacunary version formulas

Seven manuscripts are mentioned in the heading line of the example but onlythe fifth formula contains all seven manuscripts in none of the other ver-sion formulas do all the manuscripts occur Such formulas are called lacunaryThere can be several reasons why a manuscript is not mentioned eg part ofa page may be damaged or a page may be missing In such a case it is betternot to force the manuscript into the formula This type of lacuna is totally dif-ferent from a revealing text omission such a text omission occurs when somewords a phrase or even a couple of verses are missing from the text on an un-damaged page These omissions have to be put in the data base as a separatevariant reading

In the example of Table 2 we have an ordinary version difference inmanuscripts lsquoa lsquob lsquoc and lsquod in which the words ldquolanternsrdquo and ldquolampsrdquo in-terchange Manuscripts lsquof and lsquog have a revealing omission because the words


Table 2 Omissions and lacuna

Manuscript versions

lsquoa Carrying torches and lanterns (version 1)lsquob Carrying torches and lamps (version 2)lsquoc Carrying torches and lamps (version 2)lsquod Carrying torches and lanterns (version 1)lsquoe Carrying (lacunary)lsquof Carrying lanterns (omission)lsquog Carrying lanterns (omission)3 lsquoa lsquod lsquob lsquoc lsquof lsquog (version formula)

ldquotorches andrdquo are not in their texts although the lines appear to be undamagedManuscript lsquoe is unreadable and hence it is lacunary at this verse

In the version formula we have three established different readings version1 ndash lsquoa lsquod version 2 ndash lsquob lsquoc and omission version ndash lsquof lsquog Manuscript lsquoe must notbe included in the version formula because it may belong to any of the othergroups It goes without saying that processing this formula should have noeffect on the relations between lsquoe and any of the other manuscripts We willformulate this as a general rule

Manuscripts which are not mentioned in a lacunary formula should have noinfluence on the result of the computation when such a formula is processed

This rule has as a consequence that eg the distance between two manuscriptsshould not be computed as the sums of all the formulas in which thesemanuscripts do not share a reading If the distance between two manuscriptswere to be computed as the sums of all the formulas in which they differ thena heavily damaged manuscript would get positioned close to all the other textsA damaged manuscript is then easily considered central and old therefore re-liable and thus its merits could be overestimated The lacunary version rule ismeant among other things to prevent this type of bias

Moreover the rule and its consequences are obligatory if we want to uselacunary formulas It is clear that we cannot avoid lacunary information if wewant to make stable stemmas Imagine what the consequences would be ofnot allowing lacunary information if a highly damaged manuscript is foundafter the stemma was established should we throw away all the informationwhich was valid until the appearance of the extra manuscript which made mostversion formulas lacunary

If lacunary information is allowed even type 2 version formulas will nolonger separate the set of manuscripts into two distinct sets It is not clear where

E Wattel

lsquo lsquo

lsquoˆ5 lsquoˆ5

lsquoˆ3 lsquoˆ3

lsquoˆ2 lsquoˆ1

lsquod lsquoclsquod

lsquoa lsquoˆ2

lsquoˆ1 lsquoblsquoa

lsquoˆ4 lsquoˆ4

lsquoclsquob

lsquog lsquoglsquof lsquof

3 lsquoa lsquod lsquob lsquoc lsquof lsquog (version formula)

Figure 1 One formula with two different stemmas

the lacking manuscript should go This could be a major objection againstthe use of lacunary formulas for those philologists which restrict themselvesto only type 2 information However unapplicable and lacunary informationis so common in text traditions that lacunary formulas cannot be ignoredwithout loosing a large part of the available information So even the authorswho base their information processing on cladistic methodology (cf Salemans2000) cannot maintain their objections against lacunary type 2 version formu-las under all circumstances

The reason for the restriction to type 2 information can be seen in theexample of Figure 1 In this figure there are two different stemmas with sixavailable manuscripts lsquoa lsquob lsquoc lsquod lsquof lsquog and a putative archetype lsquo In orderto obtain a structure in which every internal node joins exactly three edges wehave five intermediate nodes lsquoˆ1 lsquoˆ2 lsquoˆ3 lsquoˆ4 lsquoˆ5 Suppose moreover that theformula lsquoa lsquod lsquob lsquoc lsquof lsquog is compared to both stemmas If the case is that theversion in lsquoa and lsquod is the original one it is clear that the deviation in lsquof and lsquogmust have occurred for the first time in node lsquoˆ4 and the deviation in lsquob and lsquocmust have occurred in node lsquoˆ1

Also if we have established the deviation in node lsquoˆ1 then all the descen-dants of lsquoˆ1 share this reading and the same observation holds for lsquoˆ4 But incase the version of lsquof and lsquog is the original one then the formula introducestwo different deviations in one branch of the stemma in the left tree but twoindependent deviations in the right tree In the left tree the deviation of lsquoa andlsquod occurred in lsquoˆ3 but not all decendants of lsquoˆ3 share its readings In the righttree the situation is even worse Either of the deviations could have occurred


for the first time in lsquoˆ3 but it could also be that this intermediate still has theoriginal version

These ambiguities are not solved when the original versions and thestemma are well established and so several authors claim that a formula ofthis type is not fit for tree construction Contrarily type 2 formulas do not in-troduce difficulties of this type In the left tree we can have formulas lsquoa lsquob lsquoc lsquod lsquof lsquog lsquoa lsquof lsquog lsquob lsquoc lsquod and lsquoa lsquod lsquof lsquog lsquob lsquoc In the right tree we have lsquoa lsquob lsquoc lsquod lsquof lsquog lsquoalsquod lsquob lsquoc lsquof lsquog and lsquoa lsquod lsquof lsquog lsquob lsquoc In all these cases it is possible to assign a uniquenode to the deviation whenever the original version is established and all thedescendants of that node share the deviation In the left tree this node will beeither lsquoˆ3 or lsquoˆ4 for the first formula it will be lsquoˆ2 for the second formula andlsquoˆ1 for the third formula In the right tree this node will be again either lsquoˆ3 orlsquoˆ4 for the first formula it will be lsquoˆ2 for the second formula and lsquoˆ1 for thethird formula Salemans 2000 follows this line of thought in the stemma con-stuction and he uses type 2 information only His argument is that we can besure that a clean formula will introduce a well defined node which may be heldresponsible for the deviation The sheer existence of such a node is necessaryeven if we cannot (yet) decide which version is original

If the processing faithfully observes the lacunary formula rule then it is nota big intervention to split up a formula into its constituent parts In the casethe original formula lsquoa lsquod lsquob lsquoc lsquof lsquog is split into three lacunary subformulas(i) lsquoa lsquod lsquob lsquoc (ii) lsquoa lsquod lsquof lsquog and (iii) lsquob lsquoc lsquof lsquog which are all type 2 formu-las The two stemmas of Figure 1 will be reduced to a substemma for each ofthose subformulas This is done by deleting manuscrips which are not in theformulas and contracting the internal nodes which do not distinguish betweenthe other manuscriptsThe result can be seen in Figure 2 In these substemmas

lsquolsquo lsquo lsquo lsquolsquo

lsquoˆ5lsquoˆ5 lsquo5 lsquo5lsquo2lsquoa

lsquo4lsquo4 lsquo1 lsquo4 lsquo4lsquod

lsquoglsquog lsquoc lsquog lsquog

lsquo3 lsquo2 lsquo1



lsquo5 lsquo3

lsquo3 lsquo2 lsquo3 lsquo1lsquo1

lsquoflsquof lsquob lsquof lsquoflsquoblsquoa lsquoa lsquoa lsquoblsquob lsquoclsquod lsquod lsquod lsquoclsquoc

lsquob lsquoc lsquof lsquoglsquoa lsquod lsquof lsquog lsquoa lsquod lsquob lsquoc lsquoa lsquod lsquof lsquog lsquob lsquoc lsquof lsquoglsquoa lsquod lsquob lsquoc

Figure 2 Three substemmas constructed from each of the two stemmas

E Wattel

there is always a node which can be held responsible for the deviation if one ofthe versions of the subformula is non-archetypical

For instance in the leftmost substemma node lsquo1 must be held responsiblefor the deviation and the reading shared by lsquoa and lsquod could be the originalwhile the reading of lsquob and lsquoc is probably not archetypical If we compare thisobservation to an analysis of the same lacunary subformula and the fourthstemma of this figure we get totally different results In the fourth substemmanode lsquo1 is responsible if the version in lsquob and lsquoc is not original and node lsquo2is responsible for a deviation in the manuscripts lsquoa and lsquod So either of theconcurrent readings could be archetypical The other substemmas have thesame structure as the fourth We must conclude that the use of restricted type2 formulas can be helpful to construct stemmas and that the position where adeviation occurred can be determined if there is prior knowledge as to whichversion is archetypical

The example above shows that it is safe to extract type 2 formulas from for-mulas with more alternatives Even when the number of alternatives is morethan three and the number of shared readings is higher we still encounter noadditional difficulties in contracting consistent stemmas and each contractedsubstemma has at least one node which can be held responsible for somedeviation

It is a well known problem that the number of genuine type 2 formulas isseldom sufficient for the construction of a stemma or even an unrooted un-oriented tree Several authors among them Dearing Dees and Dekker havetried to construct additional type 2 formulas by combining the available ver-sion information Salemans (2000) shows that the methods of Dearing canyield combinatorially erroneous results The methods of Dees and Dekker areintrinsically correct but since we know that version information hardly everhas absolute value even careful combination of information usually increasesthe error probability

Therefore we are strongly in favour of constructing the necessary type 2formulas only by splitting larger formulas into all possible lacunary type 2subformulas and adhering strictly to the lacunarity rule that manuscripts notmentioned must have no influence on the computation

Computational complexity

When all the version formulas have been collected and the weights (based onthe linguistic contents of those formulas) have been determined it should be a


matter of computation to find the optimal stemma This task is not as easy asit sounds by far even when computers can be used for the evaluation of treesand stemmas The first step is to decide along which lines the combinationof a version formula and a tree should receive a value We can illustrate thisas follows Given a stemma or a tree eg the stemma of Figure 1 left and aformula eg

3 lsquoa lsquod lsquob lsquoc lsquof lsquog 1

how should this combination be valued There are several possibilities for thisvaluation We could say that there is complete agreement and therefore count1 the original value of the formula On the other hand we could say that thereis agreement for each of the three type 2 subformulas and therefore count itas 3 A third possibility is to count all the mimimal type 2 subformulas whichalso happen to yield 3 Those minimal type 2 formulas always have the formlsquow lsquox lsquoy lsquoz Because there are always exactly four manuscripts involved thoseformulas are called quadruples (or quartets cf Bryant amp Steel) If we make thesame evaluation for the stemma of Figure 1 left and the formula


the result is 1 again if we take only overall agreement into consideration theresult is also 1 if we consider type 2 subformula evaluation and now as muchas 6 if we count quadruple agreement

The number of possibilities becomes even larger if we also start countingformulas which do not agree with the tree Consider eg the formula


If we consider quadruple evaluation we have three possibilities to choose anagreeing quadruple and six possibilities to choose a disagreeing quadruple andso the value would be ndash 3 For the full formula valuation we could argue thatthis formula is in total disagreement and so evaluation will yield the value ndash 1and the same is the case for the subformula valuation

Then there is a problem when large trees are matched with large formu-las and only one single manuscript is in the wrong set The effect would bedisagreement with the resulting value of ndash 1 for this small deviation which israther exaggerated A sophisticated solution for curbing this effect is to see whathappens when one manuscript is removed from the formula If we get agree-ment we count a positive fraction of the formula weight We could subtract afraction of the weight for each of the manuscripts which have to be omitted be-fore we get agreement and if for example more than one fifth of manuscripts

E Wattel

have to be removed before we get agreement the result should be negative Itis clear that there are several ways to make precise choices along those lines

A different approach is not to valuate the complete formula and the com-plete tree in one sweep but to compare a formula to each dichotomy which canbe derived from a tree This approach is used in Wattel and Mulken (1996b)If the dichotomy agrees with one of the slashes in the formula the result ispositive if a disagreeing quadruple can be selected from the formula we havea negative result If we have neither agreement nor a disagreeing quadruplewe call the combination consistent Sometimes this means that the formula isin agreement with some other dichotomy of the tree sometimes we have dis-agreeing quadruples elsewhere in the tree In Wattel and Mulken (1996b) thedisagreeing formulas receive very negative values while consistent dichotomiesreceive small positive values

When we combine the different possibilities there are at least twenty dif-ferent valuation and optimisation schemes some of which are treated in Bryantand Steel This present paper will also make clear that the comparision of pos-sible stemmas is not easy even when the optimisation scheme has been decidedupon The reason is a consequence of the stupendous number of possible treesthat have to be considered in the computations and although this numbermore or less depends on the type of valuation almost all computation schemessuffer from the same unmanageability problem

There is a compartment of information theory called computational com-plexity which estimates the number of computational steps that have to be-taken to solve some problem as a function of the input of the problem Thestemmatological optimisation problems score rather badly in this respect Theinput length of a file is about the product of the number of formulas times thenumber of available manuscripts Since the number of formulas does not causeproblems we will concentrate on the number of manuscripts to estimate thecomplexity

To make this point clear if the number of available manuscripts is fourthere are 32 possible unoriented trees and 394 possible stemmas If we in-crease the number of manuscripts to five the result is 262 different trees and4336 different stemmas So one single manuscript increases the possibilitieswith a factor of 123 resp 165 The number of trees for just eight manuscriptsis as high as 37 million and the number of stemmas is 72 million Supposingthat we have several hundreds of formulas and that we can evaluate a tree or astemma in a millisecond it will then take an hour to evaluate all the trees anda day to evaluate all the stemmas Should we find just one more manuscriptand valuate the same number of formulas the number of trees and stemmas


increases with a factor over thirty and it will take more than a day to eval-uate all the trees and more than a month to evaluate all the stemmas In thesame way every additional manuscript will increase the number of trees with alarge factor

The increase of processor speed does not give us much help At the momentthe processor speed of modern computers doubles about every two years Thismeans that it will take another ten years of processor speed improvement be-fore the full evaluation of all the trees in the case of nine manuscripts is reducedto one hour It follows that full evaluation of all the stemmas is not feasible if thenumber of manuscripts mounts to fifteen like in the ldquoLanceloetrdquo (cf Salemans2000) or the ldquoPercevalrdquo (cf Maas)

However the really bad news is that there is no algorithm which is certainto yield the optimal tree for a fixed set of formulas without considering all thepossible trees The optimal weighted tree construction problem is not the onlyone of this kind several other computational problems in science and businessapplications share this characteristic The best-known problem of the type isthe travelling salesman problem find the shortest round trip for a travellingsalesman who is going to visit his clients when all the distances between pairsof clients are given

This class of problem is called the class of Nondeterministic PolynomialOptimization problems or NPO for short In fact we should say that theyare optimisation reformulations of the Nondeterministic Polynomial problems(NP for short) where the question is lsquoIs there a solution (not) exceeding givenlimitsrsquo This class of NP problems has been studied extensively over the lastfourty years and for none of its members a fast algorithmic solution has beenfound up to now Moreover almost all members of this class NP are equallyhard to program

They share the characteristic that a fast algorithm which would solve itunder all circumstances would also guarantee a fast algorithmic solution forevery other member of the class NP These problems are called Nondetermin-istic Polynomial Complete or NPC So in a way should one solve one of thoseproblems one would solve them all Since no fast solution has ever been foundfor any member of the NPC class problem it is assumed that all these prob-lems are intrinsically difficult It is clear that the optimisation variants cannotbe easier than the corresponding problems which only ask for a solution withingiven limits

From these observations it follows without saying that we should proceedwith extreme caution In constructing a stemma we are not just looking for thebest tree from a given set of quadruples or formulas but we are looking for the

E Wattel

structure of a copying process which should arise above possible backgroundnoise We proceed in two steps First we construct a tentative initial stemmafrom the material and then we try to improve the overall structure by meansof a local search method Such a local search method make small alterations andadaptions to an initial structure in the hope of achieving improvements Theabsolute optimum will be found if the initial stucture was not too far away fromit and could be transformed into the optimal solution by a set of local searchimprovement steps Clearly we will never obtain a 100 certainty but we dothe best we can by starting with a carefully designed method for the making ofthe initial guess

In this paper we do not go into the details of stepwise improvement Watteland Mulken (1996b) gives a possible algorithmic solution to that problem inthe case of a given formula weight and some fixed valuation design In thenext section of this present paper we will improve on the initial guess of Watteland Mulken (1996b) with an initial tree building algorithm which adapts theformula weights during execution if internal inconsistencies are encountered

A pairing algorithm for the construction of initial trees

Let us have a set of version formulas over a collection of available manuscriptsIf a version in a formula is indicated as original we put an extra fictivemanuscript lsquo (indicating the possible archetype) into that version of the for-mula The aim is to construct an unoriented tree in which every internal nodeis a fictive manuscript which is directly connected to exactly three other nodesThe sigla and the putative manuscript lsquo are the terminal nodes of this treeTrees of this kind are called fully resolved (cf Bryant amp Steel)

In order to have a consistent notation we will call indications in the versionformulas of existing and fictive manuscripts symbols

The construction proceeds in the following way We work in differentstages and every stage consists of two passes In the first pass of each stage welook for the two most closely related symbols and join them together to make anew symbol The second pass consists of remaking the set of version formulasin such a way that the two closely related symbols are deleted in favour of thesingle new symbol The weight of the version formulas is then adapted accord-ing to the suitability of the pairing of those two manuscripts In the process thefictive distances between the new symbol and each of the two old symbols isaccumulated as well as the weights of the contradicting quadruples After thisstage is finished we start the next stage in which the number of symbols is


decreased by one When only three symbols are left those three are joined tomake the top symbol of the tree and again we can make a distance indicationfor those three last symbols and the top

The unoriented tree is now made by starting out from the top symbol astop node Next we take the three symbols which were joined to make the topwe make them into three new nodes and connect each of them to the top withan edge We use the computed distance as an indication for the fictive lsquolengthrsquoof these edges As long as the tree contains constructed symbols we can addtwo nodes for its two constituent symbols and connect them with edges withlength indications in accordance with the computed distances

This rather simple algorithm immediately yields an unoriented fully re-solved tree The method is quite common cf eg Wattel and Mulken (1996b)and the main sophistication should go into the decision which two symbolsare closest at a given stage Errors made in this decision tend to cumulate overthe following stages and propagate into the final tree Therefore we have tomake deliberate design choices to evaluate the algorithmic consequences of theweights of different formulas the weights of pairs inside a formula and thecomparison of the pair weights at the end of the stage to decide which pairshould be squeezed into one new symbol These are the decisions of the firststage The design choices of the second stage concern the new formula weightsafter the pair has been squeezed and the computations of distances andor theweights of the quadruples contradicting formation of the pair We will proceedthis section along these steps

First pass ndash 1 ndash algorithmic structure valuation of version formulas

We have already observed that linguistic criteria should be used to decide thedata base value of the version formulas Salemans (2000) uses exclusively type2 information except when a manuscript could be an intermediate In thatcase he also admits type 1 information These decisions are in line with theusual practice in cladistics We are well aware of the importance of type 2 in-formation but we consider it too restrictive to abandon the other informationTherefore we make the following design choices

i If a version formula contains precisely two version groups with more thanone manuscript each we consider this a type 2 to stress its importance theweight of such a formula is multiplied by 4

ii If a version formula contains precisely one version group with more thanone manuscript this is either a type 1 or a type 3 formula in order to

E Wattel

downgrade its importance the weight of such a formula is multiplied by025

iii Since this algorithm concentrates on pairs of manuscripts it seems reason-able to multiply the weight of the formula by the number of pairs that itcontains In that way the importance of the formula is directly related to itsinfluence on the pairing process

First pass ndash 2 ndash pair weights inside a version formula

It is clear that in a certain version formula pairs sharing a group should countpositive and pairs the members of which are in different groups should countnegative In a formula we will make all positive contributions add to the weightestablished in the previous paragraph In order to have the total contributionof the formula 0 the absolute value of the negative contributions should alsoadd to this total weight

For the positive contributions we will take into account the observationof Maas that a group with precisely two manuscripts is a reliable witness fora pairing construction Therefore pairs in groups with a small number ofmanuscripts should be favoured above pairs in groups with a large numberof manuscripts Because the total process will be iterated this is not con-tradictory to the observation of Dees that type 2 formulas with large groupsof manuscripts should count more heavily because they are entered into theprocess more often This situation can be easily established if we take

iv The total weight of the formula is distributed over those groups with morethan one member according to the numerical squares ng

2 of the num-ber ng of members of a group g (see example below) Then the positivecontribution of a group is evenly spread out over the pairs of that group

To give an example of this idea let a certain type 4 version formula

666 lsquoa lsquob lsquoc lsquod lsquoe lsquof lsquog lsquoh lsquoi lsquoj lsquok lsquol lsquom lsquon lsquoo lsquop lsquoq lsquor lsquos lsquot lsquou lsquov 270

has seven different readings and a final weight 270 after the multiplications ofthe previous steps Now this weight has to be distributed over the groups in theproportions 0 4 9 16 25 36 0 so the group lsquob lsquoc gets value 4

90 middot270 = 12group lsquod lsquoe lsquof gets 27 etc Now the only pair in group lsquob lsquoc gets the full 12 thethree pairs in lsquod lsquoe lsquof get 9 each the six pairs in lsquog lsquoh lsquoi lsquoj get 8 each the ten pairsin lsquok lsquol lsquom lsquon lsquoo get 75 each and the fifteen pairs in lsquop lsquoq lsquor lsquos lsquot lsquou get 72 eachIn fact a pair in a group of two gets almost double the weight a pair in a largegroup gets


In order to distribute the formula weight over the pairs in different groupsto get the negative contributions we distinguish two cases (a) one of the sym-bols is a group in itself ie it is a single reading or (b) both symbols are fromgroups with at least two members We observe that it is a bad idea to join thetwo symbols in case (b) because that would introduce a contradicting quadru-ple in the tree In case (a) we have at worst a single deviating reading which willvanishes in the next stage which is no big deal So pairs of type (b) are muchmore important and we will give them a factor of for example 16 in compar-ison with pairs of type (a) When we look in the example above we have 41pairs of type (a) and 155 pairs of type (b) and so the weight factor is to bedivided by 41 + 16 middot 155 = 2521 So the pairs of type (a) lsquoa lsquob lsquoa lsquov lsquob lsquov etc geta score of ndash 270

2521 = ndash01071 The pairs of type (b) for example lsquob lsquod lsquoe lsquog etc get ascore of ndash 270

2521 middot16 = ndash17136 We should again formulate this as a design choiceand obtain

v Pairs whose members are in two different non-single groups receive a neg-ative value which is 16 times as large as pairs with one or two members insingle reading groups

First pass ndash 3 ndash deciding upon the closest pair of symbols

By the end of this pass we have collected a sum Sp of all the positive contri-butions by pairs a sum Np of the absolute values of negative contributions bypairs This value Np can be split up into two parts Qp accounting for quadru-ple contributions and Rp representing the contributions from single readingsNow we can make some design choices to determine the pair that is to besqueezed into one symbol at this stage Possibilities are (a) minimal absolutevalue of Qp so the number of contradictions introduced is as small as possible(b) maximal value of Sp ndash Np which chooses the pair with maximal support(c) minimum of Qp divided by Sp ie the amount of contradictions is mini-mal in proportion to all contributions Our favoured choice is (d) the negativecontribution is proportionally minimal

vi The pair for whichNp

Sp + Npis minimal will be taken together to form a single

new symbol to replace the two symbols of this minimal pair

Second pass ndash 1 ndash production of new version formulas

Since in the previous pass it was decided which pair of symbols should be re-placed by a new symbol the first task of the second pass is to construct a new

E Wattel

data base from the old one which can then be used in the next stage Thisrequires two main decisions (a) how to replace and (b) how to adjust theweights These two steps are closely interrelated and they should be expressedfor every type of occurrence of the pair in the version formulas We dismissall the weight adjustments made in the previous pass The input is the originaldata base or the data base constructed in the previous stage When output isgenerated it is put into the new database We start with copying the headeromitting the two symbols of the pair and adding the new single symbol Thedesign choices (vii) through (xi) express the necessary decisions made for all ofthose different occurrences

vii If neither member of the pair occurs in the version formula it is justcopied to the output If only one member of the pair is in the formulathat member is replaced by the new symbol andthe version formula isthen just copied to the output with the original weight

viii The two symbols of the pair are members of the same group In thatcase replace the two symbols by the new symbol If the new formulastill has groups with more than one manuscript output the line with theoriginal weight

ix If one of the members of the pair is a single reading just omit it andreplace the other member of the pair in that formula with the new symbolIf there is more than one group left copy the new version formula to theoutput with the original weight if not omit the formula

x If both members of the pair are single readings just omit them If thereis more than one group left copy the rest of the version formula to theoutput with the original weight

xi If the members of the pair are in different groups and each of thosegroups contains more than one manuscript we have contradictingquadruples We then multiply the weight by a factor 04 and output thenew version formula twice (a) once omitting the first member of the pairand replacing the second member by the new symbol and (b) once omit-ting the second member of the pair and replacing the first member by thenew symbol In this way the total weight of the original formula loses afifth of its influence in the next stages This is a reasonable compensationfor the contradictory nature of this formula which has been discovered atthis stage


Second pass ndash 2 ndash computing edge distances

This step and the next step are not important for the tree construction itselfbut they will help to understand the trees We will again use proportions so wewill register contributions to the distance as well as weight contributions forthe formulas We concentrate on each member of the pair and use the originaldata base or the database constructed in the previous stage just as in secondpass step 1

xii If only one member of the pair is in a formula we skip this formula fordistance computation

xiii If both members are single readings the distance contribution is half theformula weight and the weight contribution is the full formula weight

xiv If both members are in different groups and each of those groups con-tains more than one manuscript the distance contribution is also halfthe formula weight and the weight contribution the full formula weight

xv If one member is a single reading and the other member is not theweight contribution as well as the distance contribution for this mem-ber is the full formula weight If the other member is a single readingand this member is not the distance contribution is 0 and the weightcontribution is the full formula weight

xvi If both members are in the same group the distance contribution isalso 0 and the weight contribution is the full formula weight for bothmembers of the pair

xvii The distance indication of the edge connecting a member of the pair withthe node of the new symbol is the quotient of the distance contributionsand the weight contributions (the easiest way to express this value is as apercentage)

Second pass ndash 3 ndash quadruple contradiction weights

In the same way as in the previous step we can compute a quadruple contribu-tion percentage Again we start with the original data base or the result of theprevious stage and compute a quadruple contribution as well as a weight con-tribution In this step not the symbols themselves but the pairs are importantIt should also be stressed that the edge which is involved most in the quadruplecontradictions is the edge which emerges from the new symbol in the directionof the top The design choice for the contribution is obvious

E Wattel

xviii If both members of a pair exist in a formula the weight contributionshould be the formula weight If the two are members of two differentgroups each with more than one manuscript we have a contradictingquadruple and the quadruple contribution equals the formula weightfor this version formula

xix The quadruple indication of the edge emerging from the new symbol inthe direction of the top is the quotient of the quadruple contributionsand the weight contributions and the easiest way to express this value isas a percentage

Final stage

The algorithm has to stop when there are only three manuscripts left in whichsituation we can look for the closest pair and compute distance contributionsIt has already been mentioned that those three symbols should be connected toa top node In this case there is no quadruple indication but the edges whichconnect the three symbols to the top will have received a quadruple indicationin the previous stages for the non-terminal symbols

If the data base gets empty at a certain stage or if all version formulas aretrivial it makes no sense to continue the algorithm any further The pairs couldbe squeezed in a random way to get a fully resolved tree and all the distanceindications should be put at 0 Quadruple indications should be consideredundefined If this happens the internal nodes of the tree which still exist at thatstage could be identified with the top node if we use an internal contractionmethod afterwards (cf Wattel amp Mulken 1996b)

Conclusion

Although the program described here is meant to yield an initial tree it can alsobe used to construct an unordered tree structure straight from the collection ofversion formulas It is even possible to construct a fully oriented stemma withit if we indicate the unquestioned archetypical versions in the formulas with asigla lsquo and use the position of this character as the top node of the stemma

The method described here has been used on several text traditions andyields consistent results most of the time It does not contain a contractionmethod and so we need a contraction program in cases where it is expectedthat some of the manuscripts are intermediates and when the stemma is notfully resolved


The results of the program itself are usually consistent with the clusterstructure of the set of manuscripts There is a need for a new program im-plementing stepwise refinement according to the latest ideas which could givestatistical information about the error probability of the final result

References

Bryant D amp M Steel (1999) ldquoFast algorithms constructing optimal trees for quartetsrdquo InProceedings of the Tenth Annual ACM-SIAM Symposium on Discrete Algorithms (pp147ndash155) Maryland ACMSIAM

Dearing V A (1974) Principles and practice of textual analysis Berkely Los Angeles ampLondon

Dees A (1977) Over stambomen van handschriften Forum der Letteren 1977 63ndash73Dekker M C H (1986) Reconstruction methods for deviation trees Doctoral thesis Math-

ematics and Informatics Vrije Universiteit AmsterdamHollander A A den (1997) De Nederlandse Bijbelvertalingen 1522ndash1545 PhD Thesis Vrije

Universiteit Amsterdam Nieuwkoop De GraafHollander A A den (this volume) ldquoHow shock waves revealed successive contamination

A cardiogram of early sixteenth century printed Dutch BiblesrdquoMaas P (1957) Textkritik Verbesserte und vermehrte Ausgabe LeipzigSalemans B P J (1996) ldquoCladistics or the resurrection of the method of Lachmannrdquo In P T

van Reenen amp M P van Mulken (Eds) Studies in Stemmatology (pp 3ndash70) Amsterdamamp Philadelphia Benjamins

Salemans B P J (2000) Building stemmas with the computer in a cladistics neo-lachmananway PhD Thesis Katholieke Universiteit Nijmegen Nijmegen University Press

Wattel E amp M P van Mulken (1996a) ldquoShock waves in text traditionsrdquo In P T van Reenenamp M P van Mulken (Eds) Studies in Stemmatology (pp 105ndash122) Amsterdam ampPhiladelphia Benjamins

Wattel E amp M P van Mulken (1996b) ldquoWeighted formal support of a pedigreerdquo In P T vanReenen amp M P van Mulken (Eds) Studies in stemmatology (pp 135ndash168) Amsterdamamp Philadelphia Benjamins

Trouble in the treesVariant selection and tree constructionillustrated by the texts of Targum Judges

Willem F SmelikUniversity College London

Textual history

The early textual history of Targum Jonathan to the Prophets is a matter ofconjecture Not a single complete manuscript predating the early twelfth cen-tury ce has survived Our earliest evidence consists of some isolated quotationsin classical rabbinic literature (second to eight centuries ce)1 and magic bowlsproduced between the third and sixth centuries ce2

The reasons for this lack of hard evidence are miscellaneous The practiceof burying and depositing outworn and faulty manuscripts which would notactively be destroyed because they contain the Sacred Scriptures may partly ac-count for this situation which has no doubt been aggravated by the vicissitudesof migration and persecution By the time when Aramaic had been supersededby Arabic in the Islamic countries and Indo-European languages in the Westthe Targums had largely lost their originally important place within the cur-riculum of rabbinic students and the liturgy of the synagogue Obviously thisdevelopment would not help preservation of extant manuscripts all the moreso since there may have been rather few manuscripts in the first place Becauseof the status of the Targum as oral Torah it is not impossible that the tex-tual transmission of the Targums remained predominantly oral during the firstmillennium ce3

Any construction of the textual history will thus have to make do with littleevidence By common consent dialect contents and the rabbinic regulationsfor the liturgical use of Bible translations favour an origin of the Targums tothe Torah and the Prophets in or soon after the second century ce ndash although

Willem F Smelik

the issue of dialect is still controversial4 Because of exegetical parallels betweenTargum Jonathan and later Amoraic literature we may safely assume that theTargum was subject to changes and supplements in the following three cen-turies even though the extent of these changes has still not been explored in asatisfactory way

A stemma for the available textual witnesses would ironically only coverthe part of the textual history when the heyday of this genre was long pastand the fluidity of targumic traditions had given way to a fairly stable tex-tual tradition Only 24 (nearly) complete manuscripts of the Targum to Judgesare extant nowadays a number similar to that to any of the prophetic bookshowever hundreds of manuscripts have the Targum to the lectionary readingsthe so-called haftaroth5 These manuscripts have been produced in such widelydivergent places and periods as twelfth-century Italy fourteenth-century Ger-many fourteenthndashfifteenth-century Aleppo fifteenth-century Yemen and six-teenth-century Spain Their relationships have hardly been explored6 beyondthe fairly obvious family connections of the Babylonian Yemenite Sefardi andAshkenazi manuscripts

One of the long-standing questions concerns the position of the Yemenitemanuscripts and especially the manuscript in two volumes which Alexan-der Sperber selected as the basis for his commonly used edition of TargumOnqelos and Targum Jonathan ms Or 2210-11 (British Library London)7

This manuscript was previously thought to be unique among the Yemenitewitnesses but it receives substantial support from some of the fragmentaryBabylonian witnesses widely held to represent the oldest text type for histor-ical textual and grammatical reasons8 It is plausible that his manuscript ofchoice was a crock of gold It should be noted that he gave preference to a gram-matically correct text9 so that it may not be entirely coincidental that Dalmanhad given preference to this manuscript in his grammar of Jewish-PalestinianAramaic10

In recent years some stemmatologists have focussed on the issue of vari-ant selection for the purpose of tree construction Some scholars notablySalemans11 argue against a quantitative approach which allows any variantreading to have its influence They suggest that a rigorous selection of variantsshould precede any attempt to build a stemma Other scholars would disagreearguing that all variant readings should be taken into account and that selec-tion should occur in the final stages of analysis A middle of the road approachallows most variant readings to have their influence but would attribute lowervalues to circumspect variants and higher ones to more reliable variants

Trouble in the trees

No generalizations should be applied to any given textual tradition withoutconsideration of its idiosyncrasies12 Codicological studies of medieval Hebrewliterature have shown that scribes assumed considerable freedom in revisingadapting and improving the text of their exemplars not infrequently the au-thors even appealed to scribes to correct mistakes and even on at least oneoccasion to add useful material13 Because draft versions of certain bookssometimes circulated before the final version was lsquopublishedrsquo the texts oftenexisted in multiple co-existing versions For that reason it will be useful to ob-serve justify and verify the reliability of several types of variant readings forthe Targum to the Prophets to avoid the pitfalls of either including or exclud-ing too much information To establish these categories variations introducedin dual copies of a single exemplar or in the copy of a known exemplar are ofgreat value By study of these variants we hope to disclose patterns and ten-dencies in the process of textual transmission and to identify purely randomvariations that did not carry any value in the eyes of the scribe him- or her-self Especially in dual copies of a single exemplar we may identify the kind ofvariation that the scribe would not deem very important and that I will calllsquorandomrsquo variation

A genealogy of the textual witnesses for the Jewish Aramaic Bible transla-tions is problematic because our copies reveal traces of infuences from othersources than their exemplars other copies which the scribe consulted the He-brew text which was usually transmitted within the same source and quitepossibly memorized passages14 These influences thwart a linear type of geneal-ogy where each copy is produced from a single exemplar All the same linearrelationships between the texts remain important and the extent to which weare able to retrieve such relationships should concern us here

In a previous study I have briefly explored a handful of criteria for the se-lection of variant readings in this article I will supplement and evaluate thesecriteria In tandem these two studies highlight random variation in our tex-tual witnesses in order to establish an empirical rather than theoretical basisfor the selection15 Surprisingly no such criteria have ever been set for tar-gumic literature16 This article will supplement these earlier observations onvariant selection in addition it will evaluate the value of the classificationThe evaluation will assume the form of interpreting alternative trees and shockwaves (both of which have kindly been produced by Evert Wattel of the FreeUniversity in Amsterdam)17

The core of this study thus consists of observations on the phenomenon ofrandom variation in manuscript reproduction Random variations are thosevariants that do not carry any genealogical value but may have been intro-

Willem F Smelik

duced by several scribes independently of each other in such cases agreementsbetween manuscripts would not reflect their genealogy To identify such coin-cidental variation I will draw extensively on two Spanish manuscripts and thefirst two Great Rabbinic Bibles together with their (likely) exemplar (Vorlage)It is hoped that this approach may yield a transparent verifiable categorization(or types) of variant readings Simultaneously such genealogical informationas could possibly be provided by these types of variant readings even those re-garded with the utmost reservation will be discussed and related to the variouslevels of stemmatological studies18

The Spanish manuscripts used here were written by the Jewish convertAlfonso de Zamora in 1532 and 153319 The exemplar of both apographs isunknown but may well be the lost ms that had been purchased for the Com-plutensian Polyglot20 published between 1518 and 1520 as he wrote them inAlcalaacute de Henares where he had been a professor of Hebrew since 1512 andwhere the manuscripts had been stored after publication of the Polyglot Bothcopies contain the Aramaic text of Targum Judges in the left column and itsLatin translation in the right column

The First and Second Rabbinic Bible edited by Felix da Prato and Jacob benH ayyim respectively and published by Daniel Bomberg in 1517 and 152521

represent a parent and daughter text the first edition presumably based onCodex Solger now housed in Nuumlrnberg served as the basis of the second onehowever Jacob ben H ayyim evidently consulted one or more manuscripts inaddition to his exemplar22

Random variation and tree construction

To single out variant readings that do not unequivocally carry genealogicalvalue I will explore the phenomenon of lsquorandom variationrsquo in manuscriptreproduction While isolating these phenomena I will pay attention to thelevels of inquiry at which they can still have their say because they add in-formation that we cannot afford to lose The following fields of variants weredistinguished

a Orthographyb Errorc Separation and contractiond Vocalizatione Abbreviation

f Plusg Minush Substitutioni Semantic shiftj Preposition and copulative


k Relative particlel Verbal morphologym Number

n Persono Gender

p Status

Inevitably the categorization is heuristic and the categories themselves areblurred23 This particular classification is devised to test the reliability of cer-tain variants whose reliability has been called into question in previous studieswhich tend to single out certain verbal and nominal variants24 These lat-ter variants numbered f to i are deliberatedly contrasted with the remainingcategories that are deemed to produce much less reliable results such as or-thography and morphology The following observations will only produce atentative result in order to answer two questions (1) Which variants can befruitfully employed to build a provisional stemma And (2) to what extent doall variant readings shed light on particular relations between textual witnesses

Since the first five categories have been discussed in greater detail else-where25 the following discussion will focus on the remaining categories how-ever I will summarize my findings for the first ones

Orthography

The vast majority of variant readings between the two copies of Alfonso deZamora concerns the use of vowel letters (70) The variation shows thatthese variants do not bear on his copiesrsquo relationship to their exemplar hencethey obscure rather than illuminate genealogical relationships This conclu-sion should not be applied to all textual witnesses however since the FirstGreat Rabbinic Bible and its successor the Second Great Rabbinic Bible al-most always agree in their use of vowel letters From these observations twoconclusions can be inferred spelling variants should not be admitted as evi-dence for the initial construction of the network between all textual witnessesall the same they should not be discarded altogether because they may corrob-orate our observations about certain individual relationships between textualwitnesses

Errors

Scribal errors represent unintentional changes that are quite common inmanuscript reproduction26 and may therefore have been introduced by scribesindependently of each other Common scribal errors like the omission of wordsbased on similar beginning or ending of clauses (homoioarcton homoioteleu-

Willem F Smelik

ton) and the confusion of similar letters occur in almost every manuscriptThese errors may have been transmitted down a family line when they havebeen faithfully transmitted by copyists In such cases they reveal a genealogicalrelationship between these manuscripts as for example between the closelyrelated Yemenite manuscripts Or 2371 and 1471 housed in the British Li-brary which share four unique cases of haplography27 as well as small minusesunique to them elsewhere28

Nonetheless common errors can never be taken to represent genealogicalinformation at face value for two reasons29 Scribal slips are often typical andsusceptible to repetition indeed examples of shared but no doubt unrelatederrors are found in several verses Moreover errors were susceptible to correc-tion Phonetic errors and Hebraisms could easily be reverted into the originalreading in a later copy30 and minuses could be restored on the basis of theparallel transmitted Hebrew text

Only if there is circumstantial evidence to support the assumption of a realgenealogical connection between the mss sharing a scribal error would we takethis type of evidence into consideration

Vocalization

The use of variant vocalization is fraught with difficulties The systems of vocal-ization differ the scribe did not always vocalize the text himself31 or as seemslikely in case of Alfonso de Zamora he invented the vocalization himself

Separation and contraction

The separation and contraction of prefixed morphemes such as andresulted in a variation that appears to reflect scribal conventions and

sometimes typographical ad hoc considerations rather than genealogical in-formation

Abbreviation

Abbreviations reflect scribal conventions and typographical considerationshence they vary by text They are only useful if a copyist made a mistake whenhe filled out an abbreviation in his exemplar or when the versions differ in theirreadings Some of the variants agree with each other by pure chance becausecopyists filled out abbreviated words32


Grammatical properties

Many variant readings concern a change of grammatical properties like gen-der number status and morphology Some of these variant readings are in-spired by a zeal for grammatical improvement for instance the introductionof the feminine third person plural suffix which in earlier manuscriptsseems not to have been distinguished from the masculine form 33 As gram-matical lsquoimprovementsrsquo if indeed that is what they are they may have beenintroduced independently by different scribes at different times Such exam-ples would not carry genealogical information but rather would distort ourpicture of the network between the manuscripts and editions Table 1 providesmany examples of improvements on the part of Jacob ben H ayyim who editedthe Second Rabbinic Bible for Daniel Bomberg On the other hand these datawill sometimes reveal a dialectal kinship between groups of texts as with thegender of the word which is invariably masculine in the Eastern texts butfeminine in the Western ones

Another frequent variation concerns the number of nouns after numer-als consider the following variation in Table 2 below34 Some of these textsof which we either know or suspect a close relationship nicely fit togetherbut there are many irregularities to be noted as well On the assumption thatthe Babylonian and Yemenite manuscripts represent the oldest text-type thesingular state of the noun is original and the lack of variation except for oneposition in Eb66 is significant The most likely explanation for the variationin this table is that copyists occasionally felt that a plural form was necessaryhowever they may have made their lsquocorrectionsrsquo almost unconsciously as thereare blatant inconsistencies even within a single chapter One should also takeinto account that the syntagma is followed by a plural participle in 181116 and 17 some versions correct this plural form into the singular but onceagain inconsistently35 As a consequence these lsquocorrectionsrsquo may well have beenintroduced independently from each other representing a polygenetic irrele-vant type of variant readings as some variations do indeed suggest Thus thedisagreements cannot be relied upon whereas the agreements appear to be re-liable To deal with these readings it seems advisable to create a separate classof readings so that the validity of this class including many occurrences notincluded in Table 2 can be assessed in comparison with other classes Ratherthan curbing their possible blurring effect by using value factors in computa-tional approaches or by their immediate identification as noise I would arguein favour of testing them in isolation

Willem F Smelik

Table 1 Changes of grammatical properties in Rb236

A variation of person especially in verbal forms is less easy to evaluate asthe occurrences are not frequent enough to isolate them The source of thesevariant readings is equally difficult to pinpoint The variant for


Table 2 vs 38

Variation placeSiglum 331 76 77 78 716 719 84 1811 1816 1817 2047

p s s s s s s s s s s sEb1 ndash s s s s s ndash s s s sx s s s s s s s s s s sy s s s s s s s s s s sw s s s s s s s s s s sq63 s s s s s s s s s s sEb66 s s s s s s s s pa s sK s s s s s s pa s s s sB s s s s s s s s pa s sO s pa s s s s s pa s sS s s pa s s s s s pe s sW pc pc pa s s s s s pa s sM pc pc pa s s s s s pa s sF s s s s s s s pa pa s sP s s s s s s s pa pa s sT s s s s s s s pa pa s sd s s pc s s s s pa pa s sE s s p39 s s s s pa pa s sa pc s s s s s s pa pa s sC pe s s s s s s pa pa s sA s s s s pa pa s pa s s sJ s s s s pa pe pa pa pa s sQ s s s s s s s pa pa s pa

N pa s pa s s s pe pa pa s sRb1 pa s pa s s s pe pa pa s pa

Rb2 pa s pa s s s pe pa pa s s

in 1119 for instance may well reflect influence of Deut 227on either the Hebrew or the Aramaic text or on a later corrector working fromanother exemplar with a different text The variant for in 118 is lesslikely to reflect intertextual influence here however an error is not inconceiv-able Both readings however contradict the Masoretic Text (mt) and appearin old Eastern manuscripts which is a factor to be considered as it is an oldrule of thumb that readings contradicting the mt are important for the textualcriticism of the Hebrew Bible40

Willem F Smelik

Table 3 Substitutions and semantic shifts in Rb2

Substitutions and semantic shifts

In the translation some equivalents for Hebrew lexemes do not overlap withthese lexemes on a semantic level Such equivalents are called substitutionsThey obviously represent purposeful choices in the sample texts our sin-gle example occurs in 824 More frequent is the use of an equivalent whichpartly overlaps with the alternative reading a phenomenon labelled as a se-mantic shift41 Together the examples occurring in the Second Great RabbinicBible are listed in Table 3 I will include in these categories verbs nouns andadjectives only42 conjunctions particles and prepositions will be discussedseparately

Which substitutions are susceptible to multiple introduction Some sub-stitutions reflect a surprise about the choice of translation equivalents when astereotyped equivalent was not adopted at these points variant readings occurthat prefer the more usual equivalent In fact the Masorah to Targum Onqelosseems to be designed at least in part to report exceptional translations so asto prevent alterations resulting from hyper-criticism Other substitutions arerelated to Hebrew variant readings43 these are likely to be introduced by morethan a single scribe and thus not reliable as unique mutations in the textualhistory

Similarly it stands to reason that semantic shifts or near-synonyms whichreflect dialectical developments such as for in 1644 cannot be acceptedas reliable evidence at face value obsolete words may well have been replacedby later copyists (substitution by synonyms)45 But the extent of such practicesshould be investigated and verified before assuming this hypothesis to be trueReadings introduced on the basis of a Hebrew text can easily be pointed out inthe textual history of Targum Judges whether semantic shifts due to dialecticalprocesses occur is far less clear46

A fair number of substitutions and semantic shifts consist of theologicaland exegetical changes which have been inspired by non-textual considera-


tions47 yet they are often not so obvious that they could have been introducedindependently by several scribes without one knowing about the other Thechange recorded for Rb2 in 824 is far from obvious even though it reflects anexegetical concern that may have been shared by others as well Other exam-ples are related to established exegetical discussions and may involve a series ofsubstitutions or semantic shifts a good example is a set of semantic shifts inchs 17 and 1848 The more complicated such an operation of alteration is theless likely it becomes that several scribes carried out the same activity

There is a real chance of contamination here for differences such as thesewould not go unnoticed They would often be scribbled in the margins ofmanuscripts either as an alternative reading or as a correction Whether theeditor of Rb2 for instance noticed the example listed above himself or foundthem in the margins of his (master) exemplar is unclear the source for theseparticular changes cannot readily be identified These readings result in con-tamination because they do not reflect the Vorlage of Rb1 yet they do notseem to have originated with the editor without recourse to alternative textsFor our purposes however the real issue is whether the reading is susceptibleto independent introduction by more than one scribe lsquoSimultaneous contam-inationrsquo which occured when a scribe consulted other sources in addition tohis exemplar49 is not a problem to eradicate because it is stemmatologicallyrelevant it is a phenomenon to detect and preserve Random variation is thereal culprit

In sum most of the semantic shifts and substitutions should be evaluatedon a case by case basis Those reflecting Hebrew variant readings should notbe taken into account all others should be tested The presence of polygeneticreadings is possible in this category as a whole so one must interpret cautiouslyUnfortunately the relatively small number of such readings does not allow aseparate test to be carried out on them

Pluses and minuses

Pluses and minuses are not invariably valuable so that different types of plusesand minuses should be distinguished The majority of minuses reflect typicalscribal errors especially haplography50 and a good many pluses may reflect theinfluence of the Hebrew text The following examples in Rb2 (Table 4) illustratehow limited the value of such variants may be

Three of these examples in Rb2 reflect the influence of the Hebrew text(720 722 1135) although the first one is shared by all other witnesses andmay therefore also reflect consultation of other Aramaic texts one is an error

Willem F Smelik

Table 4 Plus and minus in Rb2

on the part of the editor or typesetter (1129) another a correction of an errorin the exemplar (1131) and the sixth represents a Sefardi () influence (con-tamination 1139) None of these readings reflects a trustworthy genealogicalrelationship

The minuses and pluses in the copies of Alfonso de Zamora printed in Ta-ble 5 do not fare very well either Alfonso de Zamora strove to achieve a faithfulcopy of his exemplar51 as his corrections show (marked by lsquomgrsquo) Any minus hasusually been restored in the margin by Alfonso de Zamora himself while theequivalent of the minus in the parallel Latin translation confirms that it reallyis a scribal error52 Nonetheless many of these minuses are shared at randomby other mss which points to independently occurring but identical errors

The evaluation of the translation plays a prominent part in the creationof variant readings When later tradents readers or correctors evaluated theadequacy of the translation and felt it did not accurately reflect the Hebrewthey introduced changes or noted alternatives The phenomenon of doubletranslations allows us to detect such later alterations of the text when a copy-ist combined two alternative translations53 apparently on the assumption thatone of them perhaps written in the margin had been omitted by accident ndashor perhaps he was unsure which one to select In most cases to be sure thesealternatives were not combined into an erroneous double translation54 andthe only trace of such evaluations is a plus or minus in some of the textualwitnesses that do not appear to be mere errors55 A particularly fine exampleis found in 21 where the (majority) translation for Hebrew reads

56 some copyists or glossators apparently believed the He-brew construction should be preserved literally hence in a handful of Yemenitemanuscripts the more lsquoliteralrsquo translation replaced the longer read-ing57 whereas in other manuscripts both translations were jumbled together

58

A further source of minuses and pluses consists of actual Hebrew variantreadings that have found their way into the Aramaic text Whether these pluses


Table 5 Pluses and minuses in mss 7542 and M-1

belonged to the lsquooriginalrsquo text of Targum or have been introduced into thetext at a later stage often remains obscure59 In fact copyists often would altertheir readings adjusting their text to a Hebrew variant in their bilingual exem-plar60 Interestingly a number of such instances has been found in manuscriptsof poor grammatical quality This also illustrates that no matter how poor amanuscript may be it should not too readily be omitted from an edition evenif related manuscripts in a better condition exist

Finally some pluses may have resulted from standardized language or la-cunary exemplars In 636 the plus to is indicative of a standardcombination (K has the plus at this point) so that this reading does not nec-essarily have any genealogical value Other variant readings may have resultedfrom difficulties in reading the exemplar

Willem F Smelik

It is hard to assess the actual degree of parallelism ndash that is accidentalagreement in change ndash in a fair but consistent way The combined reading in21 which was discussed above is likely to reflect accidental agreement betweenthe texts that attest it rather than a genealogical relationship Several scribesmay have included in their main text a reading that was noted in the marginof their exemplars (namely ) On the other hand the two other readings (

and ) do reveal a genealogical relationship and wemay have a fair idea about the original reading as well as later developments sothat these instances can often still be fruitfully employed In other words partof the lsquoformularsquo is reliable and part of it is not61 For the purpose of evalua-tion a lsquosafersquo file without these instances should be preferred the informationthus left out of consideration can always be brought to bear upon our under-standing of the genealogy in detailed refinement Other cases to be excludedare the minuses that reflect typical scribal errors and the pluses that involvestereotyped phrasing will be categorized separately

Prepositions and

In the sample selection of the two copies produced by Alfonso de Zamora onlya single difference in their use of prepositions occurs which is due to scribalerror (M reading instead of interestingly the Antwerp polyglot whichis based on Mrsquos exemplar has the same mistake) Rb2 has a different prepo-sition than its exemplar Rb1 in six instances (see Table 6) Of these six noless than five agree with the Hebrew (Masoretic) Text which suggests that theparent text casted a long shadow over its translations (even if some of theseHebraisms may have been suggested by the consultation of other Aramaicversions they ultimately appear to be Hebraisms) Apart from such reversalsto Hebrew language prepositions may also have been changed for idiolectchanges or idiomatic adaptations The only instance that is not a Hebraism

Table 6 Prepositions in Rb2


Table 7 The particles in Rb2

(716) consists of a grammatical improvement None of these examples repre-sents a reliable source of information for building a stemma therefore theirimpact on tree construction should be considered separately

The plus or minus of the copulative should be studied in conjunctionwith Hebrew textual criticism which falls outside the scope of the presentstudy71 The variation in using reflects inner-Aramaic variation which is notso prone to Hebrew influence On the other hand could as easily be intro-duced as omitted in many instances as Table 7 of such changes in Rb2 mightsuggest and it seems wisest to refrain from the use of these readings in buildingthe initial stemmas

Additional types of variants

The present survey has not been exhaustive but has focussed on the mostfrequently found and relevant phenomena For example auxiliary verbs trans-position and paratextual elements have not been included

Auxiliary verbs do not occur at random in targumic literature and theirpresence or absence which is rejected as inflectional parallelism by Salemans72

should accordingly be included in the variant selectionTranspositions do not occur often in this particular textual tradition for

obvious reasons primarily because these errors would have been easily de-tected in comparison with the Hebrew text which the Targum either physicallyor functionally accompanied The Aramaic translation tends to mirror the He-brew word order faithfully73 for this reason inversion of verses is rare as well74

Paratextual elements however shed a far more interesting light on thegenealogy of manuscripts The Toseftot lsquoadditionsrsquo were often added in themargin of texts but as an alternative translation they were soon incorporatedin the body of the text here juxtaposed to the original translation there oust-ing that lsquooriginalrsquo reading As such their genealogical value and textual position

Willem F Smelik

may vary between exemplar and copy especially because the marginal readingsmay not have been written by the same scribe who signed for the body of thetext75

Evaluation

Methodology

It is one thing to argue for categories of variants but another to test them Evenif we try to keep our assumptions in check by comparing the variation betweentexts whose relation is known to us it seems worthwhile to evaluate the resultseven further

Two tools have been used in the following evaluation similarity graphsand shock waves Shock waves are used to detect possible non-linear influencesbetween the (sub)groups of textual witnesses usually called lsquocontaminationrsquoSuccessive contamination occurs when a copyist (or editor) switched from oneto another exemplar at some position in the text up to this position his copyfollows exemplar X thereafter exemplar Y To detect such shifts in relationsWattel and Van Mulken developed the instrument of shock waves which peakwherever a change of affinity between textual witnesses occurs76

Similarity graphs show the relations between the witnesses on the basisof each category and it is possible to construct several images of these wit-ness relations to see how these images vary among themselves and whereirregularities may point to important information about either my classifica-tions or the data The images or genealogical trees represent the network ofrelations between the textual witnesses Their construction follows the proce-dures as developed by Evert Wattel and they can be based both on similarityscores between witnesses and on dichotomies of pairs of witnesses the latterby quadruple calculations Thus it is possible to create a set of images for eachcategory To see the wood for the trees only a few tell-tale graphs based on thesame method will be reproduced here77 Finally the known relations betweenmanuscripts and editions as outlined above will be brought to bear upon thesetrees wherever the trees are in conflict with these data the construction musthave been wrong and since the procedures have been tested with great care itis likely that my selection of readings was unreliable

Some of the categories produced a sufficient amount of variant readings tobe processed separately but most did not Therefore they were combined intothe following groups


1 Scribal conventions orthography separation and contraction (or wordboundaries)

2 Scribal mistakes errors including erroneous changes in the suffixed pro-nouns78

3 Grammatical properties (nouns and verbs) number gender state4 Variation in other wordgroups and inflection preposition copulative con-

junction tense and conjugation5 Nouns and verbs plus minus semantic shift and substitution6 Nouns and verbs selected instances (each case that might be an error or

could reflect non-linear influences has been omitted after a case by caseevaluation)

These groups essentially contrast nominal and verbal variations with those inother wordgroups inflection errors and spelling The shock waves for all thesegroups did not display a significant shift of affinity not even on the border-lines of categories that were combined into a single file This result justifiesthe combinative approach no matter how pragmatic it is It also suggests thatthere is no case of successive contamination among any of the witnesses in-volved Nonetheless there are some interesting observations to be made onthese shock waves detailed comments will follow below wherever relevant fora category and separately in an assessment of the extent to which the traditionis contaminated

The first test that these trees and shock waves pass is based on the rela-tions between texts which are known to us the two copies made by Alfonso deZamora the Nuumlrnberger Codex and its daughter-editions and the two affili-ated Yemenite mss bearing the sigla lsquowrsquo and lsquoyrsquo in Sperberrsquos edition The graphsin which the direct relations between these witnesses have been distorted dis-credit the data on which they were based Wherever there is little hard evidencewe must resort to other ways to assess our results A heuristic way is to evaluatethe cohesion of different images which are based on a single category In somecases the trees for one category do not agree among themselves whereas inother categories the trees are much more homogeneous suggesting that theselatter categories are more reliable

Category 1 Spelling

The first category contains all orthographic variant readings and this file wasexpected to produce the poorest result Even though the picture yielded by thereadings in this category shares a few groupings with those based on the other

Willem F Smelik

Eb4

Eb91

Rb6

Rb3

Eb1

s4

q2

D

G

o

Eb66

x

p

d

q63

K

F

C

Q

O

y

Rb2

T

W

E

w

J

a

Rb1

S

A

B

P

N

M

total

15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 2700

Figure 1 Shock waves for Category 2

categories79 it is essentially flawed External evidence as well as careful com-parison with the other categories proves that a number of witnesses have endedup in the wrong position The Antwerp polyglot o based on the same exem-plar as W and M has not been grouped together with these texts The siglaEb1 Eb91 and Eb4 (all Babylonian manuscripts) have been separated from


the mainstream of Eastern manuscripts to which they undoubtedly belongLikewise but less marked Eb66 and q2 have been separated from this branchIn almost all other trees they are grouped together Similarly S has not beengrouped together with W M O as in four out of six categories Likewise theclose connection between a A B J and Q has been distorted

Although some of the information agrees with that presented by the re-maining categories the overall picture is not reliable It should be remarkedthat G and s4 two of the haftaroth contain too little evidence to producereliable results hence they float around from tree to tree

Category 2 Errors

The shock waves for all obvious scribal errors are marked by pointed waves andstable relationships in between see Figure 1 In many cases the peaks occurwhere an obvious error has been corrected in subsequent copies either of thescribersquos own accord or by following a marginal correction A point in case isprovided by some errors in Codex Solger where the First and Second RabbinicBible follow its marginal corrections Such corrected errors blur the picture asthe shock waves seem to illustrate which confirms the point that errors shouldonly be used as secondary evidence in the final evaluation of a stemma

This file results in a tree that completely distorts the picture for the Sefardimss M W O S see Figure 2 Whereas we know that M and W have beencopied by the same scribe either from the same exemplar or from each othernothing in this tree suggests as such As a result of editorial activity a numberof errors in N and Rb1 have been corrected in subsequent editions indeedthese editions have been torn apart in the present picture

The quadruple method achieved a somewhat better result but separatedM and W besides other errors such as the separation of w y even in a categorywhere I had expected them to be brought together (see Section 22) All in allthis category seems to yield the most unreliable results of all

Category 3 Grammatical properties

The third category which includes all grammatical properties like numbergender and state achieves far more stable results The Sefardi group80 the Sol-ger group the late Ashkenazi group and the Eastern subgroups agree more orless with the remaining categories

As for the shock waves none of the peaks displayed in the bottom linewas remarkable and minor peaks appear to reflect either minor divisions or

Willem F Smelik

ˆ34

15

2

5

ˆ33

542

11 37

P

J

ˆ32

33

B

23

ˆ31

1141

AD

ˆ30

4 7

Cˆ29

84

ˆ16

33

NRb1

ˆ28

16

T

4

ˆ27

14

S

3

ˆ26

9

O

3

ˆ25

32

ˆ24

2

ˆ21

3

ˆ23

3 35

W

7

ˆ9

1 6

Eˆ3

1 5

G Eb4

ˆ20 ˆ19

34

ˆ17 ˆ5

63

Kˆ18

4

ˆ1

018

30

ˆ4 Rb2

2 2

p Mˆ14

12

ˆ13 ˆ12

22

ˆ11

3

ˆ2

3

ˆ7

1 2

Eb1F

3 1

s4 q63

5 2

d o

Eb91

4 4

Eb66

2

ˆ10

2

ˆ6ˆ8

4 1

q2 x

ˆ22

44

w y

0 2

Rb3Rb6

116

Q a

Figure 2 Initial tree for Category 2

noise An example is the singular (1011) in a Yemenite subgroup (x yw q63) which agrees with a single Ashkenazi manuscript B and with the Rab-binic Bibles Rb1 and Rb2 over against the plural in all other witnessesThe reading of Rb1 can be explained as a result of an abbreviation in N its ex-emplar filled out independently Rb2 here follows Rb1 These witnesses agree


with the Hebrew text in having a singular the singular in B is another exampleof independent adaptation to the Hebrew text81

The results are flawed according to the pairing method (based on similarityscores) which separates the Eastern subgroups from one another (w y x q63and q2 Eb1 p Eb91 Eb66) In a graph produced according to the quadruplemethod this dichotomy has been removed as the agreement with graphs forthe following categories shows this must be considered an improvement82 Thedifference between the two results based on the same data is remarkable initself and suggests that the data still suffer from noise

Apart from some floating nodes which are not too problematic or vaguepositions where the data are limited (eg Eb4 11 Eb91 13 G 16) thereis some difficulty in grouping the Ashkenazi texts example a J Q AB and theBomberg group N Rb1 Rb2 Rb3 Rb6 and E K and d The grouping does notappear to be correct in its entirety in comparison to the images based on thefollowing categories as if the divisions still have to become more pronounced

Category 4 Prepositions copula relativa

The shock wave for the fourth category displayed a few peaks a fact that pointsout unexpected alliances that usually represent polygenesis For example thereading in 1121 for in some witnesses which we would not nor-mally consider to represent a group this variant renders the text smoother thanit was Hence it appears to be likely that this variant reading was introducedmore than once83 Another example concerns the phrase(722) in which four Yemenite mss read which is more an idiomaticuse of language In the same verse the geographic indication has beenchanged into in most witnesses except for a hotchpotch of eight mss thatmay well have copied the Hebrew wording in their Aramaic text84

While the graphs for this category offer a slight improvement in the East-ern grouping as well as in their cohesion a few odd positions are noteworthyq2 (32 formulas) and Eb91 (15) are dislocated and so is T (95) Thequadruple method provides a far more reliable result here and elsewhere asq2 and Eb91 here join the Eastern group This reflects the real situation bettersince these witnesses always join the main group and never disagree with ptherefore the pairing method based on similarity scores distorts the stemmahere However the quadruplesrsquo graph does not give a proper picture for Eb1(33) and d the incunable of Leiria the incunable is linked to the Sefardiwitnesses in contrast to the pictures yielded for the previous and the two fol-lowing categories85 Although we know this text was produced in Portugal

Willem F Smelik

43

ˆ34

ˆ33

7 31

a

J

3211

ˆ32

B

A

249

12

ˆ31

8

ˆ30 ˆ27

8 12

Qˆ29

14

ˆ18

7

ˆ28

66

ˆ14

14

d

7

ˆ25

107

Kˆ24

8 10

Cˆ23

12

E

7

ˆ22

5 13

ˆ20

75

ˆ17

5

ˆ16

7

54

ˆ15

7

T

4

ˆ10

2

Eb66

3

ˆ7

22

ˆ4

22 0

ˆ3

p

11

s4 Eb4

w y

11

ˆ8 x

11

ˆ11 q63

42

ˆ12

Eb1

F

P

1011

ˆ26 ˆ21

15

D

11

ˆ9

2

G

5

ˆ6

4 4

oˆ2

1 0

q2 Eb91

15

S

8

ˆ19

5 7

O

W

ˆ13

3 3

M

3 5

NRb1

ˆ5

11

ˆ1

20

Rb3

Rb2Rb6


hence a Sefardi link would not have come as a surprise the text is closer tothe NuumlrnbergndashBomberg group than to the Sefardi textual tradition The posi-tion of this edition requires further study The Antwerp Polyglot based on theexemplar from which M and W derive was disjoined from the Sefardi group inthe pairing tree and linked closer to O and S following the quadruple method


In brief the graphs seem to confirm that the results in this category are notentirely reliable

Category 5 Substitution plus minus semantic shift

More convincing are the results for the fifth and the sixth category but thedifferences between their graphs beg the question how to explain them TheEastern group of manuscripts and the Sefardi and Ashkenazi subgroups areobvious in each of the graphs but there is still some lack of clarity in the pairinggraph because w y x q63 was too far separated from the remainder p Eb1 Eb66Eb91 q2 this was resolved in the quadruple approach Nonetheless the graphsare inconsistent in their Eastern subdivisions whereas the pairing method hastwo main branches which it separates the quadruple approach resulted in onemain branch starting with w y x q63 and the remaining Eastern manuscripts asseveral sprouts

Between the various graphs some witnesses are floating The fragmentarycharacter of some texts partially accounts for this s4 (19) and Eb91 (14)may well be too lacunary to establish their exact position within the network Inany case s4 is attached to the first group according to the quadruple approachand to the second one according to the pairing method This picture is differentfrom that for the next category

Category 6 Selected substitutions pluses and shifts

In the last and very selective set of variants all readings which may have beencreated independently in more than one witness for other reasons than thecategorical ones are omitted Substitutions and semantic shift are susceptible tosecondary changes as explained in Section 27 and any of these which is likelyto have been introduced repeatedly by copyists has been omitted here Thecase for omitting minuses as evidence has been made above (28) includingpluses here obviously implies that I have made a decision in some cases that areading is unlikely to be authentic I have omitted all pluses and minuses thatcould somehow be related to a correction in the Aramaic which is based on theHebrew original all other instances were evaluated one by one

It is difficult to argue at this stage of my studies whether the refinementin this category produces an essential improvement over the former one al-though there are some indications that it does Both the pairing method andthe quadruple approach present the Eastern manuscripts as one group with twosubgroups w y x q63 and p Eb1 Eb66 q2 s4 apart from the floating manuscript

Willem F Smelik

ˆ34

43

A

15

ˆ33

29

B

29 17

ˆ31

1413

ˆ30

ˆ32

23

J

26

ˆ28

12

a

12

Q

ˆ25 F

8 9

106

Kˆ24

6 7

ˆ21ˆ23

6 8 6 6

C P

G

11

ˆ7ˆ22

7 6

ˆ18

9 6

ˆ16 ˆ15

711

q63ˆ12

3

ˆ9

2

x

2 1

w y

6 5

pˆ14

4 6

ˆ5ˆ11

3 3

Eb1ˆ6

3 1

Eb91

4 0

s4ˆ1

1

Eb66

0

q2

30

D E

Tˆ4

6 7

ˆ17 Eb4

11 32

ˆ27 ˆ26

7 10

ˆ20 S

9 6

ˆ10 O

o

2 2

ˆ8

2 0

M W

ˆ29

22 14

dˆ19

10 6

ˆ3ˆ13

5 5

NRb1

2 0

ˆ2 Rb6

0 2

Rb2Rb3


Eb91 (16) there is no real difference between these graphs The cohesion ofthe network images has thus been improved The floater s4 (12) is now firmlyplaced within the second group Eb4 (13) admittedly is floating around as ithas been doing lsquoconsistentlyrsquo Overall the picture is clearer than that of the pre-


vious category but whether we have resolved the tree or eliminated dissensionremains to be argued

A and B are more closely linked together than in the former category overagainst a J Q in the same branch of a later Ashkenazi type of text This changeconfirms how influential the presence of data can be because all the formu-las in this category are present in the former as well This influence may beexplained in two ways either the small amount of data used here distorts thepicture alternatively it can be argued that the use of unreliable data in theformer category has undesirable effects

Contamination as a characteristic

The cardiograms offer one surprising phenomenon the categories which weredeemed most reliable have the most unstable shock waves while the utterly un-reliable first category has less pointed shock waves compare Figures 1 5 and6 To some extent we should not attribute too much importance to this phe-nomenon because a lower number of readings increases the pictorial impactof the waves However there is more to this phenomenon than representationalone The shock waves for Categories 3 and 4 were also more stable than thosefor Category 5 (and 6) To put these images in perspective we may briefly con-sider the highly stable pictures achieved for some traditions and the less stablepictures for others in previous studies86 There is sufficient reason to assumethat the textual tradition of Targum Judges is essentially perhaps sui generis87

contaminated I believe that this degree of contamination bears on the stabil-ity of the shock waves Further study of contaminated and non-contaminatedtraditions should verify or falsify this assumption

The contamination is not of the successive type where a copyist (or ed-itor) switched from one to another exemplar at some position in the textRather simultaneous and incidental contamination where a copyist consultsanother exemplar either throughout the process of copying or incidentallyhave considerably influenced the text relations88 The phenomenon of par-allelism should be added to this picture in many positions a lsquoself-evidentrsquocorrection may have occured to several scribes who were quite unaware of oneanother Such activity results in agreements that do not reflect any real relationbetween the texts89

The authoritative status of the Hebrew parent text of the translation is themain reason for this high level of contamination (hereafter used as a shorthandfor contamination and parallelism) The Aramaic text was never intended to beused on its own in isolation from the Hebrew original on which it depended

Willem F Smelik

o

Eb4

D

Eb91

Rb3

G

Rb6

s4

q2

Eb1

C

B

Q

Eb66

T

A

S

J

w

Rb2

F

M

x

K

d

N

E

a

y

o

p

Rb1

W

P

q63

total

0 90 180 270 360 450 540 630 720 810 900 990 1080 1170 1260 1350 1440 1530


and with which it was often transmitted When the Hebrew and Aramaic alter-nate by a single verse in a manuscript it stands to reason that the scribe maystill have had a Hebrew word in mind when he was writing the Aramaic textSubsequent correctors have been quite active too We have already had occa-sion to refer to readings where one group of texts reads A another B and a


third one simply took them together A B (see n 53 above) Such conflationsare typical of a tradition that was subjected to continuous gauging under theinfluence of the original text variant versions and the authoritative rabbinictradition For that reason this textual tradition is contaminated to the boneswould not the purest file then produce unstable (but reliable) relationshipswhereas from a genealogical point of view fringe elements such as errors andspelling may produce a more stable (and yet less reliable) result

It is perhaps useful to point out that instances of contamination are truereflections of how texts were related it is the complexity they create that we arewary of but the readings themselves and the relations they reflect should notbe dismissed lightly While accidental agreements are genealogically irrelevantcontamination is not

Epilogue

To construe an image of the witness relations of a textual tradition we mustnot proceed blindly and include either all readings or a selection that has notbeen accounted for by philological considerations and empirical analysis Eventhough for some text traditions all variant readings seem to produce reliableresults as the high level of consistency which August den Hollander observedin the readings of Dutch Bible translations illustrates90 it goes without sayingthat such a consistency cannot be assumed by definition but must always bedemonstrated

Philological insights inform the classification and selection of variants Allvariant readings are important but some are more important than others andthe degree of difference should be worked out in detail for each textual tradi-tion The Aramaic Bible translation of the Prophets presents its own case Forthe genealogy of this particular tradition there is no watertight case to be madefrom variations of orthography vocalization separation and contraction andgrammatical properties like gender number status and morphology these as-pects vary unpredictably within (phases of) the language Variants then areto be expected against the grain of a tree Codicological arguments accountfor additional categories Errors and abbreviations in manuscripts and editionsare not unassailable as the comparison between closely related witnesses hasdemonstrated hence such variations are unreliable source materials for treeconstruction

The genealogical value of lexemic variations of nouns adjectives and verbsproved to be ambiguous Wherever the Hebrew original may have prompted a

Willem F Smelik

G

Eb4

Eb91

s4

D

Rb6

Rb3

q2

Eb1

Eb66

E

T

c

o

P

K

d

q63

F

x

B

O

Q

A

N

w

M

Rb2

J

W

y

a

Rb1

S

p

total

0 32 64 96 128 160 192 224 256 288 320 352 384 416 448 480 512 544 576


lexemic change in the Targum such changes may easily have been reproducedindependently by scribes working in different places and periods Dialecticaldevelopments may well account for a number of semantic shifts between tex-tual witnesses and so do exegetical considerations these cases should not begeneralized but have to be evaluated on an individual basis91


Determining categories on the basis of general philological and codico-logical considerations is one thing to keep all these premises in check anotherThe main thrust of this study is the evaluation of all categories of variation intwo steps First it was suggested that such variants between multiple copies ofa single exemplar or of copies produced from established exemplars wouldprovide the necessary hard evidence Needless to add such information maywell be supplemented by for example codicological data Then the categorieswere tested by a series of computations that were aiming at falsification of ourassumptions not at an optimal tree

The variants between manuscripts and editions of Targum Judges whichare valuable as a first model for the Targum of the Prophets as a whole92 even-tually confirmed the general point that constructing a tree requires rigorousselection of readings The categories of error and orthography are notoriousrogues Their graphs disagreed with the others and were incoherent The nexttwo categories of grammatical properties prepositions copula and relativaproduce ambiguous results their trees are quite close to the trees produced forthe last two categories but there is some evidence that they distort known rela-tions between textual witnesses Their graphs (including those not reproducedhere) lack the cohesion of those for the last two categories Hence it seemswisest to refrain from using their input in the initial tree construction whiletheir input should not be disregarded in a case by case evaluation The last twocategories of semantic shift substitution plus and minus produce the best re-sults in terms of hard evidence and cohesion Further testing will be necessaryto evaluate the differences which occur at this level especially between the fifthand sixth category the danger of refining onersquos statistical source materials toomuch should never be underestimated

At this juncture the question should be raised whether the problem ofcontamination and parallelism has been overcome This question can not beanswered with either lsquoyesrsquo or lsquonorsquo the procedures followed here usually aimat an optimal tree rather than a historically correct representation of text re-lations Yes the final picture could be confirmed by some hard evidence butno not all details of the picture could be confirmed by lack of evidence Thedifferences between the graphs for the fifth and sixth category do not warrantoverconfidence

A case should be made for the preservation of all readings since none ofthe categories analyzed above proved to be absolutely irrelevant Many of thecategories listed above may in possibly varying degree have been introducedby scribes independently from each other This polygenetic nature of thesecategories renders them unsuitable for general genealogical purposes but it

Willem F Smelik

should be stressed that such readings remain valuable Corrections of num-ber in nouns and verbs for example seem to have flown easily from a scribersquospen hence rendering such readings less useful for stemmatological purposesAt the same time however most of such readings will not have been changedfrom one copy to the other As a result they are useful for the detailed study ofsubgroups and the correlation of findings with the actual readings themselvesSpelling may reveal the provenance of a witness Shared errors may reveal in-termediacy in some cases Even abbreviations as demonstrated may be ofvalue here The course of individual readings can be illuminated from the mostunexpected perspectives Moreover the readings themselves may be authentic

Our images of text relations should be taken as a heuristic device to under-stand the course of variant readings93 Parallelism incidental and simultaneouscontamination lead to a variety of historical relations which these images can-not capture Only to some extent will peaks in the cardiograms reveal positionswhere such crossovers occur The contaminating readings may however pre-serve the better text even in a witness that is considered to be of less genericvalue for the lost original As a result the value of trees for individual read-ings should not be over-estimated in particular not in a highly contaminatedtradition

If trees cannot predict the authenticity of readings94 it is perhaps lessappropriate to use such trees for the selection of manuscripts in a criticaledition as has recently been proposed by Houtman for Targum Isaiah95 Arewe losing relevant readings if we select those manuscripts which are the bestrepresentatives of a (sub)group We should consider the possibility that poormanuscripts may have unique readings to offer even though or because theyare not the best representatives of their (sub)group Slightly odd and con-taminated manuscripts are very valuable for the readings they preserve Forthe Targum of the Prophets some poor copies riddled with errors springto mind all of which have some unique readings but none of which repre-sents the Ashkenazi texts very well96 Weitzman in an important study of thePeshitta the Syriac Bible translation has recently highlighted the importanceof so-called lsquopolesrsquo in a manuscript map97 this aspect should also be taken intoaccount especially when the number of complete manuscripts is rather smallas it is here Of course such selection procedures are inevitable when the num-ber of manuscripts becomes too large for comprehensive collation even thenhowever these considerations should play a prominent part

Finally it is interesting briefly to compare the results achieved here withthe excellent discussion of characteristics that Salemans has provided andwhich has inspired the present investigation98 Although I have emphasized the


phenomenon of hybridization based on the specific conditions of targumic lit-erature and have taken a different view of the importance of word order andauxiliary verbs99 the results appear to confirm his insistence on the use of vari-ant lexemes consisting of nouns and verbs only100 This is a remarkable resultand shows that philological observations may benefit from taking generalizedprinciples into account based on a different tradition and a different lan-guage Such principles however should always be evaluated by the philologistbecause each tradition has its own peculiarities as has been demonstrated

Notes

A shortened version of this paper was read at the Workshop for Stemmatology Vrije Uni-versiteit Amsterdam 13 October 2000 I would like to thank Piet van Reenen for his kindinvitation to participate in this stimulating workshop This article is also published in AS 12(London Continuum Press 2003) 247ndash287

See A Berliner Targum Onkelos (Berlin Gorzelanczyk 1884) II p 112 P ChurginTargum Jonathan to the Prophets (New Haven Yale University Press 1907 [= 1927])

See S A Kaufman lsquoA Unique Magic Bowl from Nippurrsquo JNES 32 (1973) pp 170ndash174C Muumlller-Kessler lsquoThe Earliest Evidence for Targum Onqelos from Mesopotamiarsquo JAB 3(2001) pp 181ndash198

For the interface of oral and written transmission see Y Elman and I Gershoni (Eds)Transmitting Jewish Traditions Orality Textuality and Cultural Diffusion (New Haven YaleUniversity Press 2000) M Jaffee Torah in the Mouth Writing and Oral Tradition in Pales-tinian Judaism 200 bcendash400 ce (Oxford Oxford University Press 2001) W F SmeliklsquoOrality Manuscript Reproduction and the Targumsrsquo in A den Hollander U Schmid andW F Smelik (Eds) Paratext and Megatext as Channels of Jewish and Christian Traditions(JCP Leiden E J Brill 2003) For the cultural shift towards orality in the case of early Bibletranslations see W F Smelik lsquoThe Rabbinic Reception of Early Bible Translations as HolyWritings and Oral Torahrsquo JAB 1 (1999) pp 249ndash272

See Muumlller-Kessler lsquoThe Earliest Evidence for Targum Onqelosrsquo

Based on an extensive survey of library catalogues and sporadic examination in situ Ihave compiled a comprehensive but eclectic short-list of targumic manuscripts in whichthe information of the Institute of Microfilmed Hebrew Manuscripts has later been takeninto account I would like to thank B Richler for his invaluable assistance for providing theessential catalogue card copies Short-lists for the books of Psalms Judges Isaiah and Samuelhave been produced but it is hoped that the entire list will be revised in co-operation withDavid Kroeze and Dineke Houtman (Kampen Netherlands)

With the exception of Targum Hosea no stemma has ever been produced of the ex-tant manuscripts of Targum Jonathan J Ferrer i Costa El targum drsquoOsees en tradicioacuteiemenita (Colleccioacute de Tesis Doctorals Microfitxades 869 PhD dissertation Universitat deBarcelona 1991) Recently A Houtman has presented the first results of her studies into the

Willem F Smelik

stemma of Targum Isaiah although published without a stemma see A Houtman lsquoTextualTradition of Targum Jonathan to Isaiahrsquo in J Targarona Borraacutes and A Saacuteenz-Badillos (Eds)Jewish Studies at the Turn of the Twentieth Century (2 vols Leiden E J Brill 1999) I pp145ndash153 The situation is slightly better for the targums of Job Lamentations and CanticlesF J Fernaacutendez Vallina lsquoEl Targum de Jobrsquo (unpublished PhD dissertation Universidad deMadrid 1981) A van der Heide The Yemenite Tradition of the Targum of Lamentations Crit-ical Text and Analysis of the Variant Readings (Leiden E J Brill 1981) C Alonso Fontela lsquoElTargum al Cantar de los Cantares (Edicioacuten Criacutetica)rsquo (unpublished PhD dissertation Univer-sidad de Madrid 1987) D M Stec The Text of the Targum of Job Introduction and CriticalEdition (AGJU 20 Leiden E J Brill 1994) D Shepherd lsquoBefore Bomberg The Case of theTargum of Job in the Rabbinic Bible and the Solger Codexrsquo Bib 79 (1998) pp 360ndash379

A Sperber The Bible in Aramaic (5 vols Leiden E J Brill 1959ndash1973)

Followed by the Yemenite mss and finally the Western witnesses I hope to elaborate onthese findings which I presented at the third meeting of the International Organisation forTargum Studies (IOTS) Oslo July 1998 An eventual pedigree of the available textual evi-dence will be based on almost half of the book of Judges evenly distributed over beginningmiddle and end (11ndash41 628ndash832 1010ndash148 1712ndash1824 and 2047ndash2125 307 out ofthe 618 verses or 497 made up by the book of Judges) in correspondence with acceptedpracticeThe support for ms Or 2210-11 hereafter lsquoprsquo should be explored in greater detail and set inrelief with the Western textual evidence A brief additional exploration of the evidence in thebooks of Joshua to Kings suggested the possibility that some Yemenite manuscripts otherthan Sperberrsquos base text (ms Or 2210 British Library) have been squared with the Hebrew(Masoretic) text to some extent

Sperber The Bible in Aramaic IVB p 30

G Dalman Grammatik des juumldisch-palaumlstinischen Aramaumlisch (Darmstadt Wissenschaft-liche Buchgesellschaft 1905 [repr 1905]) p xvi

B J P Salemans Building Stemmas with the Computer in a Cladistic Neo-LachmannianWay The Case of Fourteen Text Versions of Lanseloet van Denemarken (Nijmegen NijmegenUniversity Press 2000) see also Salemans lsquoCladistics or the Resurrection of the Methodof Lachmannrsquo in P van Reenen and M J P van Mulken (Eds) Studies in Stemmatology(Amsterdam John Benjamins 1996) pp 3ndash70 K-H Uthemann lsquoWhich Variants are Usefulin Discovering the Deep Structure of the Manuscript Tradition of a Text Contra a So-calledEssentially Quantitative Approachrsquo pp 249ndash261 of the same volume

For example variations of word order or omission of clauses and verses are unreliablein most targumic texts as a result of the usually parallel transmitted Hebrew parent textwhich would alert many scribes and correctors to conspicuous differences and allow themto revert these changes Thus the fifth text-genealogical rule formulated by Salemans Build-ing Stemmas pp 81ndash85 which attributes genealogical information to differences of wordorder cannot be applied to the targums without modification because in this tradition therule interferes with his first principle that a variant lsquofits inconspicuously in a text versionrsquo(p 64) The modification does not falsify Salemansrsquo principles For a fuller discussion seebelow under lsquoSubstitution and Semantic Shiftrsquo and lsquoMinus and Plusrsquo Sections 27 and 28respectively


See M Beit-Arieacute lsquoTransmission of Texts by Scribes and Copyists Unconscious and Crit-ical Interferencesrsquo in P S Alexander and A Samely (Eds) Artefact and Text The Re-Creationof Jewish Literature in Medieval Hebrew Manuscripts Proceedings of a Conference held in theUniversity of Manchester 28ndash30 April 1992 (=BJRL 753 [1993] Manchester John RylandsUniversity Library 1994) pp 33ndash51

See Smelik lsquoOrality Manuscript Reproductionrsquo pp 76ndash80

See W F Smelik lsquoHow to Grow a Tree Computerised Stemmatology and Variant Se-lection in Targum Studiesrsquo in J Cook (Ed) Bible and Computer AIBI-6 (Leiden E J Brill2002) pp 613ndash644 lsquoEmpiricalrsquo is a goal rather than an achieved result in the strict sense ofthe word because of the inevitability of some working assumptions about the texts and thelack of hard data

Houtman lsquoTextual Traditionrsquo p 148 does not mention any criteria beyond the standardexclusion of spelling and scribal error nor does she make mention of the number and typeof readings on which her study is based

These two fundamental aspects of stemmatology tree construction and shock waveswill be explained below Section 30

For a distinction between various levels cf P van Reenen and L Schoslashsler lsquoFrom Variantto Pedigree in the Charroi de Nicircmes A Typology of Variantsrsquo in Van Reenen and Van MulkenStudies in Stemmatology pp 263ndash304

The copy of the Prophets and the Writings he produced on behalf of the University ofSalamanca earned him the sum of 12 ducats as recorded in the archives of the university(see F Marcos Rodriacuteguez lsquoLos manuscritos pretridentinos hispanos de ciencias sagradas enla Biblioteca Universitaria de Salamancarsquo Repertorio de Historia de las Ciencias Eclesisticas enEspaa 2 [1971] pp 261ndash507) De Zamora was baptised in 1506 after he had begun workingfor Cisneros but the codicological aspects of his manuscripts show he was a trained scribe inthe Jewish-Sefardi tradition (so that the inclusion of his manuscript in evaluating the processof manuscript reproduction in the Jewish tradition appears to be valid) Christian patronageis revealed only by certain peculiarities the manuscripts are to be read from left to rightfollowing the direction of the parallel literal Latin translation and occasionally Aramaicwords are hyphenated which is rare in Jewish Hebrew manuscripts Comparison with otherSefardi manuscripts shows that De Zamora was faithful to his exemplar but to facilitatesingling out non-literal additions to the base text translation he sometimes inserted themarker lsquoToseftarsquo in the running textWhether both mss were copied from this exemplar or one was copied from the other isuncertain see M Taradach and J Ferrer Un targum de Qoheacuteleth Ms M-2 de SalamancaEditio princeps Texte arameacuteen traduction et commentaire critique (Le monde de la Bible 37Genegraveve Labor et Fides 1998) who argue that the differences point to a common Vorlagerather than one being copied from the other This unqualified assessment prompts the ques-tion of how many errors and accidental () differences should exist between two manuscriptsto disconnect the umbilical cord between them The intermediacy of manuscripts shouldbe argued along other ways in particular by examining the presence or absence of uniquereadings (type-1 variation see M van Mulken lsquoThe Manuscript Tradition of the Perceval ofChreacutetrien de Troyes A Stemmatological and Dialectological Approachrsquo unpublished PhDdissertation Amsterdam 1993 p 50 Salemans Building Stemmas pp 25ndash27 155ndash156 A

Willem F Smelik

den Hollander De Nederlandse Bijbelvertalingen 1522ndash1545 [Nieuwkoop De Graaf 1997]p 152 n 74) One should note that there is not much difference for our purposes whetherboth manuscripts are apographs of an exemplar now lost or whether one has been copiedfrom the other

The work for this Polyglot was launched by Cardinal Francisco Jimeacutenez de Cisneros thearchbishop of Toledo in 1502 and led to publication in 1518 although it went into circu-lation only after Pope Leo Xrsquos approval in 1520 In 1504 Alfonso de Zamora (c 1474ndash1544)was hired to produce a Latin translation of the Targums of the Latter Prophets and the Writ-ings Due to vehement opposition among others by the new archbishop of Toledo Juan deTavera neither the Targums nor their Latin translations were ever to be included in the Poly-glot with the exception of Targum Onqelos Fortunately however most of the manuscriptswere to be preserved in the biblioteca de San Ildefonse in Alcalaacute for consultation Later theywould be utilized for the Biblia Regia albeit in a purged form

D Amram The Makers of Hebrew Books in Italy (London Holland Press 1973) pp 146ndash224 J S Penkower lsquoThe Chapter Divisions in the 1525 Rabbinic Biblersquo VT 48 (1998) pp350ndash374 idem lsquoVerse Divisions in the Hebrew Biblersquo VT 50 (2000) pp 379ndash393 (383ndash384)

This conclusion was presented at the sixth meeting of the Association InternationaleBiblique et Informatique Stellenbosch July 2000 to be published in a forthcoming studylsquoTargum Judges in the Great Rabbinic Biblesrsquo

Some pluses and minuses are errors other variant readings combine several character-istics

See esp Salemans Building Stemmas

Smelik lsquoHow to Grow a Treersquo

The category of scribal errors can easily be confused with the concept of lsquoerrorrsquo in somestemmatological models in which lsquoerrorrsquo represents a non-original reading

In Targum Judges 21 14 1118 124

in 210 in 215 There are also some minuses shared by other mss

See the first text-genealogical rule as formulated by Salemans Building Stemmas pp64ndash71

For phonetic errors see Smelik lsquoHow to Grow a Treersquo for Hebraisms see for examplethe Hebraism in 828 in N and Rb1 corrected into in Rb2

See Smelik lsquoHow to Grow a Treersquo B Narkiss lsquoThe Relation between the Author ScribeMassorator and Illustrator in Medieval Manuscriptsrsquo in J Gleacutenisson and C Sirat (Eds) Lapaleacuteographie heacutebraiumlque meacutedieacutevale (Colloques internationaux du CNRS 547 Paris Eacuteditionsdu Centre national de la recherche scientifique 1974) pp 79ndash86 M Beit-Arieacute lsquoThe WormsMah zor ndash MS Jerusalem Jewish National and University Library Heb 40 7811 Wuumlrzburg(Germany) 1272rsquo in idem The Making of the Medieval Hebrew Book Studies in Palaeographyand Codicology (Jerusalem Magnes Press 1993) pp 152ndash180 (162)

For an example see below Section 33


A frequent phenomenon concerns the expression of the genitive by a construct chain orby the intermediate use of the particle Changes from the emphatic to the absolute state orvice versa occur frequently as well

See G Dalman Grammatik des juumldisch-palaumlstinischen Aramaumlisch (Darmstadt Wis-senschaftliche Buchgesellschaft 1960) p 191 (sect385)

In v 11 P T E Q Eb66 K in v 16 only P in v 17 P d K Q Rb1 Rb2 (N contains an errorhere)

All the sigla used in this study are explained in Table 8 at the end of this article

The determinative state is not followed by the particle here

The following abbreviations are used in this table s = singular p = plural (a absolutec shortened emphatic e emphatic)

E has the erratic unless this is taken as a defective spelling

These readings may reflect a Hebrew variant reading now lost Perhaps differences ofperson should sometimes be glossed over (a step which should not be obscured for obviousreasons) when variant readings belong to two different categories In 1124 for exampleall witnesses except for two mss read one ms reads another Obviously thevariant readings attest to the same verb over against all other witnesses but they do notreflect the same person

See W F Smelik lsquoTranslation and Commentary in One The Interplay of Pluses and Sub-stitutions in the Targum of the Prophetsrsquo JSJ 29 (1998) pp 245ndash260 idem lsquoConcordanceand Consistency Translation Studies and Targum Jonathanrsquo JJS 49 (1998) pp 286ndash305

The distinction between adjectives and nouns is often difficult to draw Cf SalemansBuilding Stemmas pp 85ndash89 (87 n 71)

In 29 the reading occurs in some Hebrew mss and it is supported by the PeshittaVulgate and in Targum by mss A M O S E D B Wmg (as well as the Antwerp Polyglot) theother mss support

Contrast however E van Staalduine-Sulman The Targum of Samuel (Leiden E J Brill2002) pp 158ndash159 See also C A Dray ldquoIs Subtlety in Translation the Reason for theTargumic Use of various Verbs of Fleeingrdquo AS 2 (2004) pp 25ndash35

M O Wise lsquoAccidents and Accidence A Scribal View of Linguistic Dating of the Ara-maic Scrolls from Qumranrsquo in T Muraoka (Ed) Studies in Qumran Aramaic (AbrNS3 Leuven Peeters 1992) pp 124ndash167 For a stemmatological perspective see SalemansBuilding Stemmas pp 70 236

The change of to or vice versa is not related to dialectical processes nonethelessthis type of variation appears to be rather unreliable

In this connection it should be recalled that according to Uthemann intentional changesshould not be taken into account when studying the genealogy of a manuscript traditionUthemann lsquoWhich Variants are Usefulrsquo p 257 Intentions may have been operative inde-pendent from the exemplar and therefore may have influenced several scribes independentlyof each other However it is possible that the terminology obscures more than it revealsWhatever label one adopts (random vs purposive intentional vs non-intentional or any

Willem F Smelik

other) the aim of stemmatology always is to establish the relationship between textual wit-nesses without the confusion of analogue but independent developments and hybridiza-tion That is exactly what Uthemann strives to achieve so that the confusion hinted at heremay derive from different labels rather than fundamentally opposed approaches

W F Smelik Targum of Judges (OTS 36 Leiden E J Brill 1995) pp 594ndash596

For a definition of the various types of lsquocontaminationrsquo see E Wattel and M J P vanMulken lsquoShock Waves in Text Traditionsrsquo in P van Reenen and M J P van Mulken (Eds)Studies in Stemmatology (Amsterdam John Benjamins 1996) pp 105ndash121 (105ndash106)

Such errors may when shared by manuscripts or editions point to a common ances-tor that introduced the error but they are generally speaking not a reliable indication ofgenealogical relationship

L Diacuteez Merino lsquoFidelity and Editorial Work in the Complutensian Targum Traditionrsquoin J A Emerton (Ed) Congress Volume Leuven 1989 (VTSup 43 Leiden E J Brill 1991)pp 360ndash382

The Salamanca ms contains more uncorrected minuses that are apparently uninten-tional than the Madrid ms In W eight uncorrected minuses occur in 11 17 25 37 16 25632 823 in M only three in 321 634 181

The following examples reflect the combination of two alternative translations in 635the reading in mss A J in 810 the reading in ms B in 818 the reading

in ms A in 129 the reading attested by mss N Rb1 Rb2 in 1122 the readingin the Leiria edition (d) See also Smelik lsquoOrality Manuscript Reproductionrsquo p 77

There are of course double translations which seem to be original

For example in 219 the Hebrew does not qualify the Israelite lsquopracticesrsquo as lsquoevilrsquo as doesthe Targum with the plus the absence of this plus in some mss (T Q) and Rb1 maysimply reflect an adjustment toward mt The interpretation of geographic indications forexample has sometimes been supplemented with a transliteration see for instance inRb2 and edition d in 722 This also applies to minuses the reading for in724 is closer to mt and presumably of a secondary nature See also n 53 above

This reading is found in mss p Eb3 F T P W M O S o C a E D B K Q N Rb1 Rb2 Rb3Rb6

In mss x y w q63 this is one of the occasions where Sperberrsquos base manuscript may bemore reliable than the other Yemenite manuscripts he used This reading may first have beenan alternative one noted as in the margin

In mss Eb66 A d J Theologically motivated changes are often more obvious A markedexample is to be found in 1123-24 in a ms in Jena where a neutral statement abouta non-Israelite deity was deemed to be improper and has been recast into the standarddenunciation of foreign deities

See for example the plus in 142 reflecting Hebrew

See Smelik Targum of Judges p 643 n 15

An analysis of contamination did not show a remarkable indication for pollution at thispoint The shock waves at this juncture were not marked by peaks


The standard translation is although some witnesses do occasionally readthroughout Targum Jonathan to the Prophets Even Rb2 usually has the first

reading it deviates in Josh 512 Judg 638 and 214 contrast 1 Sam 53 4 1111 18102027 3017 318 2 Kgs 815 Jer 203 Jon 47

Although G has the same preposition as Rb2 it is prefixed to a different noun Fora discussion of the translation see Smelik Targum of Judges pp 544ndash546

That is p x y w q63 Eb1 Eb66 A F T P W M O S d C a E J K Q Rb2

That is p x y w q63 Eb1 Eb66 A F T P W M O S d C a E B J K Q Rb2 The Hebrew has and the translation in most mss was corrupted into the similar reading in N

Rb2 agrees with p w q63 Eb1 A T W M O S d C a B J K Q Om

That is p x y w q63 Eb1 Eb66 A F T P W M O S o d C a E B J K Q Rb2

That is p x y w q63 Eb91 s4 A F T P q2 W M O S o d C E B J K Q G Rb2 The particle iserroneous here

That is p x y w q63 Eb1 Eb66 A F T P W M O S d C a E B J K Q Rb2 The preposition isa dittography

That is (with spelling variations) p x y w q63 Eb1 Eb66 A F T P W M O S d C a E B J KQ N Rb2

The following variations occur 816 N Rb1 Rb2 1010 N Rb1 Rb2 119 N Rb1 Rb2 122 N Rb1 Rb2 124 N Rb1 Rb2 132

N Rb1 Rb2 1314 N Rb1 Rb2

Salemans Building Stemmas pp 70 252ndash256 characteristic 4b

See Smelik lsquoTargum of Judgesrsquo Ch 3

Salemans Building Stemmas characteristics 8 and 10

Different positions apply to the insertion of the marker a paratextual elementmarking a part of the translation that has been added later on or an alternative translationadded by the scribe editor or glossator These Toseftot themselves yield interesting informa-tion as their position in the running text or margin differs as does their contents On theallocation of glosses see Smelik Targum of Judges pp 162ndash179

E Wattel and M J P van Mulken lsquoWeighted Formal Support of a Pedigreersquo in VanReenen and Van Mulken (Eds) Studies in Stemmatology pp 135ndash167 idem lsquoShock Wavesin Text Traditionsrsquo pp 105ndash121 E Wattel lsquoClustering in Stemmatological Trees How toHandle a Large Number of Versionsrsquo in Van Reenen and Van Mulken (Eds) Studies inStemmatology pp 123ndash134 Den Hollander lsquoHow Shock Waves Revealed Successive Con-taminationrsquo pp 1ndash2

A total of 12 graphs per category has been produced by Evert Wattel and all of thesehave been taken into account Wherever percentages are provided the witnesses involvedare lacunary or even fragmentary and the percentage indicates the number of formulas inwhich the witness was involved

Some suffixed pronouns may have been spelled defectively such as for

In particular such groupings as at the bottom right w y x q63 or Rb1 Rb2 Rb3 Rb6are convincing

Willem F Smelik

Due to a technical error the readings of lsquoorsquo have been largely omitted from this category(only 6 included) nonetheless this was enough for the quadruple method which groupedit correctly together with W and M

The plural appears to be more original as it reflects the collective meaning of the He-brew this is based on the considerations that this reading (a) has a greater distance to mtwhile (b) it better reflects the translation strategies of the targum Another peak appears toreflect a correct split of the Sefardi mss from the remaining witnesses That it results in apeak is not incongruous with the theory

It should be noted that in the fifth category this Yemenite subgroup was still separatedfrom the main Eastern branches by some Western texts although far less pointed

The variant is found in W M O B K the text reads

These witnesses are p Eb1 Eb66 T C J K Q

Following of course the same method of quadruple computation Both methods havedifficulties in positing K according to the quadruple method K is close to the Nuumlrnberg ndashBomberg group

Contrast the shock waves in A den Hollander lsquoHow Shock Waves Revealed SuccessiveContamination A Cardiogram of Early Sixteenth-Century Printed Dutch Biblesrsquo forthcom-ing with Wattel and Van Mulken lsquoShock Wavesrsquo p 119 For brevity of argument the shockwaves for categories three and four have been omitted

For more details see n 3 above

See now Den Hollander lsquoHow Shock Waves Revealed Successive Contaminationrsquopp 1ndash2

See esp Salemans Building Stemmas pp 67ndash71

Den Hollander Nederlandse Bijbelvertalingen 1522ndash1545 p 136

Of course taking similar changes into account lsquoindividualrsquo here does not mean lsquoatom-isticrsquo

Onqelos may carry its own characteristics in view of its even more careful editing thePalestinian versions the higher number of texts and the inclusion of the whole text in theliturgy

Cf M P Weitzman The Syriac Version of the Old Testament An Introduction (Cam-bridge Cambridge University Press 1999) pp 316ndash317

To some extent the focus on the original reading in critical editions should be calledinto question Why should later readings be less interesting

Houtman lsquoTextual Traditionrsquo

ms 11 in Goumlttweig Austria (A) ms Elf6 in Jena Germany (J) ms Or Fol 1ndash4 inBerlin (B) ms 26879 in London (a) For more details on A see Smelik lsquoOrality ManuscriptReproductionrsquo

Weitzman The Syriac Version pp 316ndash322

Salemans Building Stemmas passim


See n 12 and Section 210 above

Thus excluding morphological inflectional and orthographical variants of verb andnouns (including adjectives in targumic literature) and all other variant readings

Table 8 Sigla of used manuscripts and editions

siglum description provenance type

Manuscripts (geographical order)

a Add 26879 British Library London Ash completeB Or Fol 1ndash4 Staatsbibliothek Berlin Ash completeD parm 3188 Biblioteca Palatina Parma Ash completeJ Elf6 Universitaumltsbibliothek Jena Ash completeK Reuchlin 3 Karlsruhe Ash () completeN Cod Solger 320 Stadtbib Nuumlrnberg Ash completeQ heacutebreu 18 Bibliothegraveque Nationale Paris Ash completeC BH III Biblioteca Civica Berio Genoa Ash completeA 11 Stift Goumlttweig Ash completeG Heb A 10 Kaufman Coll Budapest Ash haftarahF UrbinatesndashVaticani 1 Vatican Ash completeP Laud Or 326 Bodleian Oxford Ash completeT Or 72 Biblioteca Angelica Roma Ash complete

E heacutebreu 75 Bibliothegraveque Nationale Paris Or complete

W 1 Biblioteca Antigua Salamanca Sef completeM 7542 Biblioteca Nacional Madrid Sef completeO Opp Add 40 75 Bodleian Oxford Sef completeS Kennicott 5 (2329) Bodleian Oxford Sef complete

p Or 2210 British Library London Yem completey Or 2371 British Library London Yem completew Or 1471 British Library London Yem completex Or qu 578 Staatsbibliothek Berlin Yem completeq632 632 R Y Kapah Jerusalem Yem completeq2 2 R Y Kapah Jerusalem Yem incompletes4 Sassoon 1154 Yem haftarah

Eb1 229 JTS library New York Bab incompleteEb4 505 JTS library New York Bab incompleteEb66 Cambridge UK amp JTS New York Bab nearly completeEb91 H olon Y L Nah um Bab incomplete

Willem F Smelik

Table 8 (continued)


Editions (chronological order)

d Former Prophets Leiria 1494 Sef completeRb1 1st Rabb Bible Bomberg 1516ndash1517 Ash completeRb2 2nd Rabb Bible Bomberg 1524ndash1525 Ash completeRb3 3rd Rabb Bible Bomberg 1548 Ash completeRb4 4th Rabb Bible Bomberg 1568 Ash completeo Antwerp Polyglot 1569ndash1573 lsquoSefrsquo completeRb5 5th Rabb Bible Bragadin 1617ndash1619 Ash completeRb6 6th Rabb Bible Koumlnig 1618ndash1619 Ash complete

Ash = Ashkenazi Bab = Babylonian Or = Oriental Sef = Sefardi Yem = Yemenite

Scribal variationsWhen are they genealogically relevant ndashand when are they to be consideredas instances of lsquomouvancersquo1

Lene SchoslashslerUniversity of Copenhagen

Introduction

The intention of this paper is to look for linguistic criteria for a distinctionbetween genealogically relevant and genealogically irrelevant scribal variationsI have tried to do so before when working on the mss of Narcisse and of theCharroi de Nicircmes (Schoslashsler 1988 1989 Van Reenen amp Schoslashsler 1996) I thenproposed to distinguish three later four levels of linguistic variation (see VanReenen amp Schoslashsler 1996 Section 3 Local variants)

1 differences in spelling and phonology2 differences in morphology and syntax3 content differences in related passages and4 content differences resulting in unrelated passages

The genealogical relevance of each level differs level 1 may present dialectallyhence possibly genealogically relevant information about rhyme and asso-nance (1996279) Like level 1 level 2 may contribute to the identification ofthe dialect of both the exemplar and a copy and may thus be genealogically rel-evant (1996279) Level 3 may offer dialectally determined lexical variation iepossibly genealogically relevant information More importantly it may permitthe establishment of the so-called ldquotype-2 oppositionsrdquo which are genealogi-cally relevant lexical variations opposing at least two mss against two othermss in exactly two groups2 eg AB versus CD (1996280) A distinction ismade between passages showing at least some resemblance (level 3) and pas-

Lene Schoslashsler

sages which are completely independent (level 4) At level 3 and 4 we may findimportant genealogically relevant ldquotype-2 oppositionsrdquo (1996281)

I have realised ndash as have most of my colleagues working in the field of stem-matology ndash that it is extremely difficult to differentiate between on the onehand variations revealing the absence of a well-established linguistic norm orliterary authorized version and on the other hand variations revealing differ-ent manuscript traditions Until recently I did not see any way to cast morelight on this problem A possible way to a better understanding of medievalvariation could be to study two copies of one exemplar made by the samescribe but such cases are rare However they do exist and Keith Busby hasrecently discovered an interesting one (1993a 1993b) mss T and V of Chreacute-tien de Troyersquos Perceval I find his argumentation absolutely convincing andI accept his results concerning the identity of the hand of the mss T and Vof Perceval According to Busby (1993a54) it is not possible to show that Tis a copy of V or vice versa it is more likely that they are copies of the sameexemplar3 In the following I will examine these two copies as illustrations ofwhat I will assume to be genealogically irrelevant variations In Section 1 I shallstudy the variations between the two mss at each level as mentioned above InSection 2 I shall compare the types of variation with those found for compa-rable mss among the nine mss of the Charroi de Nicircmes None of these msswere copied by the same hand In Section 3 I will conclude the investigationby considering whether the study of the Perceval copies can lead us to a fur-ther understanding of which types of variation are genealogically relevant andwhich are not

The two Perceval mss are from the thirteenth century and their geograph-ical provenance is Northern possibly from Oise For more details on the twomss see Busby (1993a amp 1993b) and Van Mulken (1993)

Variations between mss T and V

The two mss have been studied on microfilm the quality of which is some-times rather bad Therefore it has not been possible for me to make a completeinventory of the variations between T and V I have drawn upon Busby (1993a)who provides a most accurate listing of variations Parallel passages in the twomss are indicated in Busby (1993a51) they amount to 3098 verses Busbycounts over 350 points of difference with the exclusion of purely orthographi-cal ones on average one every 875 lines (Busby 1993a55)

Scribal variations

Variations at the level of spelling and phonology (level 1)

As established in Van Mulken (1993) mss T and V contain spellings which re-veal that the copies adapt the original champenois dialect to a northern dialect(that of Oise) especially in the rhyme position less so inside the verse Thelevel of dialectal adaptation is almost the same in the two mss ndash as should beexpected for the same scribe T has a dialectal coefficient of 68 V has a dialectalcoefficient of 6456 the maximum being 1004 Even so spelling predilectionsdiffer from one ms to the other as already observed by Van Mulken T prefersthe spellings vos nos totes V prefers vous nous toutes (Van Mulken 1993219)Additional examples are

T tor (5901) Escalibor (5902)V tour (5901) Escalibour (5902)

Most frequently the spelling is identical doutetoute (5931-2)In the following section I will examine some different spelling variations

The dialectally relevant distinction or non-distinction between an et en isdiscussed in Van Mulken (1993) We find hesitation in both mss

T anqui (6049) espanstans (6261-2) desfendreprendre (6745-6)descentbauchent (7261-2)

V encui (6049) espenstens (6261-2) desfendreprandre (6745-6)descentbauchant (7261-2)

The northern spelling ch corresponding to the central spelling c seemsmore frequent in T than in V the mss showing both the central and thenorthern spelling

T che que (5933 5943 ) por che (5976) chiteacute (6145) comenche (6216)atache (6340) charme (6340)

V ce que (5933 5943 ) por ce (5976) citeacute (6145) comence (6216) attace(6340) carme (6340)

The mss hesitate between the spellings s and z I get the impression that V hasa predilection for s

T mandezdemandez (5957-8) fazsolas (5971-2) avez (5978)V mandesdemandez (5957-8) fassolas (5971-2) aves (5978)

A frequent spelling variation without any phonological basis is that betweenu and l I have the impression that T prefers the vowel whereas V prefers theconsonant

Lene Schoslashsler

T au(s) (6260 7087 ) teus (5998) chaveus (6988)V al(s) (6260 7087 ) tels (5998) chavels (6988)

Marking of elision is unstable as remarked by Busby (1993a56) which impliesvariations of the type qursquo ndash que eg 6607 T jusqursquoanuit ndash V jusque anuit

Spelling variations between T and V are very frequent We know that thereis no important temporal difference between the two copies5 The lsquoinputrsquo di-alect being common to both copies and the lsquooutputrsquo dialects being common aswell the numerous spelling variations reflect the absence of a norm ndash even ofa strict personal norm What is permanent about our copyist is that he keepsthe same attitude towards his act of copying (see Schoslashsler 1995246) he is notmirror-copying in one ms and freely adapting in the other In both mss herespects the original rhyme and adapts the spelling inside the verse to his owndialectal habits Compare Van Mulken (1993168)

The behaviour of manuscripts t and v once again illustrates the kind of scribalconservatism with regard to rhyme known as diglossia which we encoun-tered in manuscript u for OR words in rhyme position manuscripts t and voccasionally use the lsquooursquo spelling though there are no linguistic circumstanceswhich would have obliged the scribes to modify their usual spelling These par-tial mirror-copyists too seem to have preserved the spellings of their modelsin rhyme position whereas they felt free to alter the spelling within the rest ofthe textrdquo

Variations at the level of morphology and syntax (level 2)

I will first present paradigmatic variation and afterwards discuss syntacticvariation The best known dialectally relevant morphological variations arevariations in the declension system differing forms of gender marking andvariation in the pronominal forms Conjugation variations are also found Ourtwo mss show variations in all these cases

DeclensionBoth Chreacutetienrsquos champenois dialect and the northern dialect of the 13th cen-tury are conservative with respect to declension Accordingly we find a regularsystem with additional -s-marking in the singular nominative forms of somefeminine and masculine nouns although this is more frequent and more con-sistent in feminine than in masculine nouns A few examples

Scribal variations

MASC maistre(s) (6072) traitre(s) (7559) but not although it would bepossible in the following masculine nouns sire hom prestre frere(see eg 6302 6305 6321 6415 6454 6804 9113 )

FEM raisons (7182 8020 8878) mer(s) (7590) cor(s) (8897)

GenderIn the northern dialects the feminine singular definite article may take theform le instead of the central form la Throughout the two mss the two formsalternate with la as the dominant form

T la espee (5903) la traiumlson (6095) le cort (6107) le color (6138) le pointe(6376) la crois (6496) la voie (6620) la palme (7020)

V le espee (5903) le traiumlson (6095) la cort (6107) le color (6138) le pointe(6376) le crois (6496) le voie (6620) le palme (7020)

Compare also Busby (1993a58)Another dialectally dependent gender-marking is that of feminine -e added

to adjectives of the GRANT-type The only case found here is additional -e inverse 8330 tele

Pronominal paradigmsThere are a few differing forms but they are too infrequent to permit any con-clusions Alternations are found among personal pronouns T jou (6492) on(6152) V je (6492) lrsquoen (6152) possessive pronouns T soe ndash V soie (6416)relative pronouns T qui ndash V cui (6415 6951) T celui a cui ndash V celui cui (8938)

ConjugationBesides a few straightforward copying errors such as T mirois (2nd personsingular) instead of V miroit (3rd person singular) (6678) we find only fewcases of differing forms like T averiez ndash V avriez T feiumlsse ndash V fesise (8358) Testoit ndash V ert (9146) T aiue ndash V aide (6466) T lairai ndash V laisserai (8418)

If morphological variation is rather rare syntactic variation ndash especiallyvariation in word order ndash is less infrequent as will be shown below

DeclensionAs mentioned above the declension system of both lsquoinputrsquo and lsquooutputrsquo di-alects is conservative We find very few lsquoerrorsrsquo in the use of the nominativeforms with one frequent exception terms of address are generally found inthe accusative instead of the expected nominative forms (for declension ofterms of address see Schoslashsler 1984) Examples are vassal (T amp V 6880 7014

Lene Schoslashsler

7030 ) chevalier (T amp V 7283 8421 in the rhyme ) Gauvain (T amp V 6140V 7094 8902) There are a few unexpected accusative forms instead of nom-inative forms in the subject function fief (V 7393) chevalier (T 9012) andan accusative form of the subject complement (T amp V 7013 in rhyme posi-tion V 7394) especially following the verb sembler 7188 7324 (in the rhyme)There is one strange nominative form instead of the expected accusative formin 7337-8 que il lrsquoen ront li senestres tot en ront (corrected into the accusativeform by Roach and by Busby in their editions le senestre)

TenseA well known feature of older literature is the alternation of narrative tensesespecially between the historic present and the preterite (see Schoslashsler 19731994) Busby (1993a58) mentions several cases eg sai ndash sot (6035) vint ndashvient (6036 ) fu ndash est (6260 ) fist ndash fait (9229) There are a few other unpre-dictable cases of variation eg T covenoit ndash V convient (6026) (7984) poez ndashporrez (9138) a ndash ara and also one case of change of person ie second sing ndashfirst plural (8213) mejerois ndash mengerons

We have seen syntactic phenomena here which show a certain degree ofvariation On the other hand it is also highly interesting to find syntactic phe-nomena that are stable Historians of Old French have often looked in vainfor factors and parameters determining certain variations Curiously enoughwhat is often considered as inexplicable variation or as instability due to on-going linguistic changes (like the use of the determiners the position of theadjectives the use of the pronominal subject the choice of person in addressetc) seems to be stable for our copyist I will first consider the structure of theNoun Phrase and afterwards the Verb Phrase

Noun PhraseThe use or omission of determiners is largely stable (except for the pair onlrsquoen)with some alternation between the possessive and the definite article eg(6197) T tote sa paine ndash V toute la paine

The position of the adjective does not show variation both mss have eitheranteposition as in 8083 sa lie chiere or postposition as in 6530 un palefroinoiret petit

There is no blurring of the distinction of the two demonstrative paradigmscil versus cist

Scribal variations

Verb PhraseThe expression or omission of a pronominal subject is largely stable in the twomss with some fluctuation concerning the type of pronoun personal relativeor demonstrative (see Busby 1993a57) qursquoil ndash qui (7449 ) il ndash cil and cele ndashele (168 8446) etc

There is hardly any variation in the form of address between the 2nd per-son singular and plural which have however been observed to fluctuate in anintricate way within narrative texts see Foulet (1967198ff) The only changeI have found is clearly provoked by a change of the rhyme (7419-20) T en vosfiergarder ndash V en toifoi6

The use of the subjunctive is stable I have found only four cases of differ-ence between T and V 6041 T ert (imperfect indicative) ndash V fust (imperfectsubjunctive) and 6921 T soit (present subjunctive) ndash V est (present indica-tive) (8423) T avez (present indicative) ndash V euumlssiez (imperfect subjunctive)(8457) T deuumlst estre (imperfect subjunctive) ndash V devoit estre (imperfect in-dicative) The first two types of context are known to show fluctuation ofmood as the governing verb is one expressing uncertainty eg 6040-1 De cheque mesire Gavains ertfust el chastel ne savoit mot 6920-1 Or quit je que cischevaliers soitest mors see Foulet (1967208) The last two examples showthe well known fluctuation of mood in connection with hypothesis

Concord of the past participle with the direct object of a compound tenseis a difficult matter in Old French it is hard to understand the fluctuation inconcord versus non concord However our scribe appears to know some sortof a system because he is largely consistent about it see eg 5957 6242

I have found hardly any variations in the valency patterns of the verbs Inone case (6198) there is a difference in the preposition introducing a preposi-tional object mettre sa peine T en V a querre la lance In another case (7463)I have found a difference between the prepositions de and a introducing an in-finitive clause as subject Que ne seroit pas vostre biens T de V a demorer encest rivage Fluctuation in the use of prepositions introducing infinitive clausesis in fact not infrequent see Van Reenen and Schoslashsler (1993)

In Medieval French word order is largely free One might therefore expectvariation in word order such as (S)OV ndash (S)VO ndash OV(S) In our two mssvariation in word order is nevertheless limited and mainly concerns adverbialphrases and pronouns (see however v 6560 below) This could be due to thelimitations of the octosyllabic rhyming verse At subsequent levels of variationie levels 3 and 4 though these limitations do not seem to play any role soinfluence from the metre should not be overestimated A few typical cases showthe types of variation found7

Lene Schoslashsler

(5970) T Et tenez vos mrsquoent a vilaine ndash V Et tenez mrsquoent vos a vilaineV S C V C S

(6146) T Se destorner vos en pleuumlst ndash V Se vos destorner en pleuumlstInf C C Inf

(6298) T Doivent estre hui en peneance ndash V Doivent hui estre en penitanceInf Adv Adv Inf

(6560) T Quant mesire G vint la ndash V Mesire G quant vint laConj S V S Conj V

(7404) T Que je le cheval nrsquoen euumlsse ndash V Le cheval que je ne lrsquoeuumlsseConj S O O Conj S

(8028) T Que sachiez bien je ne porroie ndash V Que bien sachiez je ne porroieV Adv Adv V

Compared to the frequency of spelling variation morphological and syntacticvariation is limited This presents a contrast with with the state of affairs at level3 where variations are much more frequent

Variations at the level of content in related passages (level 3)

Most of the differences listed and commented on by Busby (1994) concernvariations at the level of content ie lexical variation in related passages Ourmss offer a precious source for identification of synonyms or near synonyms ofOld French We find synonyms for words belonging to all word classes nounsadjectives pronouns articles conjunctions prepositions verbs and adverbsFirst I shall quote some illustrative examples of what I consider to be syn-onyms Afterwards I shall proceed to near synonyms and words or expressionshaving related meanings Finally I shall consider ellipses and variations con-cerning more than one word I first give the form found in T then the onefound in V

SynonymsNouns

(6298) peneance ndash penitance (7144) ambleuumlre ndash aleuumlre (7372) nacele ndash bargele(8868) onor ndash grant los (9136 9145) chaceor ndash coreor

Adjectives

(8464) trestot bien ndash tout le bien

Pronouns

See Section 12 above (6086) il ndash cil (7346) il ndash on

Scribal variations

Articles

See Section 12 above (6197) tote sa paine ndash toute la paine

Conjunctions

(5988) et ndash mais (6521) ou ndash quant (6609) que ndash car (7095) que que ndash coique (8401) mais que ndash fors que (8910) ains que ndash anccedilois que

Prepositions

(6036) venir el ndash al chastel (6081) venir a ndash en la tor (7069) desor ndash desus(7266) en le ndash ens le

Verbs

(5999) voist avant un pas ndash face avant un pas (6627) alever ndash eslever (7166)nrsquoatoche ndash ne toche (9160) lasser ndash pener

Adverbs

(5946) onques ndash ainc (5964) pas ndash mie (6204) hors ndash fors (6493 6767) issi ndashensi (6788) neporoec ndash neporquant (7537) molt ndash tant (8077) tant ndash molt(8215) amont ndash ccedilasus (8462) buen ndash buer (8480) si hautes ndash molt hautes

My list is not exhaustive but I believe the relative frequency of synonyms tobe representative Particles appear to be more easily interchangeable synonymsthan other words with the exceptions discussed in Section 12 Nouns adjec-tives and verbs have a more specific lexical meaning which apparently makesit difficult to have full synonyms while near synonyms or related expressionsare more frequently found see Sections 2 and 3 below

Near synonymsIt is of course impossible to draw a clear line between lsquosynonymsrsquo and lsquonearsynonymsrsquo however the following examples illustrate the latter type (6306) lagloire de Dieu ndash la gloire del Ciel (6467) veve dame ndash veve amie (5948) malvais ndashcoart (6305) sains hom ndash bons hom (8270) preus ndash grans merveillous ndash coragousforseneacute ndash molt peneacute

Words from the same semantic field or with related basic meaningsStill more differing are cases where words have distinct meanings but stillbelong to the same or a related semantic field or if more than one word isconcerned the basic meanings are at least related

Lene Schoslashsler

Variation between different kinds of trees (6676) carme ndash orme betweendifferent but related activities (6697) pensez ndash volez (7160) pot ndash sot (8944)rasamblai ndash redonai (8490) paser ndash aler (8852) vendrons ndash serons (8898) laveriteacute en savez toute ndash la veriteacute trsquoai dite toute (9102) establie ndash aramie

Even more differing are (8274) aparole ndash acole

Ellipses and variation concerning more than one word (a hemistich ora whole verse)

In many cases the copyist simply replaces a word by its synonym or by a re-lated word he may skip a word or insert another or he may even modify thewhole verse normally without seriously changing its meaning The followingexamples illustrate major modifications

(8310) T amp V jel vos dirai T sanz detriier ndash V bien volentiers(8454) T Lasses por coi somes ndash V Or mais por coi somes(8473) T celi et dist Bele or me dites ndash V pucele fait il or me dites(7406) T au chevalier faillir V tollir ne(l) doi(8522) T si srsquoest touz cois en pais V en piez tenus(8856) T et je ravrai la moie toute V ma gent trestote

Most frequently we find smaller modifications which are probably simpleerrors

(6631) T a nul sens ndash a nul tens(8344) T dont lrsquoen laisserai je issir ndash V et je lrsquoen laisserai issir(8416) T Gavains i vient si le salue ndash V Il vient vers li si le salue(8418) T car ci ne vos lairai je mie ndash V ci ne vous laisserai je mie(8423) T Certes bataille avez assez ndash V vous eussiez bataille assez(6176) T ou morir ou languir set ans ndash V morir ou languir bien set ans(8344) T dont lrsquoen laisserai je issir ndash V et je lrsquoen laisserai issir(8348) T ne vos anuit ndash V qui qursquoil anuit

Sometimes the scribe has simply forgotten one or several words as is apparentfrom the metre

(5967) T en fui ndash V en is lacking(5997) tant ne redout ndash T ne is lacking(6000) T damoise ndash V damoisele(6189) se vos la lance ndash V has skipped one of the two la-syllables(7599) cil rendroit as dames lor terres ndash T as dames is lacking

Scribal variations

Or the verse contains too many words as in V (7249) T furent vestues lespluisors ndash V furent vestues richement les pluisors

In one case only (7419-20) do we find a change of word order necessitatinga change of rhyme (see also Busby 1993a58)

ms T ms VMe porrai je en vos fier ndashO V S C Inf

Porrai me je fier en toiV O S Inf C

De mon cheval en foi garderC C Inf

De mon cheval garder a foiC Inf C

Variations of the type examined in this paragraph often provide the basisfor oppositions which are accepted as genealogically relevant If supported byother mss they may provide type-2 oppositions that can be used for establish-ing a stemma I was very surprised to see the liberty of one and the same scribecopying the same exemplar and I fear that many of the variations that we thinkgenealogically relevant could instead be independent free innovations madespontaneously by the scribes ndash ie cases of ldquomouvancerdquo (see Note 1)

Variations at the level of content in unrelated passages (level 4)

Given the likelihood that we are dealing with two copies of the same exemplarmade by the same person we should not expect to find any variations that areunrelated in content In fact we find numerous confirmations of relatednessthat have not yet been mentioned here eg common readings of a differentsort opposing T and V to other mss such as common erroneous successionof verses (61848685 62303334 6496996500 661615 66465152 666467) We also find common but not significant errors of content ndash due to thesimilarity of characters as in 8406 li coroit soz lrsquoauberc le sanc corrected byRoach and Busby to li coroit sursor le hauberc blanc and a common numberindication viic in stead of vc in 7566 corrected by Roach and Busby

However we find at least three cases of order of verses where the mss donot agree V presents the order 685152 ndash T has 685251 later T has the order72001234566a6b7 ndash whereas V has 720012346a6b7 ndash without verse7205-6 T has theorder 90923465 ndash V 90923456 Moreover we find anadditional verse in V only following 6612 and a verse in V lacking in T (8030)

Strangely enough we find some cases of genuine differences of content Avery curious case is found in 8424 where the two mss have almost oppositemeanings but one of these must be an error T se mes amis ne fust lassez ndash V

Lene Schoslashsler

se mes anemis ne fust grevez Other cases of opposite meanings are found in thefollowing examples

(8476) T ainz que je nrsquoaie V perde vostre grace(5975) T nrsquoautre folie nrsquoi pensai ndash V nrsquoonques folie nrsquoi pensai(7354) T desor la penne ndash V desoz la penne

There are cases where the basic meaning is not really opposite but only veryvaguely related

(8266) T li notoniers dont vos ai dit ndash V Li notoniers devant son lit

and the following strange example

(8486-7) T Si ne sai ou il plus bas soit Ha bele on ne porroit ce dolt ndash V Je nesai pas ou plus bas soit lrsquoiaue est trop parfonde ce dolt

In these cases we no longer have identity of meaning whatever may be the rea-son There is no reason to think that a change of relationship in the mss tookplace (see Van Mulken 1993219) and my impression that variations increasein frequency after folio 12 (verse 8000) is probably due to a sudden new inspi-ration of our scribe as he reached that part of the text The conclusion to bedrawn from the investigation of this level is that what could be considered asgenuine unrelated variations are indeed very rare

Variations between the mss A2 A3 and A4 of the Charroi de Nicircmes

In this section I shall briefly compare the types of variation found in mss T andV of the Perceval to those found for comparable mss of the Charroi de Nicircmeswhich differs in genre it is an assonanced decasyllabic epic text much shorterthan Perceval (some 1500 verses) This text has survived in nine mss of whichnone were copied by the same hand and none were copied from any of theother extant manuscripts Three mss were copied at approximately the samemoment (1300) from the same exemplar ie from the same lsquoinputrsquo dialectwhich is from the south-eastern part of northern French into the same lsquooutputrsquodialect that of Niegravevre-Allier (Ms A1 has the original lsquoinputrsquo dialect Haute-Marne slightly north-east of Niegravevre-Allier) The mss have almost the samescore of dialectally marked linguistic features 73 76 79 out of a maximum of100 The mss are largely similar and might have been copied in the same atelieraccording to Tyssens (1967) to whom I refer for further information uponthese mss In short a comparison between the variations found in T and V and

Scribal variations

of those found in A2 A3 and A4 may contribute to a better understanding ofindividual liberty in copying in the case of one scribe working several times insuccession on the same exemplar as compared to the liberty of several scribesworking on the same text

Variation at the level of spelling and phonology (level 1)

As established in my 1995 study the adaptation of the exemplar found in Char-roi is very similar to what has been found for Perceval Throughout the msswe find variations like the ones quoted below

Variation between en and an (1198) A2 Angleterre ndash A34 Engleterre(1112) A24 emdementiers ndash A3 andemantiers (507528545601 ) A23 hen-nor ndash A4 anor

Between ein and ain (1112) A2 einsi ndash A34 ainsiBetween ai and eacute (687) A23 aidier ndash A4 edier (60) A2 ferai ndash A4 fereacute

(340) A23 dirai ndash A4 direacute etc (generally A4 has a predilection for eacute) Formore details of vowel variations see Schoslashsler (1995)

In the consonant system there is much hesitation between the spellings cand qu (542) A24 coronne ndash A3 quorone (948) A4 car ndash A23 quar (1176)A3 cuens ndash A24 quens (1386) A23 corent ndash A4 queurent

There is also much variation as to simple or double marking of identicalconsonants oral as well as nasal there is alternation between the spellings land u etc

In short we get a comparable variation pattern between these independentmss to that which we found in the two Perceval copies

Variation at the level of morphology and syntax (level 2)

In the three mss of the Charroi we find variations similar to those found inthe Perceval mss The declension system of the three mss is less conservativethan the system we found in T and V In particular the scribe of A4 has greatdifficulties in trying to master the two-case system We may even find internallsquocontradictionsrsquo of case inside Noun Phrases like in 1086 A23 li glorieus lifi(l)z Sainte Marie (correct nominative forms) ndash A4 le glorieus le fiuz SainteMarie (mixture of nominative and accusative forms) We find the same alter-nation between different articles (1439) A2 li cors (correct nominative form) ndashA3 lor cors (possessive indeclinable article) ndash A4 le cors (erroneous accusativeform) The position of the adjective is identical in the three mss so is the con-

Lene Schoslashsler

cord or absence of concord in the past participle of compound tenses (a fewconflicting examples are found eg in 723)

In contrast with the Perceval mss there seems to be a slight blurring ofthe demonstrative paradigms as is apparent from 941 com cil est que en cel(A2) cest (A3) ce (A4) char veez A4 has a tendency to prefer the neutraliseddemonstrative form ce

As for tense-variation the mss of the Charroi do often vary but theynormally have the same choice of tenses except for a few cases

(188) A4 membre (present) ndash A2 membra (preterite) (237) A23 ai ocis (per-fect) ndash A4 ocis (preterite) (707) A2 ccedilrsquoa fet (perfect) A34 ce fet (present)Aymes le Viell (831) A24 done (present) ndash A3 dona (preterite) (851) A4savez (present) ndash A23 saroiz (futur)

There are a few cases of differences in transitivity eg between direct andindirect constructions such as dire ndash dire de (908) tirer ndash tirer a (1332)

Minor variations in word order and small words skipped in one or twomss are like those found in T and V eg (739) Et si te paines A24 de moimolt empirier ndash A3 molt de moi empirier

At this level once again we get a comparable variation pattern betweenthese independent mss and the two Perceval copies


There are many cases of synonyms near synonyms or related constructions Ishall quote a few illustrative cases

Nouns

(488) A24 chatel ndash A3 regneacute (753) A3 Rois Loys ndash A24 Loys Sire (880)A23 regne ndash A4 pais (1057) A3 corneles ndash A24 toreles crsquoest nom de (A23)pute gent ndash (A4) mescreant (1188) A23 fieacute ndash A4 terre (1200) A23 empire ndashA4 compeignie

proper names show often variations

(609) A24 Guielin le franc ndash A3 Guielin lrsquoenfant (952 957) A23 Ricor-dane ndash A4 Cordane TiacreFiacre (1136 1138 1155 1186 ) (1295) A3Gilebert de Faloise sor Mer ndash A24 Gilebert de Faloise le Ber (1364) A23 RaoldrsquoOmacre ndash A4 Raoul de Marche

Scribal variations

Adjectives

(621) A23 tot ndash A4 trestot

Pronouns

(370) bien vueil que tuit (A24) vos (A3) lrsquooiez

Articles

Variations between celtel ndash celetele are frequent ndash probably due to the diffi-culty in distinguishing the letters c and t eg 210 329

Conjunctions

(1449) A24 com ndash A3 quant

Prepositions

(191) A2 en ta cort ndash A4 a ta cort (635) A23 sor ndash A4 sus (several occur-rences of the alternation sussor) several occurrences of the varation betweentrusqursquoau ndash jusqursquoau eg 1005 1083

Verbs

(161) A2 refis ndash A4 en fis (425) A2 fet ndash A34 dit (601) A2 si sorrist ndashA34 si srsquoen rist (808) A23 srsquoaseons ndash A4 si seons (848) A34 aresonez ndashA2 resonez (910) A23 preismes ndash A4 veismes (1240) A34 creverent ndash A4coperent (1333) A2 errache A4 arache ndash A3 estache (1359) A23 se prist aescrier ndash A4 comanccedila a crier (1366) A23 mener ndash A4 amener

Adverbs

(461) A23 ausi ndash A4 ainsi (574) A2 molt ndash A34 si (604) A23 faintement ndashA4 faussement (989) A3 com fierement ndash A24 comfetement

In some cases the variations between the mss are more significant ndash so sig-nificant that it is difficult to speak of clear identity of content These are infact difficult border-cases Yet we still find an underlying similarity of struc-ture which indicates that these variations are related and thus differ from thoseclassified as level 4-variations see below Cases which are still related althoughdifferent in meaning are eg (917) nrsquoen iert menccedilonge (A24) oiumle (A3) dite(1159) Par voz merciz faites nos (A24) en doner (A3) entendant Simpleerrors include eg (794) A2 demander ndash A34 dementer (1277) A24 delez ndash(A3) celez (1365) A34 lrsquoapostre ndash A2 la porte

Lene Schoslashsler


The three A mss have a common exemplar a None of the daughters of aare free adaptations and no variations occur at level 4 As shown in eg VanReenen and Schoslashsler (1996) unrelated (type-2) variations are altogether rarein our text I here quote two of them in order to illustrate how different vari-ations can be when they are unrelated In both cases we have an oppositionbetween the A and the B families on the one hand and the mss C and D on theother hand

v 0029 ABCDA1 Li cuens Guillaume fu molt gentix et berA2 Li quens Guillaume fu molt gentis et berA3 lacunaA4 Li quens Guillaume fu molt gentil et berB1 Li quens Guillaume fu molt gentiz et berB2 lacunaC A son ostel descendi au degreD En la grant place est descendus li ber

v 0161 ABCDA1 Srsquoil le deffent bien en doi avoir blasmeA2 Srsquoil le deffent bien en doi avoir blasmeA3 lacunaA4 Srsquoi le deffant bien en doi avoir blameB1 Srsquoil srsquoen deffent bien en doi avoir blasmeB2 Srsquoil srsquoen defent bien en doi avoir blasmeC Srsquoil le desdist prest sui que mrsquoen combateD Srsquoo volt noier pres suis drsquoa lui conbatre

To sum up variations at level 4 ie at the level of content in unrelated pas-sages are not found between these mss copied by different scribes copyingfrom one and the same exemplar at approximately the same moment and atapproximately the same place

Conclusion

Does the study of the Perceval copies compared to the Charroi copies bringus closer to an understanding of which types of variation are genealogicallyrelevant and which are not

I will summarize what we have learnt from the study of the two sets of mss

Scribal variations

Spelling variations (ie variations at level 1) and lexical variations (ie vari-ations at level 3) are very frequent Morphological and syntactic variations (ievariations at level 2) are much less frequent As expected unrelated variationsof content (ie variations at level 4) are virtually non-existent It has been estab-lished for both sets of mss that the common dialect of the copies differs fromthe original dialect Type and frequency of variations do not however differsignificantly between the two sets When mss have made the same itinerary atapproximately the same moment we do not find essential differences in vari-ation between one copyist copying the same exemplar and different copyistscopying one and the same exemplar Absence of a norm makes individual vari-ations so frequent that independent copyists and one and the same copyistworking on the same text behave equally freely when copying The conclusionto be drawn from this comparison then is that we have no linguistic cluesfor distinguishing a single personrsquos individual variation from different personrsquosvariations ndash when copies were made at approximately the same moment and inthe same dialect Fundamental differences may arise in cases of different copy-ing attitudes or techniques (mirror copying versus adaptation transcriptionversus dictation) and when copies were made at different moments and in dif-ferent dialects But our cases with apparently the same copying attitude andcommon lsquooutputrsquo dialects do not permit me to postulate the existence of twoessentially different linguistic situations Only the identity or the difference ofthe hand itself can tell us whether the spelling variations stem from one or fromdifferent persons

On the other hand it should not be forgotten of course that external ev-idence dialectal evidence and chronological evidence are the first importantpoints of genealogical investigations ndash but these points are of no relevance herewe are talking here about specific linguistic clues in stemmatology

Sometimes one even gets the impression that scribes make spelling vari-ation a point of honour Let me quote the case of Yvain vv 2025ndash2031 withregular alternation between antel and entel that Kajsa Meyer has kindly broughtto my attention (quotation adapted here see Meyer (199569) for a diplomatictranscription of the Guiot-manuscript)

( anquel maniere)antel que graindre estre ne puetentel que de vos ne se muetMes cuers nrsquoonques aillors nel truifantel qursquoillors pansser ne puisentel que toz auos mrsquootroj

Lene Schoslashsler

antel que plus uos aim que mojentel srsquoil uos plest a deliure

The conclusions of the investigation into mss T and V and the comparisonwith the Charroi A group of manuscripts are threefold

1 Individual variation in spelling (level 1) and lexicon (level 3) is much largerthan I had previously believed it to be On the other hand morphologicaland syntactic variation (level 2) is more stable than I had expected

2 If the results of the investigation of mss T and V are representative of theirtime these findings will have implications for our understanding of theway people used their language in northern France in the 13th century8

3 The unexpected liberty of variation of lsquomouvancersquo seen in one and thesame person must have implications for our investigation of relations be-tween mss Philologists often base their arguments on lexical variations ndashbut we have seen in our two mss that lexical variation without any ge-nealogical relevance is actually very frequent Levels 2 and 4 (morphologysyntax and unrelated lexical variations) may offer better clues for the inves-tigation of manuscript variations But one of the important conclusionsto draw from this investigation is that we will have to reconsider whichvariations are genealogically relevant and which are not

I believe that these conclusions support many of the claims made by the so-called ldquonew philologyrdquo see eg Schoslashsler (forthcoming) with references

Notes

ldquoMouvancerdquo ldquole caractegravere de lrsquoœuvre qui comme telle avant lrsquoacircge du livre ressort drsquounequasi-abstraction les textes concrets qui la reacutealisent preacutesentant par le jeu des variantes etremaniements comme une incessante vibration et une instabiliteacute fondamentalerdquo Zumthor(1972507) quoted apud Mulken (199331) Cf also Micha (196669ndash70) ldquonous sommesen preacutesence drsquoune reacutefection libre constante nous sommes extrecircmement eacuteloigneacutes drsquounetradition meacutecanique du texte ougrave le copiste reproduit consciencieusement ce qursquoil a sous lesyeux rdquo

For a definition see the contribution of Wattel this volume

According to Busby (1993a) it is most likely that T and V are independent copies ofthe same exemplar see however Busby (1993b notes to 7261 7338 and 7725) for indi-cations that V could be a copy of T See also Busby (2002) Chapter 2 ldquoVarieties of ScribalBehaviourrdquo

See Dees et al (1987)

Scribal variations

See Busby ea p 54 T ndash ldquothe third or fourth quarter of the thirteenth centuryrdquo and V ndashldquoafter 1250rdquo

The relevant verses are quoted in Section 13

V = verb S = subject C = (different sorts of) complement Inf = infinitive Adv = adverbConj = conjunction O = object

It might be interesting to compare my findings with a few other texts copied several timesby the same copyist Keith Busby has kindly informed me of some fabliaux having beencopied twice eg La vieille Truande

References

I Primary sources

Perceval ndash T Paris Bibliothegraveque Nationale ffr 12576 V Paris Bibliothegraveque Nationalenouv acqfr 6614

Charroi de Nicircmes ndash A2 Paris Bibliothegraveque Nationale ffr 1449 A3 Paris BibliothegravequeNationale ffr 368 A4 Milano Biblioteca Trivulziana 1025

Busby Keith (1993b) Le Roman de Perceval ou le Conte du Graal Edition critique drsquoapregraves tousles manuscrits Tuumlbingen Niemeyer

Hanchard J (1955) Le Charroi de Nicircmes chanson de geste du XIIe siegravecle eacutedition du manuscritde Boulogne-sur-Mer Louvain

Lange-Kowal Ernst-Erwin (1934) Das altfranzoumlsische Epos vom Charroi de NicircmesHandschrift D herausgegeben mit sprachwissenschaftlichem Kommentar und GlossarJena

Meyer Kajsa (1995) La copie de Guiot fol 79v-105r du manuscrit ffr 794 de la BibliothegravequeNationale ldquoli chevaliers au lyeonrdquo de Crestien de Troyes Eacutediteacutee par Kajsa MeyerAmsterdam-Atlanta Rodopi

McMillan Duncan (1972) Le Charroi de Nicircmes Editeacutee drsquoapregraves la reacutedaction AB avecintroduction notes et glossaire Paris Klincksieck

Roach William (1959) Chreacutetien de Troyes Le Roman de Perceval ou Le Conte du Graal ParisDroz

Walker B J (1955) The Boulogne text of the Charroi de Nicircmes University of London

II Philological and linguistic studies

Busby Keith (1993a) ldquoThe Scribe of MMS T and V of Chreacutetienrsquos Perceval and itsContinuationsrdquo In K Busby et al (Eds) Les manuscrits deThe manuscripts of Chreacutetiende Troyes (pp 49ndash65) Amsterdam Rodopi

Busby Keith (2002) Codex and Context Reading Old French Verse Narrative in ManuscriptIndashII Amsterdam Rodopi

Comfort T E (1954) The Charroi de Nicircmes Old French Chanson de geste edited from theManuscript of Boulogne-sur-Mer University of Illinois

Lene Schoslashsler

Dees Anthonij et al (1980) Atlas des formes et des constructions des chartes franccedilaises du13e siegravecle Beiheft zur Zeitschrift fuumlr romanische Philologie Band 178 Tuumlbingen MaxNiemeyer

Dees Anthonij et al (1987) Atlas des formes linguistiques des textes litteacuteraires de lrsquoancienfranccedilais Beiheft zur Zeitschrift fuumlr romanische Philologie Band 212 Tuumlbingen MaxNiemeyer

Foulet Lucien (1967) Petite Syntaxe de lrsquoAncien Franccedilais Paris ChampionJodogne Omer (1956) Le manuscrit de Boulogne du ldquoCharroi de Nicircmesrdquo Publicaciones de

la Facultad de Filosofiacutea y Letras II (17) 301ndash326 ZaragozaMicha Alexandre (1966) La tradition manuscrite des romans de Chreacutetien de Troyes Genegraveve

DrozMulken Margot van (1993) The Manuscript Tradition of the Perceval of Chreacutetien de Troyes

A Stemmatological and Dialectological Approach (PhD Thesis) AmsterdamReenen Pieter van amp Lene Schoslashsler (1993) ldquoLes indices drsquoinfinitif compleacutement drsquoobjet

en ancien franccedilaisrdquo Actas do XIX Congreso Internacional de Linguumliacutestica e FiloloxiacuteaRomaacutenicas Vol V 523ndash545 La Coruntildea

Reenen Pieter van amp Lene Schoslashsler (1996) ldquoFrom Variant to Pedigree A Typology ofVariantsrdquo In P van Reenen amp Margot van Mulken (Eds) Studies in Stemmatology (pp263ndash304) Amsterdam ndash Philadelphia Benjamins

Salemans B J P (1990) ldquoText Genealogical Remarks on Lachmann Beacutedier Greg andDearingrdquo Leuvense Bijdragen 79(4) 427ndash468 Louvain

Schoslashsler Lene (1973) Les temps du passeacute dans Aucassin et Nicolete Lrsquoemploi du passeacute simpledu passeacute composeacute de lrsquoimparfait et du preacutesent ldquohistoriquerdquo de lrsquoindicatif Etudes romanesde lrsquoUniversiteacute drsquoOdense vol 5 Odense

Schoslashsler Lene (1984) La deacuteclinaison bicasuelle de lrsquoancien franccedilais son rocircle dans la syntaxede la phrase les causes de sa disparition Etudes romanes de lrsquoUniversiteacute drsquoOdense vol19 Odense

Schoslashsler Lene (1988) ldquoLa constellation de Narcisse Distribution spatiales et temporellesconstellations des manuscritsrdquo Etudes de variation linguistique offertes agrave Anthonij Dees agravelrsquooccasion de son 60me anniversaire 247ndash263 Amsterdam

Schoslashsler Lene (1989) ldquoProblegravemes de stemmatologie illustreacutes par le cas de NarcisserdquoJaarboek 1988ndash1989 167ndash174 Amsterdam

Schoslashsler Lene (1994) ldquoDid lsquoAktionsartrsquo ever dominate Verbal Aspect in Old Frenchrdquo In CarlBache Hans Basboslashll amp Carl Erik Lindberg (Eds) Tense Aspect and Action Empiricaland Theoretical Contributions to Language Typology (pp 165ndash184) Berlin ndash New YorkMouton de Gruyter

Schoslashsler Lene (1995) ldquoNew Methods in Textual Criticism the Case of the Charroi deNicircmesrdquo In J Fisiak (Eds) Medieval Dialectology (pp 225ndash276) Berlin ndash New YorkMouton de Gruyter

Schoslashsler Lene (forthcoming) ldquoThe Copyist at Work How did he Work What are theConsequences for Linguistic Research and for Editorial Policyrdquo

Tyssens Madeleine (1967) La Geste de Guillaume drsquoOrange dans les manuscrits cycliquesParis Les Belles Lettres

Zumthor Paul (1972) Essai de poeacutetique meacutedieacutevale Paris Seuil

The effects of weighting kinds of variants

Matthew Spencer Linne R Mooney Adrian C BarbrookBarbara Bordalejo Christopher J Howe andPeter RobinsonUniversity of Cambridge University of Maine Orono De Montfort University Leicester

Introduction

Manuscript production by manual copying introduced many different kinds ofvariants into texts from changes in spelling and punctuation to the insertionor deletion of whole lines or sections Are all these kinds of variants equallyreliable for stemma reconstruction and if not how should we deal with dif-ferences in reliability It will be helpful to distinguish two different classes ofvariants those that are biased with respect to the true stemma and those thatare unbiased but relatively uninformative Both classes have analogues in thereconstruction of phylogenies (evolutionary trees depicting the lines of descentof organisms) by evolutionary biologists For example the external appearanceof many organisms is often more affected by the conditions under which theylive than by their lines of descent It has been known for more than a centurythat such external characters are of little use in establishing evolutionary rela-tionships For example in Chapter XIV of The Origin of Species (Darwin 1859)ldquoNo one regards the external similarity of a mouse to a shrew of a dugong toa whale of a whale to a fish as of any importancerdquo Spelling and punctuationvariants are good examples of variants that are likely to be biased as they maybe strongly influenced by the dialect spoken by the scribe Such variants areusually excluded from stemmatic analyses (Robinson 199772ndash74) Unbiasedbut relatively uninformative variants are usually those that show a high fre-quency of change Most stemmatic analyses are based on the principle that weshould prefer a stemma requiring relatively few changes to produce the ob-

Matthew Spencer et al

served distribution of variants over a stemma that requires many changes Ifa variant arose independently many times distributed over changing groupsof manuscripts it can tell us little about the relationships among manuscriptsbecause its distribution is unlikely to be simply related to the true stemmaOn the other hand a very improbable variant gives strong evidence that allthe manuscripts in which it occurs are related by descent In evolutionary bi-ology Darwin stressed the importance of relatively constant characters overcharacters showing large amounts of variation (Darwin 1859Ch XIV)

Here we assume that biased classes of variants have been removed andexamine the circumstances under which unbiased but uninformative variantsmay cause problems for stemma reconstruction We use Lydgatersquos Kings ofEngland as an example Many of our methods are derived from evolutionarybiology We cite relevant papers in the biological literature but we translatebiological terms into their analogues in stemmatology (for example lsquotaxonrsquobecomes lsquomanuscriptrsquo and lsquophylogenetic treersquo becomes lsquostemmarsquo)

The data

Lydgatersquos Kings of England consists of 105 lines in most versions 35 manuscriptsand three early printed editions survive We transcribed all 38 surviving copiesand manually coded them so as to distinguish eleven kinds of variants (Table 1The table gives the weight for a change from the base text to each kind Forchange from one kind of variant to another we used the maximum of weightsfor each kind of variant (for example if one manuscript had a variant portionof a line and the other had that portion omitted we gave the difference be-tween these two a weight of 50) Missing data due to damage were always givenweight 0)

Each coded text contained 860 single-letter symbols We describe ourmethods and results in more detail elsewhere (Mooney et al 2001)

Determining the values that should be used to weight different kinds ofchanges is problematic If there is a priori information on the frequency or im-portance of different kinds of change one could use this information to assignweights In most cases such information will not be available at least until af-ter the stemma has been produced It may be possible to optimize the initialguesses (Ronquist 199582) but this will be difficult if there are many differentkinds of variants Our aim was to determine whether weighting different kindsof variants is likely to be important for a given text We therefore comparedstemmata constructed using two plausible sets of weights uniform weights


Table 1 Non-uniform weights used for each kind of variant

Kind of variant Weight

Stanza omitted 90Line changed completely 90Variant portion of line changes meaning 50Word change affecting rhyme 40Line or portion of line omitted 30Proper noun variant changes meaning 25Word variant changes meaning 20Major word addedomitted changes meaning 15Two (or more) words in reverse order 15Word variant without change in meaning 10Minor word addedomitted without change in meaning 5

and our subjective ideas of the importance of different kinds of changes asgiven in Table 1 We gave high weights to kinds of variants that we thought un-likely to occur independently more than once (such as a completely changedline) and low weights to kinds of variants that we thought could easily havearisen independently several times in the tradition (such as the omission of aminor word without changing the meaning of a sentence) We stress that theseweights are subjective in that other scholars might assign different values ordifferent rank orders to these kinds of variants However we analyzed severalother weighting schemes including random weights with similar results whichwe do not report here

We are therefore confident that our conclusions are not much affected bythe choice of weighting scheme We return in the Discussion to the problem ofobjective assignment of weights

One reason why weightings might make no difference is that if all variantsare transmitted via the same stemma the same stemma should be recoveredwhatever weights are used for different kinds of variants (although samplingerror in texts of finite length may introduce some disagreement) To test thisidea we also compared stemmata constructed using each of the four common-est kinds of variants in the Kings of England data major words addedomittedminor words addedomitted word variants changing meaning and word vari-ants not changing meaning We did not examine rare kinds of variants becausethere were too few data to be reliable


Tree reconstruction and comparison

We used neighbour-joining to reconstruct stemmata from matrices of pairwisedistances between manuscripts Neighbour-joining (Saitou amp Nei 1987406ndash425) is a simple clustering algorithm that sequentially separates pairs ofmanuscripts from an initially unresolved stemma such that at each step thesum of least-squares branch length estimates is minimized For biological databoth simulations (eg Saitou amp Imanishi 1989524 Tateno et al 1994273) andanalyses of real data with well-supported phylogenies (Wiens 2000624ndash625)have shown neighbour-joining to be among the most successful tree recon-struction methods For uniform weights we estimated pairwise distances asthe proportion of non-missing locations at which two manuscripts had dif-ferent coding symbols For non-uniform weights differences between codingsymbols were weighted as in Table 1

We gave high weights to kinds of changes we thought unlikely to have oc-curred independently many times and low weights to those that we thoughtcould easily have arisen independently in different parts of a manuscripttradition

Although the observed number of differences between a pair of manuscriptsignores the possibility of multiple changes at single locations such eventswould have been rare in the Kings of England tradition because the maxi-mum pairwise uniformly weighted distance was 021 With distances this smallthere will be few cases of several changes at a single location When this isnot the case one should attempt to estimate the actual number of differences(Spencer amp Howe 2001) We used three methods to check for contamina-tion split decomposition (Huson 1998 68ndash73) Mantel correlations (Upton ampFingleton 1985) between distance matrices from adjacent sections of text andthe construction of separate stemmata from separate sections of text (Robinson199775ndash79) None of these methods suggested extensive contamination so abranching stemma is probably a reasonable representation of this text tradi-tion All our stemmata are unrooted they indicate the topology of the stemmabut not the location of the archetype Stemmata could subsequently be rootedby judgements about the originality of readings palaeographic evidence orinternal manuscript information on exemplars To help illustrate the main fea-tures of the stemmata we have grouped manuscripts into categories (A B andC) suggested by manual stemmatic analysis (Mooney et al 2001)

We used two different methods to compare stemmata partition distancesand consensus stemmata The partition distance (Penny amp Hendy 198576) be-tween a pair of stemmata is a measure of their topological similarity Removing


1

1

3

2

2

3

4

4

5

5

Figure 1 Partition distance between a pair of stemmata

any edge divides a stemma into two sets of manuscripts We say that an edge iscommon to both of a pair of stemmata if in both stemmata there is an edgewhose removal divides the stemma into the same two sets of manuscripts Thepartition distance between two stemmata is simply the number of edges foundin one but not both of the stemmata (Fig 1 Each of the bold edges is present inonly one of the two stemmata For example the bold edge in the upper stemmadivides the manuscripts into the sets 12345 There is no single edge inthe lower stemma whose removal splits the data into these two sets All otheredges are present in both stemmata) We can scale this distance by the numberof edges that could possibly be found in one of two stemmata (2 (N ndash 3) fora bifurcating stemma with N extant manuscripts) to give a distance measurebetween 0 and 1 The probability distribution of partition distance is knownfor random stemmata with small numbers of extant manuscripts and can oth-erwise be obtained by simulation Thus we can determine the probability ofobserving a given partition distance under the hypothesis of no relationshipother than chance between a pair of stemmata

A consensus stemma includes an edge only if it is present in some definedproportion of the set of stemmata being compared Here we use strict consen-sus in which an edge is only included if it is present in every stemma becausewe want to indicate where disagreements occur although several other con-sensus rules are possible (Page amp Holmes 199834ndash35) Where the stemmatadisagree the consensus stemma will have a polytomy rather than a bifurcationWe do not present branch lengths for consensus stemmata as their purposehere is simply to show areas of disagreement between stemmata constructedusing different weightings


We used PAUP (Swofford 2001) for neighbour-joining calculating parti-tion distances and preparing a consensus stemma We calculated distance ma-trices and generated input files for PAUP using Matlab 53 (The MathworksInc Natick MA)

Results

Figure 2 is a stemma obtained by neighbour-joining on uniformly-weighteddistances The different fonts represent manuscript groups obtained by manualstemmatic analysis (Mooney et al 2001) A bold B italic C underlined othersnormal font Branch lengths are drawn to scale (measured in arbitrary units)The stemmata from uniform weights (Fig 2) and the weights given in Table 1(Fig 3) had many features in common although they differed in detail (ourrepresentations also differ in orientation but this has no meaning)

Both had a strong C group (manuscripts in underlined font) clearly sep-arated from the A (bold font) and B (italic font) groups (apart from themanuscripts Bodley 48 and Bodley 686 which our manual stemmatic analy-sis placed doubtfully in the C group but are here placed with the A group)The A and B groups were not distinct from each other The relative branchlengths were different in the two stemmata For example de Worde and RWyer were further away from other manuscripts when the weights in Table 1were used than when uniform weights were used This happens when differ-ent manuscripts have different proportions of each kind of variants eitherbecause each manuscript represents a small sample or because there are system-atic differences However many small groups of manuscripts appeared in bothstemmata (eg R Wyer de Worde Pynson and TCC 601 1 TCC 601 2)

Table 2 gives the distribution of scaled pairwise partition distances es-timated from 1000 random bifurcating stemmata on a set of 38 extantmanuscripts (only the values given in this table occurred) The observed scaledpartition distance between stemmata using uniform weights and weights in Ta-ble 1 was 057 Table 2 shows that a distance this small between two randomly-generated bifurcating stemmata for a set of 38 taxa would be very unlikely bychance Polytomies in the strict consensus stemma (Fig 4) show where the dif-ferences occur We remain uncertain about the relative positions of a large setof manuscripts in the A and B groups We were also unable to resolve the posi-tion of the C group relative to Ipswich Pynson de Worde and R Wyer but theC group (other than Bodley 48 and Bodley 686) was clearly separated from Aand B Several smaller groupings (eg Harley 2251 BL Ad 34360 Digby 186


RWyer

de Worde

Jesus 56IpswichLeiden

Harley 372Lansd 699

RawlC 48Caius 249

Lansd 210

Harley 7333

Egerton 1995

Linc Lat 129

TCC 601 2TCC 601 1

Fairfax 16BL Ad 310 42

NottinghamBodley 48

Pynson

Galba E VIII

Peterborough

Harley 2261

Titus DXX

Lambeth 306

Harley 2169

CUL Ad6686Stowe 69

Bodley 686Dublin 516Bodley 912RawlC 316

Buhler 17

Lansd 762Digby 186


001 distance units

Ashmole 59

Figure 2 Stemmata obtained by neighbour-joining on uniformly-weighted distances

and RawlC48 Jesus 56 Leiden Harley 372 Lansdowne 699) also occurredunder both weightings

Table 3 gives scaled partition distances between each of the four stemmatadrawn from the four commonest kinds of variants These distances were higher


Ashmole 59

CUL Ad6686Dublin 516

Bodley 686Bodley 912

LincLat 129Nottingham

Digby 186BL Ad 34360Harley 2251

Lansd 762Harley 2261

Peterborough

Lansd210

Ipswich

Pynson

RawlC 48

Caius 249Leiden

Stowe 69

Lambeth 306Jesus 56Harley 372

Lansd 699

TCC 601 1TCC 601 2

Harley 7333

Egerton 1995Titus DXXBodley 48

BL Ad 31042

Fairfax 16RawlC 316

Buhler 17

Harley 2169

01 distance units

RWyer

de Worde

Galba EVIII

Figure 3 Stemmata obtained by neighbour-joining on distances weighted as in Table 1Ms groups indicated as in Figure 2 Branch lengths to scale (in arbitrary units) but thescale is different from Figure 2


de WordeRWyer

RawlC 48 Jesus 56Pynson

Ipswich

Caius 249


BL Ad 34360Digby 186

Harley 7333

Egerton 1995

TCC 601 1TCC 601 2

Buhler 17RawlC 316

Dublin 516Bodley 686

Bodley 48

BL Ad 31042

Leiden

Harley 372

Lansd 699

Ashmole 59Bodley 912

Fairfax 16LincLat 129

CUL Ad6686Galba EVIIITitus DXXHarley 2169Harley 2261

Lansd 762Stowe 69

Lambeth 306Nottingham

Peterborough

Figure 4 Strict consensus stemma for neighbour-joining stemmata using uniformweights or weights as in Table 1 Ms groups as in Figure 2 Polytomies points wheredifferent weightings lead to different topologies Branch lengths not to scale

than those between the two different weightings of the whole data set proba-bly because they represent quite small sample sizes However they were stillunlikely to have arisen by chance (cf Table 2)

Table 2 The distribution of scaled pairwise partition distances estimated from 1000random bifurcating stemmata on a set of 38 extant manuscripts

Scaled partition distance Probability

086 601 times10ndash6



094 002097 018100 079


Table 3 Scaled partition distances between stemmata drawn using neighbour-joiningusing the four commonest kinds of variants 1 ndash major words addeddeleted 2 ndash minorditto 3 ndash word variants changing meaning 4 ndash ditto not changing meaning

1 2 3 4

1 ndash2 086 ndash3 086 080 ndash4 083 089 086 ndash

Discussion

The differences between stemmata for the Kings of England constructed usingdifferent weightings were small (Fig 4 Table 2) This was probably becauseall variants were fairly rare and all were transmitted along the same truestemma All the uniformly-weighted pairwise distances between manuscriptswere small so even the most frequent kinds of change would only rarelyshow reversals and coincident variation The pairwise scaled partition distancesamong stemmata drawn using single kinds of variants were smaller than ex-pected by chance alone (Table 3) supporting the idea of a single true stemmafor all these variants If it is not generally important to give different weights todifferent kinds of variants the task of stemma construction may be simplifiedFor example automatic coding systems would not need to distinguish differentkinds of variants and it will be easier to develop explicit mathematical modelsfor copying errors

However there will be cases in which giving different weights to differ-ent kinds of variants might lead to major differences between stemmata Ifwe were unable to identify biased variants beforehand different weightingscould lead to completely different stemmata Furthermore if there were manydifferences between manuscripts reversals and coincident variation might becommon enough that the most frequent kinds of variants were unreliable in-dicators of the true stemma A worst-case scenario combining both of theseproblems is the analysis of a large manuscript tradition extending over a verylong time period Changes in language over time could introduce nonrandompatterns of variation unrelated to descent for example in the substitution ofone word for another or in conventional word order Similarly the large num-ber of copying events could allow frequent reversals and coincident variantsBecause such cases clearly exist the choice of weights for different kinds ofvariants merits more attention than it has received so far Other situations in


which the choice of weights might affect the stemma include contaminatedtraditions in which different kinds of variants were systematically transmittedalong different pathways and traditions with so few variants that sampling er-ror becomes important A survey of the effects of weighting in different texttraditions would be the best way to test these ideas

There are several unresolved problems with the analyses we have describedhere The manual coding we used is subjective and error-prone Automaticcoding could potentially distinguish between some kinds of variants (Robinson1989103ndash104) but has not yet been employed on a large scale The choice ofweights is of more concern We arrived at the weights in Table 1 after extensivediscussion about the importance of different kinds of change and the likelihoodof scribal errors We have no way of knowing whether another set of weightsmight not be much better (although we did analyse some other sets of weightsincluding random weights with similar results) Non-textual evidence such asa full or partial stemma produced independently of textual variants (for ex-ample on the basis of scribal statements of the exemplar used when producinga manuscript Robinson amp OrsquoHara 1996126) will always be needed to distin-guish between biased and unbiased kinds of variants if each kind appears tosupport a different stemma In order to establish objective weights based onthe probabilities of different kinds of change we need a stemma but we needweights in order to obtain that stemma in the first place There are two possiblesolutions

1 Examine the frequencies of different kinds of variants on a stemma whichhas already been established by other means We would then have to as-sume that the frequencies of these kinds of variants are the same in othermanuscript traditions However a stemma (or a part of a stemma) couldonly be established independently of any kind of information on vari-ants when there is non-textual evidence for relationships (for example astatement in the manuscript of the exemplar used) Such cases are rarealthough not completely unknown (OrsquoHara amp Robinson 199355ndash56)

2 A related problem is the determination of reliable and unreliable locationsin a text One would like to give high weights to reliable locations andlow weights to unreliable locations but it is difficult to know in advancewhich locations are reliable It has been suggested that one can constructan initial stemma using some guesses at weights (uniform weights mightbe a reasonable guess) and weight locations according to their degree ofconcordance with this stemma A new stemma is then constructed andthe process repeated until the topology of the stemma stops changing Al-


though this approach sometimes improves the tree reconstruction process(Farris 1969374ndash385 Fitch amp Ye 1991147ndash154) it has been criticized onthe grounds that it is best to avoid such circularities whenever possible (Lee1999726ndash728) With either method one still has to make a subjective deci-sion about the form of the relationship between frequencies of change andweights although simulations can suggest forms that are likely to performwell (Farris 1969377ndash380)

In conclusion our stemmata for Lydgatersquos Kings of England are unlikely to besubstantially affected by the differing usefulness of different kinds of variants(other than spelling and punctuation variants which we did not code) How-ever we have identified situations in which one might need to weight somekinds of variants more highly than others in order to obtain a reliable stemmaDetermining appropriate weightings in these cases is an open problem

Acknowledgements

We are very grateful to the Leverhulme Trust for financial support

References

Darwin C (1859) The Origin of Species by Means of Natural Selection or the Preservation ofFavored Races in the Struggle for Life London Murray

Farris J S (1969) ldquoA Successive Approximations Approach to Character WeightingrdquoSystematic Zoology 18 374ndash385

Fitch W M amp J Ye (1991) ldquoWeighted Parsimony Does It Workrdquo In M M Miyamoto amp JCracraft (Eds) Phylogenetic Analysis of DNA Sequences (pp 147ndash154) Oxford OxfordUniversity Press

Huson D H (1998) ldquoSplitstree Analyzing and Visualizing Evolutionary Datardquo Bioinfor-matics 14(1) 68ndash73

Lee M S Y (1999) ldquoCircularity Evolution Systematics And Circularityrdquo Journal ofEvolutionary Biology 12 724ndash734

Mooney L R A C Barbrook C J Howe amp M Spencer (2001) ldquoStemmatic analysis ofLydgatersquos ldquoKings of Englandrdquo a test case for the application of software developed forevolutionary biology to manuscript stemmaticsrdquo Revue drsquoHistoire des Textes 31 275ndash297

OrsquoHara R amp P Robinson (1993) ldquoComputer-Assisted Methods of Stemmatic Analysisrdquo InN Blake amp P Robinson (Eds) The Canterbury Tales Project Occasional Papers Vol 1(pp 53ndash74) London Office for Humanities Communication Publications


Page R D M amp E C Holmes (1998) Molecular Evolution A Phylogenetic Approach OxfordBlackwell Science

Penny D amp M D Hendy (1985) ldquoThe Use of Tree Comparison Metricsrdquo Systematic Zoology34(1) 75ndash82

Robinson P (1997) ldquoA Stemmatic Analysis of the Fifteenth-Century Witnesses to theWife of Bathrsquos Prologuerdquo In N Blake amp P Robinson (Eds) The Canterbury TalesProject Occasional Papers Vol II (pp 69ndash132) London Office for HumanitiesCommunication Publications

Robinson P M W (1989) ldquoThe Collation and Textual Criticism of Icelandic Manuscripts(1) Collationrdquo Literary and Linguistic Computing 4(2) 99ndash105

Robinson P M W amp R J OrsquoHara (1996) ldquoCladistic Analysis of an Old Norse ManuscriptTraditionrdquo In S Hockey amp N Ide (Eds) Research in Humanities Computing 4 (pp115-137) Oxford Oxford University Press

Ronquist F (1995) ldquoReconstructing the History of Host-Parasite Associations UsingGeneralised Parsimonyrdquo Cladistics 11 73ndash89

Saitou N amp T Imanishi (1989) ldquoRelative Efficiencies of the Fitch-Margoliash Maxi-mum-Parsimony Maximum-Likelihood Minimum-Evolution and Neighbor-JoiningMethods of Phylogenetic Tree Construction in Obtaining the Correct Treerdquo MolecularBiology and Evolution 6(5) 514ndash525

Saitou N amp M Nei (1987) ldquoThe Neighbor-Joining Method A New Method for Recon-structing Phylogenetic Treesrdquo Molecular Biology and Evolution 4(4) 406ndash425

Spencer M amp C J Howe (2001) ldquoEstimating distances between manuscripts based oncopying errorsrdquo Literary amp Linguistic Computing 16(4) 467ndash484

Swofford D L (2001) Paup Phylogenetic Analysis Using Parsimony (and Other Methods)Vers 40b8 Computer software Sinauer Associates

Tateno Y N Takezaki amp M Nei (1994) ldquoRelative Efficiencies of the Maximum-LikelihoodNeighbor-Joining and Maximum-Parsimony Methods When Substitution Rate Varieswith Siterdquo Molecular Biology and Evolution 11(2) 261ndash277

Upton G amp B Fingleton (1985) Spatial Data Analysis by Example Vol I Point pattern andquantitative data Chichester John Wiley amp Sons

Wiens J J (2000) ldquoReconstructing Phylogenies from Allozyme Data Comparing MethodPerformance with Congruencerdquo Biological Journal of the Linnean Society 70 613ndash632

Cluster analysis and the Three Level Methodin the study of the Gospels in Slavonic

Dina MironovaSaint-Petersburg State University

Introduction

Historical background

The Slavonic alphabet was designed in the middle of the IX century in Byzan-tium at the court of patriarch Photius The originator was an outstandingscholar of his time Constantine the Philosopher The script in the course oftime received the name Glagolitic Fifty years later another script appearedwhich was called Cyrillic The new script was based on the Greek alphabet sup-plemented with a few letters to denote specific sounds of Slavonic Old ChurchSlavonic was one of the last literary languages in Europe which came into be-ing together with the new alphabet Along with the alphabet the Slavs receivedtheir first Bible texts

A two-century investigation of the tradition has formed the following viewupon the main stages in the history of the Gospels in Old Church Slavonic

1 The original translations are not preserved in the extant witnesses We canonly speak about a number of manuscripts which date back to the IXndashXIIcenturies These manuscripts have features that can be traced back to theoriginal translations and bear signs of a considerable revision of the end ofthe IX ndash beginning of the X centuries We shall refer to these manuscriptsas the Old Text Type A text type as E Colwell (196945 10ndash11) defines itis ldquothe largest identifiable group of related New Testament manuscriptsrdquoA group of manuscripts form one text type if they agree in a number ofvariants against other groups and if there is agreement of the group in themajority of variants

Dina Mironova

2 Slavonic literature was flourishing during the reign of tsar Simeon (893ndash927) in Bulgaria The activity of translators in this period was characterizedby mass lexical revision mostly the revision of Christian terminology Thisgroup of manuscripts is known under the name of the Preslav Text Type(after a large cultural center where many scribes worked) The Preslav TextType has more extant witnesses than the Old one

3 At the beginning of the XIII century St Savva of Serbia introduced the Typ-icon of Jerusalem in the Serbian Chilander Monastery on Mount AthosIt was meant to serve as the basis for monastery needs A Alekseev callsthis the New Liturgical Tetraevangelion (Evangelije 1998) because the termTetraevangelion had already been used for Liturgy before The New Litur-gical Tetraevangelion was based on the continuous text of the Old Text Typewith some additions from the Bulgarian revisions of the X century1

4 In the first half of the XIV century two new recensions of the Gospelsemerged in the Slavic Monasteries of Mount Athos Like the New Liturgi-cal Tetraevangelion these were Tetraevangelia with a rather stable text Theyrest upon the Christian terminology of the original translation and have nolinguistic features of the Preslav Text Type They are known as the AthoniteText Types The second of these comprises the largest number of sourcesIt achieved predominance in the XIVndashXV centuries and in the XVI cen-tury laid the foundation for printed editions Today with slight linguisticmodifications it is read during the Orthodox Liturgy

5 A new translation of the whole New Testament called the Chudov NewTestament was made in the middle of the XIV century under the supervi-sion of Moscow Metropolitan Alexios (c 1293ndash1378) It didnrsquot circulate inmany sources being forced out by the Athonite Text Types and is preservedin just a few extant witnesses We shall refer to this text type as to the Chu-dov New Testament Text Type This translation shares many features of theAthonite Text Types

The recensions of the Slavonic Gospels are so closely connected that they forma kind of textual continuum and the borders between them are vague

There is no complete data about all the Old Church Slavonic New Testa-ment manuscripts A Alekseev refers to more than 8000 XVIndashXVII centurymanuscripts and about 500 XIndashXV century Gospel manuscripts though thisnumber doesnrsquot embrace all sources (Alekseev 1999132)

The study of the Gospels in Slavonic

The research on the Slavonic Gospels Manuscripts2

The study of the Slavonic Gospels Tradition is complicated by two factorsFirstly the contaminated nature of the tradition puts a ban on considering re-lations between single witnesses and therefore hinders employing genealogicalmethods Readings that can be considered truly genealogically significant arerare and cannot form the basis for the classification3 The only possible wayto investigate the tradition is to consider relations between groups of witnesseswith the same text but never relations between single witnesses4 The arrange-ment of the tradition into groups with the same text is done on the basis ofvariation units (passages in the text where witnesses have different readings)Secondly the tradition has numerous extant witnesses Their number is esti-mated at more than ten thousand and prevents us from making full collationsThe collation of only 500 older manuscripts would take years A compromiseway out is collating corresponding parts of every manuscript each part beinglong enough to provide a sufficient number and variety of variation units5

A reliable classification of a contaminated tradition can be obtained if it isbased on all variation units found in the investigated piece of text It is highlyundesirable to choose only the most significant variation units especially be-cause we have too little information to dispense with part of it The followingtwo arguments prove the necessity to use all available material Firstly themanuscripts in each group are very closely connected and it is often difficultto point out readings which characterise it Therefore it is almost impossible todetect subgroups within one text type if not all variation units are consideredThe more variation units we have the more likely we should be to find all avail-able groupings large and small Secondly when we lack information some latemanuscripts with old text could mistakenly be positioned into the group of laterecensions as their old text was obscured by new syntactical and grammaticalfeatures in the process of copying6 Fortunately the time-consuming process-ing of all the variants is considerably facilitated by computer which saves timeand make it possible to work with any number of manuscripts

In the field of Old Church Slavonic New Testament Studies the methodof A Alekseev based on cluster analysis has been used since 1980 In 1999 EWattel tested the method he designed for stemmatological needs on some ofthe Slavonic Gospel manuscripts and the results in general coincided with thepreviously made classifications It became clear that the stemmatological ideaof building a non-oriented graph at the initial stage of stemma constructiongoes along with the principles of the study of a contaminated tradition The

Dina Mironova

research underlying this paper aims at comparing the methods of A Alekseevand E Wattel on the basis of the results of both classifications

The research comprises 531 manuscripts containing the Gospel accordingto Matthew passage 1414ndash1434 This piece of text has about 300 words andprovides 545 variation units

We shall first describe the current project on studying the Old ChurchSlavonic New Testament and then outline the essence of the methods of AAlekseev and E Wattel will be outlined After that we shall present the classi-fications provided by both methods and discuss their similarities and differ-ences According to the obtained results it should be possible to evaluate theadvantages of each formal method for Old Church Slavonic New Testamentstudies and the degree of precision demanded from the method

The Slavonic project

The research described in this paper was carried out in the framework of aproject in the St Petersburg branch of the Bible Society in Russia The projectaimed at preparing the editions of Old Church Slavonic Gospels based on allavailable sources and was launched in December 1994 by the Committee of theUnited Bible Societies Before that as early as January 1993 a large numberof linguists had started the collations in the framework of the Slavonic BibleFoundation The further work on the data and the preparation of the editionwas carried out in St Petersburg under the supervision of A Alekseev

The Gospel according to John was published in 1998 (Evangelije 1998)The critical apparatus of the edition included all the representatives of the OldText Type and a number of representatives for every other recension7 It wasfor the first time that all 30 witnesses chosen for the critical apparatus of theSlavonic Gospels were cited throughout the whole text for each and every in-stance where they had a different reading The basis for the edition is the XIcentury Marianus Tetraevangelion (Mr) The edition shows the textual historyof the Slavonic Gospels as reflected in the extant manuscripts and discoveredduring the research

The method of A Alekseev

The assumption underlying the research is that the tradition of the SlavonicGospels is controlled (Evangelije 1998 Alekseev 1985) which means that the


scribe could use more than one copy of the same Gospel for his work The textwas constantly corrected in order to be kept as close to the original as pos-sible Hence the traditionrsquos high level of stability which makes it difficult forthe contemporary scholar to trace back textuological connections among theextant witnesses It is not possible in any of the text types to find a variantwhich occurs in all the manuscripts belonging to that particular text type andnot in other text types One can only speak of variants which occur in mostmanuscripts of a given text type The same is true for the Greek New Testa-ment and A Alekseev has borrowed some ideas expressed by the AmericanNew Testament scholar E Colwell in his work with the Greek New Testament(Colwell 1969)

The theory of E Colwell in the method of A Alekseev

E Colwell (196963ndash83) shows that it is not relevant to apply a genealogicalmethod to the text type because a text type of the Greek New Testament tradi-tion is a highly contaminated group of manuscripts A genealogical methodimplies that every manuscript has one parent whereas one extant witnessof one text type can have more than one parent If we regard contaminatedsources as nodes of a tree most pairs of nodes will be connected by only oneedge (see Figure 1) There is no way to orient this stemma in such a way thatevery node has no more than one parent

m1 m2

m3

m4

m5

m6

m7

Figure 1 A stemma for manuscripts of the same text-type

Dina Mironova

E Colwell (196915ndash19) believes that the ldquoeffort to restore the text of aText typerdquo is misleading For every text type E Colwell suggests finding aver-age typical most characteristic representatives which combine in themselvesmost typical textual peculiarities of the given text type and presenting theirevidence in an apparatus If families (the smallest and most intimately relatedgroups) can be pointed out within a text type one representative of a familyshould also be a valuable addition to the critical apparatus In order to deter-mine text types within the Greek New Testament Tradition E Colwell chose aquantitative method

Similarity score

E Colwell and E Tune showed the inconsistency of using Textus Receptus asa basis for the collation of Greek New Testament manuscripts They suggestedcomparing manuscripts by pairs and thus forming a list of variation units Afterthe variation units have been found the number of readings in common forevery pair of manuscripts is counted Then they calculate the ratio of variantsin common for every pair and the variation units relevant for this pair Thisfigure is converted into a percentage Working with percentages enables themto compare complete and lacunary manuscripts on an equal basis If Ep is thenumber of variants manuscripts lsquoarsquo and lsquobrsquo have in common and Vp is thenumber of variation units relevant for the pair then the similarity score (Sp)for the pair lsquoabrsquo is

Sp = (Ep Vp) times 100

This is one of the coefficients often used in cluster analysis It requires the leastnumber of calculations and in our case it is difficult to justify the usage ofmore elaborate coefficients The percentage of readings in common for eachpair is put down in a square matrix on the intersection of lines and columnslines and columns standing for manuscripts In such a matrix E Colwell calls agroup a number of manuscripts which have a higher percent of similarity witheach other than with other sources

Data presentation

The contribution of A Alekseev was that he added one more step to themethod to facilitate the analysis He proposed rearranging the matrix so thatthe manuscripts with more closely related text would be positioned togetherand clusters of the same text type would be visible


at least one memberof the given pair already

belongs to a cluster

a new clusteris formed

the unassigned ms isjoined to the samecluster as its pair

two clusters arejoined together

do both mssalready belong

to some clusters

do they belongto the same cluster

nothing changesin the clusters

no

no

no

yes

yes

yes

Figure 2 Kuznestovarsquos algorithm for grouping manuscripts in clusters

The program was designed by E Kuznetsova (see Alekseev amp Kuznestova1987) The similarity score is measured by Colwellrsquos coefficient mentionedabove The objects are joined to the cluster by a single-link clustering crite-rion (for the terminology cf Galloway 1979) which requires a single strongresemblance of an object to any member of the group The program looks fora pair of manuscripts with the next highest similarity score The algorithm isshown in Figure 2

The clustering stops when all the objects are clustered together The se-quence of the manuscripts within the final cluster reflects the order in whichthey were clustered After getting an ordered matrix we look for borders be-tween clusters The matrix is designed in such a way that manuscripts withinone cluster and the clusters themselves are ordered by the decrease of the sim-ilarity score The border between clusters is the rise of the percent Let usconsider the example in Figure 3

The similarity score between lsquoarsquo and lsquobrsquo is 90 lsquocrsquo has an 87 and 88correspondence with each of them respectively The correspondence of lsquokrsquo withthem is 79 81 and 85 The next two mss lsquolrsquo and lsquomrsquo have 90 similarity

Dina Mironova

Cluster I manuscripts a b c

Cluster II manuscripts k l m n o

Cluster III manuscripts x y z p q

Cluster IV manuscripts u v

Manuscripts which didnrsquot join any cluster f g h i

Figure 3 The final matrix of clusters Manuscripts are ordered by the decrease ofsimilarity score

with lsquokrsquo Thus the similarity score goes from 90 down to 85 and then in-creases up to 90 among the mss lsquokrsquo lsquolrsquo and lsquomrsquo staying lower among the latterthree and lsquoarsquo lsquobrsquo lsquocrsquo (85ndash79) The first cluster we can point out is there-fore lsquoabcrsquo The second cluster comprises mss lsquokrsquo lsquolrsquo lsquomrsquo lsquonrsquo lsquoorsquo The percentage ofsimilarity goes down to 80 and then again up to 87 This is where the newcluster lsquoxyzpqrsquo starts When we compare it to the previous clusters we shouldnote that here the similarity between the first two objects is much higher thanamong the rest The fourth cluster is lsquouvrsquo with lsquof rsquo probably forming part of it


Disadvantages of the classification

In the chosen type of clustering the objects which didnrsquot join any cluster arepositioned at the bottom of the table lsquoat the peripheryrsquo Thus manuscripts lsquogrsquolsquohrsquo and lsquoirsquo do not form part of the lsquouvrsquo cluster but are united with all the clus-ters at the last stage A disadvantage of such an algorithm is that the peripheryof the table is a mixture of manuscripts mostly belonging to the periphery ofdifferent clusters In Figure 3 we can see that the similarity score of lsquogrsquo withthe cluster lsquoxyzpqrsquo (73ndash70) is higher than with its neighbours lsquoursquo lsquovrsquo lsquof rsquo lsquohrsquo(68ndash64) Manuscript lsquohrsquo has a higher degree of similarity with the clusterlsquoklmnorsquo (72ndash70) And finally manuscript lsquoirsquo has relatively the same simi-larity score with all clusters This drawback however doesnrsquot affect the resultIt has already been mentioned that the borders between clusters of the OldChurch Slavonic New Testament are vague Due to contamination differentclusters have a periphery in common and in the table manuscripts lsquogrsquo lsquohrsquo lsquoirsquocan be said to form the periphery of all the clusters above If a manuscript ispositioned at the bottom of the table by mistake it is easy to reposition it cor-rectly if needed (as in the case of manuscripts lsquogrsquo and lsquohrsquo if the variation unitsprove their belonging to certain clusters)

Stages of classification

The method provides a possibility to work with any number of manuscriptsbut the first matrix is not the final one the work is done in two stages At thefirst stage in the first matrix we look for groups with identical witnesses andsubstitute one representative for each of these groups The philologist decideswhich representative to choose At the second stage we produce a new matrixwhere one representative stands for each group of identical manuscripts In thismatrix we determine borders between clusters

The current project covered 1100 available manuscripts Most Athonitemanuscripts could be detected before the stage of computer classification Mostof the witnesses of older recensions and the older witnesses of the Athonite TextType (altogether 531 manuscripts for the Gospel according to Matthew) werechosen for the computer classification The method of A Alekseev yielded thefollowing clusters 321 manuscripts with the Athonite Text Type 49 with theOld Text Type 96 with the Preslav Text Type 4 with the New Liturgical Tetrae-vangelion and 3 with the Chudov New Testament Text Type The remaining 58manuscripts are at the periphery and should be studied by purely philologi-cal analysis Thus we have systematised the witnesses and significantly reduced

Dina Mironova

the number of those needing further investigation It should be specified herethat the classification algorithm is but a small part of a big project The com-plete philological part is done by scholars as well as the final evaluation andchecking of the programrsquos output

The method of E Wattel

The method of A Alekseev remained the only one which could be used for acontaminated tradition with large number of sources The application of themethod of E Wattel to the Slavonic material has shown that it is also possibleto turn to the tools of stemmatology It has also demonstrated the large scaleof employment for the method Since Wattelrsquos concept is in detail explained inWattel (this volume) we shall dwell only on the items which are significant forthe comparison of the methods of Alekseev and Wattel

The theory of A Dees in the method of E Wattel

Originally E Wattel based his work on the Three Level Method developedby A Dees (1975) A Dees suggests three stages on the way of building a ge-nealogical classification for the sources first define the deep structure then theintermediate structure and only afterwards the oriented stemma A deep struc-ture reflects the internal relations among the manuscripts It is a fully resolvedtree structure ndash a tree in which every extant witness is represented as a terminalnode and every other node in the tree represents a fictive (lost) manuscriptwith precisely three connections to other nodes An intermediate structure is acontracted deep structure E Wattel devised a set of algorithms to build a deepstructure and to contract it (Wattel ea 1996 this volume)

We shall see that the initial tree or deep structure he constructs is a dif-ferent representation of the same ideas which lay behind A Alekseevrsquos matrixwith clusters

For the further discussion we need to connect the concepts of variation unitand version formula A variation unit deals with the opposition of variants (egvariant lsquo1rsquo vs variant lsquo2rsquo) whereas a version formula implies opposition of oneor more manuscripts to other manuscripts For example if we have variationunit lsquoXrsquo with two variants lsquo1rsquo and lsquo2rsquo and manuscripts lsquoarsquo and lsquobrsquo share variantlsquo1rsquo while manuscripts lsquocrsquo and lsquodrsquo share variant 2 then the version formula forthis variation unit will be manuscript lsquoarsquo and manuscript lsquobrsquo vs manuscript lsquocrsquoand manuscript lsquodrsquo


Formula weight

Whereas A Alekseev refrains from weighting variation units E Wattel distin-guishes two weight constituents in every variation unit (every version formula)linguistic and computational Linguistic weight is supplied if needed in theinput data by the linguist himself It reflects the hierarchy of importance fordifferent types of variation units Computational weight is a characteristic ofthe version formula (not of a variation unit because the information aboutthe computational weight is derived from the combination of manuscripts inthe version formulas) It reflects ldquohow suitable the structure of a version for-mula is for the construction of a stemmardquo (Wattel this volume) In the originaldatabase linguistic weight for every version formula is multiplied by a compu-tational coefficient depending on the type of the formula This figure is addedup by all positive (a pair of manuscripts sharing the same variant) and all neg-ative (a pair of manuscripts having different variants) contributions of everypair of manuscripts in the given version formula Thus the original linguisticweight obtains a computational constituent After every formula has acquiredweight this weight is distributed in the formula between pairs and every pairgets weight in every formula This weight is positive if both members of thepair have the same variant in the formula and negative if they have differentvariants The weight of a pair of manuscripts with the same variant in a versionformula also depends on the number of manuscripts sharing this variant in thisversion formula and the weight of a pair of manuscripts with different variantsin a version formula depends on the overall number of pairs in this versionformula Having obtained the weight for every pair of manuscripts in everyformula we can calculate the sum of positive and negative weight contributionsfor every pair over the whole database Then we can start the classification

The distance function and the similarity score

Both methods under discussion use cluster analysis techniques E Wattel calcu-lates the distance between the objects (manuscripts) pairwise and A Alekseevcalculates the similarity score and each step in structuring the data is the searchfor the next pair of manuscripts with the smallest distance (highest similarityscore) Let us dwell upon the function of E Wattel If Np is the sum of negativeweight contributions for a pair of manuscripts lsquoarsquo and lsquobrsquo and Sp is a sum ofpositive weight contributions for lsquoarsquo and lsquobrsquo the distance function (Dp) is

Dp = Np (Np + Sp)

Dina Mironova

As we can see the sum (Np + Sp) corresponds to Vp (units of variation rele-vant for the pair) and Np corresponds to Ep (units of variation in which bothmanuscripts of the pair have the same reading) in the similarity score used byA Alekseev and while A Alekseev is always looking for a pair with the highestsimilarity score (Sp) E Wattel is looking for a pair with the minimal distance(Dp) Tests with the Slavonic Gospels have shown that both formulas gener-ally give the same results and the pair with the highest similarity score has thesmallest distance

The divergence of the algorithms starts after the first step has been madeand the first pair has been found As was described above the algorithm of EKuznetsova extracts both members of the pair from the database and keepson looking for the next closest pair The algorithm of E Wattel substitutesone new symbol for both members of the pair in every version formula of thedatabase adjusts the weights of all formulas and only then starts looking forthe next pair with minimal distance (the procedure is described in detail inWattel this volume)

Stages of classification and data presentation

The outcome is presented as a stemma (the procedure of drawing a stemmacan be found in Wattel ea 1996) The algorithm stops when there are onlythree manuscripts left if the database turns out to be empty at a certain stageor if all version formulas are trivial In the latter two cases ldquothe pairs could besqueezed in a random way to get a fully resolved treerdquo (Wattel this volume)Just like the first matrix of A Alekseev is not final and is used to detect identi-cal manuscripts and substitute them by one representative so the pilot stemmaof E Wattel is the basic tree which can be refined On the way to obtaining thefinal classification A Alekseev repeats the clustering algorithm while E Wat-tel designes a new algorithm of stepwise refinement of the initial tree Everystep consists of detaching a branch from the pilot tree attaching it elsewherein the tree and checking whether the new tree is a better reflection of the in-formation from the version formulas So far we have been dealing with treeswhere terminal nodes represented extant witnesses and intermediate nodes hy-pothetical witnesses Every intermediate node had precisely three connectionsto other nodes The latter restriction was caused by the need to reduce thecomputational complexity In the final stage of classification E Wattel designsa tree without this type of restriction After the best initial tree has been ob-tained the contraction phase begins (the procedure is described in Wattel ea1996137ndash138 157ndash161) It means that some internal nodes have a number of


connections larger than three and some extant witnesses become intermedi-ate This is a type of tree presented in Appendix 2 (below) which is the finalstemma for 74 Old Church Slavonic Gospels manuscripts Appendix 3 (below)provides the final matrix for the same witnesses

The keypoint in the possibility to compare these two methods lies in ourview on the presentation of results Remembering that the stemma of E Watteldoesnrsquot show such relations as in Figure 1 we must refrain from drawing it for acontaminated tradition Nevertheless if we regard the stemma from a differentposition and consider it only as a representation of several clusters then wecan compare it with the matrix of clusters on equal grounds In doing so weshould look for correspondence between clusters in the matrix and subtrees inthe stemma

We cannot interpret the edges connecting manuscripts in the stemma be-cause they are meaningless for a text type but we can interpret every group ofnodes which were joined into the tree at a certain stage To do that we need tointroduce one definition We shall call a subtree of a hypothetical node n (or asubtree n) a graph with the node n and all the edges incident to n with all theirincident nodes which have a number smaller than n (we shall consider any endnode or other node corresponding to an extant witness smaller than any othernode) Let us look at the final stemma in Appendix 2 The number of a hypo-thetical node equals the number of the step at which next manuscript joinedthe tree or two or more subtrees were joined together Therefore the lower thevalue of n the smaller the distance between the end nodes of n Subtree 37 isthe stemma itself It is made up of subtrees 36 34 29 and 23 Each of them isin its turn made up by two or more subtrees We expect to find such subtreeswhere all or most end nodes are manuscripts belonging to one recension Sucha subtree can be compared to clusters in the matrix

Thus we can see that the given stemmatological method itself doesnrsquot con-tradict the principles of the study of the Old Church Slavonic Gospels and canbe applied to this tradition The only real restriction to its employment is thatnot more than 250 manuscripts can be processed We have to remember thatthe weight depends on the number of manuscripts sharing one variant andthe number of manuscripts in the version formula Therefore if we process 250manuscripts instead of eg 500 we reduce the reliability of weighting algorithm


The main difference between the methods is in the possibility to trace mistakesShould a mistake propagate into the final classification it cannot be seen in the

Dina Mironova

stemma because the stemma doesnrsquot visualize information about the edge dis-tances and it could be seen in the matrix because the matrix supplies similarityscores for all pairs The latter presentation is verifiable Meanwhile if a mistakeis made in the stemma we do not know about it unless another classificationgives us a contradictory data This is the drawback of the presentation of themethod of Wattel

Results

Let us consider the classification of 74 Old Church Slavonic New Testamentwitnesses (the list of witnesses is given in Appendix 1) This sample in-cludes representatives of each recension found during the processing of 531manuscripts The Old Text Type includes 17 oldest and most famous sourcesThe Preslav Text Type includes 21 older representatives which are the core ofthe multiple group forming this recension The New Liturgical Tetraevangelionis represented by four most characteristic witnesses The Athonite Text Type in-cludes the Ostrog Bible of 1581 (OB) the most authoritative printed editionwhich strongly influenced manuscript production and 27 representatives ofthe core including several manuscripts which form subgroups A few almostidentical manuscripts were taken to show how uniform the recension is TheChudov New Testament Text Type comprises the Chudov New Testament itselfand two more sources In the end there is one peripheral manuscript (Jv)which demonstrates a mixture of text types

The terms matrix and stemma are used for the presentation of the resultsof the procedure of Alekseev (Appendix 3) and Wattel (Appendix 2) corre-spondingly In Appendix 2 intermideate nodes 1ndash37 are virtual The increasein number of these nodes corresponds to the decrease in the strength of tiesbetween the manuscripts which it connects

We shall consider subtrees 23 29 34 36 which are joined into the finaltree in node 37 We start from node 23 which is the smallest number Sub-tree 23 is formed by subtrees 22 19 and 6 All the endnodes of subtrees 19and 6 are Athonite manuscripts Outside these two subtrees no more Athonitemanuscripts are found They number 28 in all Endnode A is the cluster 21286 OB 60 285 296 46 64 323 Q These manuscripts are alike or almostalike They form the first cluster in the table as well (numbers 1ndash28) It shouldbe mentioned that the Athonite Text Type isnrsquot characterised by vivid distinctivetextual features like the Preslav Text Type or even the Old Text Type (see below)In the present material it was possible to point out eight variants typical of


this text type Examples of a few variants which characterize one or more texttypes are given in Figure 4 Peculiarities of the Athonite Text Type can be seenin lines 1415a 1415b 1422a 1424b 1428a 1433a 1434 in Figure 4 Oftenthe distinctive features of the Athonite Text Type overlap with those of otherrecensions mostly the Old one As we see here and shall observe again furtheron talking about other text types in the majority of variation units more thanone text types agree in a variant Again we understand that only a combinationof variants can describe a text type (cf Section 3 paragraph 1)

The Athonite Text Type has a number of subgroups Let us see if they arethe same both in the table and in the tree Both pictures have a group of 311HH II 54 In the stemma one can see better that manuscripts 311 and HHstay more closely together This group like other Athonite subgroups has butfew characteristic readings Specific of such subgroups is that all their membersgive the same reading in the majority of variation units

Cluster 1z Ht 42 1p (let us call it 1z) is also present in both picturesCharacteristic readings of this group for the most part coincide with the read-ings of the Athonite Text Type and in certain cases those of the Old Text Typeincluding Os In the table manuscript 185 joins this cluster too In the stemmait is positioned together with manuscripts 69 352 85 355 (we shall call them69) and a subgroup a 294 113 (we shall call it a) Judging from the figures inthe table manuscript 185 has a better position there than in the stemma Thisneeds to be verified in the text As a matter of fact this manuscript is lacunaryOne pericope of the two considered was ommited Therefore its affiliation withany subgroup is only approximate The text shows that only in one variationunit (1414ndash2) does manuscript 185 have a reading different from that of 1zthis variation being an omission of the conjunction lsquoandrsquo which often occursspontaneously In six cases 185 has a reading different from 69 and in fivecases a reading different from a In all other variation units found in the pas-sage all these manuscripts have the same readings These data are in favour ofpositioning 185 closer to 1z

Now we pass on to group 69 As one can see in the stemma manuscripts69 85 352 355 do not form a separate group They are clustered together with185 and subtree 3 (subgroup a) We have already discussed manuscript 185Now we shall try to see whether the text will confirm stronger links within 69compared to the links of 69 with other Athonite subgroups first of all a In 12variation units manuscripts 69 85 355 352 are opposed to a (in one of thesecases we have a dichotomy 69 a vs 85 355 352) Only two cases position aand 69 on the same side of a dichotomy Accordingly the output in the tableis more trustworthy

Dina Mironova

i Old Slavonic and both mean time To distinguish the two synonyms in English we de-noted them as time-1 and time-2ii We give the second variant when an important witness has a variant different from the other represen-tatives of the text type it belongs to The sigla of the witnesses are given in brackets In cases where thereis no agreement about a variant among the witnesses of the given text type both variants are presentedwithout specification of the sourcesiii spectre-1 spectre-2 spectre-3 are three synonyms for spectre in Old Slavoniciv Old Slavonic grammar allows both prepositional and non-preposotional variants here without changeof meaning

Figure 4 Examples of variations characterising different text types of the Old ChurchSlavonic New Testament Mt 1414ndash1434


It is typical of all the Athonite subgroups that most of the time they shareeither the reading of the Athonite Text Type or the reading of the Old TextType and rarely give a reading of their own There is not a single occurrence ofa reading in common for one of these subgroups and any other text type exceptfor the Athonite and the Old ones

Coming to subtree 22 we see that its non-hypothetical nodes are themanuscripts from the Chudov New Testament Text Type Note that in the tablemanuscripts Nkn and Pg have the similarity score of 96 and their simi-larity to Cd is 91 and 89 respectively This goes along with the stemmawhere Nkn even happens to be antecedent to Pg (for which there is no obviousphilological confirmation) This type of text shares a number of characteristicfeatures of other recensions (cf eg lines 1415a 1415b 1433 in Figure 4)It also adds certain characteristic variants of its own mostly lexical substitu-tions The studied piece of text gave 10 such variants (eg lines 1417a 1417b1423a 1424 in Figure 4) It is interesting to compare the different position-ing of this cluster in the table and in the stemma In the stemma it is joineddirectly to the Athonite Text Type This is quite logical because these two text-types are very close as was mentioned in Section 1 above In the matrix thecluster is positioned at the bottom away from the rest This can also be justi-fied because we are dealing with quite a special recension a new translationof the New Testament At the same time in the matrix we can see that its per-centage of similarity with the Athonite Text Type is slightly higher than withthe other ones Thereby we can see that the two different representations of AAlekseev and E Wattel supplement each other and show the relations betweenrecensions at different angles

All the endnodes of subtree 29 which we are considering next aremanuscripts of the Preslav Text Type There isnrsquot any Preslav manuscript whichis not an end node of this subtree In the table we also have only one clus-ter containing all the Preslav manuscripts of the selection (numbers 45ndash65)The Preslav Text Type has ten characteristic variants in this passage most ofwhich are lexical substitutions and the majority of them are typical of this re-cension only and not of any other (cf eg lines 1415a 1415b 1424 14261433 1434 in Figure 4) The readings do not speak in favour of pointing outsubgroups among the representatives of this recension chosen for the sampleTherefore although subtree 29 is made up by two subtrees and these in turnconsist of more subtrees we shall regard them all as one group We do notknow whether the stemma gives a correct division into small groups becausethe variations do not give information about them Therefore we cannot relyon the stemma output in the further grouping of these manuscripts

Dina Mironova

Let us now go on to subtree 34 in the stemma Its end nodes are the rep-resentatives of the Old Text Type both its core (manuscripts which were joinedtogether at an early stage of the tree construction) and periphery (manuscriptswhich were joined together at a later stage of the tree construction) By thenumbers of hypothetical nodes we can see that the manuscripts from the corein subtree 7 (Mr Gl Tp Zg Os) are rather close and the manuscripts fromthe periphery in subtrees 25 and 30 (192 Ar 122 As) are both far from thecore (subtree 7) and from each other (25 is a rather high number for a virtualnode) which doesnrsquot prevent them all from being part of the same subtree As-semanus (As) for example belongs to the core of the Old Text Type but has somany individual features that computer classification often positions it at theperiphery Among the characteristic readings of this recension we can point tolines 1415a 1415b 1422 1434 in Figure 4 Subtree 21 in the stemma also be-longs to subtree 34 but four out of its six terminal nodes (1r WW 385 387) inreality represent a different recension the New Liturgical Tetraevangelion andwe shall deal with them later

In the matrix (Appendix 3) the Old Text Type forms two clusters one inthe middle (numbers 33ndash44) and a smaller one at the bottom (numbers 66ndash69) They have more representatives including also those which stay behindsubtrees 7 25 and 30 in the stemma (Appendix 2 see subtree 36)The firstcluster is bigger than the core of the Old Text Type in the stemma it also in-cludes manuscripts As 372 122 48 Mv 225 (see nodes 32 and 8) The secondcluster includes only 4 manuscripts (Kh Ar 192 Vr) and there are no morerepresentatives of the Old Text Type outside these two clusters This is impor-tant for the comparison of the two classifications because in the stemma fourmanuscripts which belong to the periphery of the Old Text Type are not evento be found in subtree 34 They are endnodes of subtree 36 (48 Nkl Kh Vr)The reason the two classifications differ lies mainly in the peculiarities of theOld Text Type Many witnesses are scattered in the stemma and their positionis a good illustration of A Alekseevrsquos statement that the features characteristicof the Old Text Type are more difficult to describe than those of its descen-dants This phenomenon is explained by the fact that the peculiarities of thearchetypical text are more or less evenly distributed in all manuscripts includ-ing later text recensions whereas newly acquired features belong only to a partof the tradition (Evangelije 19989) Such a special text type immediately re-veals the differences between the methods implicit in the difference betweenthe similarity score and the distance function We have already stated that theperipheral manuscripts often have a mixed text of more than one recension


therefore they are not significant for the purpose of describing a recension andfinding its best representatives

When the two classifications position peripheral manuscripts differently itdoes not significantly influence the overall picture It is a different matter whena core manuscript is mispositioned Unfortunately this mistake occurs once inthe stemma Manuscript Mv is found at the periphery in subtree 36 whereasits text proves that it is one of the core manuscripts of the Old Text Type and itshould be positioned together with the core witnesses

The New Liturgical Tetraevangelion (manuscripts 1r WW 285 387) isbased on the continuous form of the Old Text Type (see Section 1) thereforeit is not surprising that these two recensions are so closely linked The NewLiturgical Tetraevangelion does have its typical readings (cf eg lines 1415b1422 1428 in Figure 4) but so few that in the stemma it is mixed with thewitnesses of the Old Text Type In the table the New Liturgical Tetraevangelionforms a separate cluster (numbers 29ndash32) although it also comes very close tothat of the Old Text

Manuscript Sav has very many specific readings of its own and is neverpositioned in any group although it belongs to the Old Text Type

Conclusion

We have shown that a stemmatological method can be applied to the studyof a contaminated tradition if we regard the stemma from a specific angle Itis a useful consideration from the methodological viewpoint for sometimeswhen a theory is not applicable to the material the methods employed by thetheory can still be used Since the tools for studying old manuscript traditionsare limited we shouldnot neglect this possibility The main argument in favourof using all available tools for those who start a classification is that identicalresults mean that we are on the right path while divergences will make theresearcher check the data carefully (as was done with manuscript 185 above)

The present research is the first attempt to apply a new method to theSlavonic Gospels On the whole our conclusion is that the output of both meth-ods is relatively the same This happens because the input is the same andbecause both methods use cluster analysis as a classification technique

Nevertheless stemmatological orientation has a certain impact on the clas-sification causing differences in the output After studying the differences wefound the following

Dina Mironova

In some cases the text proves that the clustering was correct in the ma-trix and incorrect the stemma This can be illustrated by the mispositioning ofMv in the stemma There is no other evidence for this mispositioning More-over it is not clear why it was mispositioned But since the cluster analysispositioned it correctly we must conclude that the variation material providedenough information

In other cases the clustering in the matrix is more general compared to theclustering in the tree but the variants do not provide any evidence in supportof the clustering of the tree For example in the stemma the Preslav Text Typeis subdivided into subgroups which cannot be verified When we examine thetext in the witnesses of the suggested subgroups more closely we do not findvariants which oppose this subgroup to the rest

The explanation for this situation lies probably in the introduction ofweight of a version formula Firstly the computational weight depends on thetype of the formula (type 1 type 2 etc)8 Since we cannot regard the stemmaas a stemma but deal only with the groups it provides type 3 version formulashave the same value as type 2 and type 1 do not have to be considered at allThat is why the originally ascribed computational weight can entail mistakes inthe end Secondly pair weight inside a version formula can lead to inaccuracieswhen not all the manuscripts are classified because pair weight depends on thenumber of manuscripts sharing the same variant These observations entail thefollowing conclusions

a the more precise method of E Wattel can lead to a bigger divergence fromreality in the case of a contaminated tradition (cf the mispositioning ofMv As 185 in the stemma)

b it is not always possible to give a philological explanation for a more preciseclassification We should remember that if there is no information in thetext this information cannot be obtained by calculations either

The fact that the results provided by both methods are almost identical allowsus to choose between the two methods in the future If a tradition under con-sideration contains less than 250 witnesses it is desirable to use both methodsto study it If the tradition is bigger and is contaminated then the method of AAlekseev could be sufficient If there is a possibility to check the result using themethod of E Wattel this would be a valuable addition Without the stemmawe could have overlooked such cases as the closeness of manuscript 225 to theNew Liturgical Tetraevangelion The stemma was also a good visualisation ofthe relationship of the Chudov New Testament and the Athonite Text Types But


in some cases it is important to confine oneself to rough results because roughresults are less likely to contradict the text

Acknowledgements

I would like to thank the Organizing Committee of the Stemmatology work-shops for the financial assistance to attend the workshops and Dr Wattel forhis consultations on stemmatology and hard work on the Slavonic material

Notes

The Old Text Type in Old Church Slavonic is represented by both Four Gospels (continu-ous text) and Lectionaries

For a detailed textual history of the Slavonic Bible and full bibliography see Alekseev(1999) Garzaniti (2001)

We refer to truly genealogically significant readings as the readings behind which therecan be seen ldquogenetic divergence that is the divergence of readings not merely in the order oftheir likeness but in the sense in which they have actually arisenrdquo (Greg 192731ndash32)

Under the same text in a group of manuscripts here and further on we understand onerecension or text type

We use the terminology of E Colwell where a variation unit means ldquothe length of thetext wherein our manuscripts present at least two variant formsrdquo And ldquoa variant (or variantreading) is one of the possible alternative readings which are found in a variation unitrdquo(Colwell 196997ndash100)

This was the case with two well-preserved 13th century Four Gospels with old text of theGospel according to John (Evangelije 19988)

The representatives preferred are mss with the least lacunae known in philology withthe least number of mechanical mistakes with more correct grammar with older text

For the types of variants see Greg (192718ndash21)

References

Alekseev A (1986) ldquoOpit tekstologicheskogo analiza slavianskogo Evangelia (Po spiskambibliotek Bolgarii) (Textual analysis of Old Slavonic Gospels (the study of manuscriptsfrom Bulgarian libraries))rdquo Starobulgarica X 3 8ndash19

Alekseev A (1999) Tekstologija slavianskoj Biblii (Textual history of the Slavonic Bible) KoumllnWeirmar Wien

Dina Mironova

Alekseev A amp E Kuznetsova (1987) ldquoEVM i problemi tekstologii drevneslavianskihtekstov (Computer and the problems of textual criticism of the Old Slavonic texts)rdquoLingvisticheskije zadachi i obrabotka dannikh na EVM 111ndash120 Moscow

Colwell E (1969) Studies in Methodology in Textual Criticism of the New Testament LeidenDees A (1975) ldquoSur une constellation de quatre manuscritsrdquo Meacutelanges de linguistique et de

litteacuterature offertes agrave Lein Geschiere 1ndash9 AmsterdamEvangelije ot Ioanna v slavianskoj traditsii (The Gospel according to John in the Slavonic

tradition) (1998) St PetersburgGalloway P (1979) ldquoManuscript Filiation and Cluster Analysis the Lai de lrsquoOmbre Caserdquo

La pratique des ordinateurs dans la critique des textes 87ndash96 ParisGarzaniti M (2001) Die altslawische Version der Evangelien Koumlln Weirmar WienLangbroek E A Roeleveld amp E Wattel (2002) ldquoValentin and Namelos discover their

parentage Narrative elements in the family tree of an international medieval talerdquo VUAmsterdam Also this volume (Roeleveld et al)

Reenen P van amp M J P van Mulken (Eds) (1996) Studies in stemmatology AmsterdamJohn Benjamins

Wattel E amp M J P van Mulken (1996) ldquoWeighted Formal Support of a Pedigreerdquo In VanReenen amp Van Mulken (Eds) 135ndash167

Wattel E (2002) ldquoConstructing Initial Binary Trees in Stemmatologyrdquo VU AmsterdamThis volume

Appendix 1

The witnesses

I The Old Text Type

Sigla Signature Type of Book Date

the core

Zg St-Peterburg RNB Glag 1 (Zographensis) Tetraevangelion C11Mr Moskva RGB Grig 6 (Marianus) Tetraevangelion C11Tp Moskva RGADA 3811 (Typographensis) Tetraevangelion earlyC12Gl Moskva GIM Sin 404 (Galicianus) Tetraevangelion 1144Os St-Petersburg RNB FpI5 (Ostromir Evangelion) Concise Lectionary 105657As Vatican Vat Cod Slav (Assemanus) Concise Lectionary C11Mv Belgrad Narodnij Muzej 1538 and

St-Peterburg RNB FpI83 (Miroslav Evangelion) Expanded Lectionary 1185Ar Moskva RGB M1666 (Archangel Evangelion) Concise Lectionary 1092

other manuscripts

Sa Moskva RGADA 38114 (Sava Codex) Concise Lectionary C11Kh Odessa OGNB 13 (Kokhno Gospel) Concise Lectionary C12122 St-Peterburg RNB Pog 11 Concise Lectionary C12-13Nkl Dublin Chester Beatty 23 (Nikolskoje Gospel) Tetraevangelion C14


372 Moskva GIM Uvar480 Tetraevangelion C14225 Moskva GIM Uvar93 Tetraevangelion C1548 St-Peterburg RNB QI924 Tetraevangelion C15192 Moskva GIM Hlud 16 Tetraevangelion C14Vr Sofia NBKM N19 (Vrachansko Gospel) Concise Lectionary C13

manuscript with features of different text-types

Jv Moskva RGADA 3812 (Javilovo Gospel) Tetraevangelion c1381

II The Preslav Text Type

Ju Moskva GIM Sin 1003 (Jury Evangelion) Expanded Lectionary c 1128Ms Moskva GIM Sin 1203 (Mstislav Evangelion) Expanded Lectionary c 1117Dl Moskva RGB Rum 103 (Dobrilovo Evangelion) Expanded Lectionary 1164B Moskva RGADA 3818 Expanded Lectionary 1363IA London the British Museum Add 39627

(Evangelije Tsara Ivana-Alexandra) Tetraevangelion c1356293 Moskva RGB 304 III1 Expanded Lectionary C14376 Moskva RGB Sof 3 Expanded Lectionary 1362366 Moskva GIM Sin 740 Expanded Lectionary 130743 St-Peterburg RNB FpI9 Expanded Lectionary C14118 St-Peterburg RNB Pog 18 Expanded Lectionary 1463K St-Peterburg RNB Sof2 Expanded Lectionary C14418 Moskva RGB 256106 Expanded Lectionary C1375 St-Peterburg RNB Sof5 Expanded Lectionary C14257 Moskva GIM Hlud 170d Expanded Lectionary C15Pv St-Peterburg BAN 34720 (Pivoavrov

Evangelion) Expanded Lectionary C15d Moskva NBMGU 2Bg45 Expanded Lectionary C14273 Moskva GIM Sin 71 Expanded Lectionary 1409Tr Moskva GTGK 5348 Expanded Lectionary C12-13032 St-Peterburg RNB FpI64 Expanded Lectionary late C1402 Jaroslavl JaIAMS 15690 (Spasskoje

Evangelion) Expanded Lectionary mid C13383 St-Peterburg RNB FpI15 Expanded Lectionary C14

III The Athonite Text Type

Ob A phototypic edition in 1988 (Ostrog Bible) Tetraevangelion 158121 Moskva RGB 1731 Tetraevangelion C15286 Moskva RGB 304 I46 Tetraevangelion 150060 St-Peterburg RNB Sof27 Tetraevangelion C16285 Moskva RGB 304 I66 Tetraevangelion C15296 Moskva RGB 299538 Tetraevangelion C1546 St-Peterburg RNB QI1198 Tetraevangelion C15

Dina Mironova

64 St-Peterburg RNB Sof21 Tetraevangelion C16323 St-Peterburg BAN Celepi 40 Tetraevangelion C16Q Moskva RGB 11317 Tetraevangelion C15HH St-Peterburg RNB QpI4 Tetraevangelion C15311 Moskva RGB 256119 Tetraevangelion C15II St-Peterburg RNB QpI2 Tetraevangelion C1454 Moskva RGADA 2018 Tetraevangelion C1542 St-Peterburg RNB FpI12 Tetraevangelion C14Ht Moskva RGB 304 III3 (Evangelije Hitrovo) Tetraevangelion C14-151p Jaroslavl JaIAMS 15569 Expanded Lectionary C141z St-Peterburg RNB FpI109 Tetraevangelion C14185 Moskva GIM Tchert 81 Expanded Lectionary C14352 St-Peterburg BAN 32137 Expanded Lectionary C16355 St-Peterburg BAN 321327 Expanded Lectionary C1669 Moskva RGB Sof 14 Expanded Lectionary C1685 St-Peterburg RNB FI27 Expanded Lectionary C16a Moskva NBMGU Vetk 339 Tetraevangelion C15294 Moskva RGB 299706 Tetraevangelion C15-16113 St-Peterburg RNB Pog 23 Tetraevangelion C15-16209 Moskva GIM Uv 745 Expanded Lectionary C16210 Moskva GIM Uv 704 Expanded Lectionary C16

IV The New Liturgical Tetraevangelion

WW Athos RMPA Slav 2 Tetraevangelion C13-14385 Moskva RGB 178891 Tetraevangelion C141r St-Peterburg RNB QpI44 Tetraevangelion C14387 St-Peterburg RNB Gilf 18 Tetraevangelion C14

V The Chudov New Testament Text Type

Cd a phototypic edition in 1892 Tetraevangelion C14(the Chudov New Testament)

Pg St-Peterburg RNB Pog 21 Tetraevangelion C14Nkn Moskva RGB 304 III6 (Nikonov Evangelion) Tetraevangelion 1399


Appendix 2

Stemma for 74 witnesses of the Old Church Slavonic New Testamentusing the method of E Wattel

Tr 376 d 366

D1 383 032 02257 12 273 Pv 118 43 75 418 B1t 293 Ju Ms K

48 Nk1 Kh 27 20 14 11 10 5 64 A 311 HH

31 Mv 28 15 2 2 4 54II 1z Ht 42 1p

32 29 17 9 1

36 37 19

35 34 23

Vr mm 33 24 22 6

30 25 21 7 Cd Nkm 3 185 69 352 85 355

192 Ar 122 As 13 8 372 Mr G1 Tp Zg Os Pg a 294 113

lr WW 225 387 385

Dina Mironova

Appendix 3

Classification of 74 witnesses of the Old Church Slavonic New Testamentusing the method of A Alexeev


Different kinds of tradition in TargumJonathan to Isaiah

Alberdina HoutmanTheologische Universiteit Kampen

Prolegomena

Targum Jonathan an ancient Aramaic translation of the biblical books of theProphets has since the late 1980s been the subject of a large-scale researchproject at the Theological University of Kampen (the Netherlands)1 Withinthis project as a matter of course the question of the textual history of the textis also considered2 This history is quite complicated Internal and external evi-dence strongly suggests a long formative period before the text took its presentshape and the text as we have it now contains at least two different kinds oftraditions The main text consists of a strictly edited explanatory translationwhich stays relatively close to the Hebrew source text but at some points it isinterspersed with small textual units of a quite different character In this pa-per I will examine whether both kinds of traditional material can be equallyused for the construction of a stemma First I will give a short general intro-duction to the Targum genre followed by some words on Targum Isaiah Thenthe question of method will be dealt with The piegravece de reacutesistance is the case ofthe so-called Tosefta Targums It will be considered whether or not they can beused as kinship revealing variants

Introduction3

In 586 before Common Era the Babylonian king Nebuchadnezzar destroyedJerusalem In the event several thousands of the inhabitants of the city-state ofJudea were carried off to Babylon This bitter experience had a great effect on

Alberdina Houtman

the development of the Jewish religion Without their Temple far away fromfamily and compatriots the Jews in exile had to find a new way of religiousexpression While in spiritual matters they kept strictly to their own culturalheritage in daily life they soon took over the vernacular of the Babylonian em-pire which was Aramaic Eventually even the Jews who remained in Palestineswitched to the new lingua franca As a result by the 4th century BCE Aramaichad ousted Hebrew as the daily language of many Jews

Since religious legislation prescribed that the Bible could only be read inthe Holy tongue which is Hebrew a habit developed of translating the weeklyScripture lesson into Aramaic after its liturgical reading in Hebrew In theschools Aramaic Bible translations were used to teach the children Hebrewand at home the translations were used for private preparation for the lesson inthe synagogue These Aramaic Bible translations are called Targums In Pales-tine they were originally not standardized witness the occurrence of severaldifferent translations of the Torah At a certain point in time however one ofthese Palestinian Targums found its way to Babylon and was there edited andadapted to the local needs and promoted as the one official Targum In the 5thto 6th century this revised Targum probably found its way back to Palestineand was supplemented with old local traditions These additions are generallycalled Tosefta Targums4 Tosefta being the Aramaic word for ldquoadditionrdquo In factit must be said that about this last stage the opinions differ5

The liturgical practice of Targum passed out of use in the Middle Ageswhen as a result of the Islamic conquests the vernacular of the oriental Jewsbecame Arabic Only in Yemen the tradition survives to this day

From the Targum on the Torah there remain some Palestinian versions be-side the official Babylonian version From the Targum on the Prophets howeveronly the standardized Babylonian version which is called Targum Jonathan ispreserved The mentioned additions to the official Targum may or may not bethe remnants of an older Palestinian tradition

Targum Isaiah

There is hardly a biblical book that has been more popular through the agesthen the book of Isaiah The rabbinical literature is replete with references toIsaiah and from the oldest times the liturgical readings of the Prophets theso-called haftarot which accompanied the Torah reading in the synagogueswere to a large extent taken from Isaiah6 It is therefore no wonder that acomparatively large number of manuscripts and fragments of manuscripts has

Different kinds of tradition in Targum Jonathan to Isaiah

survived of both continuous Targums of Isaiah and of haftarah collections ofvarying provenance and date

From the literature especially old catalogues we now know of 26 (almost)complete continuous manuscripts of Targum Isaiah as well as several hundredsof fragmentary manuscripts and collections of haftarot Of the 26 known con-tinuous manuscripts 22 are such that they can be used for stemmatologicalresearch7

Generally Targum text witnesses can be classified into an Eastern and aWestern tradition The Eastern tradition is quite homogeneous It consists ofthe oldest Babylonian tradition ndash of which unfortunately only very little re-mains ndash and the Yemenite tradition On the level of consonant readings thesetwo strands are very close The Western tradition is more heterogeneous butbroadly speaking it can be subdivided into an Ashkenazi8 tradition and a Se-fardi9 tradition

Method

The task at hand is to classify the extant text witnesses into a pedigree to en-able the study of the historical development something which is easier saidthan done with a large book that has survived in so many textual witnesses Wetherefore have to look for a way to construct a reliable stemma without havingto scrutinize all the textual material first A possible way to do this is to carryout a sample survey10 The sample has to meet two conditions Firstly it has tobe representative of the whole manuscript Secondly the sample has to be un-biased If we are interested in discovering the source of a certain manuscriptit is better to keep away from theologically hot items because in those casesa copyist might be inclined to tamper with his sources In the same mannerbeginnings and endings of textual units have also proved to be the object oflater rewriting more often than the core of the same units For our researchthe liturgical readings the so-called haftarot are the main textual units11 Alarge percentage of the expansive variant traditions which are generally calledTosefta Targums are found in the introductions or the terminations of the haf-tarot12 so we have to take a verse somewhere from the middle of the haftarotFortunately the haftarot are fairly evenly distributed over the book of IsaiahOnly twelve out of the sixty-six chapters of the book do not contain haftarahreadings From those chapters a verse from the middle was taken The totalsample therefore included 66 verses one from each chapter

Alberdina Houtman

The sample verses being selected the next step was the selection of thetextual witnesses to be included in the research The continuous text tradi-tion looked like a feasible start using all the complete continuous manuscriptsmentioned above supplemented with some early printed editions

As all stemmatologists know very well some tedious groundwork has to bedone before one can start with any stemmatological construction namely thecollation of the witnesses For this part of the process I used the program Col-late which was designed and produced by Peter Robinson13 This sophisticatedprogram is a great help in the dull job of collecting variant readings althoughone still has to transcribe the manuscripts first The variant readings that wereregistered in this way were evaluated with regard to their relevance Using thepossibility of data abstraction most orthographic variants can be filtered outautomatically Nevertheless a number always succeed in slipping through thenet In those cases the formulas were adjusted manually Variant readings thatproved to be obvious errors were removed from the formulas

Not every variant is equally important The formulas were therefore givena weight factor expressing their significance for the establishment of a stemmaI counted characteristic orthographic variations such as different spellings ofloanwords for 05 minor syntactical changes for 1 substitutions additionsand omissions (when evidently not due to reading errors) for 2 The reason forweighing is obvious If the technique of the copyist consisted of reading a cer-tain part of the text and then copying it from memory it is very well possiblethat he changed the orthography unconsciously It is even conceivable that hechanged it consciously If he knew the word in a different spelling and thoughthis exemplar inaccurate he might be tempted to correct it On the other handthe deviating orthography may also have been part of the exemplar So if twoor more witnesses share the same deviating orthography it cannot be rejectedaltogether as evidence for relationship14 The same holds true to a lesser de-gree for minor syntactical changes Conversely substitutions additions andomissions are in considerable measure kinship revealing This kind of varia-tion must therefore weigh more heavily in the construction of a stemma thanthe other kinds However although the choice for weighing is clear the deter-mination of the weight factor is not It must be admitted that this factor is acontrived choice because it is impossible to determine in exact measure howmuch more important one kind of variant is than the other

For the construction of the tree I adopted the Three Step Method that wasdeveloped in the 70s by Antonij Dees (Dees 1975 1976 1977) According tothis method in the first step the witnesses are clustered into subfamilies onpurely quantitative grounds In the second step witnesses that might have been


intermediary in the process of transmission are identified15 These two stepsproduce the chain of relationships that underlies the genealogical tree At thisstage the nature of the relationships between the different members is settledbut not as yet their direction This must be determined at the third step wherethe point of suspension ie the root of the tree is established on the basis ofqualitative arguments This last step is the most difficult one and all possiblemeans must be employed to arrive at a well-founded decision such as assess-ment of the origins of the variants palaeographical and codicological data andhistorical information

In recent years the Three Step Method has profited greatly from the de-velopment of a computer program that takes care of the first two steps Thisprogram was designed and developed by Evert Wattel of the Department forMathematics and Information Science of the Free University of Amsterdam16

Thanks to this program one can now work with large amounts of data Theprogram has been tested extensively in several projects and has proved itsworth17 What remains for the philologist is the challenge of evaluating theoutput and interpreting it in the light of circumstantial data Wattel processedthe data concerned in different ways and produced the underlying structureof the stemma The outcome confirmed on the whole what was known fromprevious research on other biblical books ie a clear division between East-ern and Western textual witnesses and a subdivision of the Western witnessesinto an Ashkenazi and a Sefardi branch This grouping was used as the point ofcomparison for the present research

Tosefta Targums

Whereas in general the text of Targum Jonathan is remarkably stable the seem-ingly random occurrence of Tosefta Targums is an intriguing phenomenonAlthough Targum Jonathan is more of a paraphrase than a translation it stillkeeps close to the Hebrew original If one knows the theological premises ofthe translators and the hermeneutic rules they used to interpret the biblicaltext it is nearly always possible to reconstruct the process that led to the givenparaphrase At some places however the characteristic style is broken by di-gressions that are only tangentially related to the text These digressions occurmostly within the so-called Tosefta Targums The following example may givean impression of the nature of the different kinds of tradition (Diagram 1)

The version of Targum Jonathan still clearly reflects the Hebrew text albeitwith some changes Instead of the Hebrew rendering that says that it was in the

Alberdina Houtman

Diagram 1

Isa 61 Isa 61 Isa 61Hebrew Bible Targum Jonathan Tosefta Targum1

In the year that KingUzziah died

In the year that KingUzziah was strickenwith leprosy

In the year that King Uzziah died thatmeans the year in which he was strickenwith leprosy For there are four who intheir lives are considered as dead whileuntil now they hold out one who isstricken with leprosy one whose eyes areblinded and who has no sons and onewho went down from the strength of hisproperty The prophet

said the prophet Isaiah saidI beheld my Lord I beheld the glory of the

LordI beheld the glory of the Lord

seated on a high andlofty throne

seated on a throne highand elevated in thehighest heavens

seated on His throne high and elevatedin the highest heavens

and the skirts of Hisrobe filled the Temple

and the Temple wasfilled with the splendorof His glory

and the Temple was filled with thesplendor of His glory

1 According to MS Heacutebreu 75 Bibliothegraveque Nationale Paris

year that King Uzziah died Targum Jonathan contends that it was the year inwhich he was stricken with leprosy18 Secondly Targum Jonathan paraphrasespart of the text in order to keep off an inexpedient anthropomorphic under-standing of God Thirdly the impression that Isaiah actually saw the Lord hadto be avoided This was considered theologically incorrect because in Exod3320 it says ldquoman may not see Me and liverdquo In the Tosefta Targum an expla-nation is given for the translation of ldquodeadrdquo as ldquostricken with leprosyrdquo To thisend it cites a rabbinic tradition that is also known from other sources19 Thisspecific Tosefta Targum is inserted immediately after the translation of TargumJonathan without any label In another manuscript the expansive version isgiven as the first translation while the short version of Targum Jonathan isgiven afterwards indicated as ldquoanother translationrdquo20 This is one of the strangefeatures of the Tosefta Targums They are scattered throughout the manuscripttraditions where they may be inserted in the text at the appropriate place in-dicated as Tosefta Targum or the like or not indicated at all At other timesthey are written in the margin either by the first or a second hand or gatheredtogether at the end of the main text There also existed special collections ofTosefta Targums (Bernstein 1986151 Klein 1986410 Gleszligmer 1995165)


Since the Tosefta Targums are not found in the oldest strands of the tex-tual tradition they apparently consist of extraneous material that was added tothe common text of Targum Jonathan hence the name Tosefta which meansaddition In some manuscripts they are indicated as Targum Jerushalmi ieldquoTargum of Jerusalemrdquo which points to a Palestinian origin However both thecontents of these additions21 as well as their language22 often reflect Babylo-nian influence These facts apparently contradict each other

The occurrence of Toseftot is not restricted to a specific branch of the ge-nealogical tree They occur in all textual traditions except in the Babyloniantradition (Kasher 199660ndash62) There are however Toseftot that only occur incertain geographic areas (Kasher 199660)

Many of the Tosefta Targums occur within the haftarah readings especiallywithin the festive portions and may thus attest to the instruction heard by thecongregation in the synagogue (Kasher 199616ndash18) Moreover a significantpart of them occurs at the beginnings and endings of the haftarah readings23

Since beginnings and endings of textual units have been found to be the objectof later rewriting more often than the core of the same units this makes themunfit for stemmatological purposes (Den Hollander 1997138) On the otherhand being the most substantial variants in an otherwise quite stable text tra-dition it is very tempting to use them Willem Smelik in his book The Targumof Judges employed them for stemmatological purposes and they seemed toconfirm the results that were based on other criteria (Smelik 1995129ndash153)Therefore having a stemma in hand that was based on an unbiased sample ofthe witnesses (Houtman 1999a) I decided to test the value of Tosefta Targumsas kinship revealing variants To that end I re-checked the continuous textwitnesses for the occurrence of Tosefta Targums and put them into a diagram

Results

I will now first present the grouping of the witnesses on the basis of the unbi-ased sample to set a benchmark for comparison

The first group contains the Eastern tradition The second group containsbroadly the Sefardi tradition (I will return to this) The third group contains theAshkenazi tradition and the last group finally consists of manuscripts whichcannot be convincingly placed in any of the other categories

Alberdina Houtman

Diagram 2

Eastern ldquoSefardirdquo Ashkenazi Irregularf Babylonianfragmentsv Or 2211P Heacutebreu 1325

N Solgerb First Rabbinic Bibleg Second RabbinicBible

Q Heacutebreu 18K Codex ReuchlinianusJ El F6d Add 26879

p Heacutebreu 96B Or Fol 2

e Gaster 673z Or 1474L Lutzki 239

c H 116k Kaufmann A13

F Urbinas 1C BH VA Goumlttweig 11

Y Qafih 5 H Heacutebreu 75O Opp Add 40 76r Villa Amilo Antwerp Polyglot

D Parma 3188

Eastern tradition

The Eastern manuscripts encompassing the Babylonian fragments and theYemenite manuscripts make up a strong cluster which contains two recogniz-able sub-clusters the internal connections of which are considerably strongerthan the connections with the other members MS Qafih 5 is a very young(1900) Yemenite manuscript which stands slightly apart from the rest

Sefardi manuscripts and printed editions

The Western tradition is much more heterogeneous than the Eastern traditionThe group that I have called Sefardi for convenience contains one exceptionalsub-group This consists of MS Solger the First Rabbinic Bible and the SecondRabbinic Bible It definitely belongs to the Western text tradition but scrutinyof the place of this sub-group within the tradition does not disclose a clearkinship to either the Ashkenazi or the Sefardi tradition Within the group of theSefardi witnesses H116 and MS Kaufmann A13 can be discerned as a strongsubgroup The other subgroup is less coherent

Ashkenazi manuscripts

The third group consists of the Ashkenazi manuscripts These manuscripts areby far not as closely mutually related as the manuscripts within the other twogroups They are distinguished from the other groups more through their dif-ference than through a mutual likeness Nevertheless some connections canbe observed There seem to be two sub-groups of which the first consists of


Codex Reuchlinianus MS Heacutebreu 18 MS El f6 and MS Add 26879 The sec-ond sub-group has very loose internal links It consists of MS Urbinas 1 MSBH V MS Goumlttweig 11 and MS Parma 3188

Irregular manuscripts

Finally there are two manuscripts that do not fit into one of the groups de-fined namely MS Heacutebreu 96 and MS Or Fol 2 It is difficult to account for theplace of MS Heacutebreu 96 Although it was probably written in Spain in a Sefardiscript it seems textually closer to the Yemenite than to the Sefardi text tradi-tion MS Or Fol 2 has the outward appearance of an Ashkenazi manuscriptbut its consonantal text does not conform to any of the groups defined24

We will now compare these findings with the results of the collation ofthe Tosefta Targums Two things need to be mentioned in advance Firstly noBabylonian fragments available contain the verses concerned25 Therefore theywill not be found in the table Secondly I have not included Tosefta Targumsthat are known from the marginal readings of the Codex Reuchlinianus onlybecause these unique readings were taken from different sources that have notyet been identified26 Including this material would therefore not add to thetext genealogical information

So restricting ourselves to Tosefta Targums that are not unique to CodexReuchlinianus we find eight cases One of them a Tosefta Targum to Isa 1032occurs in two quite different recensions27 I have collated these separately sothat there are nine cases to be considered

See Diagram 3 What do these data tell us If we first consider the Yemenitetradition we see that it has no Tosefta Targums integrated in the text In lsquovrsquo lsquoersquolsquozrsquo and lsquoLrsquo there is a Tosefta Targum to 1032b but it is not part of the origi-nal text In lsquovrsquo lsquozrsquo and lsquoLrsquo it is added in the margin while in lsquoersquo it occurs as anannex on a separate page Apparently the Yemenite tradition like the Babylo-nian tradition did not include Tosefta Targums originally It is a known factthat the publication of the Rabbinic Bibles in the 16th century strongly influ-enced the Yemenite tradition We see accordingly that the added material ofthe Yemenite MSS coincides with the First Rabbinic Bible (lsquobrsquo)

The Sefardi tradition is characterized by the Tosefta Targums on Isa 614915 4924-25 and 5010-11 lsquocrsquo is exceptional since it has no Tosefta Targumsat all28 Except for lsquoHrsquo none of the Sefardi manuscripts has a Tosefta Targumto Isa 1032 Within the Sefardi tradition lsquorrsquo and lsquoorsquo have the same occurrenceof Tosefta Targums This is not surprising because lsquorrsquo was probably the ex-

Alberdina Houtman

Diagram 3

+ presentndash absentm present in marginsim text not available or illegible

emplar of lsquoorsquo which is the Antwerp Polyglot (Diacuteez Merino 199483ndash85 Stec19947ndash16)

The First and Second Rabbinic Bibles (lsquobrsquo and lsquogrsquo) and lsquoNrsquo which is the ex-emplar of the First Rabbinic Bible (Houtman 1999b191ndash202) share the sameversion of Isa 1032 As already indicated above on the basis of the unbiasedsample research the place of this small group within the Western tradition isproblematic The pattern of Tosefta Targums being dissimilar to the patternsof both the Sefardi and the Ashkenazi group confirms this finding

The Second Rabbinic Bible is sometimes said to be only a slightly revisededition of the First Rabbinic Bible We see however that it includes ToseftaTargums that were not present in the first edition This indicates that the editorof the Second Rabbinic Bible Jacob ben H ayyim who came as a refugee fromSpain to Italy incorporated the traditions of his home country

Five out of the eight Ashkenazi manuscripts share the same version of Isa1032 of which four also contain the Tosefta Targum to Isa 103329 Obviouslythese two form a pair30 It is remarkable that MS Or Fol 2 (lsquoBrsquo) which hasthe outward appearance of an Ashkenazi manuscript also has this typicallyAshkenazi combination of Isa 1032a combined with a Tosefta Targum to Isa1033 This evidence may push it more strongly towards the Ashkenazi grouplsquoFrsquo and lsquoCrsquo share the otherwise unique Tosefta Targums to Isa 661-2 and 6623Although except for Isa 661-2 in lsquoCrsquo they are not incorporated in the runningtext but are appended at the end of the text by the same hand as the basic textthey may be considered an integrated part of this tradition

lsquoprsquo is exceptional in its unique combination of the ldquoSefardirdquo Tosefta Tar-gum to Isa 61 and a marginal reading of the Tosefta Targum to Isa 1033 This


unusual combination confirms the results based on the unbiased sample thatthis is probably a contaminated tradition

So far for the data that confirm the earlier results There are however alsoelements that contradict the general tendencies

ndash The Ashkenazi manuscripts lsquoQrsquo lsquoKrsquo and lsquoArsquo are extraordinary because theirbasic text contains no Tosefta Targums at all The same holds for the Sefardimanuscript lsquocrsquo In an automated comparison these MSS would probablyhave been grouped with the Eastern tradition on the basis of the (original)absence of Tosefta Targums

ndash The ldquoAshkenazirdquo version of Isa 1032 occurs unexpectedly in the Sefardimanuscript lsquoHrsquo

ndash On the other hand the decidedly Ashkenazi manuscript lsquoDrsquo contains aTosefta Targum to Isa 61 which seemed to be a Sefardi characteristic

Conclusions

What can we conclude concerning the use of Tosefta Targums as relation-ship revealing variants Because of the mentioned contradictions it seems notadvisable to include the Tosefta Targums in the automatic procedure for theproduction of the underlying structure of a stemma On the other hand theyare too important to be neglected If used with care they can be of aid in thereconstruction of the textual history For example in the case of the SecondRabbinic Bible it seems that the editor who generally followed the text of theFirst Rabbinic Bible slavishly made an exception for the Tosefta Targums thatwere known to him from his home country In cases of complete conformityof the occurrence of Tosefta Targums a close relationship is probable like be-tween lsquoNrsquo and lsquobrsquo and lsquorrsquo and lsquoorsquo where it is a matter of direct dependence Inthe case of lsquoFrsquo and lsquoCrsquo there is no direct linear dependence but the relationshipbetween these manuscripts is certainly closer than between the other Ashke-nazi manuscripts which are very heterogeneous The strange combination ofTosefta Targums in lsquoprsquo confirms the results of the unbiased sample research thatthis must be a contaminated tradition

The name Tosefta means lsquoadditionrsquo Perhaps that is how we should usethem as additional evidence in the reconstruction of the textual historyWithin the Three Step Method it should therefore belong rather to the thirdstep than to the first two steps

Alberdina Houtman

Acknowledgments

The investigations were supported by the Foundation for Research in the Fieldof Humanities which is subsidized by the Netherlands Organization for Scien-tific Research

Notes

The project includes the production of a bilingual concordance as well as a seriesof monographs Bilingual Concordance (1995ndash) Smelik (1995) Van Staalduine-Sulman(2002) the present author works on the Targum of Isaiah

Houtman (1999a 1999b 2000) See also the publications of our former team memberW F Smelik Smelik (2002) and his contribution in the present volume

The following overview is largely based on Alexander (1998 esp 247ndash250)

The designation ldquoTosefta Targumrdquo is somewhat biased and therefore misleading Butsince this designation is commonly used in scholarly literature on the subject I decidedto adopt the term

See the useful overview of previous research into date and provenance of the ToseftaTargums in Smelik (199577ndash85)

Although the details of the history and the content of the lectionaries are still underdebate the broad outline is clear According to Wacholder (1966) 25 of the haftarot in theannual cycle were taken from Isaiah while in the triennial cycle almost half of the haftarotare from Isaiah (he counts 187 different haftarot including the festival readings of which 93are taken from Isaiah) According to Ch Perrot (Perrot 141ndash143) the book of Isaiah is readon 97 of the 158 regular Sabbaths of the triennial cycle

Two are lost one is too much damaged to be used and one is only available in the formof a carbon block

Roughly concerning the regions of Germany France and Eastern Europe

Roughly concerning the regions of Spain Portugal and North Africa

The problems of using a sample to represent whole manuscripts are outlined by B MMetzger (Metzger 1992181)

The division into chapters is a rather late Christian invention and therefore irrelevantfor our purposes Bishop Stephan Langton introduced the division at the beginning of the13th century in the Latin Bibles Gradually this division also entered the Jewish Bibles SeeGinsburg (189725ndash31)

See Kasher (198576) and Kasher (199617) A Shinan noticed the same phenomenonfor the weekly reading of the Torah the sedarim See Shinan (1987106ndash107) and Shinan(199226ndash31)

Robinson (1994)


For the evaluation of orthographic differences for stemmatological purposes see Smelik(2002)

If a MS has few unique readings if any it may have been intermediary between someother MSS

See eg Wattel (1996)

Eg Van Mulken (1993) Den Hollander (1997) De Visser-van Terwisga (1995ndash1999)see also the contribution of A Roeleveld E Langbroek and E Wattel in the present volume

The Aramaic does not use the word ldquoleprosyrdquo but the euphemistic expression ldquohe wasstricken with itrdquo This expression is generally used in connection with leprosy The employedverb is related to the Hebrew word that was used in II Chron 2620 where the story ofUzziahrsquos leprosy is related The story concludes there with the words ldquofor the Lord had struckhimrdquo The reason that the translator chose not to translate the Hebrew literally was thataccording to the rabbinic tradition this verse describes the beginning of Isaiahrsquos propheticactivities Since it is known that Isaiah prophesied during the reign of Uzziah he could notpossibly have had his first prophetic vision on the day Uzziah died This is an example ofharmonization of Scripture which is a regular principle in the Targums

Eg Babylonian Talmud Nedarim 64b

MS Heacutebreu 96 Bibliothegraveque Nationale Paris

Bacher (18744) Epstein (189550)

See Tal (1975191ndash200) Kasher (199614ndash16)

See above n 12

Here as elsewhere this situation may be restricted to only part of the manuscript For theBook of Judges for instance Smelik has shown a relationship of this manuscript with CodexSolger MS El f6 and MS Heacutebreu 18 See Smelik (1995142ndash147) Also Kasher classifies it asan Ashkenazi manuscript See Kasher (1996303)

So unfortunately we cannot check Kasherrsquos assertion that this tradition does not containTosefta Targums for the case of Isaiah

For the unique character of the glosses in the Codex Reuchlinianus see Smelik (1995170ndash175)

Actually there are three recensions of which one occurs in the Codex Reuchlinianus onlySee Kasher (1996151ndash155)

Ie for Targum Isaiah Some of the other books of the Prophets do contain ToseftaTargums See Luzzatto (1844132ndash137) Sperber (IVb139ndash140)

Were we to include MS Or Fol 2 on the basis of its outward appearance as an Ashkenazimanuscript the numbers would be 6 out of 9

Kasher counts them together as one TT

Alberdina Houtman

References

A Bilingual Concordance to the Targum of the Prophets (1995ndash ) Project Director JohannesC de Moor Chief Editors Willem F Smelik and A Houtman 20 vols to date LeidenBrill

Alexander P S (1988) ldquoJewish Aramaic Translations of Hebrew Scripturesrdquo In M J Mulderamp H Sysling (Eds) Mikra Text Translation Reading and Interpretation of the HebrewBible in Ancient Judaism and Early Christianity (pp 217ndash253) Assen Van GorcumPhiladelphia Fortress

Bacher W (1874) ldquoKritische Untersuchungen zum Prophetentargumrdquo Zeitschrift derdeutschen morgenlaumlndischen Gesellschaft 28 1ndash72

Bernstein M J (1986) ldquoA New Manuscript of Tosefta Targumrdquo In Proceedings of the NinthWorld Congress of Jewish Studies Jerusalem August 4ndash12 1985 4 vols I (pp 151ndash158)Jerusalem World Union of Jewish Studies

Dees A (1975) ldquoSur une constellation de quatre manuscritsrdquo In A Dees et al (Eds)Meacutelanges de linguistique et de litteacuterature offerts agrave Lein Geschiere par ses amis collegravegeset eacutelegraveves (pp 1ndash9) Amsterdam Rodopi

Dees A (1976) ldquoConsideacuterations theacuteoriques sur la tradition manuscrite du lai de lrsquoOmbrerdquoNeoplilologus 60 481ndash504

Dees A (1977) ldquoOver stambomen en handschriftenrdquo Forum der Letteren 18 63ndash78Diacuteez Merino L (1994) ldquoTargum Manuscripts and Critical Editionsrdquo In D R G Beattie

amp M J McNamara (Eds) The Aramaic Bible Targums in their Historical Context (pp51ndash91) Sheffield JSOT Press

Epstein A (1895) ldquoTosefta du Targoum Yerouschalmirdquo Revue des Etudes Juives 30 45ndash51Ginsburg Ch D (1897) Introduction to the Massoretico-Critical Edition of the Hebrew Bible

photostatic reprint of edn London 1897 New York KtavGleszligmer U (1995) Einleitung in die Targume zum Pentateuch Tuumlbingen MohrHollander A A den (1997) De Nederlandse bijbelvertalingen 1522ndash1545 Nieuwkoop De

GraafHoutman A (1999a) ldquoTextual Tradition of Targum Jonathan to Isaiahrdquo In J Targona Borraacutes

amp A Saacuteenz-Badillos (Eds) Jewish Studies at the Turn of the 20th Century 2 vols I (pp145ndash153) Leiden Brill

Houtman A (1999b) ldquoTargum Isaiah According to Felix Pratensisrdquo Journal for the AramaicBible 1(2) 191ndash202

Houtman A (2000) ldquoPlanning a New Targum Edition Look before You Leaprdquo Journal forthe Aramaic Bible 2(2) 213ndash231

Kasher R (1986) ldquoThe Aramaic Targumim and their Sitz im Lebenrdquo In Proceedings of theNinth World Congress of Jewish Studies Jerusalem August 4ndash12 1985 4 vols I (pp75ndash85) Jerusalem World Union of Jewish Studies

Kasher R (1996) Targumic Toseftot to the Prophets Jerusalem World Union of JewishStudies

Klein M L (1992) ldquoTargumic Toseftot from the Cairo Genizahrdquo In D Muntildeos Leoacuten (Ed)Salvacioacuten en la Palabra TargumndashDerashndashBerith En memoria del profesor Alejandro DiacuteezMacho (pp 409ndash418) Madrid Ediciones Cristiandad


Luzzatto S D (1844) ldquoNachtraumlgliches uumlber die Thargumimrdquo Wissenschaftliche Zeitschriftfuumlr juumldische Theologie 5 124ndash137

Metzger M (1992) The Text of the New Testament Its Transmission Corruption andRestoration (3rd ed) Oxford At the Clarendon Press

Mulken M van (1993) The Manuscript Tradition of Chreacutetien de Troyes A Stemmatologicaland Dialectological Approach (Doct Thesis Vrije Universiteit) Amsterdam

Perrot Ch ldquoThe Reading of the Bible in the Ancient Synagoguerdquo In Mulder amp Sysling (Eds)Mikra (pp 137ndash159)

Robinson P M W (1994) Collate Interactive Collation of Large Textual Traditions Version2 Oxford University Centre for Humanities Computing

Shinan A (1987) ldquoSermons Targums and the Reading from Scriptures in the AncientSynagoguerdquo In L I Levine (Ed) The Synagogue in Late Antiquity (pp 97ndash110)Philadelphia American Schools for Oriental Research

Shinan A (1992) The Embroidered Targum The Aggadah in Targum Pseudo-Jonathan of thePentateuch Jerusalem Magnes

Smelik W F (1995) The Targum of Judges Leiden BrillSmelik W F (2002) ldquoHow to Grow a Tree Computerised Stemmatology and Variant

Selection in Targum Studiesrdquo In J Cook (Ed) Bible and Computer ndash Proceedings ofthe 6th AIBI Congress ndash Stellenbosch 17ndash21 July 2000 (pp 495ndash518) Leiden Brill

Sperber A (1992) The Bible in Aramaic 4 vols IVb (pp 139ndash140) (2nd ed) Leiden BrillStaalduine-Sulman E van (2002) The Targum of Samuel Leiden BrillStec D M (1994) The Text of the Targum of Job Leiden BrillTal A (1975) The Language of the Targum of the Former Prophets and its Position within

Aramaic Dialects Tel Aviv Tel Aviv UniversityVisser-van Terwisga M de (1995ndash1999) Histoire ancienne jusqursquoagrave Ceacutesar 2 vols Orleacuteans

ParagdigmeWacholder Ben Zion (1966) In J Mann (Ed) The Bible as Read and Preached in the Old

Synagogue 2 vols I (p xxxii) Cincinatti Hebrew Union CollegeWattel E (1996) ldquoClustering in Stemmatological Trees How to Handle a Large Number

of Versionsrdquo In P van Reenen amp M van Mulken (Eds) Studies in Stemmatology (pp123ndash134) AmsterdamPhiladelphia John Benjamins

Valentin and Namelos discovertheir parentage

Narrative elements in the family treeof an international medieval tale

Annelies Roeleveld Erika Langbroek and Evert WattelAmsterdam Vrije Universiteit Amsterdam

Introduction

During our explorations of Middle Low German texts we came across a talewhich is traditonally called lsquoValentin and Namelosrsquo for short Apart from twoversions in Middle Low German this tale has come down to us in three otherlanguages in verse and prose and the more we looked into these different ver-sions the more the narrative and linguistic aspects and the family history ofthe tale started to intrigue us Stemmatological research into the widely dif-fering and not easily comparable renderings proved to involve feeling our wayand testing the ground with each successive step and required the applicationof unique and innovative methods

The tale

The story of Valentin and Namelos contains several narrative motives oftenfound in medieval verse epics The main characters are twin brothers whoare unaware of each otherrsquos existence because they were left as foundlingsValentin is found by a princess and grows up at court Namelos (lsquoNo-namersquo) isadopted by a she-wolf and grows up in the wilds When they meet in a fighta powerful bond instantly manifests itself After a great many adventures theydiscover that they are brothers and find their royal parents The fictional narra-


tive background is the courtly world of king Pippin of France king Crisosmusof Hungary and the battles against the (Muslim) heathens

The manuscripts

The tale has come down to us in four languagesdialects in verse and prose(we do not take into consideration the tale of ldquoValentin et Orsonrdquo of whichmany versions in French and English exist in early prints but which differconsiderably from our tale) We are dealing with

N I and N II ndash Middle Dutch 2 verse fragments of 176 lines each c 1350N III ndash Middle Dutch verse fragment of 395 lines of which 16 are barely leg-ible and 22 are completely illegible 1340ndash1360K ndash Middle Mid German verse fragment of 52 lines date unknownS ndash Middle Low German verse 2291 lines c 1450H ndash Middle Low German verse 2613 lines c 1476ndash1481B ndash Middle Mid German prose 1465Z ndash 3 Old Swedish mss prose 16th century

This would seem to be a sufficient number of texts for the drawing up of a pedi-gree and several scolars have tried to do so since the tale has been the subject ofstudy (Seelmann 1884xii f Beta 19078 Karg 1924229 Dieperink 1933157Langbroek amp Roeleveld 1987 ii) Matters are much more complicated than theyappear however

a S H B and Z have come down to us more or less complete but S and Hare in verse B and Z are prose This makes looking for variant readingsdifficult and speculative it is not always possible to distinguish betweenreadings produced by the constraints of rhyme and readings which varyfor other reasons

b Like S and H manuscripts N I II and III and K are in verse but onlyfragments of them survive and unfortunately K does not cover the sameground as the Middle Dutch fragments The situation is roughly as follows(see Diagram 1)1ndash25 are the episodes of the story as defined by Dieperink (19335ndash11)B and (apparently) N contain only 22 epsiodes italics indicate verseasymp H S and Z break off an incident in episode 3 and resume it in episode 5 H S and Z break off an incident in episode 4 and resume it in episode 10$H S and Z break off an incident in episode 9 and resume it in episode 16

Valentin and Namelos discover their parentage

Diagram 1 The episodes

c The Middle Dutch fragments render the narrative material in a very de-tailed and lenghty manner using 5 to 7 times as many lines of verse torelate the same episode as S or H which makes it impossible to do any-thing like matching line for line Prose recension B while less expansivethan the Middle Dutch fragments is still much more elaborate than S Hand Z and has unique elements and details

d Not the least complicating factor is that four languagesdialects are in-volved Middle Dutch Middle Mid German Middle Low German andOld Swedish

The methodological approach

In order to obtain a pedigree which reflects the historical connections betweenthe recensions we are using the stemmatological construction programmes de-scribed by Wattel and Van Mulken (1996a b) These programmes require adatabase consisting of formulas each of which describes a version differencebetween the recensions under consideration

A line of the database contains (1) a textual position (2) a version differ-ence and (3) an optional weight factor for the version difference Where noweight factor is given a weight of 10 is assumed While the database is beingconstructed it is also possible to use category indicators instead of weight fac-tors weight factors can then be assigned to the categories afterwards and ifnecessary be changed without having to reconstruct the whole data base

When the database is finished the programmes will supply a pedigree aset of shock waves and centrality scores for the manuscripts as described byWattel and Van Mulken (1996a b) The stemma is the most important resultfor most purposes but it must be borne in mind that a stemma is an abstrac-


tion an instrument for the understanding the historical process and not anend in itself

The unusual text tradition of Valentin and Namelos

Applying stemmatological methods to the Valentin and Namelos manuscriptsturned out to be very complicated So far different texts have always been com-pared with respect to variant spellings rhyme and metre number of verseschoice of words and grammatical structures It was impossible to do this forour unusual set of texts due to the substantial differences in language lengthand style

However in all the recensions the narrative does in fact develop along ap-proximately the same lines and identical story elements are found in all ofthem It occurred to us therefore that we should attempt to compare nottextual variants but variants in story elements and narrative development Itturned out that this procedure yielded material just as suitable for mathemati-cal processing as textual variants

Classifying the material

We restricted our variant apparatus to the following categories

category classification weight factor1 important semantic differences 302 minor semantic differences 203a inversions over several lines of text 223b inversions within one line or over adjacent lines 124 important grammatical structures1 105a important interpolations or omissions 215b minor interpolations or omissions 116 differences in personal or place names 15

Of great importance in computer aided stemma construction is the valua-tion of the version differences Each of the categories mentioned was given aweight factor which is listed in the last field of the table Important differenceswere given a weight factor 10 larger than the corresponding minor differencesThese valuation weights are usually a matter of personal experience and exper-tise and of the text tradition involved and small differences in weight will notinfluence the final stemma construction


If a version difference occurs uniformly through an entire section (as hap-pens with personal names) a version formula is included in the data base twoor three times eg at the beginning at the end and sometimes in the middleof that section

When the database is finished the type of version information will notinfluence the results the programme does not discover that it is not deal-ing with the usual information on spelling rhyme and metre grammar andchoice of words

The programmes of Wattel and Van Mulken (1996a b) are especially de-signed to be able to handle missing information During the evaluation of aversion formula (a line of the database) by the programmes some recensionsand pairs of recensions count positive (because they agree) some negative (be-cause they do not) but the sums of weights and influences is always 0 Wherean item is missing in one recension it will have no influence and the recen-sion in question will end up in the intermediate range in the evaluation of theversion formula for that item In the end the position of a recension in thepedigree will not be biased by its number of occurrences in the version formu-las Naturally the results will be more reliable for recensions which occur oftenin every section of the database for the text tradition

The most important type of information for every pedigree constructionare the so-called lsquotype 2rsquo version differences the abcd type These are versiondifferences in which there are at least two different readings with at least twomanuscripts for each reading This is also the case for the Valentin and Namelostexts although many of our version formulas do not contain type 2 informa-tion However during the computer processing of the different parts of the database the amount of information was sufficient

An illustration

For the comparisons we chose the episodes for which we have Middle Dutchfragments at our disposal so that we had five renderings to work on N S H Zand B Our methods and strategies in finding formulating and assessing vari-ants is exemplified in the list of variants we eventually drew up for the episodefor which we had fragment N I at our disposal

Note that in H and Z the episode is split up after an introductory reporton the situation it breaks off to be continued much later on in the tale In S thefolio which should contain a similar introduction is missing but the episode


must have been split up in the same way as in H and Z for later on in the taleonly the rest of the episode occurs

Lines h 649-58 + 1458-1535 s 1202-1276 z 4022-27 + 926-9612 b 8940-9027 n 1-176Number of lines h 10 + 78 s 75 z 6 + 91 b 34 n 176 lines is lsquoarchetypicalrsquo information (see below under 7)

H versions type description of difference651 hznb 2 G in love with Ph ditto wants to marry her651 nbhz 5b - - - G is a traitor651 hzbn 5a - - - G pulls Phrsquos chin she knocks out 3 of his teeth655 hzbn 1 G feels love Ph doesnrsquot notice Ph doesnrsquot want him655 bn 2 Ph doesnrsquot respond hopes to be united with husband1458 hszbn 1 episode resumed (long interval) episode unbroken1458 hsbnz 5b - - - G is a fool1460 hszbn 5a Ph stays loyal and constant - - -1460 hsz 5b - - - Ph refuses to become Grsquos lover1460 hszbn 5a - - - G is angry swears he will be revenged1460 hsznb 1 - - - ldquoPh acts as if she were a nunrdquo1460 hsznb 5a G says Ph was not banished home unjustly - - -1460 hszn 3a G says it much later G says it here1460 hszbn 5a - - - remark by Lica G predicts she will regret it1464 hsznb 2 G brooding on revenge G thinks of plan to kill Ph1464 hszbn 5a - - - contents of plan auctorial remark Lica in danger1467 hszb 2 early one morning one night1481 hszb 5b G goes to Ph G goes to Ph finds her asleep1481 hszbn 5b he is holding the knife - - -1481 hszbn 2 stabs princess to death accomplishes his design1481 hszbn 2 puts knife Phrsquos hand cunningly puts knife Phrsquos hand1490 hszbn 5b - - - G goes to bed awaits daybreak gets up1490 hszbn 1 G says he has been dreaming king been dreaming1490 hsznb 5b - - - G wants king and himself to go and investigate1497 hszbn 5b king is frightened - - -1497 hsnzb 5b - - - king gets up1497 hsbnz 5b - - - king gets dressed1500 hszbn 5b - - - Ph is holding the knife in her hand1500 hszbn 5b - - - Ph is still asleep1504 hsznb 3a G Ph is guilty evidence of knife here ditto later1508 hszbn 5a G wakes Ph brutally king cries out Ph wakes1508 hszbn 5b - - - Ph sees Lica is dead1512 hsznb 5b G asks how Ph could do such a thing - - -1512 hszn 3a G asks question here G asks question later


1512 hszbn 5b G says Ph should be burned at the stake - - -1525 hsznb 2 Ph had the knife Ph had the knife so she should die1525 hszbn 5a - - - G points out that wound is bleeding again1525 hszbn 5b G binds the knife into Phrsquos hands - - -1529 hszbn 5b Ph is dragged about - - -1529 hszb 3a Ph dragged about here Ph dragged about earlier1529 hszb 1 G drags Ph from room ditto by hair knights pity1529 hszbn 5a - - - passage of further accusals and denials1531 hszbn 5b Ph is very sad - - -1533 hznsb 5a Ph is brought before the law - - -1533 hzn 2 G brings Ph before court G brings Ph into ldquothe hallrdquo1533 hszbn 5b - - - the court sits1533 bn 2 king has case judged king explains about murder

what punishment649-58 en 1458-1535 (ie through the whole episode)

hszbn 6 king of Arabia no name king of Arabia is Saluberhszbn 6 his daughter no name his daughter is Licahsznb 6 Phila Philomena Philamenahszbn 6 Gawyn Glutes

We made similar lists of version differences for the episodes which are coveredby fragments N II and N III in the same way

The Middle Dutch fragments

It is significant that N I and N II were discovered in the same context were writ-ten in the same hand and were once parts of one single manuscript while N IIIwas discovered elsewhere and was not from the same manuscript At the timewhen fragment N III was found it was thought to be from a different textualtradition than the other two fragments (De Vreese 1892145ndash146) De Vreesecomes to his conclusion mainly on the grounds that the order of events is ap-parently different in this fragment Namelos goes off to look for his wife beforeValentinrsquos wedding while in N II (and B H S and Z) she comes to look for himright at the end of the tale As N III is only a fragment no conclusions can bedrawn from this apparent difference there is nothing to prove that Namelosrsquojourney is not cancelled or given up prematurely for some reason or anothernor is it impossible that a similar episode also occurred in the text of which weonly know the small fragments N I and N II The facts that king Crisostomushad married again and that horses and swords are mentioned by name in NIII only were additional grounds for De Vreesersquos conclusion Again we do not


know that these facts and names did not also occur in the text of which N I andN II are fragments B H S and Z are shorter and much shorter recensions ofthe tale so no conclusions can be drawn from their accounts about whetheror not the much more expansive Middle Dutch fragments belong to one singletextual tradition

Our stemmatological investigations now seem to show however that thethree Middle Dutch fragments are part of one and the same textual tradition

Fragment K

In fact for a comparison of textual variants of at least four texts in the tra-ditional manner we did have a small passage where we might be able tocompare S H and K (verse) and Z (prose) Z appears to be or derive froma translation of a Middle Low German exemplar very similar to S and H(Dieperink 193331ff 152ff Seelmann 1884XIIf) and when we comparedthe 3 rhyming texts S H and K (Langbroek amp Roeleveld 1998a) we foundthat K probably also derives from a similar text However K is a fragmentof only 52 lines and comparison with S and H shows that the translator orcopyist eyeskipped 12 lines in the middle of the fragment In addition onetext is prose while the others are verse and the texts are in three different lan-guages We therefore had to do a great deal of puzzling over how to comparethe differences fairly

In fact we had to invent an additional version difference category Therewere lines where the prose reading of Z was so different from the verse readingsof H S and K that they could not actually be compared With some diffidencetherefore we introduced category 7 lsquosingle version informationrsquo and gave it aweight of 20 comparable to that of minor semantic differences

Our variant apparatus now fell into to the following categories

category classification weight factor1 important semanctic differences 302 minor semantic differences 203a inversions over several lines of text 223b inversions within one line or over adjacent lines 124 important grammatical structures2 105a important interpolations or omissions 215b minor interpolations or omissions 11


6 differences in personal or place names 157 single version information 20

The versions for a few lines will illustrate our procedure

stands for a reading which cannot be compared

H versions type description of difference1186 hskz 7 al stille also stille 1186 hsk 2 al stille also stille1186 hskz 5b do - - -

1187 hskz 7 scherent scheumes

1187 hsk 1 scherent scheumes

1188 hksz 7 dat har all blot daz hor al blot dat har 1188 hks 3b dat har all blot daz hor al blot dat har1189 hksz 2 dar daz doch1189 hszk 5b was he was1189 hksz 1 in groter notin grossir not varin wyl groter war1190 hkzs 5b sach sach ok1190 hksz 7 sunder wan al sunder wan 1190 hks 5b sunder wan al sunder wan1191 hksz 5a cruce roslasht kors1192 hksz 2 stolte ritter Falantin1192 hksz 5a - - - men art1194 hksz 5b vil wol wol1195 hksz 7 wart des wart 1195 hks 5b wart des wart1195 hskz 7 gemeyt bereyt 1195 hsk 1 gemeyt bereyt1197 hszk 2 do lerde he he larte yn1197 hksz 2 vp voten up den voten 2 foslashter

This procedure is again very unusual but was also suggested by our unusualtextual material and the results agree perfectly with what could be expectedwithout the help of modern stemmatological methods

Archetypical information

There were a few places where we found so-called archetypical informationie information which we judged to be close to a hypothetical original recen-sion of the tale Our criteria for the cases where we considered informationto be archetypical were perhaps subjective but by this time we knew our ma-


terial down to the smallest detail and our considerations cannot but have acertain validity An example of archetypical information is the distribution ofthe crowns in B and probably N In N II the number of illegible lines is notlarge enough to have contained the news of Pippingrsquos death and the ensuingdistribution of all the crowns which takes place at this point in H S and Z butnot in B B puts the distribution of the crowns at the very end of the tale byway of happy endings and rightful rewards for all concerned This is surely theoriginal and logical order of the narrative

Not surprisingly we found archetypical information in the Middle Dutchfragments and in one case in B with their more elaborate recensions of thestory but our choice was never for more ample information only but also formore logical information

The archetypical variants function in the procedure as if they were foundin a fictitious additional manuscript the results position the lsquoarchetypersquo inthe pedigree

The automatic comparison and the results

During the automatic comparison of the variant formulas over the text theprogramme executes several passes

The central recension

The first pass locates the theoretical graphical centre of the text tradition Thistheoretical centre will be the recension which has the largest number of read-ings in common with the others and is in that respect central to the texttradition This does not usually mean that it is an especially good recensionor that it is close to the archetype but the centre is a popular recension Inthis case the centre turned out to be H the recension in the Hamburg codexknown as the lsquoHartebokrsquo Close to central is S the recension in the Stockholmmanuscript which is very similar to H in other respects as well

The distance distributions

In the second pass the list of versions is used to compare each recension withall the others At this stage the programme proceeds from the beginning of arecension and works towards the end The resulting information is presented asa collection of graphical visualisations for each recension Clearly these graphs


can only represent points were information is there to be obtained and so theycontain many gaps

The vertical variable in each separate graph ranges between +100 (fortotal agreement) to ndash100 (for total difference) For recensions which agree inabout half of the version formulas (= lines of the database) the means has ascore of 0

The horizontal variable in these graph gives the numbers of the versionformulas The different sections of version formulas are separated by dottedlines These sections correspond to the episodes of the tale for which we hadthe Middle Dutch and Middle Mid German fragments at our disposal (see forthe full list of episodes Diagram 1 under 2)

N K N N NI I III II13 14 13 19 22

B 13 13 19 22Z 9 14 16 22 25S 9 14 16 22 25H 9 14 16 22 25

Recensions H S Z B + NIversion formulas 1-12 89-137

Recensions H S Z + Kversion formulas 13-88

Recensions H S Z B + N IIIversion formulas 138-164

Recensions H S Z B + N IIversion formulas 165-199

In the description of the stemmatological construction programmes in Watteland Van Mulken (1996) using the line numbers in the text was seen as themost obvious choice but for the present text tradition with its huge differencesbetween lengthy and concise recensions it makes more sense to use the linenumbers in the database ie the version formula numbers This alternative hasno influence on the passes of the programmes

At this stage the behaviour of the graphs makes it possible to locate the cor-rupted and contaminated parts of the text and spot positions where possibledifferences or switches occur in different exemplars as to quires or chapters

The collection of graphs for recension H is presented in Figure 1 and thegraphs for the Middle Dutch fragments N I N II and N III are given in Figure 2

Figure 1 shows that S is very closely related to H in all the sections wherean N recension exists However in one position H is closer to N than to S atthe end of section N I

From Figure 2 we can conclude that B is the closest neighbour of N butthe similarity between H and S is much stronger than the similarity between Nand B the graph for the comparison of H to S (in Figure 1) dips less often andless low than that for the comparison of N to B (in Figure 2)


lsquoNlsquoH

lsquoBlsquoH

lsquolsquoH

lsquoKlsquoH

lsquoZlsquoH

lsquoSlsquoH

100 maxndash88 meanndash100 min


100 max68 meanndash100 min


ndash100 meanndash100 min


with lsquoNI with lsquoNI with lsquoNIIwith lsquoNIIIvariants with lsquoK

Comparing lsquoH to lsquoN

Comparing lsquoH to lsquoB

Comparing lsquoH to lsquo

Comparing lsquoH to lsquoK

Comparing lsquoH to lsquoZ

Comparing lsquoH to lsquoS

10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200

Figure 1 Distance distributions for H

lsquoBlsquoN

lsquolsquoN

lsquoKlsquoN

lsquoZlsquoN

lsquoSlsquoN

lsquoHlsquoN


100 max100 mean





Comparing lsquoN to lsquoB

Comparing lsquoN to lsquo

Comparing lsquoN to lsquoZ

Comparing lsquoN to lsquoH

Comparing lsquoN to lsquoS

10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200

Figure 2 Distance distributions for N I N II and N III


There is no information about the silimarity of K and N as the investigatedsections do not overlap at any point

The shock waves

In the next pass of the programme the lsquoshock wavesrsquo are constructed (Wattel ampVan Mulken 1996) These shock waves are low at positions where the interrela-tions between the recensions are stable ie where most or all recensions agreewith at least one other recension without switching from one to another Theshock waves peak at positions where corruptions contaminations and shiftsoccur the interrelations between recensions are unstable as recensions do notagree or switch from agreeing with one to another of the other recensions Asour material was so unusual we decided to execute this pass of the programmealso and find out what information the shock waves could give us

Figure 3 gives the results of this passAs can be seen from the shock waves the sections covered by N I and N II

are the least stable while the section with N III is totally stable The shock wavesalso disclose that recension B is not very stable throughout and that between

lsquoN

lsquoB

lsquo

lsquoZ

lsquoK

lsquoS

lsquoH

total


10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200

Figure 3 The shock waves


H K

S Z

Diagram 2 Version formulas 12-88

S

H Z B N

Diagram 3 The other sections

version formulas 12 and 88 recension H is totally stable and that fragment Kis rather unstable

In the version formulas we sometimes added a reading that gave us theimpression that it was original (see under 7) We introduced these readingsas a fictitious additional recension which was processed in exactly the samemanner as the existing recensions so that differences and shock waves were alsoobtained for Since all but one of these archetypical readings correspond tothe readings in N the graph in which is compared to N is high whenever itexists and the shock wave on never peaks

The initial trees

In the next pass trees are constructed for the different sectionsThe simplesttree possible is the quadruple

The section between version formulas 12-88 does not pose any problem Sand H position themselves to one side K and Z to the other side of a quadruple(Diagram 2)

For the other sections where K of course is lacking we get S and H at oneend with Z close to this pair and N and at the other end while B positionsitself somewhere in between (Diagram 3)

The final pedigree

The two trees (Diagram 2) and (Diagram 3) in 84 have now to be merged intoone single pedigree If we put H at one end and N at the other end of the texttradition it is clear that on the path from N to H we first meet links to fictitious and to B and lastly a link to S (Diagram 4)


HS B

N

Diagram 4 H and N at the respective ends

S

H

ZK

B N

Diagram 5 Adding K and Z

lsquo

4

3

2

1

1

3

7

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2

Figure 4 A possible pedigree

However we have to make two extra attachments a link to K and a link toZ It was not clear from our first results whether K should be attached closerto H or to Z the quadruple H-S-K-Z remains unresolved in this combinedpedigree There may be a combined attachment to the line from N to H whichforks into Z and K (Diagram 5)

From this we constructed a possible complete pedigree in Figure 4The small digit in each edge (= connecting line ldquobranchrdquo in the tree)

indicates the distance expressed by that edge it should be used only as aninidication and has no objective meaning


lsquo

5

4

3

2

1

1

3

6

1

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2

Figure 5 The final pedigree

Node 2 the one in the middle of the plot with four links represents theunresolved quadruple mentioned above

When the extra category of version formulas was added which expressesthe undisputed differences between prose (Z) and verse (H S and K) as de-scribed in section 6 the programmes instantly split node 2 and positioned Kcloser to H and S When more formulas were added to express the fact that thenumber of episodes in B and apparently also in N differs from the number ofepisodes in H S K and Z the programmes put B closer to N We now obtainthe final pedigree (Figure 5)

Validity of narrative elements in stemma construction

The results of the computer aided stemma construction show that it is possibleto use not only textual but also narrative variants as a basis for an investigationinto family relationships between texts The results of our investigations intonarrative elements agree with what we had expected from other indications


and indeed assumed in our earlier publications (Langbroek amp Roeleveld 19971998a b) see also under 11

The other evidence investigations into rhyme

It is generally agreed that the tale of Valentin and Namelos was not originallywritten in Middle Low German but as it remained unclear from what languageor dialect it was translated into Middle Low German we earlier investigatedthe origins of the rhymes in the Stockholm recension (Langbroek amp Roeleveld1998a and 1998b) Two recent editions of the Stockholm recension of Valentinand Namelos were in existence (Geeraedts 1984 Langbroek amp Roeleveld 1997)which should guarantee reliable readings The rhymes were investigated with-out any preconceived notions about their origin and lsquoretranslatedrsquo into MiddleHigh German Middle Mid German and Middle Dutch to trace perfect rhymesThe principles of this work are based on the work of Th Klein (Klein 1997)

From a total of 2291 lines of verse in S we could use 2274 lines or 1137rhyming pairs the rest are orphans The assessing and allotting as to possi-ble original languages or dialects was done on morphological-phonologicaland on lexical grounds Most rhyming pairs turned out to be neutral mean-ing that lsquoretranslatingrsquo them into Middle High German Middle Mid Germanand Middle Dutch yielded perfect rhymes of existant words in all three lan-guagesdialects This was the case for 828 per cent of the rhyming pairs nota surprising percentage for such close linguistic relatives Of the remainder7 per cent yielded perfect rhymes in Middle Dutch only and 77 per cent inboth Middle Dutch and Middle Mid German Only 18 per cent argued ex-clusively for a Middle High German origin In summary a percentage of 905of all the rhymes would be consistent with a Middle Mid German (or MiddleFranconian) exemplar (or possibly 923 per cent if the Middle High Germanpercentage of 18 is added) As many as 982 per cent of the rhymes however isconsistent with a Middle Dutch exemplar Whether the same percentages holdfor the Hartebok recension is not yet certain For comparison of H with S weused our transcript from microfilm Our very recent diplomatic edition of theHamburg Hartebok manuscript will have to be consulted for conclusive results


Notes

Category 4 applies only in cases where the different grammatical structures in questionare possible in all the languages involved it does not in fact occur outside the comparisoninvolving fragment K (see under 6)

Category 4 applies only in cases where the different grammatical structures in questionare possible in all the languages involved it did in fact occur only once

Text editions

N I and II (Staatsbibliothek Preussischer Kulturbesitz Berlin Germ fol 751 3) ndash MiddleDutch verse 2 fragments of 176 lines each c 1350 editions Kalff 1886 p 204ndash220CD-rom Middelnederlands 1998 Rijmteksten

N III (Bibliotheek der Rijksuniversiteit Gent nr 27499) ndash Middle Dutch verse fragment of395 lines of which 16 are barely legible and 22 are indecipharable 1340ndash1360 editionsDe Vreese 1892 Van der Schaaf 1991 CD-rom Middelnederlands 1998 Rijmteksten

K (Kgl Bibl Kopenhagen lost) ndash Middle Mid German verse fragment 52 lines date un-known editions R Nyerup Deutsches Museum 1784 vol II p 91ndash93 Seelman 1884

S (Kgl Bibl Stockholm Cod Holm Vu 73 fol 1r-33r) ndash Middle Low German verse 2291lines c 1450 editions Geeraedts 1984 Langbroek amp Roeleveld 1987

H (Hamburg Staats- und Universitaumltsbibliothek Cod 102c in scrinio known as thelsquoHartebokrsquo fol 33r-75v) ndash Middle Low German verse 2613 lines between 1476ndash1481editions Staphorst 1731 Langbroek amp Roeleveld 2001

B (Stadtbibliothek Breslau fol 304 13-38b) ndash Middle Mid German prose 1465 editionSeelman 1884

Z (Kgl Bibl Stockholm) ndash 3 Old Swedish mss prose 16th century edition Wolf 1934

References

Beta E (1907) Untersuchungen zur Metrik des mittelniederdeutschen Valentin und NamelosDoct Thesis Leipzig

Cd-rom Middelnederlands Woordenboek en teksten (1998) Den Haag AntwerpenDieperink G J (1933) Studien zum Valentin und Namelos Ein Beitrag zur Geschichte

der literarischen Beziehungen zwischen Flandern Middel- und Niederdeutschland undSchweden zur Zeit der Hanse Haarlem

Geeraedts L (1984) Die Stockholmer Handschrift Cod Holm Vu 73 Edition undUntersuchung einer mittelniederdeutschen Sammelhandschrift Niederdeutsche StudienVol 32 Koumlln Wien

Kalff G (1886) Middelnederlandsche epische fragmenten (Reprint Arnhem 1967)Karg F (1924) Die altschwedische Erzaumlhlung von Valentin und Namelos Festschrift fuumlr E

Mogk Halle


Klein Th (1997) ldquoDie Rezeption mittelniederlaumlndischer Versdichtungen im Rheinland undAugustijns lsquoHerzog von Braunschweigrdquorsquo Amsterdamer Beitraumlge zur aumllteren Germanistik47 79ndash107

Langbroek E amp A Roeleveld (1997) Valentin und Namelos Mittelniederdeutsch und Neu-hoch-deutsch Herausgegeben uumlbersetzt und kommentiert von Erika Langbroek undAnnelies Roeleveld unter Mitarbeit von Arend Quak Amsterdamer Publikationen zurSprache und Literatur 27 Amsterdam Rodopi

Langbroek E amp A Roeleveld (1998a) Valentin bekommt einen Gefaumlhrten Ein Vergleichder Reimpaare in den Handschriften S H und K Amsterdamer Beitraumlge zur aumllterenGermanistik 50

Langbroek E amp A Roeleveld (1998b) Wie reimen sich die Nachbarn Eine Untersuchungnach den urspruumlnglichen Reimen in lsquoValentin und Namelosrsquo in der StockholmerHandschrift Cod Holm Vu 73 Niederdeutsches Jahrbuch 121

Langbroek E amp A Roeleveld (Eds) with I Biesheuvel amp H Kienhorst (2001) Het HartebokHs Hamburg Staats- und Universitaumltsbibliothek 102c in scrinio MiddeleeuwseVerzamelhandschriften uit de Nederlanden VIII Hilversum Verloren

Schaaf K van der (1991) Valentijn ende Nameloos III een paralleleditie van eenMiddelnederlands fragment waarin een handschriftbeschrijving en een onderzoeknaar de relatie van de Middelnederlandse Valentijn en Nameloos-versie met deMiddelnederduitse Assen (typescript)

Seelmann W (1884) Valentin und Namelos die niederdeutsche Dichtung die hochdeutscheProsa die Bruchstuumlcke der mitelniederlaumlndischen Dichtung nebst Einleitung Bibliographieund Analyse des Romans Valentin amp Orson Niederdeutsche Denkmaumller Vol 4 NordenLeipzig

Staphorst N (1731) Hamburgische Kirchen-Geschichte Teil I Bd 4 HamburgVreese W de (1892) ldquoEen nieuw fragment van Valentijn en Nameloosrdquo TNTL 11 140ndash162Wattel E amp M van Mulken (1996a) ldquoShock Waves in Text Traditionsrdquo In P van

Reenen amp M van Mulken (Eds) Studies in Stemmatology (pp 105ndash121) AmsterdamPhiladelphia John Benjamins

Wattel E amp M van Mulken (1996b) Weighted Formal Support of a Pedigree In P vanReenen amp M van Mulken (Eds) Studies in Stemmatology (pp 135ndash167) AmsterdamPhiladelphia John Benjamins

Wolf W (1934) Namnloumls och Valentin Kritische Ausgabe mit nebenstehender mittelnieder-deutscher Vorlage Samlingar utgiva av Svenka Fornskrift-Saumlllskapet Uppsala

Index

Aabbreviation 170 172 186accidental variation 127ndash128

131ndash132 134 142adaptation 61 187 209 219 223addition 137ndash138 140 142additional contribution 57ndash58 82admission criteria 145ndash146apparatus 96 244 246 288 292archetype 26ndash27 152 158 230 294

archetypical information 290293ndash294

hyparchetype 27 60artefact 74 84 129 137 199ashkenazi (tradition) 168 185ndash187

189 191 196 206 273 275ndash279281

athonite text types 242 260attestation 14ndash16 19 21 29 33ndash36

38ndash43 45 55 61ndash62 64 67 7077 79ndash80

ausgangstext 25author 14 25ndash26 79 90 113 119

200autograph 25ndash27auxiliary verbs 181 197

Bbias 151bible (edition) 100 104ndash107 110

133 167 169ndash171 173 175ndash176185 193 196ndash200 206 241 244254 261 263 270 274 276ndash279

bishop of Capua cf Victor 129boniface (St) 130

byzantinebyzantine text 43 87ndash91 94 96byzantine tradition 17ndash18 49

87byzantine witnesses 22 43ndash44

61

Ccardiogram 99ndash100 104 106ndash107

109ndash111 204categories 169 171 176 182ndash185

187 191 193 195 201 204chains of coherencies 34change in relationships 106 116chudov New Testament text type

242 249 254 257 263church Fathers 18 21ndash22 78circle 25 50 67ndash72 74 83 134

circular argument 25circular development of variants

74circular edges 67 72circular (genealogical)

relationship 51 68 74cladistic 5 152 198classes of variants 227ndash228classification 169 171 193 243

249ndash254 258ndash260 266 288 292cluster 165 246ndash249 254ndash255

257ndash260 276cluster analysis 241 243 246 251

259ndash260codex Cassellanus 130ndash135 137

139ndash140 142codex Fuldensis 129ndash133 135 137

Index

codex Sangallensis 130ndash131coherence coherency 13 30ndash33 35

37ndash41 44ndash45 54 63ndash65 7072 77 79ndash81 89 139

chains of coherencies 34coherence-based genealogical

method 34coherent chain 43coherent field 34directed coherency 33field coherency 34genealogical coherence 30ndash31

33 41 44 77 79 81imperfect coherency 40 80perfect (genealogical) coherency

40ndash41pre-genealogical coherency 33stemmatic coherency 33 37undirected coherency 63 67 72

coincidental correspondence 16 30coincidental emergence of variants

24commentary 19 49computational complexity 147 154

156 252conflation 193conflicting data 60connective variants 28ndash29 32consensus stemma 230ndash232 235contamination 14 16ndash18 22ndash24 30

34 43 46 48ndash53 55 64115 134 178 193 195 202230 249

contaminated tradition 13 1551 60 76ndash77 196 243 250253 259ndash260 279

incidental contamination 8109 191

multi-stage contamination 5067

simultaneous contamination 8116 177 196

successive contamination99ndash100 104 109 116ndash117182ndash183 203ndash204

contraction 147ndash148 164 170 172183 193 252

convergent evolution 7 9ndash10

coptic 18 79 89

copulative 170 181 183

copyists 128 134

correction 21 119 136 172177ndash178 185 189 191

corrector 16 175

critical apparatus 244 246

Ddata presentation 246 252

deep structure 198 250

design choice 161 163

deviation 152ndash155

dialectical development 176 194

diatessaron 129ndash132

dichotomy 150 156 187 255

dictation 223

direct copy 130ndash133

direct (genealogical) relationship37ndash38 70

direction

directed coherency 33

directed edge 30

direction of variants 114

undirected coherency 63 67 72

undirected edge 64ndash67 72

dissimilarity 101

distance

distance distribution 100

distance function 251 258

distance matrix 5

partition distance 230ndash232 235

distribution formula 101

DNA 3ndash5 7ndash8 10

double translation 178

Eearly version 18 21ndash22 78

eastern tradition 271 275ndash276 279

Index

edge 14 49 55 60 77 81 120ndash121147 159 163ndash64 231 245254 299

circular edges 67 72directed edge 30undirected edge 64ndash67 72

Editio Critica Maior (ECM) 1417ndash18 27 61 77 87 97

error(s) 4 26ndash28 54 91 129 132134ndash138 159 178 181 183193 196 199 202 211216ndash217 221 236 272

printing errors 136scribal error 90 134 171ndash172

177 180 185 200 237evolution 3 4 7ndash10exemplar 26ndash27 49 78 89 92 99

110 116 123 133 169ndash172 175177ndash180 182 184ndash186 188 191195 199ndash201 207ndash208 217ndash219222ndash224 237 272 278 292 301

Ffalsifiability 76faulty reading 78field coherency 34flow cf textual flowformula weight 146 155 158 161

163ndash164 251fragment 52 63 74 286 289

291ndash292 298 302fragmentary witness 71fragmentation 21 70 74 82

Ggender 171 173 183 185 193

210ndash211genealogical coherence 30ndash31 33

41 44 77 79 81genealogical relevance 101 207 222

224genealogies 132general textual flow 33 37genuine variants 18 29 54 78

global stemma 29ndash30 33ndash34 37 4346 49 51 53 58 63ndash64 70ndash7174ndash76 81 83

global textual flow 34 37ndash38graph 30 64 72 77 81 83ndash84

101ndash103 110 187 189 243 253295 298

Hhaftarot(h) 185 270ndash271 280harmony tradition 131 133heading line 100ndash101 148ndash150heart beat 104ndash105 108 111heliand 130ndash131homoioarkton 14homoioteleuton 14 136ndash138 142hyparchetype 27 60

Iimperfect coherency 40 80incidental contamination 8 109

191inconsistency 89 114ndash115 123 146

246indirect relationship 70 84initial text 25ndash27 29 35 42 45ndash46

52 61 63 81ndash84initial tree initial stemma 147 158

164 186 188 190 195 237 250252

intermediarityintermediary node 30 60

62ndash67 71ndash72intermediate structure 250

Kkings of England 7ndash9 228ndash230 236

238kinship revealing 269 272 275koine 28 87 90ndash92 96ndash97

Llacuna(e) 18 69 82 150ndash151 222lanceloet van Denemerken 128

Index

lapsus 27latin 18 89 129ndash133 139ndash140 170

178 199ndash200 280old Latin Harmony 131

lectio difficilior 41 79lectionaries 18 21ndash22 78 261 280level of content 214 217 220 222level of morphology and syntax 210

219level of spelling 209 219levels of linguistic variation 207lexical variation 207 214 223ndash224lexicon 224linguistic variation 207local search 147 158local stemma(ta) 13ndash16 29 34ndash35

38ndash40 61ndash62 68ndash70local textual flow 34 42lydgate 7ndash9 228 238

Mmajority reading 87ndash88 90majority Text 88mantel correlation 230marginal text 78marginal variants 21masorah 176maximal contribution 58maximum likelihood 5ndash6mediate priority 69ndash70mediate relationship 70minimal contribution 57ndash58 82minus 170 178 181 183 189 195

198mirror copying 223missing links 23modern editors 132morphology 171 173 193 207 210

219 224morphological variation 211

213 223mouvance 117 207 217 224multiple stemma 118multi-stage contamination 50 67mutation 4ndash5

N

narrative development 288

neighbour-joining 230 232ndash236

new liturgical tetraevangelion 242249 254 258ndash260 263

new philology 224

new Testament 13 17ndash18 24ndash2528 49 61 87 89ndash91 104106ndash109 129 140 241ndash246 249254 256ndash257 260 263ndash266

node 152ndash154 158ndash159 163ndash164245 252ndash254 257ndash258 300

intermediary node 30 6062ndash67 71ndash72

terminal node 75 250

noise 104 114 147 158 173186ndash187

non-ancestor 30ndash31 60

non-coincidental correspondences30

nondeterministic 157

non-direct relationship 30ndash31

nonsense reading 127 138 141

number 171 173 183ndash185 193 196

O

old Latin Harmony 131

old Testament 104ndash107 111 204

old text type 241ndash242 244 249254ndash255 257ndash259 261

omission 14ndash15 21 82 128137ndash138 140 142 150ndash151 171198 212ndash213 229 255

optimal stemma 147 155

optimal substemma 30ndash31 3351ndash52 55ndash56 61ndash63 65 67 7180 83ndash84

optimization 157

orientation 119ndash121 259

oriented stemma 164 250

original reading 28 43 88 119ndash120172 180ndash181 204

Index

original text 14 24 29 41 88 90119 179 193 277

origin of the reading 114orthodox corruption 89orthography 82 170ndash171 183 193

195 272orthographica 28 54orthographicals 128

Pparallelism 114 119ndash120 127ndash128

136 138ndash141 180ndash181 191195ndash196

paratextual element 181 203parsimony 5ndash6particular textual flow 33 37partition distance 230ndash232 235peak 104 108 110ndash111 182 204

297perfect (genealogical) coherency

40ndash41periphery (of a cluster) 249

258ndash259philoxeniana 79phylogenetic 4ndash5 7ndash8 10 228picardism 122place of variation 13 20 27ndash32

34ndash35 56 62ndash63 77plus 170 178ndash179 181 183 189

195polynomial 157posterior cf also prior 31 33 37

51ndash52 55ndash57 59ndash60 63 67ndash7176 84

potential ancestor 31 33 37ndash4042ndash43 46 49ndash51 55 57 59ndash6267ndash68 77 80ndash83

pre-genealogical coherency 33predominant textual flow 46ndash48 59

81 84preposition 170 180 183 203preslav text type 242 249 254 257

260 262printing errors 136

prior cf also posterior 31 33 3751ndash53 55 57ndash60 62ndash63 67ndash7176 82ndash84

priority 55ndash56 69ndash70punctuation variants 227 238

Qquadruple 122 155ndash156 161

163ndash164 182 185 187ndash189 204298ndash300

quire separator 110 117 122

Rrandom variation 169ndash170 177reading

faulty reading 78majority reading 87ndash88 90nonsense reading 127 138 141original reading 28 43 88

119ndash120 172 180ndash181 204origin of the reading 114variant reading 138 150 168

187 201 261recombination 7ndash8 10relationship

change in relationships 106116

circular (genealogical)relationship 51 68 74


indirect relationship 70 84mediate relationship 70non-direct relationship 30ndash31

resolved tree 159 164 250 252

Ssahidic version 79sample survey 271scribe(s) 8ndash10 14 23 25 27ndash28 54

89 92 96ndash97 135ndash136 138169ndash173 176 182 191ndash192194ndash196 198ndash201 203

Index

208ndash210 213 216ndash219222ndash223 227 242 245

scribal error 90 134 171ndash172177 180 185 200 237

scribal variations 207search algorithm 147sefardi (tradition) 168 178 185

187ndash189 199 204 206 273275ndash279

semantic shift 170 176 183 189195 198

separation 170 172 183 185 193shock wave 102ndash104 108 110ndash111

147 187 298similarity graph 106 182similarity score 246ndash249 251ndash252

257ndash258simultaneous contamination 8 116

177 196single version information 292ndash293spelling variants spelling variation(s)

171 203 209ndash210 214 223ndash224227

split decomposition 5 6 230stability 55ndash57 67 70 72 80 84

191 245stemma(ta) cf also tree 3 8ndash16 24

59 72ndash73 77 100ndash101106ndash107 109 114 123127ndash128 140 142 145ndash148151ndash158 168 171 181 185187 197ndash198 217 227ndash229236ndash238 243 245 250ndash255257ndash260 265 269 271ndash273275 279 287ndash288 300

consensus stemma 230ndash232235

global stemma 29ndash30 33ndash3437 43 46 49 51 53 5863ndash64 70ndash71 74ndash76 81 83

initial tree initial stemma 147158 164 186 188 190 195237 250 252

local stemma(ta) 13ndash16 2934ndash35 38ndash40 61ndash62 68ndash70

multiple stemma 118optimal stemma 147 155stemmatic coherency 33 37substemma 30ndash31 33ndash34

37ndash38 46 49 51ndash52 54ndash5860ndash63 65ndash67 71 74ndash7580ndash81 83ndash84 153ndash154

stepwise refinement 147ndash148 165252

story elements 288substitution 142 170 176 183 189

195subtree cf also substemma 253successive contamination 99ndash100

104 109 116ndash117 182ndash183203ndash204

syntactic variation 210ndash211 214223ndash224

syriac 18 79 89 131ndash133 139 196204

Ttargum 167ndash170 176 179 181 191

194ndash204 269ndash271 273ndash275277ndash281

targum Jonathan 198 269tosefta Targum 274 277ndash280

terminal node 75 250test passages 17 78 87 92ndash93

95ndash97teststellen 17tetraevangelion 244 262ndash264

new liturgical tetraevangelion242 249 254 258ndash260 263

text of the author 25 119text type 168 241ndash246 249

253ndash263textual flow 33ndash34 36ndash38 40ndash43

45ndash49 51ndash53 55ndash57 59 6365 67 70ndash72 80 83ndash84

strength of the textual flow 56textual flow diagram 41 45ndash46

81textus receptus 90 246

Index

three level method three step method114 119 241 250 272ndash273 279

transcription 116 223transposition 7 181tree cf also stemma 4ndash6 8 10 13

46 48 71 81 114ndash115146ndash148 150 152ndash159 161163 167 168 170 181 185191 193 199ndash200 228 230238 245 252ndash255 258 260272ndash273 275 285 298ndash299


resolved tree 159 164 250 252subtree cf also substemma 253unoriented tree 154 158ndash159

type 0 (variation) 19type 1 (variation) 19 150 159 260type 2 (variant variation

oppositions) 19 22 101 146150ndash155 159ndash160 207ndash208 217260 289

type 3 (variation) 19 150 159 260type 4 (variation) 19 150 160 249

Uundirected coherency 63 67 72undirected edge 64ndash67 72unit of variation 27ndash28unoriented tree 154 158ndash159

Vvaluation 146 155ndash156 158ndash159

288variant variation(s) 14ndash17 23

25ndash26 33 36ndash41 43ndash44 4649ndash73 75ndash76 79ndash84 87ndash9294ndash96 100 104ndash106108ndash110 115 119ndash120 123141 145 149 157 172ndash175182ndash183 189 193 195 198205 212 218ndash220 229232ndash233 236ndash237 241

243ndash246 250ndash252 254ndash257269 272ndash273 275 279 292294 300

accidental variation 127ndash128131ndash132 134 142

circular development of variants74

classes of variants 227ndash228

coincidental emergence ofvariants 24

connective variants 28ndash29 32


genuine variants 18 29 54 78

(levels of) linguistic variation207

lexical variation 207 214223ndash224

marginal variants 21

morphological variation 211213 223

place of variation 13 20 27ndash3234ndash35 56 62ndash63 77

punctuation variants 227 238

random variation 169ndash170 177

scribal variations 207

spelling variants spellingvariation(s) 171 203209ndash210 214 223ndash224 227


type 0 (variation) 19

type 1 (variation) 19 150 159260

type 2 (variant variationoppositions) 19 22 101146 150ndash155 159ndash160207ndash208 217 260 289



unit of variation 27 28

variant reading 138 150 168187 201 261

Index

variant selection 167ndash169 181199

variants in story elements andnarrative development 288

variation in word order 211213

versional variants 78

version formula 145ndash146 149ndash152155 159ndash160 162 164 250ndash253260 289 295


victor bishop of Capua 129vocalization 170 172 193vulgate 87 129 131ndash133 139ndash140

201

Wweight 101 158ndash164 228ndash229

237ndash238 253 260 272287ndash288 292

formula weight 146 155 158161 163ndash164 251


Edited by

Pieter van Reenen


Margot van Mulken

With the assistance of Annelies Roeleveld

John Benjamins Publishing Company

AmsterdamPhiladelphia












Table of contents

Prologue vii




Gerd Mink





Margot van Mulken



Ulrich Schmid



Willem F Smelik

Table of contents


Lene Schoslashsler



Dina Mironova




Index 305

Prologue




Prologue




Local stemmata







Prologue





Textual variation



Prologue







Prologue

Weighting variants





Prologue






Margot van Mulken

Notes




P I




Introduction



















4

1 1

11

W

X

Y

Z










W

X

Y

Z


W

X

Y

Z



Stemmatic analysis








Harley 2261

Lambeth 306

Peterborough

Buhler 17


RawlC316Ashmole 59

Nottingham



Titus DXXTCC 601 2

TCC 601 1



Bodley 48Bodley 686

Harley 372Pynson

LeidenLansd 699

Jesus 56Caius 249













Conclusions


Acknowledgements


References




















Introduction




Gerd Mink




b d

c














b d1 e

c d2 f




Gerd Mink







James 41210-12 a 01 02 03 81 1735 2344d P74 218 1718

James 5745 a P74 03b 02 81 218 1718 1735 2344e 01





Gerd Mink







1000

900

800

700

600

500

400

300

200

100

01 2 3 4 5 6 7 8 9 10 11 12 13 14 15


nu

mbe

r of

occu

rren

ces








Gerd Mink

nu

mbe

r of

occu

rren

ces

nu

mbe

r of

occu

rren

ces

450

400

350

300

250

200

150

100

50

0

800

700

600

500

400

300

200

100

0

1 2 3 4 5 6 7 8 9

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19 20 21

10 11 12 13 14 15 16 17 18 19 20

number of variants






Letter of James

about 1740 words











Gerd Mink






20

18

16

14

12

10

8

6

4

2

0

nu

mbe

r of

man

usc

ript

s

1 2 3 4 5 6 7 8 9 10

century

P100

P1000103

01802002504404903302

04048

P74




Gerd Mink












Initial text (A)



Gerd Mink












Places of variation


Gerd Mink














Attestation




Gerd Mink



Intermediary nodes


Optimal substemma







Prior vs posterior


Potential ancestor


Gerd Mink













Gerd Mink


Procedures






James 31222-30

o

a

b

c d g

fh1

h2 i k l mj e





Gerd Mink














Gerd Mink








variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8






Gerd Mink


variant a

variant a

variant b

witnesses

witnesses

witnesses

a

b2

b1

1 2 3 4

5

9

6

10

7

11

8

12

c









variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8





Gerd Mink



A

01

1175

2492

1243 1852

1448

429

17352

2344

307

206 522 1292 1490 2200 453 720 918 1678 24644

4243

01422 0204

1047

18422 254121838

63 884 3264 3983 4422 9154 996515244 21476 25444467 629 642

6234

2818

2805941799 614 1831 630

2412

69

61 1848 218 808 1661 254

621

1359 1127

436 1067 1409 1718 1563 2374










Gerd Mink

14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

642

218

Byz

5 69 88 323 442 915 1832 2718

25412 87621359

1563 4361718 10672374 1409

623623 6212464

2805

12432

813

03

02044 2344 21386142

2492 1505

11752 17392 18522012

57 33 18902412

2464

2805

945 1241 16114

1 Peter 41624-28

variant a

variant b

1359 2541






14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

1 Peter 41624-28

variant b

3073

14485

180 453 468 720 918 1678 2186 2818 12922

17354

429

206 522 1490 2200

424

025

22982

variant a variant b

A



Gerd Mink









1827

2180

2242

3515

2420

8004

966

518

4518

7418

7516

095

400

322

1390

104

1842

643

2544

8832

132

660

799

612

5113

6717

6526

7413

5911

2761

418

90

254

4318

5062

324

6412

9725

4118

3862

117

5161

1837

1661

876

1832

2494

2652

1563

1718

2374

2412

330

2718

2805

1270

1598

1893

436

1067

1409

2243

378

2147

1595

1066

0142

1853

1848

020

618

3893

181

252

319

323

431

442

456

459

676

915

999

1501

2423

2523

218

808

1799

1831

630

1292

2138

2495

398

018

056

6931

246

761

762

963

164

215

0917

2927

7420

652

214

9022

0015

05

P20

3394

197

424

1840

2718

S42

916

11

0204

417

3523

4418

045

346

872

091

816

7821

8621

9728

1824

9202

4694

512

4122

9814

48

P74

P10

001

0304

025

8130

711

7512

4317

3918

52

A

048

365

Figu

re16

P

redo

min

antt

extu

alfl

owth

rou

ghw

itn

esse

sof

Jam

es

Gerd Mink

A

03

P100 048 01 1175 1852

025 1448

81 1243 307

46804173502

P74 2344 1739

2298

945

1241

33


most probable












Gerd Mink

B

C

D


B

C

D

E


B

C

D

E



according to

C BD B CD B

E D BE C D

B


C EE C











Gerd Mink



A

03









A 1175 025 03 1739 01

686926

694920

648915

670902

677896

648871

XXXXX

XXXXX

XXXX

XX

XX

XX

XX

XX

X

X

SUM

722732732734736

734735





Gerd Mink













Gerd Mink



A 1175 025 1739

1243







c


c



A 1175 025 1739

1243




686926729729

5740

2

694920385013

8723

1

6489154100427ndash8677

3

677896463040

2708

2




Gerd Mink


2142-10

21516

2198-14

2262-4

03 11751243

04

11751243

04

03 1175 1243 2492

variant c

variant b

variant e

variant b

1175 1243

stemma (section)

03

1175

1243

042492











Gerd Mink

BB

CC DD

C lt D












Gerd Mink

d

a e

bc


B

03

04

03 lt 04











Gerd Mink

variant b

variant a witness B

witness C witness D


C D

C D

intermediary node

C D








AA 1243 0303 18521739

041175





Gerd Mink

04 1175

A

1852 1243

04 1175

1739

03










Circular edges



Gerd Mink

witnesses B C D E

directioncases



C gt DD gt C

D gt EE gt D

E gt CC gt E

21

32

21

aaaaaaaa

abbbbbbb

babbbcalsquoalsquo



bb balsquo alsquo

c aldquo

C

B

D

E







witnesses C D E

direction cases


local stemmata

C gt DD gt C

D gt EE gt D

E gt CC gt E

21

21

21

aab

bb

bbaa

abbbaa

C D E

D ED E

C

(x)aaa

abbb c(y)

bcc

a

b

C gt DD gt C

D gt EE gt D

E gt CC gt E

41(1)

21(2)

31

21(1)

21(2)

21

aab

bb

c

bbaacc

abbbaa

C D E

dir cases A cases B





Gerd Mink







A A1243 1175 025 0403 1739 031852

1243 173904

A 04 1175

04 1175

A

185203

1739

025 1175

1243

04




Gerd Mink

circular edges

B

D

C

intermediary node

D B

B

C

D







circular edges

intermediary nodes

C D D B B C

B

D

C

B

C

D


04 1175

1243 04 04 1739 1739 1243

03A

18521175

025

04 1243 1739

A

03 025 1175

173912431852

04




Gerd Mink












Gerd Mink

AA 1175 02503A

A

1739

17391739

1243117503

1175

1243

5

8

1

025

03

4

6

7

23

1 973 19 02 925 55 03 926 54 04 908 47 125 904 31 306 920 29 287 896 35 328 915 29 26




Conclusion




Notes





631




Cf Figure 36

Gerd Mink



Cf Aland (1987 1991 1993 1998ndash1999)


















Mink (200052)

Cf Mink (1993489)








Ibid





Cf P74 see note 10

Gerd Mink




























Gerd Mink



Cf paragraph 410




























Cf paragraph 53


Gerd Mink









References




















Introduction




Klaus Wachtel









Klaus Wachtel














Klaus Wachtel













4 908 973 2 1 92 (129141) 67 (23)8 974 65 2 8 98 (150153) 100 (11)

774 1 2 98 (150153) 100 (11)1168 3 1 98 (150153) 67 (23)1514 1 3 98 (150153) 100 (22)

10 908 342 1 48 100 (1212) 100 (11)895 2 1 99 (151153) 100 (1414)1091 2 1 98 (150153) 92 (1213)1194 2 1 98 (150153) 92 (1213)1517 2 1 96 (145151) 77 (1013)2676 3 1 96 (142148) 91 (1011)248 2 1 93 (142152) 100 (99)

11 961 1212 2 1 99 (152153) 100 (55)570 1 3 99 (151152) 100 (55)1207 1 3 99 (150151) 100 (66)200 2 2 99 (150152) 100 (66)944 2 3 99 (150152) 100 (55)905 2 1 99 (149151) 100 (66)1444 2 3 99 (148150) 100 (44)188 2 1 98 (150153) 100 (55)1179 3 1 98 (150153) 100 (44)1351 2 14 98 (4748) 100 (11)2682 2 131 97 (3132) 100 (11)29 3 3 97 (147152) 100 (44)148 3 2 96 (129134) 100 (55)

13 757 543 2 1 93 (119128) 80 (2025)828 3 1 91 (138152) 79 (2633)788 3 1 89 (125140) 89 (2326)346 2 1 89 (7382) 87 (1315)

14 987 140 1 6 100 (153153) 100 (22)1343 1 6 100 (153153) 100 (22)2224 1 6 100 (153153) 100 (22)2522 1 6 100 (153153) 100 (22)95 1 6 100 (150150) 100 (22)405 1 20 100 (106106) 100 (11)123 2 10 99 (152153) 100 (22)208 2 6 99 (152153) 100 (22)1080 2 10 99 (152153) 100 (22)1191 2 10 99 (152153) 100 (11)1225 1 11 99 (152153) 100 (22)1585 1 9 99 (152153) 100 (11)



Klaus Wachtel

Table 1 (continued)


2509 2 10 99 (152153) 100 (11)584 2 10 99 (151152) 100 (22)1058 2 15 99 (104105) 100 (22)

15 935 2562 1 1 99 (151153) 100 (88)1439 2 1 98 (149152) 90 (910)1163 1 1 98 (145148) 90 (910)53 3 1 95 (146153) 75 (68)902 3 1 95 (146153) 86 (67)2374 2 1 95 (143151) 100 (44)2502 2 2 94 (141150) 100 (22)

18 974 35 1 40 100 (153153) 100 (44)55 1 40 100 (153153) 100 (44)128 1 40 100 (153153) 100 (44)201 1 40 100 (153153) 100 (44)479 1 40 100 (153153) 100 (44)480 1 40 100 (153153) 100 (44)645 1 40 100 (153153) 100 (44)696 1 40 100 (153153) 100 (44)769 1 40 100 (153153) 100 (44)789 1 40 100 (153153) 100 (44)867 1 40 100 (153153) 100 (44)928 1 40 100 (153153) 100 (44)955 1 40 100 (153153) 100 (44)1023 1 40 100 (153153) 100 (44)1072 1 40 100 (153153) 100 (44)1117 1 40 100 (153153) 100 (44)1147 1 40 100 (153153) 100 (44)1339 1 40 100 (153153) 100 (44)1401 1 40 100 (153153) 100 (44)1493 1 40 100 (153153) 100 (44)1496 1 40 100 (153153) 100 (44)1550 1 40 100 (153153) 100 (44)1560 1 40 100 (153153) 100 (44)1584 1 40 100 (153153) 100 (44)






Table 2

Ms MT Group Pattern

18 974 322 555 1004 150483 967 322 555 1004

1279 1504141 967 1612B 322 555 1004 1504147 961 263 322 555 1004B 1504

693155 961 322 555 1004 1504

992 1513167 954 273 322 555 1004B 1504

603664

189 935 123 322 555 1004B 1504394 693 1323

894 1343246 974 322 555 1004B 1504386 967 322 555 1004 1504

963394 967 322 555 1004B 1504

1183547 966 322 555 1004 1504

693553 967 322 555 1004 1504

762685 967 322 555 1004B 1504

1525689 961 92 322 555 1004B 1504

123763 967 322 555 1004B 1504

762781 967 322 555 1004B 1504

548797 954 182 322 555 1004 1504

322B 6912D547


Klaus Wachtel





Conclusions







Notes

ECM IV 19972003









Klaus Wachtel


References














Introduction1







A A den Hollander

















A A den Hollander

F

D

C

B2

B1

A4

A3

A2

total

A1

47ndash2

ndash44ndash99

37ndash88

10078

89ndash39

67ndash39

89ndash39

89ndash39

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500

25

ndash92

ndash39

97

26

23

27

24






F

D

C

B2

B1

A4

A3

A2

total

A1

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500







A A den Hollander









III II II VI VI IV


CvR

SM

JvL

WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 180 360 540 720 900 1080 1260

1525

1544

1527

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525



A A den Hollander















A A den Hollander


0 120 160 200 240 280 320 360 40040 80

IX IIII IX(B)

MC

AvB

JvL

JvL

CL

JvG

JvL

MHvH

HPvM

JvG

JvL

JvL

JvG

AvB

AP

AvB

SM

AvB

WV

CvR

HvR

CvR

WV

JvG

CvR

MKGvdH

DP

HF

total

1539

1525

15232

15231

1524

1525

1526

1527

1543

1524

1522

1523

15262

1523

1525

1524

1545

1524

1529

1525

1525

1528

1528

15261

1526

1525

1523

1525





Conclusions






A A den Hollander


Notes



























References




A A den Hollander





WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 20 40 60 80 100 120 140 160 180 200 220

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525




Introduction




Margot van Mulken










C A

R B T P S












Contamination


Margot van Mulken











The quire separator







Margot van Mulken

C A

R S P B


C A

R B T P S


C A

R

B

T

P

S






Table 2






Orientation



Margot van Mulken

Table 3


Table 4









Table 5


Table 6


Table 7




Conclusion


Margot van Mulken





Notes








asbcprT 1190 1380 1386 1409 1487 1535 1572 1612 1640 2368 26282 2666 28002987 3063 3086 3118 3162 3362 3555 3602 3743 3948 4133 4380 44464695

apsbcrT 1344 1590 1613 1633 1638 2360 2512 2628 2945 3105 3420 3478 35453631 3651 3657 3776 3824 4400 4476 4607

apsTbcr 1151 1201 1244 1250 1285 1330 1348 1403 1521 1544 1666 1719 17221741 1744 1745 1754 1797 2247 2394 2405 2428 2588 2601 2679 26812703 2726 2836 2919 2921 3026 3253 3430

abpsTcr 1296 2402 3106 3237 40822 1232 1244 1322 1544 1602 1719 2375 23942428 30862 3577 37072




References





Margot van Mulken








Editions






P II

Textual variation

Genealogy by chance



Introduction



Ulrich Schmid



Genealogy by chance





Ulrich Schmid




Genealogy by chance



Ulrich Schmid





Genealogy by chance


Ulrich Schmid





Genealogy by chance



a Grein p 130







c Grein p 132




Ulrich Schmid

e Grein p 191




f Grein p 192


g Grein p 259


h Grein p 260


i Grein p 261




Genealogy by chance







Ulrich Schmid









Genealogy by chance







Ulrich Schmid




Genealogy by chance



Notes



Cf Petersen (1994)



Fischer (1963546)



Ulrich Schmid



Rathofer (1971 1972 1973)









References



Genealogy by chance














Introduction




E Wattel















E Wattel





Material











E Wattel








Manuscript versions








E Wattel

lsquo lsquo

lsquoˆ5 lsquoˆ5

lsquoˆ3 lsquoˆ3

lsquoˆ2 lsquoˆ1

lsquod lsquoclsquod

lsquoa lsquoˆ2


lsquoˆ4 lsquoˆ4

lsquoclsquob


















lsquo5 lsquo3





E Wattel


















E Wattel













E Wattel















E Wattel
























E Wattel


















E Wattel



Final stage



Conclusion





References













Textual history




Willem F Smelik











Willem F Smelik











n Persono Gender

p Status



Orthography


Errors


Willem F Smelik




Vocalization





Abbreviation






Willem F Smelik




Table 2 vs 38






Willem F Smelik











Pluses and minuses



Willem F Smelik






58






Willem F Smelik



Prepositions and













Willem F Smelik


Evaluation

Methodology













Category 1 Spelling


Willem F Smelik

Eb4

Eb91

Rb6

Rb3

Eb1

s4

q2

D

G

o

Eb66

x

p

d

q63

K

F

C

Q

O

y

Rb2

T

W

E

w

J

a

Rb1

S

A

B

P

N

M

total

15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 2700






Category 2 Errors







Willem F Smelik

ˆ34

15

2

5

ˆ33

542

11 37

P

J

ˆ32

33

B

23

ˆ31

1141

AD

ˆ30

4 7

Cˆ29

84

ˆ16

33

NRb1

ˆ28

16

T

4

ˆ27

14

S

3

ˆ26

9

O

3

ˆ25

32

ˆ24

2

ˆ21

3

ˆ23

3 35

W

7

ˆ9

1 6

Eˆ3

1 5

G Eb4

ˆ20 ˆ19

34

ˆ17 ˆ5

63

Kˆ18

4

ˆ1

018

30

ˆ4 Rb2

2 2

p Mˆ14

12

ˆ13 ˆ12

22

ˆ11

3

ˆ2

3

ˆ7

1 2

Eb1F

3 1

s4 q63

5 2

d o

Eb91

4 4

Eb66

2

ˆ10

2

ˆ6ˆ8

4 1

q2 x

ˆ22

44

w y

0 2

Rb3Rb6

116

Q a










Willem F Smelik

43

ˆ34

ˆ33

7 31

a

J

3211

ˆ32

B

A

249

12

ˆ31

8

ˆ30 ˆ27

8 12

Qˆ29

14

ˆ18

7

ˆ28

66

ˆ14

14

d

7

ˆ25

107

Kˆ24

8 10

Cˆ23

12

E

7

ˆ22

5 13

ˆ20

75

ˆ17

5

ˆ16

7

54

ˆ15

7

T

4

ˆ10

2

Eb66

3

ˆ7

22

ˆ4

22 0

ˆ3

p

11

s4 Eb4

w y

11

ˆ8 x

11

ˆ11 q63

42

ˆ12

Eb1

F

P

1011

ˆ26 ˆ21

15

D

11

ˆ9

2

G

5

ˆ6

4 4

oˆ2

1 0

q2 Eb91

15

S

8

ˆ19

5 7

O

W

ˆ13

3 3

M

3 5

NRb1

ˆ5

11

ˆ1

20

Rb3

Rb2Rb6











Willem F Smelik

ˆ34

43

A

15

ˆ33

29

B

29 17

ˆ31

1413

ˆ30

ˆ32

23

J

26

ˆ28

12

a

12

Q

ˆ25 F

8 9

106

Kˆ24

6 7

ˆ21ˆ23

6 8 6 6

C P

G

11

ˆ7ˆ22

7 6

ˆ18

9 6

ˆ16 ˆ15

711

q63ˆ12

3

ˆ9

2

x

2 1

w y

6 5

pˆ14

4 6

ˆ5ˆ11

3 3

Eb1ˆ6

3 1

Eb91

4 0

s4ˆ1

1

Eb66

0

q2

30

D E

Tˆ4

6 7

ˆ17 Eb4

11 32

ˆ27 ˆ26

7 10

ˆ20 S

9 6

ˆ10 O

o

2 2

ˆ8

2 0

M W

ˆ29

22 14

dˆ19

10 6

ˆ3ˆ13

5 5

NRb1

2 0

ˆ2 Rb6

0 2

Rb2Rb3











Willem F Smelik

o

Eb4

D

Eb91

Rb3

G

Rb6

s4

q2

Eb1

C

B

Q

Eb66

T

A

S

J

w

Rb2

F

M

x

K

d

N

E

a

y

o

p

Rb1

W

P

q63

total

0 90 180 270 360 450 540 630 720 810 900 990 1080 1170 1260 1350 1440 1530






Epilogue




Willem F Smelik

G

Eb4

Eb91

s4

D

Rb6

Rb3

q2

Eb1

Eb66

E

T

c

o

P

K

d

q63

F

x

B

O

Q

A

N

w

M

Rb2

J

W

y

a

Rb1

S

p

total

0 32 64 96 128 160 192 224 256 288 320 352 384 416 448 480 512 544 576








Willem F Smelik







Notes








Willem F Smelik
















Willem F Smelik































Willem F Smelik






































Willem F Smelik































Willem F Smelik

Table 8 (continued)







Introduction




Lene Schoslashsler






Scribal variations














Lene Schoslashsler








Scribal variations












Lene Schoslashsler







Scribal variations







Lene Schoslashsler










SynonymsNouns


Adjectives


Pronouns


Scribal variations

Articles


Conjunctions


Prepositions


Verbs


Adverbs





Lene Schoslashsler










Scribal variations












Lene Schoslashsler










Scribal variations












Lene Schoslashsler










Nouns




Scribal variations

Adjectives


Pronouns


Articles


Conjunctions


Prepositions


Verbs


Adverbs



Lene Schoslashsler






Conclusion



Scribal variations





Lene Schoslashsler







Notes





Scribal variations





References

I Primary sources














Lene Schoslashsler





















Introduction





The data


















1

1

3

2

2

3

4

4

5

5






Results





RWyer

de Worde


Harley 372Lansd 699

RawlC 48Caius 249

Lansd 210

Harley 7333

Egerton 1995

Linc Lat 129

TCC 601 2TCC 601 1


NottinghamBodley 48

Pynson

Galba E VIII

Peterborough

Harley 2261

Titus DXX

Lambeth 306

Harley 2169

CUL Ad6686Stowe 69


Buhler 17

Lansd 762Digby 186


001 distance units

Ashmole 59





Ashmole 59






Peterborough

Lansd210

Ipswich

Pynson

RawlC 48

Caius 249Leiden

Stowe 69


Lansd 699

TCC 601 1TCC 601 2

Harley 7333


BL Ad 31042

Fairfax 16RawlC 316

Buhler 17

Harley 2169

01 distance units

RWyer

de Worde

Galba EVIII



de WordeRWyer


Ipswich

Caius 249



Harley 7333

Egerton 1995

TCC 601 1TCC 601 2

Buhler 17RawlC 316


Bodley 48

BL Ad 31042

Leiden

Harley 372

Lansd 699




Lansd 762Stowe 69


Peterborough








094 002097 018100 079



1 2 3 4


Discussion











Acknowledgements


References
























Introduction





Dina Mironova












Dina Mironova













m1 m2

m3

m4

m5

m6

m7


Dina Mironova


Similarity score




Data presentation









to some clusters



no

no

no

yes

yes

yes





Dina Mironova














Dina Mironova









Formula weight




Dp = Np (Np + Sp)

Dina Mironova












Dina Mironova


Results









Dina Mironova







Dina Mironova








Conclusion




Dina Mironova









Acknowledgements


Notes









References



Dina Mironova










Appendix 1

The witnesses

I The Old Text Type


the core



other manuscripts













Dina Mironova








Appendix 2


Tr 376 d 366

D1 383 032 02257 12 273 Pv 118 43 75 418 B1t 293 Ju Ms K

48 Nk1 Kh 27 20 14 11 10 5 64 A 311 HH

31 Mv 28 15 2 2 4 54II 1z Ht 42 1p

32 29 17 9 1

36 37 19

35 34 23

Vr mm 33 24 22 6

30 25 21 7 Cd Nkm 3 185 69 352 85 355


lr WW 225 387 385

Dina Mironova

Appendix 3





Prolegomena


Introduction3


Alberdina Houtman





Targum Isaiah






Method


Alberdina Houtman









Tosefta Targums



Alberdina Houtman

Diagram 1




















Results



Alberdina Houtman

Diagram 2









D Parma 3188

Eastern tradition














Alberdina Houtman

Diagram 3













Conclusions



Alberdina Houtman

Acknowledgments


Notes













Robinson (1994)











See above n 12








Alberdina Houtman

References




































Introduction


The tale




The manuscripts




























An illustration

















Fragment K






























N K N N NI I III II13 14 13 19 22

B 13 13 19 22Z 9 14 16 22 25S 9 14 16 22 25H 9 14 16 22 25











lsquoNlsquoH

lsquoBlsquoH

lsquolsquoH

lsquoKlsquoH

lsquoZlsquoH

lsquoSlsquoH














10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200


lsquoBlsquoN

lsquolsquoN

lsquoKlsquoN

lsquoZlsquoN

lsquoSlsquoN

lsquoHlsquoN


100 max100 mean










10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200




The shock waves




lsquoN

lsquoB

lsquo

lsquoZ

lsquoK

lsquoS

lsquoH

total


10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200



H K

S Z


S

H Z B N




The initial trees




The final pedigree



HS B

N


S

H

ZK

B N


lsquo

4

3

2

1

1

3

7

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2






lsquo

5

4

3

2

1

1

3

6

1

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2












Notes



Text editions








References






Mogk Halle













Index





189 191 196 206 273 275ndash279281










61












Index










34 43 46 48ndash53 55 64115 134 178 193 195 202230 249








coptic 18 79 89


copyists 128 134


corrector 16 175








dichotomy 150 156 187 255

dictation 223



direction


directed edge 30




dissimilarity 101

distance



distance matrix 5







Index

edge 14 49 55 60 77 81 120ndash121147 159 163ndash64 231 245254 299








163ndash164 251fragment 52 63 74 286 289


Ggender 171 173 183 185 193






101ndash103 110 187 189 243 253295 298






164 186 188 190 195 237 250252






Index










N




new philology 224





noise 104 114 147 158 173186ndash187






number 171 173 183ndash185 193 196

O







optimization 157




Index

original text 14 24 29 41 88 90119 179 193 277
































Index




187ndash189 199 204 206 273275ndash279

semantic shift 170 176 183 189195 198





171 203 209ndash210 214 223ndash224227















syriac 18 79 89 131ndash133 139 196204

Ttargum 167ndash170 176 179 181 191











Index





































variant reading 138 150 168187 201 261

Index








201


237ndash238 253 260 272287ndash288 292













Table of contents

Prologue vii




Gerd Mink





Margot van Mulken



Ulrich Schmid



Willem F Smelik

Table of contents


Lene Schoslashsler



Dina Mironova




Index 305

Prologue




Prologue




Local stemmata







Prologue





Textual variation



Prologue







Prologue

Weighting variants





Prologue






Margot van Mulken

Notes




P I




Introduction



















4

1 1

11

W

X

Y

Z










W

X

Y

Z


W

X

Y

Z



Stemmatic analysis








Harley 2261

Lambeth 306

Peterborough

Buhler 17


RawlC316Ashmole 59

Nottingham



Titus DXXTCC 601 2

TCC 601 1



Bodley 48Bodley 686

Harley 372Pynson

LeidenLansd 699

Jesus 56Caius 249













Conclusions


Acknowledgements


References




















Introduction




Gerd Mink




b d

c














b d1 e

c d2 f




Gerd Mink







James 41210-12 a 01 02 03 81 1735 2344d P74 218 1718

James 5745 a P74 03b 02 81 218 1718 1735 2344e 01





Gerd Mink







1000

900

800

700

600

500

400

300

200

100

01 2 3 4 5 6 7 8 9 10 11 12 13 14 15


nu

mbe

r of

occu

rren

ces








Gerd Mink

nu

mbe

r of

occu

rren

ces

nu

mbe

r of

occu

rren

ces

450

400

350

300

250

200

150

100

50

0

800

700

600

500

400

300

200

100

0

1 2 3 4 5 6 7 8 9

1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19 20 21

10 11 12 13 14 15 16 17 18 19 20

number of variants






Letter of James

about 1740 words











Gerd Mink






20

18

16

14

12

10

8

6

4

2

0

nu

mbe

r of

man

usc

ript

s

1 2 3 4 5 6 7 8 9 10

century

P100

P1000103

01802002504404903302

04048

P74




Gerd Mink












Initial text (A)



Gerd Mink












Places of variation


Gerd Mink














Attestation




Gerd Mink



Intermediary nodes


Optimal substemma







Prior vs posterior


Potential ancestor


Gerd Mink













Gerd Mink


Procedures






James 31222-30

o

a

b

c d g

fh1

h2 i k l mj e





Gerd Mink














Gerd Mink








variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8






Gerd Mink


variant a

variant a

variant b

witnesses

witnesses

witnesses

a

b2

b1

1 2 3 4

5

9

6

10

7

11

8

12

c









variant a

variant b

witnesses

witnesses

a

b

1 2 3 4

5 6 7 8





Gerd Mink



A

01

1175

2492

1243 1852

1448

429

17352

2344

307

206 522 1292 1490 2200 453 720 918 1678 24644

4243

01422 0204

1047

18422 254121838

63 884 3264 3983 4422 9154 996515244 21476 25444467 629 642

6234

2818

2805941799 614 1831 630

2412

69

61 1848 218 808 1661 254

621

1359 1127

436 1067 1409 1718 1563 2374










Gerd Mink

14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

642

218

Byz

5 69 88 323 442 915 1832 2718

25412 87621359

1563 4361718 10672374 1409

623623 6212464

2805

12432

813

03

02044 2344 21386142

2492 1505

11752 17392 18522012

57 33 18902412

2464

2805

945 1241 16114

1 Peter 41624-28

variant a

variant b

1359 2541






14482 17352 22982

A

025 307

180 453 468 720 918 1678 2186 2818

12922

429

206 522 1490 2200424

1 Peter 41624-28

variant b

3073

14485

180 453 468 720 918 1678 2186 2818 12922

17354

429

206 522 1490 2200

424

025

22982

variant a variant b

A



Gerd Mink









1827

2180

2242

3515

2420

8004

966

518

4518

7418

7516

095

400

322

1390

104

1842

643

2544

8832

132

660

799

612

5113

6717

6526

7413

5911

2761

418

90

254

4318

5062

324

6412

9725

4118

3862

117

5161

1837

1661

876

1832

2494

2652

1563

1718

2374

2412

330

2718

2805

1270

1598

1893

436

1067

1409

2243

378

2147

1595

1066

0142

1853

1848

020

618

3893

181

252

319

323

431

442

456

459

676

915

999

1501

2423

2523

218

808

1799

1831

630

1292

2138

2495

398

018

056

6931

246

761

762

963

164

215

0917

2927

7420

652

214

9022

0015

05

P20

3394

197

424

1840

2718

S42

916

11

0204

417

3523

4418

045

346

872

091

816

7821

8621

9728

1824

9202

4694

512

4122

9814

48

P74

P10

001

0304

025

8130

711

7512

4317

3918

52

A

048

365

Figu

re16

P

redo

min

antt

extu

alfl

owth

rou

ghw

itn

esse

sof

Jam

es

Gerd Mink

A

03

P100 048 01 1175 1852

025 1448

81 1243 307

46804173502

P74 2344 1739

2298

945

1241

33


most probable












Gerd Mink

B

C

D


B

C

D

E


B

C

D

E



according to

C BD B CD B

E D BE C D

B


C EE C











Gerd Mink



A

03









A 1175 025 03 1739 01

686926

694920

648915

670902

677896

648871

XXXXX

XXXXX

XXXX

XX

XX

XX

XX

XX

X

X

SUM

722732732734736

734735





Gerd Mink













Gerd Mink



A 1175 025 1739

1243







c


c



A 1175 025 1739

1243




686926729729

5740

2

694920385013

8723

1

6489154100427ndash8677

3

677896463040

2708

2




Gerd Mink


2142-10

21516

2198-14

2262-4

03 11751243

04

11751243

04

03 1175 1243 2492

variant c

variant b

variant e

variant b

1175 1243

stemma (section)

03

1175

1243

042492











Gerd Mink

BB

CC DD

C lt D












Gerd Mink

d

a e

bc


B

03

04

03 lt 04











Gerd Mink

variant b

variant a witness B

witness C witness D


C D

C D

intermediary node

C D








AA 1243 0303 18521739

041175





Gerd Mink

04 1175

A

1852 1243

04 1175

1739

03










Circular edges



Gerd Mink

witnesses B C D E

directioncases



C gt DD gt C

D gt EE gt D

E gt CC gt E

21

32

21

aaaaaaaa

abbbbbbb

babbbcalsquoalsquo



bb balsquo alsquo

c aldquo

C

B

D

E







witnesses C D E

direction cases


local stemmata

C gt DD gt C

D gt EE gt D

E gt CC gt E

21

21

21

aab

bb

bbaa

abbbaa

C D E

D ED E

C

(x)aaa

abbb c(y)

bcc

a

b

C gt DD gt C

D gt EE gt D

E gt CC gt E

41(1)

21(2)

31

21(1)

21(2)

21

aab

bb

c

bbaacc

abbbaa

C D E

dir cases A cases B





Gerd Mink







A A1243 1175 025 0403 1739 031852

1243 173904

A 04 1175

04 1175

A

185203

1739

025 1175

1243

04




Gerd Mink

circular edges

B

D

C

intermediary node

D B

B

C

D







circular edges

intermediary nodes

C D D B B C

B

D

C

B

C

D


04 1175

1243 04 04 1739 1739 1243

03A

18521175

025

04 1243 1739

A

03 025 1175

173912431852

04




Gerd Mink












Gerd Mink

AA 1175 02503A

A

1739

17391739

1243117503

1175

1243

5

8

1

025

03

4

6

7

23

1 973 19 02 925 55 03 926 54 04 908 47 125 904 31 306 920 29 287 896 35 328 915 29 26




Conclusion




Notes





631




Cf Figure 36

Gerd Mink



Cf Aland (1987 1991 1993 1998ndash1999)


















Mink (200052)

Cf Mink (1993489)








Ibid





Cf P74 see note 10

Gerd Mink




























Gerd Mink



Cf paragraph 410




























Cf paragraph 53


Gerd Mink









References




















Introduction




Klaus Wachtel









Klaus Wachtel














Klaus Wachtel













4 908 973 2 1 92 (129141) 67 (23)8 974 65 2 8 98 (150153) 100 (11)

774 1 2 98 (150153) 100 (11)1168 3 1 98 (150153) 67 (23)1514 1 3 98 (150153) 100 (22)

10 908 342 1 48 100 (1212) 100 (11)895 2 1 99 (151153) 100 (1414)1091 2 1 98 (150153) 92 (1213)1194 2 1 98 (150153) 92 (1213)1517 2 1 96 (145151) 77 (1013)2676 3 1 96 (142148) 91 (1011)248 2 1 93 (142152) 100 (99)

11 961 1212 2 1 99 (152153) 100 (55)570 1 3 99 (151152) 100 (55)1207 1 3 99 (150151) 100 (66)200 2 2 99 (150152) 100 (66)944 2 3 99 (150152) 100 (55)905 2 1 99 (149151) 100 (66)1444 2 3 99 (148150) 100 (44)188 2 1 98 (150153) 100 (55)1179 3 1 98 (150153) 100 (44)1351 2 14 98 (4748) 100 (11)2682 2 131 97 (3132) 100 (11)29 3 3 97 (147152) 100 (44)148 3 2 96 (129134) 100 (55)

13 757 543 2 1 93 (119128) 80 (2025)828 3 1 91 (138152) 79 (2633)788 3 1 89 (125140) 89 (2326)346 2 1 89 (7382) 87 (1315)

14 987 140 1 6 100 (153153) 100 (22)1343 1 6 100 (153153) 100 (22)2224 1 6 100 (153153) 100 (22)2522 1 6 100 (153153) 100 (22)95 1 6 100 (150150) 100 (22)405 1 20 100 (106106) 100 (11)123 2 10 99 (152153) 100 (22)208 2 6 99 (152153) 100 (22)1080 2 10 99 (152153) 100 (22)1191 2 10 99 (152153) 100 (11)1225 1 11 99 (152153) 100 (22)1585 1 9 99 (152153) 100 (11)



Klaus Wachtel

Table 1 (continued)


2509 2 10 99 (152153) 100 (11)584 2 10 99 (151152) 100 (22)1058 2 15 99 (104105) 100 (22)

15 935 2562 1 1 99 (151153) 100 (88)1439 2 1 98 (149152) 90 (910)1163 1 1 98 (145148) 90 (910)53 3 1 95 (146153) 75 (68)902 3 1 95 (146153) 86 (67)2374 2 1 95 (143151) 100 (44)2502 2 2 94 (141150) 100 (22)

18 974 35 1 40 100 (153153) 100 (44)55 1 40 100 (153153) 100 (44)128 1 40 100 (153153) 100 (44)201 1 40 100 (153153) 100 (44)479 1 40 100 (153153) 100 (44)480 1 40 100 (153153) 100 (44)645 1 40 100 (153153) 100 (44)696 1 40 100 (153153) 100 (44)769 1 40 100 (153153) 100 (44)789 1 40 100 (153153) 100 (44)867 1 40 100 (153153) 100 (44)928 1 40 100 (153153) 100 (44)955 1 40 100 (153153) 100 (44)1023 1 40 100 (153153) 100 (44)1072 1 40 100 (153153) 100 (44)1117 1 40 100 (153153) 100 (44)1147 1 40 100 (153153) 100 (44)1339 1 40 100 (153153) 100 (44)1401 1 40 100 (153153) 100 (44)1493 1 40 100 (153153) 100 (44)1496 1 40 100 (153153) 100 (44)1550 1 40 100 (153153) 100 (44)1560 1 40 100 (153153) 100 (44)1584 1 40 100 (153153) 100 (44)






Table 2

Ms MT Group Pattern

18 974 322 555 1004 150483 967 322 555 1004

1279 1504141 967 1612B 322 555 1004 1504147 961 263 322 555 1004B 1504

693155 961 322 555 1004 1504

992 1513167 954 273 322 555 1004B 1504

603664

189 935 123 322 555 1004B 1504394 693 1323

894 1343246 974 322 555 1004B 1504386 967 322 555 1004 1504

963394 967 322 555 1004B 1504

1183547 966 322 555 1004 1504

693553 967 322 555 1004 1504

762685 967 322 555 1004B 1504

1525689 961 92 322 555 1004B 1504

123763 967 322 555 1004B 1504

762781 967 322 555 1004B 1504

548797 954 182 322 555 1004 1504

322B 6912D547


Klaus Wachtel





Conclusions







Notes

ECM IV 19972003









Klaus Wachtel


References














Introduction1







A A den Hollander

















A A den Hollander

F

D

C

B2

B1

A4

A3

A2

total

A1

47ndash2

ndash44ndash99

37ndash88

10078

89ndash39

67ndash39

89ndash39

89ndash39

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500

25

ndash92

ndash39

97

26

23

27

24






F

D

C

B2

B1

A4

A3

A2

total

A1

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500







A A den Hollander









III II II VI VI IV


CvR

SM

JvL

WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 180 360 540 720 900 1080 1260

1525

1544

1527

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525



A A den Hollander















A A den Hollander


0 120 160 200 240 280 320 360 40040 80

IX IIII IX(B)

MC

AvB

JvL

JvL

CL

JvG

JvL

MHvH

HPvM

JvG

JvL

JvL

JvG

AvB

AP

AvB

SM

AvB

WV

CvR

HvR

CvR

WV

JvG

CvR

MKGvdH

DP

HF

total

1539

1525

15232

15231

1524

1525

1526

1527

1543

1524

1522

1523

15262

1523

1525

1524

1545

1524

1529

1525

1525

1528

1528

15261

1526

1525

1523

1525





Conclusions






A A den Hollander


Notes



























References




A A den Hollander





WV

JvL

HPvM

HvL

JvL

HPvM

JvL

WV

JvL

WV

WV

WV

DP

JvL

HvR

total

0 20 40 60 80 100 120 140 160 180 200 220

1542

1542

1541

1538

1535

1535

1534

1533

1532

1532

1528 [1531]

1528

1527

1526

1525




Introduction




Margot van Mulken










C A

R B T P S












Contamination


Margot van Mulken











The quire separator







Margot van Mulken

C A

R S P B


C A

R B T P S


C A

R

B

T

P

S






Table 2






Orientation



Margot van Mulken

Table 3


Table 4









Table 5


Table 6


Table 7




Conclusion


Margot van Mulken





Notes








asbcprT 1190 1380 1386 1409 1487 1535 1572 1612 1640 2368 26282 2666 28002987 3063 3086 3118 3162 3362 3555 3602 3743 3948 4133 4380 44464695

apsbcrT 1344 1590 1613 1633 1638 2360 2512 2628 2945 3105 3420 3478 35453631 3651 3657 3776 3824 4400 4476 4607

apsTbcr 1151 1201 1244 1250 1285 1330 1348 1403 1521 1544 1666 1719 17221741 1744 1745 1754 1797 2247 2394 2405 2428 2588 2601 2679 26812703 2726 2836 2919 2921 3026 3253 3430

abpsTcr 1296 2402 3106 3237 40822 1232 1244 1322 1544 1602 1719 2375 23942428 30862 3577 37072




References





Margot van Mulken








Editions






P II

Textual variation

Genealogy by chance



Introduction



Ulrich Schmid



Genealogy by chance





Ulrich Schmid




Genealogy by chance



Ulrich Schmid





Genealogy by chance


Ulrich Schmid





Genealogy by chance



a Grein p 130







c Grein p 132




Ulrich Schmid

e Grein p 191




f Grein p 192


g Grein p 259


h Grein p 260


i Grein p 261




Genealogy by chance







Ulrich Schmid









Genealogy by chance







Ulrich Schmid




Genealogy by chance



Notes



Cf Petersen (1994)



Fischer (1963546)



Ulrich Schmid



Rathofer (1971 1972 1973)









References



Genealogy by chance














Introduction




E Wattel















E Wattel





Material











E Wattel








Manuscript versions








E Wattel

lsquo lsquo

lsquoˆ5 lsquoˆ5

lsquoˆ3 lsquoˆ3

lsquoˆ2 lsquoˆ1

lsquod lsquoclsquod

lsquoa lsquoˆ2


lsquoˆ4 lsquoˆ4

lsquoclsquob


















lsquo5 lsquo3





E Wattel


















E Wattel













E Wattel















E Wattel
























E Wattel


















E Wattel



Final stage



Conclusion





References













Textual history




Willem F Smelik











Willem F Smelik











n Persono Gender

p Status



Orthography


Errors


Willem F Smelik




Vocalization





Abbreviation






Willem F Smelik




Table 2 vs 38






Willem F Smelik











Pluses and minuses



Willem F Smelik






58






Willem F Smelik



Prepositions and













Willem F Smelik


Evaluation

Methodology













Category 1 Spelling


Willem F Smelik

Eb4

Eb91

Rb6

Rb3

Eb1

s4

q2

D

G

o

Eb66

x

p

d

q63

K

F

C

Q

O

y

Rb2

T

W

E

w

J

a

Rb1

S

A

B

P

N

M

total

15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240 255 2700






Category 2 Errors







Willem F Smelik

ˆ34

15

2

5

ˆ33

542

11 37

P

J

ˆ32

33

B

23

ˆ31

1141

AD

ˆ30

4 7

Cˆ29

84

ˆ16

33

NRb1

ˆ28

16

T

4

ˆ27

14

S

3

ˆ26

9

O

3

ˆ25

32

ˆ24

2

ˆ21

3

ˆ23

3 35

W

7

ˆ9

1 6

Eˆ3

1 5

G Eb4

ˆ20 ˆ19

34

ˆ17 ˆ5

63

Kˆ18

4

ˆ1

018

30

ˆ4 Rb2

2 2

p Mˆ14

12

ˆ13 ˆ12

22

ˆ11

3

ˆ2

3

ˆ7

1 2

Eb1F

3 1

s4 q63

5 2

d o

Eb91

4 4

Eb66

2

ˆ10

2

ˆ6ˆ8

4 1

q2 x

ˆ22

44

w y

0 2

Rb3Rb6

116

Q a










Willem F Smelik

43

ˆ34

ˆ33

7 31

a

J

3211

ˆ32

B

A

249

12

ˆ31

8

ˆ30 ˆ27

8 12

Qˆ29

14

ˆ18

7

ˆ28

66

ˆ14

14

d

7

ˆ25

107

Kˆ24

8 10

Cˆ23

12

E

7

ˆ22

5 13

ˆ20

75

ˆ17

5

ˆ16

7

54

ˆ15

7

T

4

ˆ10

2

Eb66

3

ˆ7

22

ˆ4

22 0

ˆ3

p

11

s4 Eb4

w y

11

ˆ8 x

11

ˆ11 q63

42

ˆ12

Eb1

F

P

1011

ˆ26 ˆ21

15

D

11

ˆ9

2

G

5

ˆ6

4 4

oˆ2

1 0

q2 Eb91

15

S

8

ˆ19

5 7

O

W

ˆ13

3 3

M

3 5

NRb1

ˆ5

11

ˆ1

20

Rb3

Rb2Rb6











Willem F Smelik

ˆ34

43

A

15

ˆ33

29

B

29 17

ˆ31

1413

ˆ30

ˆ32

23

J

26

ˆ28

12

a

12

Q

ˆ25 F

8 9

106

Kˆ24

6 7

ˆ21ˆ23

6 8 6 6

C P

G

11

ˆ7ˆ22

7 6

ˆ18

9 6

ˆ16 ˆ15

711

q63ˆ12

3

ˆ9

2

x

2 1

w y

6 5

pˆ14

4 6

ˆ5ˆ11

3 3

Eb1ˆ6

3 1

Eb91

4 0

s4ˆ1

1

Eb66

0

q2

30

D E

Tˆ4

6 7

ˆ17 Eb4

11 32

ˆ27 ˆ26

7 10

ˆ20 S

9 6

ˆ10 O

o

2 2

ˆ8

2 0

M W

ˆ29

22 14

dˆ19

10 6

ˆ3ˆ13

5 5

NRb1

2 0

ˆ2 Rb6

0 2

Rb2Rb3











Willem F Smelik

o

Eb4

D

Eb91

Rb3

G

Rb6

s4

q2

Eb1

C

B

Q

Eb66

T

A

S

J

w

Rb2

F

M

x

K

d

N

E

a

y

o

p

Rb1

W

P

q63

total

0 90 180 270 360 450 540 630 720 810 900 990 1080 1170 1260 1350 1440 1530






Epilogue




Willem F Smelik

G

Eb4

Eb91

s4

D

Rb6

Rb3

q2

Eb1

Eb66

E

T

c

o

P

K

d

q63

F

x

B

O

Q

A

N

w

M

Rb2

J

W

y

a

Rb1

S

p

total

0 32 64 96 128 160 192 224 256 288 320 352 384 416 448 480 512 544 576








Willem F Smelik







Notes








Willem F Smelik
















Willem F Smelik































Willem F Smelik






































Willem F Smelik































Willem F Smelik

Table 8 (continued)







Introduction




Lene Schoslashsler






Scribal variations














Lene Schoslashsler








Scribal variations












Lene Schoslashsler







Scribal variations







Lene Schoslashsler










SynonymsNouns


Adjectives


Pronouns


Scribal variations

Articles


Conjunctions


Prepositions


Verbs


Adverbs





Lene Schoslashsler










Scribal variations












Lene Schoslashsler










Scribal variations












Lene Schoslashsler










Nouns




Scribal variations

Adjectives


Pronouns


Articles


Conjunctions


Prepositions


Verbs


Adverbs



Lene Schoslashsler






Conclusion



Scribal variations





Lene Schoslashsler







Notes





Scribal variations





References

I Primary sources














Lene Schoslashsler





















Introduction





The data


















1

1

3

2

2

3

4

4

5

5






Results





RWyer

de Worde


Harley 372Lansd 699

RawlC 48Caius 249

Lansd 210

Harley 7333

Egerton 1995

Linc Lat 129

TCC 601 2TCC 601 1


NottinghamBodley 48

Pynson

Galba E VIII

Peterborough

Harley 2261

Titus DXX

Lambeth 306

Harley 2169

CUL Ad6686Stowe 69


Buhler 17

Lansd 762Digby 186


001 distance units

Ashmole 59





Ashmole 59






Peterborough

Lansd210

Ipswich

Pynson

RawlC 48

Caius 249Leiden

Stowe 69


Lansd 699

TCC 601 1TCC 601 2

Harley 7333


BL Ad 31042

Fairfax 16RawlC 316

Buhler 17

Harley 2169

01 distance units

RWyer

de Worde

Galba EVIII



de WordeRWyer


Ipswich

Caius 249



Harley 7333

Egerton 1995

TCC 601 1TCC 601 2

Buhler 17RawlC 316


Bodley 48

BL Ad 31042

Leiden

Harley 372

Lansd 699




Lansd 762Stowe 69


Peterborough








094 002097 018100 079



1 2 3 4


Discussion











Acknowledgements


References
























Introduction





Dina Mironova












Dina Mironova













m1 m2

m3

m4

m5

m6

m7


Dina Mironova


Similarity score




Data presentation









to some clusters



no

no

no

yes

yes

yes





Dina Mironova














Dina Mironova









Formula weight




Dp = Np (Np + Sp)

Dina Mironova












Dina Mironova


Results









Dina Mironova







Dina Mironova








Conclusion




Dina Mironova









Acknowledgements


Notes









References



Dina Mironova










Appendix 1

The witnesses

I The Old Text Type


the core



other manuscripts













Dina Mironova








Appendix 2


Tr 376 d 366

D1 383 032 02257 12 273 Pv 118 43 75 418 B1t 293 Ju Ms K

48 Nk1 Kh 27 20 14 11 10 5 64 A 311 HH

31 Mv 28 15 2 2 4 54II 1z Ht 42 1p

32 29 17 9 1

36 37 19

35 34 23

Vr mm 33 24 22 6

30 25 21 7 Cd Nkm 3 185 69 352 85 355


lr WW 225 387 385

Dina Mironova

Appendix 3





Prolegomena


Introduction3


Alberdina Houtman





Targum Isaiah






Method


Alberdina Houtman









Tosefta Targums



Alberdina Houtman

Diagram 1




















Results



Alberdina Houtman

Diagram 2









D Parma 3188

Eastern tradition














Alberdina Houtman

Diagram 3













Conclusions



Alberdina Houtman

Acknowledgments


Notes













Robinson (1994)











See above n 12








Alberdina Houtman

References




































Introduction


The tale




The manuscripts




























An illustration

















Fragment K






























N K N N NI I III II13 14 13 19 22

B 13 13 19 22Z 9 14 16 22 25S 9 14 16 22 25H 9 14 16 22 25











lsquoNlsquoH

lsquoBlsquoH

lsquolsquoH

lsquoKlsquoH

lsquoZlsquoH

lsquoSlsquoH














10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200


lsquoBlsquoN

lsquolsquoN

lsquoKlsquoN

lsquoZlsquoN

lsquoSlsquoN

lsquoHlsquoN


100 max100 mean










10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200




The shock waves




lsquoN

lsquoB

lsquo

lsquoZ

lsquoK

lsquoS

lsquoH

total


10 20 30 40 50 60 70 80 900 100 110 120 130 140 150 160 170 180 190 200



H K

S Z


S

H Z B N




The initial trees




The final pedigree



HS B

N


S

H

ZK

B N


lsquo

4

3

2

1

1

3

7

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2






lsquo

5

4

3

2

1

1

3

6

1

6

lsquoN

1

lsquoZ

8

lsquoS

2

lsquoH

2

lsquoK

3

lsquoB

2












Notes



Text editions








References






Mogk Halle













Index





189 191 196 206 273 275ndash279281










61












Index










34 43 46 48ndash53 55 64115 134 178 193 195 202230 249








coptic 18 79 89


copyists 128 134


corrector 16 175








dichotomy 150 156 187 255

dictation 223



direction


directed edge 30




dissimilarity 101

distance



distance matrix 5







Index

edge 14 49 55 60 77 81 120ndash121147 159 163ndash64 231 245254 299








163ndash164 251fragment 52 63 74 286 289


Ggender 171 173 183 185 193






101ndash103 110 187 189 243 253295 298






164 186 188 190 195 237 250252






Index










N




new philology 224





noise 104 114 147 158 173186ndash187






number 171 173 183ndash185 193 196

O







optimization 157




Index

original text 14 24 29 41 88 90119 179 193 277
































Index




187ndash189 199 204 206 273275ndash279

semantic shift 170 176 183 189195 198





171 203 209ndash210 214 223ndash224227















syriac 18 79 89 131ndash133 139 196204

Ttargum 167ndash170 176 179 181 191











Index





































variant reading 138 150 168187 201 261

Index








201


237ndash238 253 260 272287ndash288 292
