Studies In Stemmatology II

Studies in Stemmatology II Studies in Stemmatology II Edited by Pieter van Reenen August den Hollander Margot van Mu...

161 downloads 668 Views 12MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Studies in Stemmatology II

Studies in Stemmatology II Edited by

Pieter van Reenen August den Hollander Margot van Mulken With the assistance of Annelies Roeleveld

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

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 den Hollander and Margot van Mulken. p. cm. Includes bibliographical references and indexes. 1. Manuscripts. 2. Manuscripts, Medieval. 3. Transmission of texts. I. Reenen, Pieter Th. van. II. Hollander, A. A. den III. Mulken, Margot van Z105.S782 2004 091-dc22 isbn 90 272 3222 9 (Eur.) / 1 58811 535 6 (US) (Hb; alk. paper)

20040100159

© 2004 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microﬁlm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

Prologue

vii

I. Stemmatological methods and techniques Parallels between stemmatology and phylogenetics Christopher Howe, Adrian Barbrook, Linne Mooney, and Peter Robinson Problems of a highly contaminated tradition: the New Testament: Stemmata of variants as a source of a genealogy for witnesses Gerd Mink

3

13

Kinds of variants in the manuscript tradition of the Greek New Testament 87 Klaus Wachtel How shock waves revealed successive contamination: A cardiogram of early sixteenth-century printed Dutch Bibles August den Hollander

99

The manuscript tradition of the Cligés of Chrétien de Troyes: A stemmatological approach Margot van Mulken

113

II. Textual variation Genealogy by chance! On the significance of accidental variation (parallelisms) Ulrich Schmid Constructing initial binary trees in stemmatology Evert Wattel Trouble in the trees! Variant selection and tree construction illustrated by the texts of Targum Judges Willem F. Smelik

127 145

167



Table of contents

Scribal variations: When are they genealogically relevant – and when are they to be considered as instances of ‘mouvance’? Lene Schøsler The effects of weighting kinds of variants Matthew Spencer, Linne R. Mooney, Adrian C. Barbrook, Barbara Bordalejo, Christopher J. Howe, and Peter Robinson Cluster analysis and the Three Level Method in the study of the Gospels in Slavonic Dina Mironova Different kinds of tradition in Targum Jonathan to Isaiah Alberdina Houtman

207 227

241 269

Valentin and Namelos discover their parentage: Narrative elements in the family tree of an international medieval tale Annelies Roeleveld, Erika Langbroek, and Evert Wattel

285

Index

305

Prologue

The publication of this volume of Studies in Stemmatology is the second in a series. Its predecessor was published in 1996 and opened the most actual state of the art in stemmatology to a broad audience.1 That volume not only aimed at giving scholars access to modern stemmatological methods and techniques, but also at illustrating how profitable the application of these methods might be for their future work. The first volume was very well received by stemmatologists all over Europe and also gave an impulse to new research, as several articles in Studies in Stemmatology II clearly illustrate. The contributions to this present book partly proceed from those of the first volume. Most of them are the result of the on-going scholarly debate on stemmatology of recent years. Several of the contributions to this volume were presented on 13–14 April 2000, during the NOSTER-conference at the Netherlands Institute of Advanced Studies (NIAS) at Wassenaar, and on 13 October 2000, during the Stemmatology Conference at the Vrije Universiteit in Amsterdam. Some others are the result of the annual colloquia of stemmatologists, held at the Vrije Universiteit in Amsterdam.2 The object of this second volume of Studies in Stemmatology is the evaluation of the most recent methods and techniques in the field of stemmatology, as well as the development of new ones. The book is largely interdisciplinary in character: it contains contributions from scholars from classical, historical, biblical (Smelik, Houtman, Den Hollander), medieval and modern language studies, as well as from mathematical and computer scientists (Wattel) and biologists (Howe e.a.; Spencer e.a.). Various manuscript traditions are dealt with here: some of them within one field of language (Van Mulken, Schøsler), some multilingual (Roeleveld/Langbroek/Wattel), the last group of course requiring a special methodological approach to the establishment of variants. Other traditions were very extensive, e.g. the New Testament manuscripts (Mink, Wachtel) and the Old Church Slavonic manuscripts (Mironova). The contributions in the book have been divided into two sections. The first section deals with various stemmatological methods and techniques. The second section focusses more specifically on the various problems concerning textual variation.

 Prologue

Stemmatological methods Not a bifurcating tree? Christopher Howe, Adrian Barbrook, Linne Mooney and Peter Robinson present a relatively new stemmatological approach. They explore the similarities between the evolution of DNA sequences and the changes occurring in manuscript traditions. They show how the techniques of evolutionary biology can be applied to stemmatic analysis and how a number of features of manuscript traditions have clear parallels in genetics. They conclude that the process of incorporation into DNA mirrors the incorporation of changes into the manuscripts. It follows that programs for phylogenetic analysis of sequence data can be exploited for stemmatic analysis of manuscript tradition. For this approach the Splits Tree program is used, which has the advantage that it does not presuppose, as many methods do, that the tree is a bifurcating one.

Local stemmata? Methodologically new, and very promising, is the contribution of Gert Mink. In his contribution he broaches the problem of a text tradition of many hundreds of manuscripts in which hardly any type-2 variants are found and contamination is the rule. This is the case in one of the catholic letters of the New Testament, the Epistle of James. In such a tradition existing methods cannot be applied. Instead, Mink has two working hypotheses on which his approach is based: a. If more than one exemplar was consulted by a scribe, the exemplars are closely related. b. Variants are analysed one by one in local trees. Within the local trees the direction of the changes can be determined: they can be oriented by establishing which variant derives from which other variant. When groups of local trees are oriented in the same direction, parts of global trees can be constructed.

Reduction of witnesses Klaus Wachtel also deals with this extremely large textual tradition. In his contribution he shows how the number of manuscripts to examine can be reduced before 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 quantities must be accepted before the building of a stemma starts. By distinguishing two groups of manuscripts in the New Testament tradition Wachtel succeeds in doing so. The two groups distinguished are the Majority group and the Byzantine group. If two or more manuscripts are almost alike, there is no need for further analysis. By applying this approach Wachtel succeeds in reducing the numbers considerably, without the risk of excluding manuscripts which contain crucial textual information. The resulting group forms the input of Gert Minks analysis.

Dealing with successive contamination A illuminating example of how profitable the application of modern stemmatological tools can be is given in the contribution of August den Hollander. One of the complex problems a philologist has to deal with is a contaminated text tradition. In the first volume of Studies in Stemmatology Wattel and Van Mulken offered the instrument of the so-called shock waves (cardiograms) as a help to reveal successive contamination in a text tradition which is rather entangled. In his contribution Den Hollander shows how the application of this instrument indisputably revealed successive contamination in the textual tradition of early sixteenth-century printed Dutch Bibles. In her contribution to the present volume, Margot van Mulken shows that the output of the quire separator developed by Wattel (see first volume) may have serious consequences for the further treatment of the stemmatological process. When the separator indicates successive contamination, as in the case of the Cligés, it may be necessary to presuppose a multiple orientation of the stemmata. However, in the case of the Cligés, all the archetypes can be found in the neighbourhood of one manuscript, which fortunately reduces the complexity 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, who systematically reviewed various types of variant readings used in genealogical studies, and offered strict text-genealogical rules in order to exclude possible





Prologue

variants caused by accidental variation.3 In his contribution, Schmid illustrates the implications of applying Salemans’ rules to a text tradition: on the one hand they would exclude too much, leaving out many genealogically ‘valid’ variants, on the other hand even the variants that would be included on the basis of Salemans’ rules, still contain parallelistic readings. Therefore, Schmid concludes, 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. Textual scholars do not generally agree on which type of variant readings are suitable for the construction of a stemma, and which are not. In his contribution Evert Wattel argues for the acceptance of as many version formulas as possible, in addition to expressing the reliability of the variants by adding more or less weight to the so-called version formulas. His main focus is on the computational problems of constructing a stemma on the basis of the collective formulas. Along with a more general methodological discussion, he dwells on specific problems such as lacunary version formulas and the computational complexity.

Categories of variants In his contribution Willem Smelik deals with variant selection and tree construction in the text tradition of Targum Judges. The core of this study consists of observations on the phenomenon of random variation in the manuscript reproduction. To identify random or coincidental variation, he suggests a transparent, verifiable categorisation of variant readings. Secondly, he discusses the possible genealogical information of these various types of variants in great detail. Further, Smelik draws stemmata for each type or group of types of variants. Finally, comparison of these stemmata reveals which types of variants turned out to be genealogical relevant in his textual tradition, and which not. Lene Schøsler compares the categorised variants of two closely related manuscripts of the Perceval and four (five?) manuscripts of the Charroi de Nîmes. The first two were copied by the same scribe, with perhaps a difference in time. Assuming that the scribe copied twice from the same exemplar, it is remarkable that the variations found between the two manuscripts and those in the tradition of the Charroi de Nîmes are hardly different. In other words, whether the same scribe copies the same manuscript twice or different scribes copy a manuscript may not necessarily result in more variants.

Prologue

Weighting variants? In their contribution Matthew Spencer, Linne Mooney, Adrian Barbrook, Barbara Bordalejo, Christopher Howe and Peter Robinson attempt to increase the chance of reconstructing correct stemmas by categorizing variants into ten different kinds, such as: “line changed completely”, “word change affecting rhyme”, “word variant, changes meaning”, “minor word added or omitted, without changing meaning”. On the assumption that not all kinds of variants are equally reliable, the more a category of variants is reliable the more weight it is assigned. On comparison between stemmata of the 55 manuscripts and three printed versions of Lydgate’s Kings of England, the choice of weights appeared relatively unimportant. However, the authors expect that this may be different in larger textual traditions. The method used to reconstruct the stemmata was neighbour-joining: a simple clustering algorithm which sequentially separates pairs of manuscripts from an initially unresolved stemma. Dina Mironova deals with the problem of a textual tradition of many manuscripts. She compares two different formal genealogical methods in her study of the Gospels in Slavonic: cluster analysis (Alexeev) and the Three Level Method (Wattel). Her research comprises no fewer than 531 manuscripts, still presenting, however, a rather stable text. The large number of witnesses impelled her to work with groups of manuscripts as a way of reduction. Alexeev’s method turned out be less accurate, but more economical, since it is still easier to apply to large traditions. Wattel’s method is, however, more accurate, and forces the scholar to formulate precise classifications or explicit philological labellings. Despite the difficulties with extremely large textual traditions, according to Mironova his method is to be preferred when variants should be evaluated (weighted).

Exclusion of variants Dineke Houtman studies the textual history of Tosefta Targum Jonathan, an extended Aramaic Bible commentary. She focuses on the question of how to deal with this type of text in stemmatological research, especially when comparing it with the Hebrew Bible text and its paraphrasing Aramaic Targum text. All three types of text represent different stages in the textual history. The text of the Targum remains close to the Hebrew Bible, the text of Tosefta Targum, however, gives a more free rendering. Houtman concludes that including textual 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 great precaution.

Alternative classification of variants Annelies Roeleveld and Erika Langbroek, in cooperation with Evert Wattel, deal with the text tradition of ‘Valentin and Namelos’. Its extant manuscripts are written in no fewer than four languages/dialects: Middle Dutch, Middle Low German, Middle Mid German and Old Swedish. Other differences between the texts are also considerable: some are in verse, others in prose; some are much more lengthy and elaborate in their descriptions than others; some have been preserved in fragments only. The problem in this tradition is what sort of variant to group in comparable units. The authors develop a classification in which the number of incidents, the order of incidents and the detail in the description of incidents play a part. Although this notion of variant goes beyond the traditional view, their approach shows that satisfactory results can be obtained. The resulting stemma plausibly shows that the Middle Dutch versions are the more original. Rhyme analysis had already pointed in the same direction; the original language was most probably Middle Dutch. The editors hope that this second volume of Studies in Stemmatology will inspire scholars like the first volume did, and stimulate the development of new methods and strategies aiming at further control of variation and contamination in (large) text traditions. Pieter van Reenen August den Hollander Margot van Mulken

Notes . Studies in Stemmatology (1996). Pieter van Reenen & Margot van Mulken (Eds.). Amsterdam/Philadelphia: John Benjamins Publishing Co. . At this place we wish to thank NIAS, NOSTER and the Faculty of Arts of the Free University for their willingness to support the various stemmatological meetings in the past years. . B. J. P Salemans (2000). Building Stemmas with the Computer in a Cladistic, NeoLachmannian, way. The Case of Fourteen Text Versions of Lanseloet van Denemerken (diss. Nijmegen). Nijmegen.

P I

Stemmatological methods and techniques

Parallels between stemmatology and phylogenetics Christopher Howe*, Adrian Barbrook*, Linne Mooney, and Peter Robinson University of Cambridge* / University of Maine, Orono / De Montfort University, Leicester

.

Introduction

The work and ideas we discuss here are part of a project entitled “STEMMA – Studies on Textual Evolution of Manuscripts by Mathematical Analysis”, funded by the Leverhulme Trust. This project aims to apply the techniques of evolutionary biology to the analysis of manuscript traditions. In particular we are interested in the application of computer programs developed for evolutionary biology to the study of manuscripts. In this paper we explore the similarities between the evolution of DNA sequences and the changes occurring in manuscript traditions. We will show how the techniques of evolutionary biology can be applied to stemmatic analysis and how a number of features of manuscript traditions have clear parallels in genetics. Another paper in this volume (Spencer et al.), to which this chapter should serve as an introduction, discusses more specific issues in this work. The computer programs we are using were developed for research in biology, so their application to manuscript stemmatics requires some knowledge of the underlying biology. A more detailed discussion can be found in biological textbooks (e.g. Voet et al. 1999). DNA (deoxyribonucleic acid) is comprised of four kinds of unit. These are collectively called nucleotides (more fully deoxyribonucleotides), and the four kinds are adenosine, guanosine, cytidine and thymidine deoxyribonucleotides. They are more conveniently designated A, G, C and T. DNA molecules are composed of chains of such nucleotides, and the order 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 thus the structure and function of proteins, which are the main functional entities in cells. For example, enzymes, which catalyse the reactions of metabolism, are proteins. In most systems, the DNA chains come in pairs – forming the double helix made famous by Watson and Crick (Watson & Crick 1953). The sequence of nucleotides in one strand determines the sequence in the other. Thus A in one is always opposite T in the other, G in one opposite C in the other. As cells divide, their genetic information has to be duplicated. This process is termed DNA replication, and the aim is to make an identical copy of the parental molecule. However, errors can occur in the replication process, so a parental molecule with sequence, say, ACGGTACTAG TGCCATGATC

might give rise to two daughters, one of which had the same sequence, and the other of which had a different sequence, say ACGGCACTAG TGCCGTGATC

Here, the T in the fifth position in the upper strand has been replaced by a C, and in the lower strand an A has been replaced by a G. We say that a ‘mutation’ has 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 mutations in their DNA. The longer it is since two species had a common ancestor, the more different (in general) is their DNA sequence. So we can use the differences in DNA sequence among species as a way of inferring their evolutionary relationships. We can recover something akin to a family tree, showing which species share a common ancestor to the exclusion of others. This is called a phylogenetic tree. Recovering a phylogenetic tree using a given DNA sequence for a group of species requires both a model for sequence evolution and a method for tree recovery. Models of sequence evolution are in effect a set of assumptions about how the sequences change. They may be explicit (in which case the computer program 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, much longer sequences would be used): W X Y Z

AAAAAAAA GGAAAAAA CCTTTTAA CCTTCCAA

The distance matrix would be W W – X – Y –

X 2 – –

Y 6 6 –

Z 6 6 2

and the tree inferred would be: 1 W

1

Y

4 X

1

1

Z

Figure 1. Hypothetical example of a distance matrix analysis.

they may be implicit, in which case one may be unaware that these assumptions are being made. It is particularly important to realize that, even if invalid assumptions are being made, it will still usually be possible to recover a phylogenetic tree. However, the tree may be incorrect as a result of making the wrong assumptions. Examples of the assumptions that may be made as part of a model include the relative frequencies of certain kinds of mutation (often termed the transition/transversion ratio); independence of mutations (i.e. that mutation at one position does not affect the chance of a mutation at another position); or identical distribution (that the same regions of a sequence are potentially able to mutate in all the organisms being considered). Evolutionary biologists use several methods for tree recovery, including distance matrix, parsimony, maximum likelihood (reviewed by Felsenstein 1988; Beanland & Howe 1992) and split decomposition methods (Huson 1998). With distance matrix methods, one calculates a matrix showing the number of differences between pairs of sequences, and determines the tree which has the best fit to this matrix. An example is given in Figure 1. With parsimony methods we prefer the tree which requires the fewest mutations. This is in effect a cladistic analysis. So, for example, with the following DNA sequences (showing only one of the two strands):





Christopher Howe et al.

W X Y Z

AAGCCAAT TAGCCAAT CGCTTGGT GGCTTGGT

Positions 2–7 would require fewest mutations if species W and X shared most recent common ancestry to the exclusion of Y and Z. Then, the most parsimonious tree would be as shown in Figure 2A. Maximum likelihood attempts to find the tree that has the highest probability of generating the data (i.e. sequences) observed. This approach can be very computationally intensive. With split decomposition, as implemented in the program SplitsTree we consider possible splits between the sequences, e.g. (WX, YZ) or (WY, XZ) etc. and look at the number of nucleotide positions consistent with each split. Split decomposition has a number of potentially desirable features. It does not presuppose, as many methods do, that the tree is a bifurcating one. It allows recovery of trees where one individual can have many descendents. It also allows conflicting information to be shown – where there are some positions in the data which are not consistent with the preferred tree. These appear as boxes in the output, giving a network rather than a tree. So the example given in Figure 2B would arise where there was some signal linking W with Y and X with Z, as well as the signal linking W with X and Y with Z. W

Y

X

Z

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

Y

X

Z

Figure 2B. Other tree-recovery methods. 2B shows the result of a hypothetical split decomposition analysis where there is support both for WX, YZ groupings and for WY, XZ groupings.

Parallels between stemmatology and phylogenetics

. Stemmatic analysis The techniques which are used to recover evolutionary trees can be applied in principle to datasets derived from manuscript traditions, using changes between texts in the same way as evolutionary biologists use changes between the DNA sequence of different organisms (Platnick & Cameron 1977; Cameron 1987; Lee 1989; O’Hara & Robinson 1989; Barbrook et al. 1998). We are applying this to a range of texts, such as John Lydgate’s 15th century poem, Kings of England, which exists in over 30 manuscript versions comprising a set of stanzas describing the Kings of England from William the Conqueror onwards. Thus we have variants such as: Worthy to stand among the worthy nyne Able to stand among the worthi nyne Able to stande among the worthyes nyne

for the same line in different texts. (These are variants of line 96 of the poem in Oxford, Bodleian, Rawlinson C.316, fol. 122v; Cambridge, Jesus College Q.G.8 (56), fol. 47v; and Ipswich, County Hall Deposit C4/4, Percyvale’s Great Doomsday Book, Bk. VI, fol. 239v.) We can encode this information to produce a dataset resembling a nucleotide sequence, (but using any letters or numbers rather than just A, C, G or T) to represent different readings at a given position, and use this directly as an input to a phylogenetic program such as SplitsTree. Thus we recently produced a stemmatic analysis of the extant manuscript versions of Lydgate’s Kings of England based on evolutionary biological techniques, which replicated stemmatic analysis done by traditional methods (Figure 3). Many of the manuscripts grouped together in the traditional stemmatic analysis were also grouped together in the phylogenetic analysis. Some were grouped together in the former, but not in the latter. This might simply reflect a need for more information in the dataset used. All groupings in the phylogenetic analysis were consistent with those in the traditional analysis. This study is described in more detail elsewhere (Mooney et al. 2001). It is remarkable how many parallels there are between the evolution of genetic material and the changes occurring in manuscripts (Howe et al. 2001). These include recombination, convergent evolution and transposition. We will look at these parallels in turn. Genetic recombination is the process whereby two different copies of a gene come together to produce a hybrid called a recombinant. The first part of the recombinant comes from one precursor gene, and the second from the other. The parallel in manuscript traditions is the change of exemplar de-





Christopher Howe et al. Harley 2261 Lambeth 306 Harley 7333 Egerton 1995

Peterborough Nottingham

Harley 2251 BL Ad 34360 Titus D.XX TCC 601 2

Buhler 17 CUL Ad6686

Ashmole 59

Dublin 516

Bodley 48 Bodley 686 Bodley 912 Fairfax 16

TCC 601 1 Harley 372 Pynson Leiden Lansd. 699 Jesus 56 Rawl.C.316 Caius 249

0.01 distance units Linc.Lat 129

Figure 3. SplitsTree diagram of relationships among 27 most complete manuscripts and early printed copies of Lydgate’s Kings of England. Bold indicates group A, italics B, underlined C, suggested by manual analysis (Mooney et al. 2001).

scribed as “successive contamination” (Wattel & van Mulken 1996), when a scribe changed the text from which he was copying part of the way through. We found examples of this in the analysis of the Prologue to the Wife of Bath’s Tale, in that the position of some texts on the stemmatic tree changed when an analysis performed on the first part of the data was compared with one done on the second (Barbrook et al. 1998). Programmes are being developed to identify recombination breakpoints in genes, and these may be useful in stemmatic analysis (e.g. Holmes et al. 1999). As well as generating hybrids where there is a clear breakpoint, recombination can give rise to more complex products, where the resulting DNA sequence is a mosaic of the two parental versions of the sequence (e.g. Medgyesy et al. 1985). This process is analogous to simultaneous (where the scribe used several exemplars simultaneously) and incidental contamination (where the scribe used a single exemplar first and then modified the text afterwards with other exemplars) (Wattel & van Mulken 1996). Transposition of genetic information is when material from one source is inserted somewhere quite different. This is a feature of the life cycle of some viruses, for example. If the source of inserted DNA is a different species it would

Parallels between stemmatology and phylogenetics

be regarded as ‘lateral gene transfer’. We have found parallels in the Kings of England tradition where there is contamination of one tradition with material from another. Thus, the verse relating to William I begins typically: This myghti William Duk of Normandye, As bokes old makith mencioun, By just title and by his cheualrye, Made kyng by conquest of brutes Albyoun... (British Library, Harley 2251)

There exists another fifteenth-century poem on the Kings of England, not by Lydgate, which typically has this for William I: At Westmyster William icrowned was The furst day of Cristemas; A gret thyng after he dude thanne; Made the kyng of Skottys his legeman...

(Bodleian, Ashmole Rolls 21)

However, within this second tradition is a text with a clear example of transposition from the first: This myghtty William duke of Northmandy, That by juste tytill and also by chyualery Conquered this land and king bycome And the kyng of Scotts he made his legeman...

(Bodleian, Bodley 131)

Convergent evolution is when the same change occurs independently in different lineages. Thus, for example an AT base pair might replace a GC base pair at the same position independently in two different species. If this occurs frequently enough, evolutionary tree-building may be misled and species with a large number of convergent changes will be grouped artefactually closely. Convergent evolution is comparable to convergence or parallelism in manuscript traditions (Salemans 1996). So, for example, scribes working independently but in the same geographical area might alter words to fit their own dialect. An example of this might be the substitution of ‘kirk’ for ‘church’ in northern England and Scotland. Thus the same change may happen in two or more manuscripts not as a result of common ancestry, but as a result of having been produced in the same part of the country.





Christopher Howe et al.

. Conclusions The process of incorporation of changes into DNA mirrors the incorporation of changes into manuscripts. For this reason, programs for phylogenetic analysis of sequence data can be exploited for stemmatic analysis of manuscript traditions, and we believe the SplitsTree program has particular advantages. Just as phenomena such as recombination, transposition and convergent evolution may pose problems for the evolutionary biologist, there are closely parallel problems in stemmatic analysis. We hope that the development of techniques in one discipline to deal with these problems will help in their solution elsewhere.

Acknowledgements We thank Matthew Spencer and Barbara Bordalejo for helpful discussions, and the Leverhulme Trust for financial support.

References Barbrook, A. C., C. J. Howe, N. Blake, & P. Robinson (1998). “The phylogeny of the Canterbury Tales”. Nature, 394, 839. Beanland, T. & C. J. Howe (1992). “The inference of evolutionary trees from molecular data”. Comp. Biochem. Physiol., 102B, 643–659. Cameron, H. D. (1987). “The upside-down cladogram: problems in manuscript affiliation”. In H. M. Hoenigswald & L. F. Wiener (Eds.), Biological metaphor and cladistic classification: an interdisciplinary perspective (pp. 227–242). London: Frances Pinter. Felsenstein, J. (1988). “Phylogenies from molecular sequences: inference and reliability”. Ann. Rev. Genet., 22, 521–565. Holmes, E. C., M. Worobey, & A. Rambaut (1999). “Phylogenetic evidence for recombination in dengue virus”. Mol. Biol. Evol., 16, 405–409. Howe, C. J., A. C. Barbrook, M. Spencer, P. Robinson, B. Bordalejo, & L. R. Mooney (2001). “Manuscript evolution”. Trends Genet., 17, 147–152. Huson, D. H. (1998). “Splitstree: a program for analyzing and visualizing evolutionary data”. Bioinformatics, 14, 68–73. Lee, A. R. (1989). “Numerical taxonomy revisited: John Griffith, cladistic analysis and St. Augstine’s Quaestiones in Heptateuchem”. Studia Patristica, 20, 24–32. Medgyesy, P., E. Fejes, & P. Maliga (1985). “Interspecific chloroplast recombination in a Nicotiana somatic hybrid”. Proc. Natl Acad. Sci. USA, 82, 6960–6964.

Parallels between stemmatology and phylogenetics

Mooney L. R., A. C. Barbrook, C. J. Howe, & M. Spencer (2001). “Stemmatic analysis of Lydgate’s Kings of England: a test case for the application of software developed for evolutionary biology to manuscript stemmatics”. Revue d’Histoire des Textes, 31, 275– 297. O’Hara, R. & P. Robinson (1993). “Computer-assisted methods of stemmatic analysis”. In N. Blake & P. Robinson (Eds.), The Canterbury Tales project occasional papers, Vol. 1 (pp. 53–74). London: Office for Humanities Communication Publications. Platnick, N. I. & H. D. Cameron (1977). “Cladistic methods in textual, linguistic, and phylogenetic analysis”. Syst. Zool., 26, 380–385. Salemans, B. J. P. (1996). “Cladistics or the Resurrection of the Method of Lachmann”. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 3–55). Amsterdam: John Benjamins. Voet, D., J. G. Voet, & C. W. Pratt (1999). Fundamentals of Biochemistry. New York: John Wiley & Sons. Wattel, E. & M. J. P. van Mulken (1996). “Shock waves in text traditions”. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 105–121). Amsterdam: John Benjamins. Watson, J. D. & F. H. C. Crick (1953). “Molecular structure of nucleic acids: a structure for deoxyribose nucleic acid”. Nature, 171, 737–738.



Problems of a highly contaminated tradition: the New Testament Stemmata of variants as a source of a genealogy for witnesses Gerd Mink Westfälische Wilhelms-Universität, Münster

. Introduction It is a well-known fact that the textual tradition of the New Testament poses a formidable challenge in the way of textual criticism and edition technique. The number of manuscripts that have come down is large; some 5600 known copies so far, although most of the older ones, in particular, have been lost. The real cause of the problems, however, is the vast degree of contamination. For such a textual tradition, existing methods of reconstructing the tradition are not sufficient and other approaches have to be developed. This study attempts to show a new way (i) of finding and evaluating the genealogical data that can be used to construct a stemma of such texts, and (ii) of constructing a stemma that reflects all genealogical data.1 The concept of coherence (cf. paragraph 4.11) will be essential for the analysis of genealogical relationships. The method is based on two design choices: (1) Instead of trying to start with the construction of overall structures of the relations between witnesses, the first step is to construct local stemmata of variants. Local stemmata consist of trees based upon just one place of variation, and not more than one. If possible, a local tree is constructed for each place of variation.

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



Gerd Mink

(εις εστιν ο νομοθετης) (“one is the lawgiver”) variants: a. και κριτης b. και ο κριτης c. κριτης d. omission local stemma of variants: a James 4:12/10-12

b

και κριτης “and judge” “and judge” “and the judge” “judge”

d c

Figure 1. A local stemma of variants

Following the custom of the Editio Critica Maior (ECM, cf. the next paragraph) the places of variation will be referred to by chapter and verse as well as word address.2 Lower-case letters indicate the variants. In the example from James 4:12/10-12 the best hypothesis is that variant a represents the original text, since the word κριτής (‘judge’) is very important for the author’s argument in the context. The omission in variant d, however, is easily explained as caused by homoioteleuton.3 Variant b occurs in rather unimportant witnesses and only adds the article to variant a.4 Variant c occurs in a single witness only;5 this witness is quite distant from even its closest relatives,6 which are part of the d attestation.7 Even more distant are the relatives that belong to the a attestation. Apparently, variant c is flawed, as the word καί (‘and’), which is in fact indispensable in the context, is missing. The stemma rests on the assumption that variant c is based upon variant d and corrected the omission in accordance with variant a. In that process, καί would have been overlooked, a typical error during revisions. However, it is also possible (hence the interrupted edge from d) that the only basis was variant a, and καί was omitted on account of the similarity of its initial letter with that in κριτής (homoioarkton); yet the closest relatives of the variant c witness (631) are all in the d attestation.8 (2) The contamination in the tradition is viewed as a process. The assumption is that, if contamination occurs, it emerges from those texts which were at the disposal of the scribe, i.e. texts in his direct environment, i.e. texts which are, for the most part, closely related with each other.

Problems of a highly contaminated tradition

James 5:7/45

“The farmer waits for the precious fruit being patient about it” (εως λαβη προιμον και οψιμον) (“until he receives an early and late one” / “until he/it receives something early and something late”) variants: a. (εως λαβη) omission (προιμον και οψιμον) (“until he receives an early and late one” / “until he/it receives something early and something late”) b. (εως λαβη) υετον (προιμον και οψιμον) “(until he/it receives early and late) rain” c. (εως) υετον (λαβη προιμον και οψιμον) “(until he/it receives early and late) rain” d. (εως λαβη) καρπον (προιμον και οψιμον) “(until he receives early and late) fruit” e. (εως λαβη) καρπον τον (προιμον και οψιμον) “(until he receives the early and late) fruit” f. (εως λαβη) και τον (προιμον και οψιμον) “(until he) also (receives) the (early and late one) a local stemma of variants:

c

b

d1

d2

f

e

Figure 2. Local stemma of variants: a more complex case

First of all, another example will demonstrate the importance of a close relationship between witnesses for the construction of a local stemma. After that, the two examples will be used to highlight the different textual developments, so typical of a contaminated tradition, in a given witness and its closest relatives. In Figure 2 the interpretation of variant a is not entirely clear. Is the meaning ‘the early and late fruit’, or something different? The vast majority of witnesses has variant b, resulting in a text of clear meaning. Because so many witnesses have this reading, relationships with witnesses of all variants can be found. Yet, it can be presumed a priori that the oldest layer in the attestation of variant b is to be found in some of its best-known witnesses, 02, 33, 81, 1852. These are more closely related to the witnesses of variant a than to others. 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 in the b attestation. Variant c is not represented in an early layer of tradition. For variant 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 related with each other; nor is there any special connection with the only witness of variant e (01, first hand). Variant d seems to have emerged on several occasions, by introducing the ‘fruit’ from the preceding context into the variant. 1175 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 witnesses have their closest relatives in the b attestation. One would, therefore, trace back their 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 question remains open. The witness of e (01) has its closest relative (03) in the a attestation (the relatives in the b attestation are quite remote). And even if variant f is not grammatically impossible, it is probably an error, which could theoretically have arisen from variants b, d, or e. The closest relatives of the single witness of f (69) are in the b attestation, from which, accordingly, it is being derived. It is a prerequisite for constructing a local stemma like the one in Figure 2 that the relationships between the witnesses as based on their degrees of agreement are known. In addition, one’s provisional assessments of the age and quality of a number of well-known witnesses is taken into account. Close relationships between witnesses alone do not say anything about the direction of textual development. Yet, if in a relationship the ancestor and the descendant can be successfully determined, further possibilities arise to ascertain a local stemma. Thus, in the example the first assumptions were confirmed; additionally, the ancestors of 1175 (variant d1) were successfully located in the a attestation,9 and d1 was accordingly derived from it (see Figure 2). Arguments for the genealogical connections between variants are based partly on their content and linguistic form, and partly on the relationship between their witnesses. The example of Figure 2 makes it clear how these relationships can explain why some witnesses share the same variant and why one variant could arise from another. Yet, a lack of relationship between witnesses can also reveal coincidental correspondences (cf. variant d). Knowledge of the correspondences between witnesses provides a first overview, which must then be examined and supplemented against information about their genealogical relationships. A closer look at the two instances under consideration reveals, in the different attestations of the variants, the traces of contamination among related

Problems of a highly contaminated tradition

witnesses. Witness P74 has 8 close relatives, which contain both instances: 02, 81, 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 4:12/10-12 a d James 5:7/45 a b e

01. 02. 03. 81. 1735. 2344. P74. 218. 1718. P74. 03. 02. 81. 218. 1718. 1735. 2344. 01.

Incidentally, it has emerged that all the close relatives mentioned have more older variants than P74.11 Therefore, P74 is probably not the ‘inventor’ of one of the variants at the two places of variation; on the contrary, its text is based on variants which are found in different close relatives with more older variants, and, consequently, we see the traces of contamination.

.

The Editio Critica Maior of the New Testament

The Institute for New Testament Textual Research in Münster, founded in 1959 by Kurt Aland, has by now accomplished – apart from a few exceptions – the task of collecting the basic material of the textual tradition of the New Testament. Nearly all known manuscripts of the Greek New Testament are now available on microfilm. The next aim was to sift the material in an intelligent way, to let new views about important manuscripts find their way into the minor editions of the institute, and finally to present an Editio Critica Maior, which does justice to present-day standards of knowledge and methodology. A first stage of work on the ECM has been the research, published since 1987, on the texts of the Greek manuscripts and their relevance (‘Text und Textwert’).12 This was intended to separate the majority of manuscripts containing the relatively uniform text which was standard at the end of the Byzantine tradition from the still large number of manuscripts which must be considered relevant on account of their deviations from the majority text. The basis was a collation of all available manuscripts according to a system of places of variation. These so-called test passages (‘Teststellen’) were known for the fact that they were different in the newer and older text forms. Research into ‘Text und Textwert’ made it possible to draw on particularly those manuscripts for the ECM, which do not contain the uniform text from the end of the textual tradition. Nevertheless, 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 five sections: the Gospels, the Acts of the Apostles and the Catholic Letters (which nearly always follow Acts in the manuscripts, making up, together with Acts, the corpus of the so-called Apostolos), the Letters of Paul and the Revelation of John. The Catholic Letters were the first field of research in the ECM project.13

. The Letter of James: Manuscripts and variants The Catholic Letters, starting with the Letter of James, appeared to be especially suitable to begin work on the ECM with. The number of manuscripts examined in the ‘Text und Textwert’ project (552) is of the same order as that for Acts, which have normally been copied together with the Catholic Letters. It is smaller than that for the Letters of Paul (about 750), and a great deal smaller than that for the Gospels (e.g. 1787 for the Gospel of Luke). Nevertheless, the number of witnesses relevant for the textual tradition is in no way smaller. The degree of contamination made the Catholic Letters very attractive for methodological investigation. As the material for the Letter of James was the first to be completely available, it has been researched most extensively. Unless otherwise indicated, the following refers to the Letter of James. As the ECM is mainly interested in the text of the first millennium, we excluded nearly all the uniform witnesses representing the final state of the Byzantine tradition from our study of James, i.e. 371 manuscripts of a total of 535. In addition, 19 lectionaries were selected for the ECM. Yet, I did not use them for my genealogical study because they cannot really be compared with the other manuscripts; they do not contain the full, continuous text, but lessons selected to be read during services. The remaining 164 Greek manuscripts contain the continuous text, with minor omissions occurring repeatedly. Some manuscripts have also suffered more substantial damage, resulting in lacunae occurring throughout the text. Yet, a number of manuscripts is quite badly fragmented: 10 contain less than 150 out of the 761 places of variation, in some of them no more than a couple of these places has survived. The important manuscript 04, at least, contains more than half of the text. The text of James contains about 1740 words; the exact number depends on textual decisions. The selected 164 Greek manuscripts, including the fragmentary ones, present 2132 genuine variants at 761 places of variation.14 Only variants were counted that appear in the first hands of manuscripts containing the continuous text.15 If the variants from Church Fathers and lectionaries and from early versions (Latin, Coptic, Syriac etc.) of the New Testament are

Problems of a highly contaminated tradition

number of occurrences

1000 900 800 700 600 500 400 300 200 100 0 1

2

3

4

5

6

7

8

9

10 11 12 13 14 15

number of witnesses exceeded in an attestation (x10)

Figure 3. Distribution of size of the attestations

included, the total number for these 761 places is 2349.16 Since many of these 761 places comprise more than one word and since the text consists of about 1740 words, it follows that about half the text is subject to variation. The variants include all types, except for type 0 variations.17 Type 3 variations 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 checks revealed about 40% type 4 variations, 30% type 1 variations and 25% type 2 variations. The remainder was made up of type 3 variations. The distribution was atypical in some passages. A single witness is often responsible for the fact that type 1 variations occur more frequently in a major section of the text. On the other hand, longer units of variation in which several changes are mutually interdependent in all probability lead to type 4 variations. The values in Figures 3 and 4 are based on the variants of the 164 witnesses at the 761 places of variation, with each witness being represented by the first hand in a manuscript, and by the text (lemma) in a commentary.18 In Figure 3 the horizontal axis gives the number of witnesses (to be multiplied by 10) which is exceeded in an attestation. The vertical axis tells how often such an attestation occurs (e.g. about 700 attestations cover more than 110 witnesses).19 Figure 3 gives an indication of how the quantity of the attestations is distributed over their total number. The total number of variants and there-



Gerd Mink

450

number of occurrences

400

800

number of occurrences



350 300 250 200 150 100

700

50

600

0

1

500

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21

number of variants

400 300 200 100 0

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20

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 (of up to 10 witnesses) are the most frequent, for there are only 965 instances of more than 10 witnesses. It is not surprising that this result corresponds to a high frequency of very large complementary attestations (more than 120 witnesses, 676 instances). In comparison with that, medium and large attestations make up only a relatively small part (e.g. there are only 128 instances of 41–120 witnesses). Figure 4 (top) shows how many variants (horizontal axis) occur in how many places of variation (vertical axis). In Figure 4 (bottom) the horizontal axis shows the number of variants which is exceeded in one place of variation, and the vertical axis gives the number of places of variation with this number of variants. Thus, while at the top it is shown that there are about 125 places with 3 variants,20 the lower diagram demonstrates that in about 150 places there are more than 3 variants.21 More than half of the places of variation have only 2 variants. More than 6 variants occur in 25 instances, more than 10 in only 4. A comparatively large part is made up of places of variation with 3 to 6 variants.

Problems of a highly contaminated tradition

Letter of James about 1740 words 535 Greek manuscripts available 164 manuscripts selected 761 places of variation 2132 variants of the 164 manuscripts 143 cases identical attestation 800 occurrences 104 cases identical one witness attestation 685 occurrences 39 cases of identical more-than-one-witness-attestation 115 ocurrences

Figure 5. The Letter of James; some numbers

The existence of 59 places of variation with only one variant is caused by deviating variants stemming from lectionaries, Church Fathers, or early versions which are assumed to be based on Greek exemplars at these places, but cannot be traced back to the first hands in the 164 manuscripts. The following numbers are based on the same data as Figure 3 and Figure 4: the variants of the 164 witnesses in 761 places of variation. If corrections, marginal variants, variants in commentaries, and the evidence from lectionaries, Church Fathers and early versions were included in the analysis, not only the total number of variants would be much larger, but the number of occurrences of identical attestations would also be smaller. The 2132 variants represent 2132 attestations of which 1332 are unique, i.e. there is no other attestation containing only the same witness or the same combination of witnesses. These unique attestations include those in the 59 places mentioned above.22 The chances of finding identical attestations among these are slight, as in each of the 59 places a number of witnesses is lacking from the 164 (due to fragmentation, unmotivated omission of larger passages or the like). There are 143 different cases in which identical attestations occur more than once. The sum of all the occurrences of these 143 cases is 800. Out of these 143 cases of identical attestations 104 concern attestations comprising only one witness. The sum of all their occurrences is 685. The 39 remaining cases relate to multiple-witness-attestations. The sum of their occurrences is 115. Of these 39 cases, 12 concern large complementary attestations in two neighbouring





Gerd Mink

places, either to only one Greek witness differing from the entire remainder of the textual tradition, or to witnesses from the field of lectionaries, Church Fathers or early versions. These 12 cases correspond to 24 occurrences. When the 12 cases are subtracted from the 39 cases of multiple-witness-combinations, two-witness-combinations are typical among the remaining ones (21 out of the 27 cases): they occur up to 6 times, typically twice or 3 times. Even wellknown pairs of manuscripts participate in two-witness-attestations only 5 or 6 times. There are only 6 cases (apart from the large complementary attestations mentioned above) in which more than two witnesses are combined. The richest 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 genealogical patterns of the textual history of this tradition, we cannot base our research on identical combinations of witnesses, since they play only a very modest role. Typically, attestations do not have duplicates containing the same witnesses. The number of type 2 variants, the corner stone in the approach of Salemans (2000), is too small. Secondly, the James text is highly contaminated. It is therefore unavoidable that we focus our research on the analysis of contamination.

. Contamination as a process In a dense tradition, it is typical of contamination that a witness shares most of its variants with its closest relative and if it deviates from this relative the variants concerned can be found in other close relatives. In the text of James, contamination is the result of small steps. That these steps are small is visible only if the number of witnesses of the tradition that have been preserved is large. However, where the proportion of witnesses that are not preserved is high, contamination does not appear to be the result of small steps, as so many intermediate witnesses are missing. If the density of a tradition is very high (as it was in the middle ages from the 11th century onwards), nearly all the witnesses have very close relatives. The agreement values are typically high, between 94% and 98%, even if the most uniform Byzantine witnesses are excluded. This implies that typically only 15–45 places of variation saw a change during the step from one preserved witness to the most closely related one that has also survived. Contamination in this context occurs in very small steps, and the steps would be even smaller if all the manuscripts had been preserved. The pattern is as follows: the two or more manuscripts involved in a copying process are among those that are most closely related,23 and even if they are lost,

Problems of a highly contaminated tradition 20

number of manuscripts

18 16

018 020 025 044? 049 033

14 12 10 8

P100? 01 03

6 4

02 04 048 P74

P100?

2 0 1

2

3

4

5

6

7

8

9

10

century

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 the witnesses that have been preserved. It is rather unlikely that the comparatively few manuscripts to survive from the 9th century and before are representative of the totality of the manuscripts of that time; a considerable number must have been lost. Figure 6 shows the manuscripts containing James and written in the first millennium that are not too fragmented.24 Although we do not know how many links are missing between them, we can be sure that a part of the variants found only in the later manuscripts is a reflection of older variants which have not been preserved in the surviving manuscripts of the first millennium. Indeed, for this period we must assume a vast number of missing links. Therefore, the nearest preserved relatives of a witness are not as closely related to it as is the case in later times. As a consequence, agreement values are lower than for the later manuscripts, with values of more than 90% being already quite good. Contamination in this context looks more radical, because of the many missing links. All those very closely related manuscripts which a given scribe might have consulted have disappeared. At first sight, contamination here appears to have had a much greater impact than in the later manuscripts, but it may still be assumed that among the older textual tradition, too, it has followed the normal pattern: the manuscripts of a given copying process must have been among the most closely related ones. But in this case they have all disappeared. We will, however, find a 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 is important to make a distinction between manuscripts and their texts. It is impossible to find the genealogical relationship between manuscripts if most of them have not survived. It is merely possible to uncover the genealogical structure of the preserved texts. In this context, therefore, the text is the witness and not the manuscript.25

. A brief outline of the method: Some definitions It may be helpful to present a brief outline of the method I have developed,26 as far as it is relevant for the present subject, for it is quite distinct from the general practice of editors. Nearly all New Testament textual critics accept that the stemma of a tradition contaminated to such a degree as the Letter of James cannot be reconstructed. However, the question of what is the original text for a given passage is widely discussed, independent of what the overall stemma might look like, and every editor has to decide what text he will provide. Willingly or not, he must make genealogical decisions concerning the individual places of variation, as soon as he settles for one of the variants as being the original text at these places. The objective of my method is a comprehensive theory of the structure of the textual tradition, with special regard to the problems of contamination and the coincidental emergence of variants. The result will not necessarily be a stemma in the traditional sense of a graphic representation. I conceive of the stemma here in a more complex sense. It does not only connect witnesses; the data representing the contents of those connections and the data designating the quality of those connections predominantly determine the stemmatic hypothesis. Usually, a stemmatic representation only displays information about the direction of development between two witnesses. Yet, the underlying data are the real substance of a stemma. They are, in a more complex sense, the nucleus of the stemma.27 Graphical representations can show no more than aspects of this complexity. My method has three elements: observations, assessments of what was observed, and structural connections based on these observations and assessments, according to rules based on a model of manuscript tradition. These rules have to be in harmony with a model of the average situation in the history of transmission of a given text. Such a model is necessary in order to have criteria to assess which results are more probable than others. The model de-

Problems of a highly contaminated tradition

pends on what we know about the particular ways of transmission, the habits of 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. “Some essential assumptions are considered more probable than their contrary (without excluding that the contrary might happen at some points): (i) a scribe wants to copy a manuscript with fidelity; primarily the scribe does not want to create new readings; (ii) if the scribe introduces other readings,28 they come from another source (normally a manuscript); (iii) if the scribe uses more than one source, few rather than many sources will be used; and (iv) the source copies have closely related texts rather than less related ones.”29 A comprehensive theory results from the total of observations and assessments and allows us to examine the plausibility of each assessment and of each textual decision in the light of an overall view. This is very important for the editors of the text, for there is a circular argument typical of textual criticism: Witnesses are important for reconstructing the initial text, and they are important because of the high number of agreements with the reconstructed initial text. In other words: witnesses are good because of their good variants, variants are good because of their good witnesses. This circle cannot be avoided, but it has to be controlled. We need a method, therefore, which can provide an overall view of the consequences of all the decisions we take, so that also the overall plausibility of what we are doing can be examined.30 In the present method this is done through an iterative process, especially designed to perform this examination. At first, I would like to explain some central concepts of the method, which itself will be outlined subsequently. . Initial text (A) The initial text is a hypothetical, reconstructed text, as it presumably existed, according to the hypothesis, before the beginning of its copying. In a hypothesis, which wants to establish the genealogical relationship between the witnesses, the initial text corresponds to a hypothetical witness A (‘Ausgangstext’). The initial text is not identical with the original, the text of the author. Between the autograph and the initial text considerable changes may have taken place which may not have left a single trace in the surviving textual tradition. Even if this is not the case, differences between the original and the initial text must be taken into account. On the other hand, the initial text is not simply a reconstruction on the basis of the surviving variants, which best explains the emergence of the variants





Gerd Mink

and thus represents the archetype of the tradition. Instead, several hypotheses are possible about the beginnings of the tradition. The simplest working hypothesis must be that there are no differences between the original and the initial text (except for inevitable scribal slips). In that case, the reconstruction of the initial text is not only determined by the subsequent tradition (which text form could have been derived from which?), but also by the author’s intentions as they come to light in the totality of what we know about him (is a variant 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 and the initial text, who might possibly have merged several writings of an author into one. Or there may have been more than one initial text, possibly even going back to more than one autograph if, for example, the author issued several versions of his work. I should also like to point out in this context that the notion of text needs a more exact definition. In textual criticism it is mostly used for what ‘was written down’ in the manuscript, strictly speaking for the sequence of characters the copyist wrote down. On the other hand, the text should convey a meaning. This aim, however, is not directly achieved by the written characters. They represent different things in different languages. A copyist may grasp the sense of a text and copy this sense. Even if he intends to copy the text true to the letter from the exemplar, he may, at the level of the characters, produce variants which emerge from the meaning of the text. But a copyist may also, particularly in long or difficult sentences, copy groups of words, or line by line, while losing the overall meaning of the text. In the case of very difficult words he might even copy character by character, which may lead to readings which hardly make any sense, and even to spectacular errors. Now, the problem is that there is no obvious difference between variants introduced on purpose by copyist, and variants produced accidentally, as long as these variants are meaningful in the widest sense. The text as carrier of meaning is therefore modified by the variants – introduced intentionally or unintentionally by the copyist –, and there is no way of differentiating between intentional and unintentional variants. If, however, a copy contains errors, i.e. readings which clearly do not make sense, these do not necessarily modify the text as a carrier of meaning. The copyist had no intention of changing the text. The copy therefore only contains an erroneous representation of the same text at the level of the characters, which nevertheless sometimes renders it unrecognisable as a carrier of meaning. But not only errors at the level of the characters are possible. After all, any linguistic element can be affected by mistakes of this sort, such as errors of

Problems of a highly contaminated tradition

concord and case, also omissions of key words for the context, which happen accidentally and against a scribe’s intention (lapsus). Even the first exemplar of the entire textual tradition, even the very autograph, may have contained such errors. The text as the actual carrier of meaning, however, would not necessarily be affected by these errors, it would only be represented defectively. Consequently, the ECM records the witnesses of erroneous readings as witnesses for the variants which they represent, albeit defectively. There is even an example where the best witnesses omit a negation (2 Peter 3:10/48-50). Although the preceding passage speaks of the passing away of the heavens, and the dissolution of the elements, and the following verses presuppose the dissolution of heaven and earth (for a new heaven and a new earth are waited for), quite superior witnesses here have the reading ‘the earth and all the works that are therein will be found (εὑρθήσονται)’,32 when logic demands ‘will not be found (οὐχ εὑρεθήσονται)’. The meaning, as a result, is extremely problematic; to my mind the reading does not make sense and must therefore be erroneous.33 Unquestionably, the hyparchetype of all these witnesses did not have the negation. Now, there are two variants (ἀφανισθήσονται ‘they will disappear’,34 and κατακαήσεται ‘they will be burned up’),35 which presuppose and express more graphically a text containing the negation: οὐχ εὑρεθήσονται ‘they will not be found’. Although it is not preserved in any Greek manuscript, it is probable that the initial text had the negation. Even if these variants which indirectly confirm the negation did not exist, the assumption should still be that the initial text contained the negation required by the sense of the text, even though the negation is not in the graphemic representation of the archetype.36 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 two different variants occur in a place of variation; the maximum in James is 24 variants. A place of variation may comprise more than one word, but it can also be the space between words. Ideally, it covers a logical unit of variation. This means that mutually interdependent changes to a text should belong to one unit of variation (e.g. if a subject and correspondingly the predicate are put in the singular). A unit of variation can also be postulated when a group of words presumably belonged together in a copyist’s view (e.g. if a word group consisting of article/particle/noun shows changes in different combinations for the article/noun and for the particle). Sometimes, very pragmatic considera-





Gerd Mink

tions might be adduced to determine a unit of variation, so as to enable the comparison of all texts at a certain place. Places of variation may also overlap. In one place of variation the question may be e.g whether a rather large group of words has been omitted or not; yet another instance of variation may result from variants within that group of words whenever it was not left out. The number of variants offered by the first hand of a witness corresponds to the number of places of variation which have been preserved in the witness. . Variants vs. readings; connective variants “A reading is the generic term for the wording of a textual unit in which a manuscript is distinguished from one or more or from all other manuscripts. A variant refers to one of at least two readings of the same textual unit which is grammatically correct and logically possible. Errors are readings which do not fulfil these criteria .”37 Errors are usually deemed as the variant they represent incorrectly. In some cases, if the corresponding variant is no longer available, the error has to be corrected and incorporated as another variant. It is theoretically possible that new variants originate specifically from an error. In practice, however, at least in the New Testament tradition, this seems to be rare, as the errors are usually corrected into the underlying variants in the further tradition. “Alternative and orthographically possible forms of the same variants are classed as orthographica.”38 Thus in the verb λαμβάνω (‘I take’) the μ, which is really a formative element of the present tense stem, in Koine Greek often gets infixed in other stems. The future form, apart from the actually correct λήψομαι, is frequently λήμψομαι. The two forms are treated as equal and interchangeable. Or the particle ἄν indicating a prospective meaning after relative pronouns and relative adverbs, is replaced by ἐάν in later Greek with the same function and position – a phenomenon caused by the fact that in classical Greek instead of ἐάν (‘if ’) people also wrote ἄν. Any scribe could adapt these things to his usage, without affecting the quality of the copy. Certain morphological divergences are considered equivalent to orthographica. They have it in common with actual orthographica that in the copyist’s view they were also interchangeable, without affecting the quality of a good copy, e.g. if instead of a strong aorist (e.g. ἐγενόμην ‘I became’) he would use a mixed aorist (ἐγενάμην). Often the notion of ‘reading’ is used in a more general way, it is frequently taken to mean ‘variant’. It is very common to speak of an original reading, from which variants were derived in the course of a text’s transmission. Yet,

Problems of a highly contaminated tradition

neutrally speaking, even the reading which is held to be part of the original text, is nothing more than one variant in the textual tradition. For every textual tradition it is necessary to determine – in accordance with language, historical period and literary genre – which readings are to rank as genuine variants. Only variants are the basis of a genealogical relationship of witnesses. If a variant contributes to genealogical coherencies (see paragraph 4.11), it is called a connective variant of the witnesses concerned. Their agreement in this variant is assumed to be not coincidental because (i) they usually have variants in common or (ii) the character of the variant argues against multiple independent emergence (see paragraph 6). . Witness vs. manuscript A manuscript is the physical carrier of the text. A manuscript has properties which can be defined paleographically and codicologically. The text in a manuscript may be considerably older than the manuscript itself. At the very latest the text was produced at the same time as the manuscript. The witness of a variant is the text, not the manuscript.39 A one-witness-attestation is the attestation of a variant found in the text of one manuscript only. A two-witnessattestation is the attestation of a variant found in the text of two manuscripts. Whenever numbers are cited in this study, which are generally used to denote manuscripts, they are used here to designate the texts transmitted in them, not the manuscripts as their physical carriers. The hypothetical witness A represents the hypothetical initial text (see paragraph 4.1). . Attestation Attestation is the total of all the witnesses presenting a certain variant at any one given place of variation. Consequently, the number of variants equals the number of attestations. . Local stemma vs. global stemma A local stemma is a stemma representing the presumed genealogical relationship between variants at one place of variation. It is a key notion in this study. A global stemma is a stemma representing the genealogical relationships between witnesses. It illustrates the overall genealogical hypothesis. A global stemma can 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 variation according to the relationships between their variants, as represented in the local stemmata. It must convey the genealogical coherencies (see paragraph 4.11) of the attestations in any one place of variation correctly. It must reflect all the changes between witnesses and all the non-coincidental correspondences in every place of variation. Meeting these conditions a global stemma is true. A global stemma is the superset of all the optimal substemmata, i.e. it consists of all their nodes (including any intermediary nodes) and edges. It is possible to have more than one global stemma. The number of global stemmata depends on the number of cases in which there are two or more substemmata of equal likelihood, and on the number of alternative substemmata in each case. For any one set of substemmata, containing not more than one substemma per witness, however, only one global stemma is possible.40 . Intermediary nodes Normally, the nodes in a stemmatic graph stand for witnesses, whose genealogically relevant contents are made up of all their variants. Witness x and witness y are immediately connected by an arrow (= a directed edge).41 An intermediary node does not represent a witness, but signifies only a subset of the variants of witness x and/or witness y which are to be connected with each other. It forms part of the connection between witnesses x and y. On the purpose of intermediary nodes, see paragraphs 7 and 8. . Optimal substemma A substemma is part of a global stemma. It links one descendant with its hypothetical ancestor or ancestors (sources of contamination). A substemma is optimal if the smallest possible number of ancestors can explain all variants of the descendant. The ancestors have to exhibit a very high degree of genealogical coherence (see paragraph 4.11) with the descendant, in order to exclude coincidental correspondences. Consequently, they are to be looked for among the potential ancestors (see paragraph 4.10). If at a place of variation a descendant corresponds only to a witness of lesser coherence, the nature of the variant must be used to verify whether the variant in question actually links the two witnesses, or whether the correspondence of variants is only a matter of coincidence. There are, however, two cases in which a connection with a non-ancestor is effected by way of an intermediary node. (i) A connection between a de-

Problems of a highly contaminated tradition

scendant and a non-ancestor may be compulsory if the latter offers the only possible prior variant at some place of variation, although it is not a potential ancestor (for the full issue, see paragraph 7). If this connection were left out, not all the variants of the descendants would be explained correctly, i.e. in accordance with the local stemmata. The substemma would not be optimal. (ii) The agreements between witnesses with a high degree of undirected genealogical coherence (see paragraph 4.11) are not covered by stemmata based on potential ancestry. The connection between such witnesses can be effected by way of intermediary nodes (or undirected edges). Such connections must also be integrated into the optimal substemma (for the conditions of this see paragraph 8). An optimal substemma is true if it is compatible with the genealogical coherencies (see paragraph 4.11) of the attestations in any one place of variation and, therefore, reflects all the changes between the witnesses concerned and all their non-coincidental correspondences. . Prior vs. posterior The local stemmata represent a hypothesis about which variant arose from which. There, each source variant is prior, the one developed from it posterior. In the comparison of pairs of witnesses the proportion of prior variants and posterior variants plays a decisive role. One of the key-questions to be asked is, in how many places does witness x have a prior variant from which the posterior variant of witness y was derived, and in how many places the opposite is true. The result reveals witnesses that are predominantly prior or predominantly posterior. . Potential ancestor Potential ancestors are all those witnesses which show a higher proportion of prior variants than posterior variants in comparison with a given witness. Consequently, this also includes witnesses not represented in a stemma or substemma as ancestors, because they are not needed to explain the variants of a descendant.





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 within the entire tradition.42 For all the witnesses closest relatives can be found. Between a witness and its closest relatives there is the highest degree of coherence. These closest relatives in turn have their own closest relatives, so that chains of coherencies develop. Thus the particular coherency within each pair of witnesses, within groups and also within the attestations of variants can be evaluated. This way, coherence – represented by coherencies – can be analysed at each place of variation as well as in the entire tradition. Coherencies between witnesses may be qualified as good or poor. Whenever a coherency is mentioned without any qualification, it refers to a useful coherency, i.e. one that is high enough to indicate a closer (genealogical) relation. The absence of coherence consequently means that there is no useful coherency, because it is too poor. The assessment of the quality of the coherency between witness x and witness y depends on the coherencies pertaining to the immediate genealogical environment of witness x and y. The relevant factor in the assessment of the quality of the coherency are the percentage values of the agreements between each of these witnesses and its closest relatives, and not the absolute values, as the number of comparable places of variation changes from pair to pair of the witnesses to be compared. A witness can very effectively be characterised by the percentage of the highest agreement (with its closest relative) and the manner in which the values of agreement (with other relatives) decrease from there. A witness may, for example, have so many individual variants that its text differs very often from the main body of the tradition. In such a case, the rate of agreement with the closest relative may perhaps be only 89%–87%,43 which under these circumstances would still be useful values. A similar situation may occur if the text of a witness was compiled from other texts, which are themselves not very closely interrelated. The case is different if none of the close relatives of a witness have survived (especially so, if a witness is based on the oldest layer of the textual tradition). In such cases one may have to fall back on the range of 92%–89% agreement to discover traces of the lost relatives. If a large number of relatives has agreement rates in the range of 96%–93%, however, a percentage of 89% in another witness is unlikely to yield more usable coherencies, as it is not likely that connective variants will be found there that are not to be found in the large number of closer relatives.44 This holds true unless variants which cannot have developed independently of each other compel one to take lower coherencies into account.

Problems of a highly contaminated tradition

If the assessment of the relationship between witnesses is done only on the basis of their agreements, it is classified as pre-genealogical coherency. This type of coherence is a criterion for determining the probability whether the witnesses allow to take into account a genealogical relationship between variants. Normally, a lack of pre-genealogical coherence within an attestation implies a coincidental multiple development of conform variants. If it is possible to include genealogical data (from the local stemmata) in an assessment of the relationship, it can be referred to as genealogical coherency.45 Here, two points are essential: (i) Based on the number of prior and posterior variants (cf. the local stemmata), the relative positions of the witnesses in the general textual flow (see paragraph 4.12) determine a predominant direction of textual flow between the witnesses (directed coherency – the normal situation). Directed coherencies determine which witnesses are potential ancestors of another witness. (ii) In addition, the rates of correspondence are important in estimating the probability of stemmatic coherencies between pairs of witnesses. Genealogical coherence presupposes pre-genealogical coherence. A special case is the undirected genealogical coherency. It exists between closely related witnesses, which accordingly have a very high pre-genealogical coherency but there is no predominant direction of textual flow. An undirected genealogical coherency is different from a pre-genealogical coherency on account of the fact that in the former case the basis are pre-genealogical and genealogical data. Stemmatic coherence is found in the global stemma and the optimal substemmata. A given stemmatic coherency designates the ultimate stemmatic relationship in an optimal substemma (see paragraph 4.8). Stemmatic coherence presupposes genealogical coherence or at least one connective variant (cf. paragraph 4.3). . Textual flow: General/particular, global/local The genealogical relationships between witnesses are a reflection of the development of the text. There is a textual flow connecting the witnesses. The general textual flow leads from earlier to later textual states. Consequently, each witness can be assigned a relative position compared to any other witness within this general textual flow. A particular textual flow exists between witnesses in the relative positions of potential ancestor and descendant. Here the textual flow is determined by the variants the descendant shares with the ancestor, and by the variants of the ancestor from which new variants in the descendant evolved. They determine the direction of the textual flow. The particular textual flow





Gerd Mink

between ancestors and a descendant becomes part of the global textual flow when ancestors and descendant in a substemma are part of a global stemma. By contrast, local textual flow between ancestors and a descendant, is effected at one place of variation, on the basis of the relationships between their variants in the local stemma and the relationship of the witnesses according to the particular or, if possible, global textual flow between them. Every local textual flow must be reflected in the global textual flow, and vice versa. The local textual flow within the attestation of a variant or for an entire place of variation may be represented in textual flow diagrams. These may be based on genealogical or stemmatic coherencies.

. Procedures It is of central importance to pay attention to coherencies and to the textual flows which rest upon them. The following insight is fundamental for the coherence-based genealogical method: In a textual tradition where all the copies have survived and where the source, or (in case of contamination) the sources, are also known, as well as the origin of every reading in every copy, the genealogical interrelationships between all the variants at any place of variation must appear in a global stemma of the witnesses as genealogical relationship between coherent fields of relationships between witnesses. Conversely, the relationship between each descendant and its ancestor, or (in case of contamination) at least a subset of relationships between it and its ancestors, should appear at any place of variation, namely as the relationship between witnesses sharing the same variant there, or as the relationship between witnesses between which a change of the variant took place which supports the ancestor-descendant relationship. Since every descendant may be an ancestor in relationship to other witnesses, chains of coherencies are formed within attestations. These chains help to find out about unique or multiple emergence of variants. Chains of coherencies connect attestations of different variants where a change of text took place between witnesses.46

Consequently, the places of variation are the points of departure for discovering, by way of the relationships between variants, something about the relationships between witnesses.47 The procedures on which my genealogical studies are based are briefly summarised below.48

Problems of a highly contaminated tradition James 3:12/22-30 a

b

?

c

o

j

h1

d

e

h2

f

g

i

k

l

m

Figure 7. A local stemma of variants

. Construction of local stemmata Local stemmata of variants must be created, including, where possible, the variant that is assumed to be the initial text. This procedure makes use of all the methods and knowledge provided by textual criticism. Two examples have been set out in the introduction. If it is not possible to establish a local stemma, the place of variation does not qualify for further evaluation until a solution has been found. It is not necessary to enter all the variants into the local stemma at all costs. Further evaluation requires clear statements. Either a variant is designated as genealogically dependent on another variant (or on several, if they explain the variant together, e.g. by merging), or the origin of a variant is designated as uncertain. In the latter case the variant is not taken into consideration for further evaluation, unless it should be possible to make a clear statement about it later. All cases in which no definitive statement is possible must remain neutral in the further evaluation. Figure 7 shows the possibilities within a local stemma; the letters represent the variants. h1 and h2 indicate that the coherency of the attestation of variant h is imperfect (see below and paragraph 5.3) and therefore the variant is assumed to have emerged twice, d is considered a mixture of a and h1, and the question mark indicates the questionable source of a variant. Variant n does not appear in the stemma, as its only witness is a lectionary. The underlying type 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 in brackets): 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 be used because the relevant text is missing from them or uncertain. When local stemmata are constructed – and this is very important – pregenealogical coherencies of witnesses within an attestation or between attestations must be verified. Pre-genealogical coherencies are based only on the agreement of the witnesses and do not provide genealogical information. If not all the witnesses cohere like a chain or a net, the hypothesis is that the variant has 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 attestation of another variant, a genealogical relationship between the respective variants is improbable. . Analysis of textual flow The genealogical links between variants become an integral part of the data sets which contain all genealogical information pertaining to each variant, see Figure 8. Figure 8 is a screenshot of a database section; the first seven fields are address fields; the labez field contains the letter addresses of variants and is up-

Problems of a highly contaminated tradition

dated in labneu for genealogical reasons (for instance two variants are treated as one or one as two);49 the fields Q1, Q2, Q3 contain the letter addresses of the source variant(s) of the variant in the Labneu field; the fields Vq1, Vq1a, etc., Vq2, Vq2a, etc. contain the letter addresses of the variants that are regarded as source variants of the variant in the fields Q1, Q2 etc. Consequently, the witnesses do not only concern variants, but variants which are genealogically linked to other variants and their attestations. These direct links between witnesses give a first idea of the general textual flow throughout the tradition by providing some basic information on the genealogical relationship between witnesses. Above all, we can see which witnesses may be potential ancestors of a given witness, and the particular textual flows connecting them. The general textual flow corresponds to the development of the text (i.e. the variants) throughout its history. This development can be demonstrated at every passage of the text in local stemmata of variants. There are different aspects of this textual flow: a general genealogical one and stemma-orientated aspects, cf. paragraph 4.12. The general genealogical aspect concerns two questions: What may the position of a witness in the general development be, and which role does it play in the general textual flow? The position of a witness within the general textual flow, compared with the position of another witness, can be determined by the ratio of two values. The first value (cf. for instance the xausy fields in Figures 21, 23, 35) is the number of variants that are posterior to those of the witness compared, the second one (the yausx fields) is the number of variants that are prior. These values also determine whether a witness is a potential ancestor of another (cf. paragraph 4.10). The stemma-orientated aspects concern the particular textual flow and the global textual flow between witnesses within a stemma or substemma. If there is a genealogical hypothesis on the relation between two witnesses, it is based on the agreements and the direction of the textual flow between these witnesses.50 A witness is a potential ancestor of another one if the textual flow runs predominantly from that witness to the other one. The potential ancestors are classified according to their degrees of agreement. In Figures 9, 21, 23 and 35, we find a witness under the heading of Zeuge1 the potential ancestors of which are to be determined. These can be found under the heading Zeuge2 if the arrow in the Richtg field is pointing to the left.51 The more a potential ancestor agrees with its potential descendant (cf. the Proz1 field),52 the more probable is a direct genealogical relationship (= stemmatic coherency) in the global stemma.53 A direct 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 ancestors of 2412

as an intermediary witness. Lists like those in Figures 9, 21, 23 and 35 allow for an assessment of the probability whether a potential ancestor will become an ancestor in a stemma or substemma and whether global textual flow will arise in consequence. . Analysis of genealogical coherencies at places of variation Now an analysis of the genealogical coherencies within one attestation and between attestations is possible. If close genealogical coherencies connect the totality of the witnesses without any break the supposed coherency of an attestation is perfect. The hypothesis is: the variant has evolved only once. An example of this is to be seen in Figure 10. The local stemma of variants outlines the hypothesis that b derives from a. In order to verify the genealogical coherencies within one attestation, the potential ancestors of every witness in the attestation have to be determined. Every witness is thus assumed to be a descendant. If witnesses are found which offer the same variant among the potential ancestors which possess a sufficient degree of agreements, the ancestor with the highest degree of agreement is chosen because it makes a direct genealogical relationship more probable. Whether a degree of agreement is high enough is to be read from the way the values decrease (cf. paragraph 4.11).54 In Figure 9 on the left, for example, values of > 89,5% have to be considered as sufficiently high, as 1243 offers a relatively old text form (which follows from the range of witnesses in the Zeuge2

Problems of a highly contaminated tradition

a

stemma:

perfect coherency

b variant a

witnesses

1

2

3

4

variant b

witnesses

5

6

7

8

Figure 10. Example of perfect coherence

field) and, therefore, many more closely related potential ancestors have been lost. In Figure 9 on the right, even a value of e.g. 92,5% is inadequate, as quite a number 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 variants which may indicate a genealogical coherency are likely to be found.55 The object of the process is to arrive at evaluations of the local stemmata of variants through evaluation of the genealogical coherencies. The quality of genealogical coherencies actually allows for rather accurate predictions about the probability that the genealogical coherencies will result in stemmatic coherencies (cf. paragraphs 4.11 and 5.5). When genealogical data yield undirected genealogical coherencies of witnesses (cf. Figure 9, 21, 23, and 35, where the Richtg field is empty), these coherencies will also lead to coherencies in the attestations if they are sufficiently high (cf. paragraph 8). If the procedure has been completed for all the witnesses of the variant and the result is a chain or net of genealogical coherencies ranging over the whole attestation, the coherency of the attestation is perfect. The witnesses are all genealogically interrelated. The variant did not emerge repeatedly by chance. In Figure 10 the arrows for variant b in each case point from the potential ancestors towards the descendants. For witness 5 no potential ancestor was found in the attestation of variant b. All the other witnesses are only eligible as descendants there. Witness 5 should therefore be the oldest text to display this variant. Its potential ancestor must be identified in a different attestation. In the example it should be witness 3 for variant a. This should be the genealogical coherency between variants a and b. If there is no potential ancestor for witness 5 with variant a, but for example with a variant c, this local stemma of variants must be false. (The other witnesses for a have not been assigned ar-





Gerd Mink

imperfect coherency

stemma:

a c

b1 b2 variant a

witnesses

1

2

3

4

variant b

witnesses

5

6

7

8

variant a

witnesses

9

10

11

12

Figure 11. Example of imperfect coherence

rows, as the example only concerns the perfect coherency in variant b and the connection with variant a). The coherency is imperfect if not all the witnesses of the attestation can be connected by close genealogical coherencies. The hypothesis is: the variant has emerged repeatedly. Logically, we have two or more variants with coincidentally the same text. In Figure 11 the examination of the genealogical coherencies by the method described resulted into two groups in the attestation of variant b. Witnesses 5 and 6 have a coherency, and so have 7 and 8. There is, however, no further genealogical coherency for witness 5 in the attestation of b. The same is true for witness 7. The coherency of the witnesses of variant b is imperfect. Witness 5, for example, has its closest potential ancestor with variant a, witness 7 with variant c. Variant b, consequently, emerged twice. It is therefore split into b1 and b2.56 Another case of imperfect coherency, different from the one in Figure 11, is not crucial for the first analysis. In Figure 12 the direct ancestors of witnesses 5 and 7 are both to be found with variant a. Here the local stemma of variants would be the one of Figure 10. With the aid of data as found in the examples in Figure 9 the genealogical coherencies of the attestations of variants can now be drawn up and assessed in textual flow diagrams. The directed genealogical coherencies will form the basis, as they are the only ones giving direction to the textual flow. If they are not of sufficient quality, undirected genealogical coherencies must be used but as a surrogate only, or fictitiously imperfect coherencies may materialise.

Problems of a highly contaminated tradition

imperfect coherency

a

stemma:

b variant a

witnesses

1

2

3

4

variant b

witnesses

5

6

7

8

Figure 12. Example of imperfect coherence

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 under Proz1),57 which make it possible to judge whether there is sufficient genealogical coherence.58 A textual flow diagram now establishes the connections between witnesses and their most closely related potential ancestors within the attestation, by means of directed edges (e.g. Figure 13). It is also possible to show by which coherencies the attestations of different variants are connected (e.g. in Figures 14, 15). In Figure 13 such a textual flow diagram is presented. It shows an example of perfect genealogical coherence within an attestation.59 The variant concerned qualified as lectio difficilior. The argument ‘lectio difficilior’ is generally used if the variant is regarded as so difficult that a copyist could not have invented, but only simplified it. If a variant is qualified as lectio difficilior, it is rather uncredible that it could have emerged more than once. In the case of 1 Peter 1:6/18 there are essentially two variants, a and b, which may present the original text. Three further variants derive from them. Variant a shows a participle (λυπηθέντας) in the accusative. The participle correlates to a remote accusative (ὑμας) in 1 Peter 1:4/24, although only slightly earlier, in 1 Peter 1:6/6, the persons involved are the subject. Variant b relates the participle to this subject and correctly puts it in the nominative (λυπηθέντες). Without any doubt, variant a constitutes the lectio difficilior, and the change to b is natural and obvious. So far, variant b has been considered as the original variant, as variant a was not believed to be correct because of its attestation. Figure 13 shows the attestation of variant a. As the data necessary for the assignment of genealogical coherencies for 1 Peter are not yet completely available, and were largely lacking at the time when the text was established, it was





Gerd Mink A

genealogical coherency of the attestation of 1 Peter 1:6/18 variant a (accorging to the genealogical results in James)

01 1175 1243

2492

17352

1852

2344

1448

429

307

206 522 1292 1490 2200 1799

614

1831

453

630

720

918

1678 24644 2818

6234 2805

94

2412 4243

01422 0204 63 69 884 3264 3983 4422 467 629 642 9154 9965 15244 21476 25444 1047 18422 1838 621

61

1848 218 808

1661 254

25412 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 pregenealogical data already known for 1 Peter,60 as they are based exclusively on the agreements. The examples in Figures 13, 14 and 15 make it very clear that thereby a good prediction for the genealogical coherencies was also possible for 1 Peter. For each witness in Figure 13 the most closely related potential ancestor within the attestation was looked for. A, the hypothetical initial text, for exam-

Problems of a highly contaminated tradition

ple, is the closest potential ancestor of quite a number of witnesses (01, 1175, etc.). In principle, it is therefore possible to consider this variant for the original reading. There are also coherent chains (e.g. A – 1852 – 1448 – 429 – 206 – 1799) ranging over the complete attestation. If no superscript is given after the number of the witness, it means that the potential ancestor is the one with the highest degree of agreement altogether (not only within the attestation), and therefore has a very high probability of being ancestor in a global stemma (level 1). This is the case for the vast majority. The superscript numbers indicate lower levels. Of these, levels of 2 to 5 may – generally – still be considered as good, depending on the proximity of the levels. The minimum here, a level of 7 for witness 104, is still acceptable, since quite a number of potential ancestors with almost the same level exists for this witness. It is striking that the superscript numbers predominantly appear with witness 424 and the series below it (0142, 020, etc.), thereby reflecting an area of especially vigorous contamination surrounding the development of the uniform Byzantine text. The main body of the Byzantine witnesses have variant b. For the witnesses followed by a superscript number in Figure 13 the altogether closest potential ancestor is to be found in the b attestation. Most of them are characterised by the fact that their more closely related potential ancestors differ considerably from one another (which is otherwise atypical). The verification of the genealogical coherencies does not actually prove that variant a is original, as the corresponding test for variant b would result into a comparable picture. But the genealogical coherencies in Figure 13 demonstrate, first of all, that derivation of a from A, and therefore its originality, is possible, and, secondly, that no multiple coincidental emergence must be assumed. A further example may demonstrate the advantage of observing the textual flow and coherencies in a contaminated textual tradition (Figures 14, 15).61 There are two main competing variants in 1 Peter 4:16/24-28: a (τῳ μέρει τούτῳ),62 and b (τῳ ὀνόματι τούτῳ),63 cf. Figure 14. So far, variant b has always been favoured and seen as the original because of its attestation in witnesses counting as outstanding: 03, 01, 02, 1739. Variant a, in contrast, is essentially attested by Byzantine witnesses, which have generally been judged to be of inferior 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 witnesses for variant b. Yet variant b, which had been considered original in view of its attestation, can hardly have served as the source for a. The meaning of variant a is not immediately clear and a represents a far more difficult reading. Yet variant b does not irritate and its phrasing agrees perfectly with a neighbouring





Gerd Mink 1 Peter 4:16/24-28

variant a

A

variant b 03

025

012 11752 17392

14482

17352

18522 180 453 468 720 918 1678 2186 2818

12432 945 1241 16114 813 2492

307

429

206 522 12922 1490 2200

424

1505

22982

Byz

642 ?

044 02 2344 ? 57 33 623 2464 2805

6142

2138

5 69 88 323 442 915 1832 2718

218

2412 1890 1359

25412

623 2464

1563 1718 2374 436 1067 1409

621 1359 8762 2541

2805

Figure 14. Multiple genesis of variant b

passage (1 Peter 4:14/6-10).64 There are no grounds, however, for changing b into a. It appears more reasonable to assume that b has developed from a. But is this possible considering the attestations? The genealogical coherence for variant a has proved to be very good, and a diagram very much like the one in Figure 13 can be drawn up (symbolised by the rectangular frame on the upper right in Figure 14; Byz = Byzantine witnesses),65 for variant b the genealogical coherence is definitely imperfect. Part of the witnesses for b show unmistakable genealogical coherencies with several witnesses 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, but none of their most closely related potential ancestors does. In Figure 14, next to the large group on the left, which derives from A through 03 and presents perfect coherence within itself, a large series of witnesses (to the right, under the frame) happen to read the same variant b. If variant b is original, variant a must have been derived from b, and in the third instance several times variants must have been derived from a which happened to have the same phrasing as b. This is quite conceivable, because b, as indicated, presents an obvious reading in the context.

Problems of a highly contaminated tradition A

1 Peter 4:16/24-28 variant b 025

307

14482

17352

180 453 468 720 918 1678 2186 2818 424

22982 429

206 522 12922 1490 2200

A variant a

variant b

025 14485 22982 3073

429

180 453 468 720 918 1678 2186 2818

206 522 12922 1490 2200

17354 424

Figure 15. Differing textual flows with changing initial text hypothesis

This offers an alternative, as shown in Figure 15. Following the coherence examination both alternatives are possible. Figure 15 shows the differences which 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 change of position of A. In addition, there are insignificant changes within the attestation of variant a. The top diagram shows a number of witnesses deriving the original variant directly from A. If the variant is no longer original, as in the bottom diagram, the textual flow has to explain the association of the witnesses in bold print by way of another witness, as A is no longer available. In this case it can easily be done through 025, although that changes the position of 1735 considerably.





Gerd Mink

That such diverging textual flow diagrams are at all possible is due to the degree of contamination, since it requires more than one ancestor in a substemma. To establish such a textual flow diagram the witnesses have to be presented according to the highest levels of probability that one is ancestor to the other (cf. the witnesses without superscripts). Only very rarely is the second to fifth highest level of probability needed (cf. the superscript numbers following some witnesses). . Process of approximation After as many local stemmata of variants as possible have been drawn up for the places of variation, the entire genealogical data and the analysis of genealogical coherencies contribute to revising the local stemmata and the cases unsolved so far. Then new genealogical data can be produced, which may result in further revisions, until a sufficiently stable overall genealogical hypothesis has been achieved. This repetitive procedure is necessary, because the knowledge that will only be the final result of the entire process would be required for the assessment of the genealogical relationships between the variants. An overall hypothesis can therefore only be acquired through approximation and iterative revision of all the intermediate results. I am referring to the circular reasoning which cannot entirely be avoided in textual criticism, but has to be controlled (cf. paragraph 4). Before an overall hypothesis can be achieved, the following point is very helpful. . Construction of graphs based on predominant textual flows Since we have now a selection of potential ancestors and the necessary information to assess the probability that they are ancestors in a global stemma, some hypotheses concerning the textual flow through the witnesses can be proposed. The predominant textual flow (first level of probability,67 Figure 16),68 the textual flow with the highest probability and the textual flow which is second in probability are shown in the top area of the tree (Figure 17).69 In Figures 16 and 17 ‘A’ is the artificial witness of the initial text containing all the readings that are assumed to be initial readings in the local stemmata of variants. The introduction of A is the logical consequence of making textual decisions.70 In the second line of Figure 16 (P74-1852) the witnesses which have the hypothetical initial text as closest potential ancestor appear. Comparing the diagram with usually established stemmata the derivation of such a large number of witnesses may come as a surprise. But the result is based on two arguments.

01 81

307

1840

38

1448

1611

206 522 1490 2200

429

1852

1505

1595

2805 1270 1598 1893

436 1067 1409

2541 1838 621 1751

2243

378 2147

61 1837 1661 876 1832 2494 2652 1563 1718 2374 2412

400 322 1390 104 1842 643 2544 88 321 326 607 996 1251 1367 1765 2674 1359 1127 614 1890

623 2464 1297

5

Figure 16. Predominant textual flow through witnesses of James

330 365 2718

43 1850

2718S

1739

0246 945 1241 2298

1243

93 181 252 319 323 431 442 456 459 676 915 999 1501 2423 2523 218 808 1799 1831 630 1292 2138 2495

018 056 69 312 467 617 629 631 642 1509 1729 2774

197 424

2492

1175

398

94

1827 2180 2242 35 1524 2080 049 665 1845 1874 18751609

254

025

33

18

04

180 453 468 720 918 1678 2186 2197 2818

03

P20

02 044 1735 2344

P100

1066 0142 1853 1848 020 6

048

P74

A

Problems of a highly contaminated tradition 



Gerd Mink The two most probable ancestors

A 03 P100

048

01

1175

1852

025 81 02 P74

1448 1243

1735

307 04

468 1739

2344

2298

33 945

most probable second most probable

1241

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

Firstly, derivation of a predominant textual flow from hyparchetypes is incompatible with the principle of simplicity, which forbids unnecessary assumptions. The problems that conventionally require the usage of hyparchetypes are solved by a different method, that derives the textual flow from several ancestors in accordance with the contamination (cf. Figures 28, 36, 40). Secondly, the large number of witnesses deriving directly from A is a reflection of the fact that from the first millennium of the textual tradition only a very small percentage of witnesses has been preserved. For, in this case, the direct predominant textual flow means nothing else but that of the many witnesses which stood between 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 so far have been rated less important, like 025, appear, or even an almost disregarded witness like 307. In contrast, an illustrious witness like 02 is missing; it appears in the third line, deriving from 81. Traditionally, this relationship has been assumed to be the other way round, presumably because the text of 02 is preserved 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 than 02. However, what is not visible in this case, is that 02 nonetheless has readings

Problems of a highly contaminated tradition

deriving from A in 28 instances, which do not occur in 81 and therefore cannot derive from 81 (this is reflected in Figure 17, edge from A to 02). As the witnesses in the second line have A as the potential ancestor with the highest degree of agreement, they are of eminent importance for the reconstruction of the textual tradition.71 The diagram (Figure 16) demonstrates impressively that there is a continuous flow through 307, 468, 424, and 617 from the old text into the extensive Byzantine tradition. The right hand branch of the diagram shows how the way leads to the witnesses of another well-know text form (HK) through 1852 and 1448. Incidentally, it is striking that nodes with many edges emanating from them often display numbers belonging to commentary manuscripts (307 in the second line, 424 in line four, 617 in line five). Commentary manuscripts contain, apart from the commentaries, the continuous New Testament text. It is unlikely that these commentary manuscripts served as exemplars for manuscripts containing the continuous text. But it is very plausible that that text form was chosen as the basis for a commentary which was the most highly esteemed at that particular time and place. Accordingly, it would also have been used as an exemplar in the scriptoria. Figures 16 and 17 are not part of a global stemma, they give an overview of textual flows which in the case of Figure 16 should be found, with a very high level of probability, in a global stemma as well. It is possible that in Figure 17, the second most probable textual flow might not explain more variants of a descendant than the most probable one, and therefore becomes superfluous in the global stemma. But this is not very probable, and so a good overview of the position of the witness in the textual history has been achieved. Figure 17 should be compared with the actual results in the optimal substemmata in Figures 28, 36, and 40.

. Typical problems of contamination Figure 17 already points to some of the problems to be expected in a contaminated textual tradition. Very often we find relationships like the following (cf. Figure 18): D originates from C and from B, from which C originates. Probably, the differences between B, C, and D are not very big. Usually the members of a substemma have a high degree of agreement. This is a reflection of the circumstance that in a region with a dense tradition there exists a stable idea about 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 if C altered variants of B and if D merged B and C and if D altered variants of B and if E merged D and B and if E altered C-variants included in D according to B

B

C

D

then C may not be potential ancestor of E E may be potential ancestor of C E

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

The problems arise when there is a chain of relationships like the one shown in Figure 19. In this case it is clear that B and D are potential ancestors of E. But it is also possible that C is not a potential ancestor of E according to the definition of potential ancestry.72 If C altered some variants of B, and if D merged B and C while altering some variants of B, and if E merged D and B, altering the special C variants included in D in accordance with the corresponding B variants, then C is not

Problems of a highly contaminated tradition

a potential ancestor of D, and even a circular relationship becomes probable: E looks like a potential ancestor of C. The condition for this is that E, while it merges D and B, retains enough posterior variants compared to D. This problem will be dealt with in more detail below (cf. paragraph 9). The example demonstrates that edges in a stemma may be misleading if we do not pay attention to the variants whose transmissions and developments are represented by them. A situation like that in Figure 20 is possible, because usually a number of the variants of the ancestor in a contaminated tradition are posterior to the corresponding variants of the descendant, and a number of the variants of the descendant are prior to those of the ancestor. (Figure 23 shows the potential ancestors of witness 1243 in the Zeuge2 field. In the yausx field the number of cases in which 1243 has a prior variant compared with the potential ancestors is 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 ‘true’), the global stemma may become false if indirect connections are interpreted. A graphic representation like Figure 19 can be read as a group of substemmata. The directed edges indicate only the predominant direction of the textual flow between two witnesses and what textual flow is necessary for the explanation of the textual state of a descendant. A stemma which does not give information on details of the textual flow (i.e. which variants it contains) cannot be used to determine the relation between nodes which are indirectly connected. The data underlying and qualifying a specific stemmatic connection must therefore be available for verification and assessment. It is also evident that the edges in the substemmata are of particular importance as carriers of information. A substemma representing a contaminated tradition must relate a descendant to an optimal combination of ancestors. A combination is optimal if it explains the totality of the variants of the descendant and, if this condition is met, is as small as possible. The requirement is that the substemma is true for every passage of text. In the case of Figure 19 each variant of D agrees with B or agrees with C or has developed from the corresponding variant in B or from the corresponding variant in C. At least one of these possibilities must apply to every passage. The following example shows an optimal substemma, in which the connection is true. It is simple and does not contain contamination. It concerns the best witness we have for the Catholic Letters, 03 (the manuscript is datable to the 4th century). In Figure 21 the witnesses compared are in the Zeuge1 and Zeuge2 fields. Further information is to be found





Gerd Mink

Zeuge1 and Zeuge2 – the witnesses being compared; Richtg – the predominant direction of the textual flow; Proz1 – agreement in percentages; Kon – agreement in absolute numbers; Proz2 – precentage of cases in which Zeuge1 has a variant which represents a further development of a variant of Zeuge2; Xausy – the same in absolute numbers; Proz3 – percentage of cases in which Zeuge2 has a variant which represents a further development of a variant of Zeuge1; Yausx – the same in absolute numbers; Proz4 – percentage of cases in which the source of a variant in Zeuge1 is doubtful; Qfragl – the same in absolute numbers; Kv – cases in 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, which is hypothetical, and fragments 0166 and 0173. They are potential ancestors, because compared to 03 more variants of these manuscripts are prior than posterior. Thus, the textual flow predominantly runs towards 03. The textual flow between fragment P23 and 03 has no direction. The resulting substemma is simple (Figure 22): A is the ancestor of 03, because in all cases where A and 03 disagree, the variants in 03 were created on the basis of the corresponding readings of A without agreeing with the variants of 0166, 0173, or P23. These witnesses do, indeed, contain prior variants compared with 03 (cf. the xausy field in Figure 21), but in each case they are identical with the A variants. Therefore, these witnesses are not needed in a substemma which explains all the variants of 03. (This conclusion cannot be drawn from the values in the list; we have to know the variants at the places of variation concerned.) It is possible that the text of 03 is also contaminated, although no further source of contamination has been preserved.73 The substemma in Figure 22, like all substemmata, is based on all the places where the local stemmata offer a hypothesis about the relationship between the variants of the descendant and the other variants. Places which remain unsolved or where the descendant has uncertain relations with other variants are therefore not included. For 03 this is the case in one place (cf. Qfragl in Figure

Problems of a highly contaminated tradition

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

686 92,6%

694 92,0%

648 91,5%

670 90,2%

677 89,6%

648 87,1%

SUM

X X X X X

X X X X X

X X X X

X X

X

722 732 732 734 736

X

X X

X X

X

X X

734 735

X

Figure 24. Combinations of potential ancestors of witness 1243

21). I will return to this when I discuss the problem of prior variants found only in non-ancestors (cf. paragraph 7). Another example concerning the top area of the global stemma demonstrates contamination. We see 1243 (Figure 23; the manuscript is dated to the 11th century) and its potential ancestors from A to P100. The first three witnesses are fragments and the textual flow between 1243 and each of them has no direction. Their agreements with 1243 are included in the agreements of the other witnesses. Figure 24 shows potential ancestors of 1243, arranged in order of the degree of agreement (not the absolute number of the agreements); in the next lines the numbers and percentages of agreeing variants are presented, and in the right hand column the total number of variants explained by the agreements with at least one witness within some combinations (marked with crosses) can be seen.





Gerd Mink

1243 has 756 places of variation (i.e. 756 variants; 761 is the maximum possible in James). A is the most closely related witness in this group, agreeing at 686 places. The percentage refers to the totality of places of variation extant in 1243 and in the witness compared. The agreements with A or 1175 explain 722 variants in 1243. If we combine A, 1175, and 025, we can explain 732 variants, but adding 03 does not result in a greater number.74 This method allows to test each contribution of potential ancestors to the explanation of the variants of 1243. The top values are shown in Figure 24. The number of 736 variants is not exceeded. Normally, variants not explained by agreements with an ancestor must be secondary, altering a variant of at least one of the ancestors.75 Most of the variants of 1243 are explained by the combination A–1175– 025–1739–01 (Figure 24, marked line). But are 5 ancestors really necessary? The overall agreement of 01 (87,1%) is rather poor. We have to ask, therefore, whether agreements may be coincidental. I summarise the principles of the procedure. A substemma links a descendant with its hypothetical ancestors, which are to be found among the potential ancestors. The substemma, as defined, is optimal if it can explain the textual state of the descendant from as few ancestors as possible. These ancestors will be very similar to the descendant and, consequently, its variants must derive from the agreements with the ancestors in as many places as possible. This is in accordance with the basic model (cf. paragraph 4). Their relationship with the descendant must therefore be based on the highest possible values of agreement. Lower values raise the question whether the agreements are coincidental. What values are to be regarded as high or low depends on the rate at which the percentage agreement values (Proz1 field) in lists like the one in Figure 23 decrease.76 In cases of doubt the decision whether a variant leads to a stemmatic connection between two witnesses has to rest on the character of the variant. A fundamental remark is necessary at this point. The entire procedure is based on all the genuine variants, excluding errors and orthographica. There is no selection of variants assumed to be more important than others. In most cases a scribe copied a manuscript probably using not more than one source copy, and he fulfilled his task very well. In the view of the scribe there existed no variant. He transmitted important and unimportant passages without making any difference. A decision between two variants was only possible if the scribe knew another variant. It is a feature of closely related witnesses that they have important as well as unimportant variants in common. Whether a variant is connective (and not accidental) depends on the degree of agreement of two witnesses or on the character of the variant. If two closely related witnesses agree on an unimportant variant, the variant is connective. The coherence within

Problems of a highly contaminated tradition

an attestation is the most important hint in this respect. We cannot, therefore, decide at the very beginning which variants are connective. If two witnesses not closely related agree, the variant is normally not connective. If these witnesses agree on a significant variant, which is unlikely to have emerged more than once without genealogical dependence, the variant is connective. Such a variant is not possible without contamination. As a witness with relatively few agreements cannot explain enough variants of a descendant, it can only be one of several ancestors. All the variants of a descendant which are explained by a new member of a combination of ancestors have to be examined, especially if no high degree of agreement speaks for genealogical relationship. Is it necessary to assume that the witness is an additional ancestor? If we do assume a combination of ancestors, we must analyse the special role of each ancestor and figure out what the consequences are if we exclude it from the combination. In order to make an optimal 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 potential ancestors and showing the genealogical relations of the variants is the starting point for establishing the optimal substemma. The database allows to observe what variants of a potential ancestor are suitable for explaining the variants of the descendant. In this way the examination of their wording also becomes possible. The resulting substemma (Figure 26) only gives the information that one of the following possibilities is true: whatever passage we check, (a) at least one of these 4 ancestors agrees with 1243 – this is true for 734 places of variation – or (b) if none agrees, at least one has a prior variant explaining a posterior variant in the descendant – this is true for the remaining 22 places of variation. We cannot, however, see the importance of each ancestor. To each edge some values can be attached (Figure 27 and 28). Of these values, those which exclusively pertain to the agreements have a direct influence on the composition of the substemma. The others indicate the respective qualities of the stemmatic connections and make it possible to assess how sensitive the connections are to changes in the local stemmata of variants. Such changes affect the ratio of prior and posterior variants which is critical to potential ancestry. The stability of the textual flow from ancestor towards descendants must be > 0. In the case of very low values a change may cause the value to become < 0 and the direction of the textual flow to reverse. The witness which was an ancestor may become a descendant. Figure 28 shows a substemma and values qualifying the edges. The values of agreement (cf. strength1 in Figure 27) and priority (included in strength2





Gerd Mink

Anfadr, Endadr – beginning and end of the place of variation; Hsnr – basic witness of the comparison; Labneu – code of its variant; Rg – relation between the variant of the basic witness and the variant of the following ancestor; = – agreeing variants; < – variant of the basic witness posterior; > – variant of the ancestor posterior; ? – relation questionable; - – no relation possible; Mrg – mediate relation; A, Ms1175, Ms025, etc. – names of potential ancestors.

Figure 25. Database section containing the genealogical relationships of the variants of potential ancestors of 1243 with the variants of 1243 A

1175

025 1739

1243

Figure 26. Optimal substemma of witness 1243

in Figure 27) demonstrate the strength of the textual flow. It depends on the character of the witnesses and their place in the textual tradition if values of agreement are considered to be high (cf. paragraphs 3 and 4.11). The priority value need not be high. Very high values argue against close relationships, as they lower the number of agreements. The stability value gives an idea of the reliability of the ancestor-descendant relationship.

Problems of a highly contaminated tradition

The textual flow from a potential ancestor to a descendant strength 1 = a ×c100 (a + p) × 100 strength 2 = c (p – s) × 100 stability = c a: number of agreements c: number of passages where comparison is possible p: number of prior variants of the potential ancestor s: number of posterior variants of the potential ancestor

Figure 27. Properties of textual flow

agreements, absolute: – (%): priority variants (%): stability of the textual flow: additional contribution: maximal contribution: minimal contribution:

686 92,6 7,29 7,29 5 740 2

694 92,0 3,85 0,13 8 723 1

648 677 91,5 89,6 4,10 4,63 0,42 0,40 7–8 2 677 708 3 2

A

1175

025 1739

1243

Figure 28. Values qualifying the edges in the substemma of 1243

Another value shows the additional contribution of an ancestor within a combination, its relevance for it. It indicates the number of variants of the descendant which could not be explained without this witness.77 Witness 1739, for example, explains 2 additional variants, where it does not agree with the other three ancestors. The value of minimal contribution is very interesting. Unlike the value of additional contribution, it concerns cases where coincidental agreement is improbable on account of the character of the variants. In the case of both the additional variants, 1739 is a necessary ancestor of 1243 because no other ancestor explains the variants of 1243 and the variants are also connective due to their character.78





Gerd Mink

genealogical coherencies in small attestation in James 2:14/2-10

variant c

stemma (section) 1243

03

03

1175 04

2:15/16

1175

variant b 1175

1243

1243 04

2:19/8-14

2492

variant e 1243 1175 04

2:26/2-4

variant b 03

1175

1243

2492

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

As a result the maximal contribution of 1739 is 708 (agreements, plus any prior variants), and the minimal contribution is 2. In the case of A there are 5 additional contributions. Only 2 of these can be explained exclusively by A.79 The minimal contribution of 1175 is 1 or even 0, for the only reading of 1175 qualifying as minimal contribution could be a simple independent orthographic agreement (as the variant of 1243 could be).80 Nevertheless, it is a member of the group of ancestors. If we were to remove 1175 from the group, assuming the agreements to be coincidental (in spite of their number), 8 variants of 1243 (cf. additional contribution in Figure 28) would remain unexplained by agreements, and we would have to search among the potential ancestors for another witness or other witnesses to explain these variants. But should we find such a witness, its general agreement with 1243 would be lower than that of 1175, and in case of two or more witnesses being necessary for the explanation of the 8 variants under discussion, the group of ancestors in the substemma would increase. In either case we would have an inferior choice and it would not be more likely that the agreements were not coincidental than in the case of 1175. It may be interesting to look at some of the additional contributions where attestations displaying the assumed relationship of 1175 and 1243 contain only

Problems of a highly contaminated tradition

a small number of witnesses. On the right hand side of Figure 29 we see an extract of a stemma. On the left hand side there are variants with their attestations. The edges show their genealogical coherencies, which are completely in harmony with the assumed stemma (cf. Figure 43 and the substemmata in Figures 36 and 40). Although 1175 has not been qualified as necessary at these places of variation because there are other witnesses (except for the second example), and although the variants are not especially significant,81 there is no doubt that 1175 participates in the textual flow, as it usually does. The examples support the general genealogical relations very well. A last question is why 01 is not needed as a further ancestor although it provides two additional instances of agreement (cf. Figure 24, marked line). The reason is that the agreements are rather coincidental in view of the kind of variants and the lower general agreement of 01.82

. Prior variants found only in non-ancestors If any two witnesses of a highly contaminated textual tradition are compared at random almost every single one has prior and posterior variants as compared to the other. Though prior variants prevail in potential ancestors, they also contain a (sometimes only slightly) smaller number of posterior variants. The opposite is true for witnesses to which the predominant textual flow is directed. They are, therefore, not potential ancestors. If the tradition is dense and every witness has a series of closely related potential ancestors, it will be easy to find the prior variants among them which explain the posterior variants in a descendant. Generally speaking, this is also true for a less dense tradition, but in such a tradition the prior variant which corresponds to a posterior one may only occur in one witness, which is not a potential ancestor. In the case of two witnesses C and D, which both can be traced back to an ancestor B, a very simple stemma evolves (Figure 30, left). D is assumed to have incorporated more changes from B than C has. This means that D cannot be a potential ancestor of C. Let us now suppose that there are places where C has variants which evolved from variants found only in D. Yet their number is so small that D is still not a potential ancestor of C. C has, e.g., 10 posterior variants compared to B, D has 30 compared to B, C has 5 posterior variants compared to D, D has 25 compared to C. For the 5 places where C has posterior variants compared to D, the left hand stemma of Figure 30 does not apply.





Gerd Mink B

B C
C

D

C

D

Figure 30. Intermediary node in case D has prior variants compared to C without being its ancestor

A correct stemma should therefore also represent the fact that prior variants are found in a witness which is predominantly not prior and therefore not a potential ancestor. The solution is an intermediary node, pertaining only to prior variants in a non-ancestor (cf. Figure 30, right). If there were no more surviving witnesses than B, C and D, it would be possible to establish the connection between C and D via a hyparchetype. If the connection between B, C and D is perceived as a substemma within a highly contaminated tradition, however, it would not be possible to establish a well founded hyparchetype, if the hyparchetype is in turn also the product of contamination. The intermediary node does not represent a hyparchetype. A hyparchetype would be a hypothetical witness, comparable to other witnesses, but only hypothetical, and would essentially contain variants at all the places of variation. The intermediary node only contains the variants which are relevant in this case. It is not advisable to merge the intermediary node into the edge pointing from B to D, as it might just as easily be situated on the way from B to C, and as it could in principle also be the product of contamination. The intermediary node in this case is to be understood as follows: There are posterior variants in C deriving from prior ones which are found in the non-ancestor D and which evolved from variants in B. C and D are now represented as products of contamination. This may be accurate. The intermediary node may also, as described, be part of the linear development from B to one of its descendants, so that only one of the two witnesses is a result of contamination. Also, in this respect an intermediary node must therefore not be interpreted in the same way as a main node. The stemmatic representation, therefore, is not a definite hypothesis about the location of contamination in the textual tradition and it raises various possibilities. The intermediary node helps to solve a problem which is the result of conflicting data (prior and posterior variants in one and the same witness) and ultimately of contamination and loss of intermediary witnesses.

Problems of a highly contaminated tradition

For an illustration I go back to Figures 21 and 22. The conclusion was that, on the basis of all the places where a certain variant could be assigned to A, and the variant of 03 could be put into a local stemma, the ancestor of 03 in an optimal substemma was A and no other witness. Yet, according to Figure 21 (cf. Qfragl) there is a place where the relationship between 03 and A is doubtful, in James 2:4/2-4. The variants and witnesses in this place are: a: οὐ διεκρίθητε ‘have you not made distinctions?’ (rhetorical question): 01. 02. 03 (additional to the reading of the continuous text). 04 and many others b: οὐχὶ διεκρίθητε (same meaning as variant a): 2544 c: διεκρίθητε ‘you have made distinctions’: 03 (in the text). 1852 d: καὶ οὐ διεκρίθητε (same meaning as variant a): 025 and many others e: καὶ διεκρίθητε (same meaning as variant c): 322. 323. 629

In the ECM variant a was the text established by the editors, although it could well be a simplification compared to d.83 At the time, the quality of the attestation of variant d (essentially characterised by 025 and Byzantine witnesses) did not appear sufficient for assuming it to be the initial text. Other editors following the traditional view of textual history acted likewise. So far, the secondary rise of variant d remains unexplained. The first word in variant d is καί, normally meaning ‘and, also’. Neither these meanings nor any of the other documented meanings make a sense here. The text preceding the place is a rather long conditional clause, interrupted by two short phrases in direct speech. The following apodosis is then introduced in a completely non-Greek manner by καί, a phenomenon which is, however, known from the realm of semitising Greek as ‘καί apodoseos’ and was possibly applied as well in James 4:15.84 It is rare in the New Testament. Later copyists cannot have been familiar with this usage, and the word must have appeared superfluous to them. Further genealogical investigations have revealed that in the attestation of variant d genealogical coherencies are present which could lead, by way of witnesses 025 and 307, to A, the initial text. A is potential ancestor of both 025 and 307, with the highest degree of agreement (cf. Figure 16). Consequently, variant d could also qualify as initial text. As variant a is easily explained as an adaptation of variant d to common usage and a number of witnesses for variant a are easily traced back to A, and as conversely variant d cannot be explained from variant a, variant d was subsequently declared the initial text.85 Figure 31 shows the resulting local stemma.





Gerd Mink d ?

a

c

b

e

Figure 31. Local stemma of James 2:4/2-4 B

03 < 04 03 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 a simplification of variant e with the objectionable καί left out in the same way as variant a probably did on the basis of variant d. This, however, is not very likely, for the witnesses of e are not potential ancestors and quite remote from 03 as witness of variant c. Within the attestation of c, by the way, 03 is ancestor of 1852. Derivation of variant c from variant d or variant a is more likely. The last word before the place of variation is μου. For variant d the resulting context is μου καὶ οὐ διεκρίθητε, for variant a μου οὐ διεκρίθητε. The reason for variant c appears to be a very common scribal error: the copyist eyeskips from one sequence of characters (here ου in μου) to an identical sequence later on in the text (here οὐ before διεκρίθητε) and continues copying from that point. With both 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 no potential ancestor of 03 in its attestation. Yet its witnesses would have a prior variant compared to 03 which is not found in any of the potential ancestors of 03. For variant a 03’s closest relative is 04. A correct stemma should therefore include the fact that there are prior variants in a witness which is predominantly not prior. The solution is an intermediary node pertaining only to this one variant (Figure 32). This will ensure that 03 can at the same time be an ancestor of 04 (see Figure 40). If variant c is derived from variant d, the substemma of Figure 22 is true.86

Problems of a highly contaminated tradition

Witnesses which are connected by an undirected genealogical coherency are also to be classed as non-ancestors. No direction of textual flow can be detected between them, for both contain equal number of prior and posterior variants. Let us assume that this applies to witnesses C and D in the example in Figure 30 (right). An analogous picture would result if D has a prior variant which is essential to explain a variant of C. At the same time it is conceivable that C has a variant which is essential to explain a variant of D. In that case there would not only be an intermediary node D > C like the one in Figure 30, but also an intermediary node C > D. P23, for example, is a fragment. However, it actually provides text at 49 places of variation, but there are mostly only one or two witnesses which differ from the initial text at these places. Thus, P23 has undirected coherencies with an abundance of witnesses. Compared to the initial text P23 differs only once: variant ff in James 1:17/40-46. This variant is represented by an erroneous reading in P23, which is probably based on variant d. The witnesses for d are 01 (first hand) and 03. There are also undirected coherencies with these. As in the previous example, 01 and 03 should be linked with P23 by way of an intermediary node in an optimal substemma on account of this prior variant.87 P23 in turn has a prior variant compared to 01 and another one compared to 03.88 But in both cases P23 agrees with the initial text A. A is ancestor in the substemmata of 01 and 03, P23 is therefore in no way necessary to explain a variant in 01 or 03.89

. Undirected genealogical coherencies At the beginning the search for the optimal substemmata is based on the potential ancestors only, because between them and the descendants directed genealogical coherencies exist and only they offer any information about the direction of the textual flow. When a substemma is found which explains all the variants of the descendant with as few ancestors as possible, the possibility remains that the substemma does not explain a number of close genealogical coherencies resulting from frequent agreements of witnesses within attestations. These coherencies may not be indicated by other substemmata either and so not be reflected in the global stemma. This phenomenon is caused by the undirected genealogical coherencies. Let us assume a substemma consisting of an ancestor B and a descendant C. A further substemma links ancestor B with descendant D. The two substemmata are compatible with the situation at a place of variation as represented in





Gerd Mink

variant a

variant b

witness B

witness C

witness D

Figure 33. Undirected genealogical coherency between C and D within an attestation intermediary node relation without direction C=D C

D C

D

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 developed their variants independently from the variant in B as the diagram seems to suggest. They would coincidentally have the same variant. If we further assume a close undirected genealogical link between C and D (cf. the link between C and D in Figure 33), it is impossible that they have developed the same variant coincidentally. This fact has to be reflected in a global stemma. When we find such witnesses with close undirected genealogical relationships (cf. the undirected edge in Figure 34, left hand graph), intermediary nodes are apparently needed for an exact description of the relationship. The supposition is that witnesses C and D agree in variants which have not emerged independently but stem from common ancestors, and the condition is that one or more attestations exist which contain neither a (surviving) common ancestor of C and D nor a witness which is a descendant of C and an ancestor of D or the reverse. These intermediary nodes are not hyparchetypes in a traditional sense, but should better be understood as a special kind of connection. The content of such a node as in Figure 34 (right hand graph) is all the agreements between C and D. Thus C and D have their ancestors, and the intermediary node establishes a link to the ancestor or ancestors (= sources) of the common variants of C and D. If the content of the intermediary node is the result of contamination, witness B is only one of the ancestors of the intermediary node (cf. Figure

Problems of a highly contaminated tradition

Figure 35. Potential ancestors of witness 1175. See Figure 21 for field legends A

03 1739

1175

A

1243

03

1852

04

Figure 36. Connection between two substemmata through an undirected edge

33). The intermediary node has to be incorporated into the substemmata. If the substemmata of both witnesses are true, the ancestors of the intermediary node have to be the intersection of the witnesses which are ancestors in the two substemmata, or a subset of the intersection. For illustration I take two witnesses which are closely related, but there is no predominant direction of textual flow between them: 04 and 1175 (cf. Figure 35, no arrow in the Richtg field, in accordance with the equal values in the xausy and yausx fields).90 They are shown with their optimal substemmata in Figure 36. 04 and 1175 have 12 places where they have secondary variants in common, but considering the good coherency between the two witnesses, this is no accident. Due to the old text form the agreement values (Proz1 field) begin at relatively low percentages. According to the optimal substemma 1739 is descendant of 04 (Figure 40, top right) and ancestor of 1175 (Figure 36). At 7 out of the 12 places 1739 reads the same variant as 04 and 1175.91 The hypothetical route of the variants at these 7 places could therefore be 04 → 1739 → 1175. At 2 out of the 12 places 03 reads the same variant as 04 and 1175,92 in whose respective substemmata 03 is the ancestor (cf. Figure 36). The route of the variants could therefore be 04 ← 03 → 1175. 3 places now remain,93 where 04 and 1175 agree, but in the same attestations no ancestors of at least one of them are available which could connect these witnesses. The substemmata in Figure 40 are true without a link between 04 and 1175, inasmuch as the variants at





Gerd Mink A

03 1852

1243

04 = 1175

1739

04

1175

Figure 37. Resolution of the undirected edge in Figure 36

the three places are further developments of the source variants offered by at least one of the relevant ancestors found in the substemmata of 04 and 1175.94 However, the relationship between 04 and 1175 would not be expressed by the substemmata. Rather, the fact that there is no stemmatic link between 04 and 1175 at the three instances in question, would promote the false impression that the two witnesses coincidentally share the same variant. Figure 37 shows the result for 04 and 1175 following the model of Figure 34. There is no undirected connection between 04 and 1175. Their common variants are explained with the aid of the intermediary node connecting the sources of the common variants with 04 and 1175 and connecting 04 and 1175 in the same way. As a result of the intermediary node one of the stemmatic coherencies and thereby ancestors may become superfluous, as in this case ancestor 1739 in the substemma of 1175. If there were no instances which make 1739 an ancestor in the substemma of 1175 apart from the 7 instances where the witness plays the role described above (04 → 1739 → 1175) this would be correct. Yet there are a large number. The connection of two witnesses by way of an intermediary node and the incorporation of the node into the relevant substemmata (cf. the substemma of 04 in Figure 40) unquestionably calls for renewed examination and possibly fresh optimising of the stemmata. It is also possible that an intermediary node is only needed from the perspective of one of the witnesses it is going to connect. This is the case for attestations containing no further witnesses genealogically connecting the witnesses (in contrast to the example 04 → 1739 → 1175) and having no common ancestors (in contrast to the example 04 ←

Problems of a highly contaminated tradition

03 → 1175), but an ancestor for one witness only, and if no places are left in which neither witness has an ancestor in the same attestation. Indeed, for 04 and 1175 such a place remains.95 If there were no such place, 04 would have had an ancestor in each place that agrees with 1175, and should not have needed an intermediary node in contrast to 1175. In this case, the edge from this node to 04 could be marked in order to indicate this fact. An optimal substemma can also be constructed for an intermediary node. The ancestors can be found amongst the ancestors common to the substemmata of the descendants involved. The node representing the agreements of 04 and 1175 requires A and 03 as ancestors (cf. Figure 37). As a simpler solution but with a certain loss of information, instead of an intermediary node an undirected edge could be used, cf. paragraph 9 and Figure 43. For intermediary nodes based on undirected coherencies the same restrictions apply as for nodes based on prior variants in non-ancestors. As for the latter, the stemmatic representation opens up various possibilities for the location of contamination in the textual tradition (see paragraph 7). At any rate, we must keep in mind that in the case of undirected coherencies the stability 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 research is necessary. Figure 20 showed how the problem may emerge. If the variants of an ancestor were mixed with the variants of the ancestor’s ancestor, over several generations of copying, thereby reintroducing older variants again and again, this may result in a descendant so rich in older variants that it becomes a potential ancestor of a mediate ancestor. A circle materialises based on contamination in multi-stage phases. But are there more preconditions than multi-stage contamination? Figure 38 refers back to Figure 20. The assumed variants of the 4 witnesses at 8 places of variation are shown in a matrix of 8 lines (in a frame). The variants are in accordance with the requirements of Figure 20. On the right are the local stemmata presupposed at the 8 places of variation. This offers the information how often a witness has a prior variant compared to another. The values are listed on the left. C has prior variants in two cases where D has the respective posterior variants (cf. lines 1 and 6 of the matrix). From the values the predominant textual flows follow, C > D, D > E, E > C.96 From the tex-





Gerd Mink B C

witnesses: B, C, D, E

E

direction/cases C>D D>C

2 1

D>E E>D

3 2

E>C C>E

2 1

variants: a, b, c, a‘, a“

D

B

C

D

E

a a a a a a a a

a b b b b b b b

b a b b b c a‘ a‘

b b a a b a‘ a“ a“

local stemmata for line: 1–5 6 7, 8 a a a

b

b

c

a‘

b

a‘

a“

Figure 38. Circular edges

tual flows follows the potential ancestor of each of the witnesses and a circle materialises. The local stemmata at the first 5 places allow a statement, in each and every case, whether the variant of a witness is prior or posterior to the variant of another witness. The local stemmata at places 6, 7 and 8, however, do not allow such a deduction for each pair of variants, as not all the variants are directly connected. If there were only direct connections like in the first stemma, the circle would no longer occur.97 The question now arises, whether the enrichment with older variants in a branch of the textual tradition is the only cause for a circle or whether other conditions may lead to circles even in small areas of a stemma. From Figure 38 it already appeared that not all the local stemmata there belonged to type a → b, and that this made a circle at least possible. Figure 39 represents a circle of 3 witnesses. The circle is supposed to be part of a stemma. In the upper half of Figure 39 it is assumed that there is only one possibility for a local stemma; it leads from variant a to variant b. The matrix of variants in the middle section, and the information obtained from it, how often each witness has a prior variant compared to another witness (see direction/cases), demonstrates the circular relationship of the 3 witnesses. The local stemma of this simple structure does not allow a circular connection between the witnesses (see above) if all the places in the matrix can only be filled by a or by b. Three positions (the small boxes in the matrix) remain empty in order to obtain the values (cf. direction/cases) needed for a circle. If one of the boxes were to be filled by a or b, there would be no circle. In fact, in the

Problems of a highly contaminated tradition C variants: a, b, c, (x), (y)

witnesses: C, D, E D

direction / cases C>D D>C

2 1

D>E E>D

2 1

E>C C>E

2 1

E

C

D

E

a a b £ b b

b b a a £ £

a b b b a a

dir. / cases A / cases B C>D D>C

4 1(1)

2 1(1)

D>E E>D

2 1(2)

2 1(2)

E>C C>E

3 1

2 1

local stemmata:

C

D

E

a a b c b b

b b a a c c

a b b b a a

a

b

a

a

b

b

c

a

b

(x)

(y)

a

c

b

c

Figure 39. Circular edges in consequence of lacunae (top) or different types of local stemmata

situation where at each place there is just a variant a and a variant b, a circle is only possible if there are appropriate lacunae in the witnesses. It follows that if there are no lacunae, the matrix can only be completed with variants leading to the same circular edges if there are other types of local stemmata. The bottom half of Figure 39 shows a matrix of variants in which the boxes (i.e. the lacunae) of the top matrix have all been replaced with variant c. We assume that at the first three places (= lines of the matrix) the first type of local stemma evidences the genealogical relationship between the variants, but at the last three places the second type of local stemma does. There c derives from b. The a witnesses have no prior variants compared to the variants of the c witnesses. A variant, after all, is only prior compared to a posterior variant which developed from it (cf. paragraph 4.9); c, however, derives from b, not from a. a has mediate priority at most. On the left, under ‘dir./cases A’, the relationship between witnesses is shown from the perspective of the priority and posteriority of their variants. The values are based on the assumption that only the first and the second types of local stemma are true. The values without brackets indicate the number of prior variants. Again a circle results, C > D, D > E, E > C. The values in brackets





Gerd Mink

show the number of indirect relationships occurring here if one witness reads a and the other c. If we now assume that the last three lines of the matrix do not conform to the second type of local stemma but to the third, the resulting values which are found under ‘dir./cases B’ correspond to the values in the top part of Figure 39. The circle remains. The changes reflect the fact that c can no longer be traced back to b. The mediate priority of a is maintained (values in brackets). It proceeds 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 the matrix, again the same values as in the top of Figure 39 ensue. As a and c now derive from a common source variant x, rendered by witnesses not being part of the circle, no mediate priorities are left at all. It would now be possible to consider the use of the mediate relationships in the construction of a stemma. But there are serious methodical objections to this. Rightly, only direct relationships between variants in the local stemmata are used to determine potential ancestors and textual flows. To determine the genealogical coherency between two witnesses their degree of agreement is essential first of all and secondly the relation between the number of their prior and posterior variants, i.e. only data based on identity of variants or direct genealogical relationships between them. It is only this direct relationship which makes it at all possible to conduct coherence tests at individual places of variation,98 and to find local chains of coherencies incorporated into a global stemma and subsets of the global stemma represented in diagrams which show the textual flow within an attestation.99 It is clear that circles can materialise as a result of different circumstances. The complexity is much higher in reality than in the examples in Figures 38 and 39 – more witnesses, more variants, therefore more complex local stemmata, the several types of which (see Figure 39 bottom half) get mixed up, and that is without taking into account the lacunae. In this situation circles may emerge, but need not. Under the conditions of Figures 38 and 39 a circle does not materialise if variants within the same line of the matrix change places for the same local stemmata. This is of course due to the very small difference between the number of prior and posterior variants of one witness in the circle compared to another. In fact, the smallest difference between pairs of witnesses concerned determines the stability of a circle, and the differences are often very small indeed. The variants have to be distributed among witnesses in a very particular way for a circle to materialise. How frequently this may happen also depends on the frequency of non-direct relationships between variants, and the degree of fragmentation of witnesses.100

Problems of a highly contaminated tradition A

1243 03 1852 04 = 1175

A

04

1175

025 1739

A

1243

04

03

1739

A

1852

025

03

1175

1739 1243 04 = 1175 04

Figure 40. Top: substemmata; bottom: their combination with a circle

Badly fragmented witnesses certainly present a very special risk. The genealogical relationship between a fragmentary witness and another witness is determined on the basis of its limited text supply. As the distribution of prior and posterior variants in a text can be very different, it is possible that an entirely different picture would have resulted from the complete witness, if it had been preserved. Witness 04, in which roughly the first 3 out of 5 chapters of James have been preserved as a continuous text, is a good example how the inclusion of such texts causes the emergence of several circles. Figure 17 presents the two most probable genealogical relations in the top area of a global stemma. This tree suggests that in a global stemma 04 will be ancestor of 1739, 1243 ancestor of 04. Yet, in the substemma of 1243 we find 1739 as an ancestor of 1243 (Figures 28 and 40). Figure 40 represents the optimal substemmata of 04, 1243 and 1739. At the bottom we see their combination, containing the connections in the substemmata. The connection of the intermediary node with 1175 as well is implied. The circle which was to be expected on the basis of Figures 17 and 28 does indeed materialise (04–1739–1243–04, cf. the bold edges). 04 is only available at 482 out of 761 places of variation, whereas the other witnesses can be used almost in their entirety. If only the first three chapters of James instead of five are taken as a basis – i.e. roughly the quantity of text preserved in 04 – the textual flow between 1739 and 1243 changes direction,101 and the circle disappears.102 As Figure 36 shows, 1739 is also an ancestor in the optimal substemma





Gerd Mink intermediary node D³B circular edges

B

B C C D

D

Figure 41. Simplified model for the resolution of circular edges

of 1175. Here again a circle will materialise in which 04 participates. It includes the other circle (04–1739–1175–1243–04). The direction of the textual flow between 1175 and 1739 would change likewise for the first three chapters.103 The problematic side effects of including fragmentary witnesses are evident. They are aggravated by the fact that the stability of the textual flow between both 1175 and 1243 and 1175 and 1739 is very poor.104 Of course, 04 is too large and too important in the textual tradition not to be considered. The example was chosen deliberately, also in order to illustrate how important it is to have access to values and facts behind a stemmatic graph. How is such a circular structure to be pictured? Figure 41 (left hand diagram) shows a simple model of circular edges. The circle can be resolved by introducing an intermediary node containing the variants of D which are the basis 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. Consequently, we have to treat all the witnesses in the same way. The result is Figure 42 (bottom). The left hand stemma in Figure 43 shows the combination of the substemmata of 04, 1243, and 1739, following the model of Figure 42. The circle 04-1739-1243-04 is now resolved. In addition, the undirected coherency between 04 and 1175 is respected (cf. Figures 36 and 37). The number of edges may be smaller if the edges from A to 04, 1243, or 1739 are “empty” (not transporting variants, for all the variants pass the relevant intermediary node). In Figure 43 the right hand stemma can be interpreted as representing the same facts as the left hand one with its intermediary nodes. There is a circle connecting 1739, 1243, and 04. 1175 and 04 are connected by an undirected edge. The logical meaning of the right hand diagram is the same as that of the left hand diagram, the disadvantage is only that there are no intermediary nodes with their special contents and ancestors. On the other hand, the cir-

Problems of a highly contaminated tradition B

circular edges

C D

C³D

D³B

B³C B

intermediary nodes

C D

Figure 42. Complete model for the resolution of circular edges A

03

A 025

04 = 1175 1852

03

025 1175 1739

1175

1243

1852 1243 ³ 04 04

04 ³ 1739 1243

1739 ³ 1243

04

1739

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

cles in the left hand diagram are more difficult to discern. Yet the right hand stemma provides less information. However, full information can be found in neither of the two stemmata. A stemma like the one on the left is only useful for specific purposes and even then one has to know the way that each of the variants follows through the stemma. Further problems arise when more circles materialise in the vicinity of a witness, as is indeed the case for 04 (see above). They have not been taken into account in Figure 43. The number of intermediary nodes increases. Yet the new nodes and edges may be simplified by combining nodes which are not contradictory. This will have consequences for the definition of the simplicity of a





Gerd Mink

(sub)stemma. When intermediary nodes depending on circles start accumulating, the limits of depiction are reached fairly soon. On the other hand, the circles from the underlying data are often easily made out. In reality, circular genealogical relationships are impossible. However, a stemma does not represent historical reality, but structures obtained from the available data. Yet, concerning the data, the relationships of fragments are essentially complicated by the fact that witnesses with different text lengths can only be compared to a certain extent. In any event the result must be interpreted with caution, both if the character of the whole witness is judged from that of the fragment, or if the complete witness is only assessed in the section which is covered by the fragment. Thereby, a circle depending on fragmentation is an artefact. Nevertheless, according to the rules the substemmata involved 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 places of variation cannot be found. Again, the background data are important for estimating to what extent artefacts contribute to the result. Matters are different with circles depending exclusively on the model of Figure 20. Those circles correspond to a circular development of variants. If we do not know the actual chronological order of the different textual states, we can only detect a circle. Rules for the construction of a global stemma which prevents the emergence of circles will have to result from the respective model of the textual tradition. I do not consider this possible. If the data suggest a circular development it should be represented as circular.

. The construction of a global stemma The global stemma is the superset comprising all the optimal substemmata as subsets including the intermediary nodes incorporated into them or the corresponding connections.105 The stemmatic coherencies between descendant and ancestors ensue from the optimisation of the substemmata. Even if intermediary nodes are not hyparchetypes and sometimes comprise only one variant, their role is technically comparable to that of the main nodes, insofar as they occupy the same position as ancestors if incorporated into a substemma. Thereby, the stemmatic connections on the level of the substemmata are established. They contain all the necessary connections and they contain only those. Each ancestor (apart from A) in a substemma will be a descendant in another substemma. Each descendant will either be an ancestor in one or

Problems of a highly contaminated tradition

more substemmata, or, if not an ancestor, represent a terminal node in the global stemma. The connections in a global stemma are thus absolutely determined by the connections in the substemmata. Consequently, only one single global stemma is possible on the basis of one set of optimised substemmata (with any intermediary nodes). Optimising on the level of the global stemma is impossible. Every change in the substemmata leads to a change in the global stemma, the change, however, being absolutely limited to the range of the substemmata involved. If there are alternative substemmata of the same quality, there are also alternative global stemmata of equal quality. Their numbers may easily mount up. If there is another alternative substemma for a given descendant, the number of possible global stemmata increases only by 1, but it doubles with every alternative substemma for further descendants. In a very dense textual tradition, with texts agreeing almost completely, the case will repeatedly occur that an ancestor in a substemma can be replaced by another without affecting the quality of the substemma. The same can happen with fragments mainly rendering places of variation at which only few witnesses depart from the mainstream of the tradition. Consequently, they will have numerous close relatives agreeing at the same high degree. In addition, it must not be assumed that the character of the lost part of the witnesses corresponds sufficiently to the preserved part if there is a greatly limited or very small number of places of variation. Nevertheless, it is not necessary to refrain from the use of fragments in stemmata at all, but it is essential to know their precise data. It is not possible to refrain from using a witness as important as 04 (see above) in the construction of the stemma, although two chapters of James are missing. Yet 04 belongs to an area of the textual tradition in which nearly all witnesses are lost. The smaller fragments are very unlikely to appear as ancestors anyway, as they explain too few variants in a descendant. It would not be very worthwhile to construct all possible stemmata. As optimising takes place at the level of the substemmata and not in the course of the creation of the global stemma, connections in the global stemma would not change if it is incomplete. It is, therefore, possible to construct parts of a global stemma and leave out those areas where there are alternatives. The remaining connections will still remain true. The informational value of an area in the stemma with many equally good alternatives is meagre at any rate. Yet, on the other hand it may be desirable to see what the development in a specific area is like, cum grano salis. This is possible if all the alternatives lead to very similar results in any case. It would then be sufficient to mark the connections for





Gerd Mink A

03

A

03 1739

025 1739

1175

A

1175

1243

A 1 1: 97,3%; 19; 0 2: 92,5%; 55; 0 3: 92,6%; 54; 0 4: 90,8%; 47; 12 5: 90,4%; 31; 30 6: 92,0%; 29; 28 7: 89,6%; 35; 32 8: 91,5%; 29; 26

03

2

3

1739 5

4

7

025

1175 8

6

1243

Figure 44. Part of a global stemma, compiled from substemmata. Left: qualification of edges: agreements in %, number of prior and posterior variants of the ancestor.

which there are alternatives. This is particularly useful where in a stemma areas without alternatives are separated from each other by an area with alternatives. As the completeness of a stemma is no prerequisite for the accuracy of the connections it is, of course, possible to extract any section of the global stemma, e.g. the top of the stemma, the stemma of a group, without constructing the complete stemma. Figure 44 demonstrates that there is no other possibility for constructing this section of the global stemma on the basis of the substemmata. If any ancestors are available here, then all of them are available. If there are no ancestors available, as with 025 and 1739, they are situated outside the section of the global stemma. If the global stemma were complete, the section of the stemma in Figure 44 would not change.106

. Conclusion It is very clear that the genealogical structure of a contaminated tradition cannot be revealed if evaluations of variants and automatic procedures are not alternated permanently. Accordingly, three levels of falsifiability have to be accounted for: the level of facts, the level of evaluation of readings, and the level of procedures and their rules. Furthermore we have to be aware of the undesirable artefacts caused by method, point of view, and way of visualisation.107

Problems of a highly contaminated tradition

The user of a stemma has some very simple questions. Why is there an edge connecting x and y, and what is the reason for its direction? How reliable is the connection? He needs more information than stemmata usually provide. The stemmatic graph of a contaminated tradition is only one possible surface of a very complex situation. The facts behind the stemmatic connections are extremely important and must be accessible for assessment. The genealogical evaluation of the facts is the basis for their graphic visualisation, if the connections rendered in it are to represent genealogical directions. If a stemmatic graph is the visualisation of a hypothesis and if it is a surface showing only certain aspects of the material, different methodological approaches will be useful in order to recover the patterns within the preserved material. It is all the more important, then, that the user of stemmatic representations should know how to read them, and understand which interpretations are justified and which are not with respect to the theoretical background and all the procedures which contribute 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 ECM where the variation starts, and, if not identical, the word with which the variation ends; the address 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 omitted variant. . The witnesses are 467, 643 and 1848. . 631. . The degree of relationship is determined by the degree of agreement. Cf. for this paragraph 4.11 (pre-genealogical coherence). . According to the pre-genealogical coherencies. Nearest relative here is 424 with only 85.9% correspondence. Also when the genealogical coherence (cf. paragraph 4.11) is taken into 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 disappear from the stemma. The relationship between the a witnesses and the c witness would not change, the relationship between the d witnesses and the c witness would become only slightly remoter (some 0.2 percentage points). Actually, a choice has to be made between the two 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: 89.6% with 02. The relative remoteness of the next relatives is caused by a larger number of peculiar variants attested only in P74 or shared only coincidentally with more remote witnesses. . They are potential ancestors, cf. paragraph 4.10. In this context, the dating of the manuscripts is not important, cf. paragraph 4.4. . Cf. Aland (1987, 1991, 1993, 1998–1999). . Cf. Aland, Aland, Mink, and Wachtel (1997, 2000, 2003). . The numerous orthographic modifications and 592 faulty readings, which almost always occur in conjunction with a correct variant were not counted as genuine variants, cf. paragraph 4.3. The places of variation include also 59 places where only variants attested by lectionaries, Church Fathers or (mostly) early versions deviate from the mainstream of the tradition. In the case of Church Fathers and versions variants, as a matter of fact, were only recorded, if they, certainly or presumably, rest upon a Greek exemplar in a now lost Greek text 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 (2000: 24–25). . Again not including corrections, marginal text and commentaries, as they do not respresent continuous text. . To be exact, 704 attestations. The dots in the diagram correspond with the following values (from left to right): 969, 875, 832, 804, 794, 782, 766, 753, 734, 722, 704, 676, 629, 533, 154. . To be exact, 124 places. The dots in the diagram correspond with following values (from left 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 (from left 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 relationships between the 164 Greek manuscripts involved, they are nevertheless included in the genealogically relevant data, as they are needed for the genealogical assessment of the texts of the lectonaries, Church Fathers and in particular the Greek exemplars of the early versions. The versions can then be subjected to the same procedures as the Greek witnesses. Cf. Spencer, 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 4.4. . Cf. also the preliminary study Mink (1993) and Mink (2000). Mink (1993) represents the state of that time and could only be based on the material in the test passages (see paragraph 1; 25 instead of 761 places). . Cf. e.g. the case of 1243, paragraph 6. . Readings here meaning variants.

Problems of a highly contaminated tradition . Mink (2000: 52). . Cf. Mink (1993: 489). . An author does not necessarily always do justice to his intentions. He may formulate a text which is barely understandable, or which requires a great intellectual effort on the part of the reader, while to the author himself its meaning is absolutely clear. A copyist is more likely to produce variants here, in order to replace a lectio difficilior, according to his understanding, at any rate. He might also, as he does not understand the text himself, produce a variant which makes no sense and thus generate a lectio difficilior as compared to the original. . Amongst others 01, 03, 025, 1175, 1739T, and 1852. . If εὑρεθήσονται is accepted without the negation, exegetists and translators attempt to assign meanings to the word which are not recorded anywhere else. Many also have realised the absurdity of this variant and proposed a multitude of imaginative conjectures, cf. Metzger (1971: 706). . 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 age of the text form: 02, 33, 81, 307, 2298, 2344. Further variants (κατακαήσονται, καήσονται) have the same meaning. . ‘They will not be found’ is actually attested in non-Greek text traditions. It is regularly found in the Sahidic version, in the witness of the Coptic dialect V, and also in some of the witnesses of a Syriac version (Philoxeniana). Either the original variant truly survived here – which is at any rate quite possible for the Sahidic – or this variant is also already a conjecture. An old conjecture is already to be found in P72: εὑρθήσονται λυόμενα ‘they will be found as dissolved ones’. . Aland, Aland, Mink, and Wachtel (1997: 16*) . Ibid. . The distinction between manuscript and text is essential. Cf. Mink (2000: 52): “The manuscript having a palaeographical date gives us only the terminus ante quem non of the text”: there is no earlier attestation of the text. It is also possible to speak of the terminus post quem 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 (2000: 53), on genealogical coherence: “In a system whose constituents are not independent of one another, those constituents must cohere in a definable manner, if we are to understand that system. Coherence within a group of witnesses means that the members of the group are connected by immediate genealogical relations.” The latter is true if we know the ways of transmission. Normally they are not known, and only part of the witnesses have been preserved. In such a case, coherence is represented by hypothetical coherencies within groups, and the genealogical relations are immediate only because a part of the tradition is lost. . 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 5.3. . Translated from Mink (2002). Expansive treatment in Mink (2003: 41–46). . For places of variation, see 4.2. . The present study uses the data available in November 2003. For limitations, see Note 56. . If very similar variants obviously emerge repeatedly from each other or respectively if there is an imperfect coherence within the attestation. . For the special problem of undirected genealogical coherencies cf. paragraphs 4.11 and 8. . 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 4.11. The degree of agreement does not conclusively determine the probability, it only influences it, for it does not allow to discern whether the textual flow between a potential ancestor x and a descendant z is included in a textual flow between a potential ancestor y and the descendant z. In such a case x would not be ancestor in an optimal substemma. Cf. paragraphs 4.8 and 6. . Smaller fragments must be ignored. Their agreement percentages are often extremely high, as they preserve areas in which only few witnesses deviate from the mainstream. This is true for the major part of the text of James (cf. paragraph 2). It is impossible to say how such a text would look like if it had been preserved in complete form. Even a partially fragmented text like 04, which nevertheless renders more than half of the text in contiuous form and cannot be left out of consideration because of its genealogical importance, presents some difficulties, see paragraph 9. . Further examples of the evaluation of similar lists in Mink (2003: 53–56). . Cf. h1 and h2 in Figure 7. In the most cases, a certain number of erroneous assessments of coherencies in the first step does not affect the overall results very much. If we do not split a variant logically into two or more variants despite the imperfect coherency, the number of agreements of each of the pairs of witnesses in the attestation will increase by 1 due to this variant. If we do split the variant, only the number of agreements of pairs in the same coherency chain will increase by 1. But if the number of agreeing variants of a pair is large, there is a high coherency of those witnesses even if some local coherencies are not realised, and if the number is small, coherence remains low even if some local coherencies are ascribed erroneously. – As regards James, the coherencies within attestations will need a final revision based on the last genealogical coherencies and particular textual flows. Until now, attestations have been split only in cases of obvious imperfect coherencies. A revision is expected to result in splitting more attestations and strengthen the tendencies which can be observed now. In some cases of poor stability of textual flow, ancestor-descendant relations may change. . Under kon the numbers of agreements are listed.

Problems of a highly contaminated tradition . The lists for two quite differing witnesses have been taken as an example here. 1243 has few potential ancestors, which indicates a position in the upper regions of a global stemma. The list for 2412 has been shortened and in reality counted 138 lines. From the top of the list it is obvious that the well-known witness 614 is the most closely related potential ancestor of 2412. The next lines document that the well-known HK group is to be counted among the potential ancestors. . Witness A, the root of the tree, is hypothetical, cf. paragraph 4.1. . Cf. Aland, Aland, Mink, and Wachtel (2000: 21*–22*). The same procedure was not suitable for 1 John because the witnesses are on the average closer to the initial text A and to each other than in James or the Letters of Peter. Therefore, it was necessary to start at those places of variation that can be assessed easily. In this way, determining potential ancestors and predominant textual flows could be based on 621 out of 761 places of variation. The results are preliminary, of course. Yet, they proved to be plausible and sufficient for a first evaluation of genealogical coherence and construction of textual flow diagrams. Cf. Aland, Aland, Mink, and Wachtel (2003: 29*). . Cf. for Figure 14 Mink (2003: 60–62). Figure 11 is a pre-form of Figure 14 here. . “Yet if he suffers as a Christian, he should not be ashamed; but he should glorify God in this part(?) / in this case(?) / on this behalf (?)”. The word μέρος has many meanings. . “Yet if he suffers as a Christian, he should not be ashamed; but he should glorify God in this name”. . “If you are reviled in the name of Christ, you are blessed”. . 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 elsewhere in the diagram with another question mark. . The level of probability depends on the position of a potential ancestor in a list with descending rates of agreement, cf. Figures 9, 21, 23, 35, and paragraph 5.3. . A few witnesses are lacking in Figure 16. Fragments P23 and 0166 contain very little text, and present few places of variation in these passages. Thus, no directions of predominant textual flow emerge. Fragments P54, 0173 and 1846 contain very little text, too. Here, however, there are too many witnesses with equally strong textual flows directed towards these fragments. With 2718S, the predominant textual flow from two directions is equally strong and has been drawn with interrupted lines. A pre-form of this graph was earlier made available for a lecture of Barbara Aland, which was printed in the Korean Journal of Biblical Research (Aland 2000). . A section of Figure 17 including the textual flows which are third in probability is to be found in Mink (2003: 57 Note 17). Figures 16 and 17 do not include intermediary nodes (cf. paragraph 4.7). Usually the first 3 to 5 levels of probability are important when we are looking for the sources of the textual state of a descendant. In cases of variants which are connective because of their character and if the difference between the levels of probability is rather small, lower levels may be used to give a better choice. Cf. the method for constructing substemmata, paragraph 6. If the differences between the levels are small, the number of witnesses 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 importance, since the variants in those witnesses could be expected to be part of the initial text; see Aland, Aland, Mink, and Wachtel (2000: 21*–22*). . Cf. paragraph 4.10. . Otherwise one would have to assume that 19 variants (cf. Figure 21, xausy field) emerged on the way from A to 03 without any intermediate stages, or without any contamination in any of the non-preserved intermediary witnesses. . 03 actually agrees with 1243 in more places than 025 does, but the degree of kinship is smaller. This degree is, however, decisive because of the respective number of places of variation where a pair of witnesses can be compared. A certain degree of fragmentation is normal. 1243 can be compared with 03 in 743 places, with 025 in 708 places. Only stronger fragmentation leads to less significant values resulting from the comparison, and moreover lessens the chance that the witness will become ancestor in substemmata, as it explains too few places of the descendants. . If the tradition is wide-spread, it may happen in a small number of cases that a descendant has a prior variant compared to all potential ancestors. Cf. paragraph 7. . For the conditions cf. paragraph 4.11. . For 025 an additional contribution of 7–8 is indicated. This is based on the fact that at one place it is not clear which variant should be the reading of the initial text A, and 025 makes an additional contribution only if it does not read the same variant as A (James 5:10/26-32). . The common omission of the sentence’s initial conjunction ὅτι in James 1:23/2 was regarded as not coincidental. The same applies to the addition of ἔργων in James 3:17/33. . James 2:3/50-56 (ὑπό, not ἐπί), 2:13/18 (κατακαυχ α ται, not –σθε or –σθω). . James 2:15/16 variant b: aorist participle λιπόμενοι vs. present participle λειπόμενοι; the pronunciation of the two words was the same. There is another place where a minimal contribution could be considered: James 5:20/20-28. 1243 has σώσει τὴν ψυχὴν ἐκ θανάτου. I derive this from σώσει ψυχὴν ἐκ θανάτου, the variant of 1175 and of no other ancestor. But although evolvement of the variant of 1243 from σώσει ψυχὴν αὐτου ἐκ θανάτου is less probable, it is not completely impossible. There are two more witnesses for the variant of 1243: 049 and 2492. 1243 is the closest potential ancestor of 2492. 049, however, is not part of the environment of 1243 at all (only 87% agreement) and only coincidentally reads the same variant, which goes back to the same source variant. . James 2:14/2-10 variant c: omission of an article which is not obligatory; 2:15/16 variant b: aorist participle λιπόμενοι vs. present participle λειπόμενοι – the difference is just one character, the pronunciation of the two words was the same; 2:19/8-14 variant e: reversal of words; 2:26/2-4 variant b: omission of γάρ ‘for’. . In two additional places 01 could explain the variants of 1243 (cf. Figure 24). In James 5:9/6-10, 1243 reads variant c: κατ’ ἀλλήλων ἀδελφοί, 01 has the same with slightly different orthography: κατὰ ἀλλήλων ἀδελφοί. Compared to variant a of the ancestors A, 03, 025 (ἀδελφοί κατ’ ἀλλήλων) this is a reversal of word order, which could have independently

Problems of a highly contaminated tradition

emerged several times. In James 5:12/40 (variant d) 01 (hand 1) and 1243, like several witnesses, change the James text (variant a, the reading of the ancestors of 1243) into the much better known parallel in Matthew 5:37, a variant which is not connective. . Cf. Aland, Aland, Mink, and Wachtel (1997: 28). . Cf. Blass and Debrunner (2001) §442, especially Note 14. . Cf. Aland, Aland, Mink, and Wachtel (2000: 22* Note 4). . The condition is that d has the initial text. If a is regarded as the initial text, the problem is 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 a large number of equally good hypotheses. . 01 has variant b in James 1:17/38. 03 has variant b in James 1:11/44. P23 in both cases has variant a. . The graph might be simplified as a result, by attempting to link 01 or 03 directly to P23, even though there is no direction of textual flow. Such a decision might, however, have far-reaching consequences. In this case, it would not only be decisive for an immediate and directed stemmatic connection whether a directed textual flow can be detected between two witnesses, but also whether the variants which are not in accordance with the direction of the textual flow are still needed for the construction of a stemma. It is better to postpone such simplifications until the consequences for the basics of the method and the construction of the global stemma are clear. . 04 is rather badly fragmented (482 out of 761 places of variantion are extant), cf. paragraph 9, so it is impossible to conjecture what the result would be, if the complete text would have survived. In the context of the principles of this procedure, however, this is not important. . James 1:18/26-28 (variant b), 1:20/12-14 (variant b), 1:22/14-16 (variant b), 1:26/2-4 (variant b), 2:13/20 (variant b), 2:26/27 (variant b), 3:4/32-42 (variant b). . James 2:14/2-10 (variant c), 2:16/42-46 (variant b). . James 2:15/3 (variant c), 2:18/24-34 (variant b), 2:19/8-14 (variant e). . At 2 out of these 3 places 1243, an ancestor in the substemma of 04, even reads the same variant as 04: James 2:18/24-34 (variant b), 2:19/8-14 (variant e). But 1243 is not an ancestor of 1175. . James 2:15/3 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 enrichment of a branch of the tradition with older variants, such a circle actually materialises in the construction of a group of substemmata. I am still expecting this to happen, as the problem arises in diagrams representing textual flows like in Figure 17, but with a larger range and taking into account potential ancestors with lower levels of probability. . Cf. paragraph 5.3. . If the local and the global textual flows do not correspond in this way, the global stemma is falsified. Cf. paragraph 4.6.





Gerd Mink . In James there are 1723 genealogical relationships between variants. 209 of these permit indirect relationships, as they determine source variants of source variants. The number of witnesses concerned by this is relatively small, however. A pair of witnesses can on average be compared at 641 places. In only 4.5 instances on average this does not result in direct genealogical relationships between variants of a pair. 697 places on average are comparable if fragments are excluded which are only comparable at 50 places at most. In that case 4.9 places on average do not produce direct genealogical relationships. . The direction is determined by the values found in Figure 23 in the xausy and yausx fields. For 1739 the ratio of the values is 35:32. This implies a predominant textual flow from 1739 towards 1243. If only Chapters 1–3 are taken as a basis, the ratio is 14:19, so the textual flow changes direction. . This does not mean that 1243 becomes an ancestor in an optimal substemma for Chapters 1–3, as the textual flow from 1243 towards 1739 is included in the textual flow from A towards 1739 and from 03 towards 1739. . Instead of 31:30 (in Figure 35) the ratio is now 18:21. . Cf. for the concept ‘stability of textual flow’ also paragraph 6. . Cf. paragraphs 8 and 9 and Figure 43. . A section in a stemmatic graph rather creates the possibility of adding information to the edges. The values displayed are typical for the top of the stemma: the percentage of agreements is not particularly high. The unusual role of 03 becomes clearer through the figures. The number of posterior variants of the ancestors compared to the prior ones indicates the relative quality of the edges. . For instance, in the method discussed the fact that there is one initial text may be an undesired artefact in some other traditions. Another (negative) artefact is the fact that a relationship in which, as a result of contamination, an immediate descendant acquired more prior variants than one of the immediate ancestors would not be detected. In this case descendant and ancestor may even change places.

References Aland, B. (2000). “Die editio critica maior des Neuen Testaments. Ihre Anlage, ihre Aufgabe, die neu entwickelten Methoden der Textkritik”. Journal of Biblical Text Research, 7, 7–23. Aland, B., K. Aland, G. Mink, & K. Wachtel (Eds.). (1997). Novum Testamentum Graecum, Editio Critica Maior, ed. by Institute for New Testament Textual Research. Vol. IV, Catholic Letters, Installment 1, James. Stuttgart: Deutsche Bibelgesellschaft. —— (2000). Novum Testamentum Graecum, Editio Critica Maior, ed. by Institute for New Testament Textual Research. Vol. IV, Catholic Letters, Installment 2, The Letters of Peter. Stuttgart: Deutsche Bibelgesellschaft. —— (2003). Novum Testamentum Graecum, Editio Critica Maior, ed. by Institute for New Testament Textual Research. Vol. IV, Catholic Letters, Installment 3, The First Letter of John. Stuttgart: Deutsche Bibelgesellschaft.

Problems of a highly contaminated tradition

Aland, K. et al. (Eds.). (1987). Text und Textwert der griechischen Handschriften des Neuen Testaments, Bd. 1. Die Katholischen Briefe. ANTT 9–11. Berlin, New York: De Gruyter. —— (1991). Text und Textwert der griechischen Handschriften des Neuen Testaments, Bd. 2. Die Paulinischen Briefe. ANTT 16–19. Berlin, New York: De Gruyter. —— (1993). Text und Textwert der griechischen Handschriften des Neuen Testaments, Bd. 3. Die Apostelgeschichte. ANTT 20–21. Berlin, New York: De Gruyter. —— (1998–1999). Text und Textwert der griechischen Handschriften des Neuen Testaments, Bd. 4. Die synoptischen Evangelien. ANTT 26, 28, 30. Berlin, New York: De Gruyter. Blass, F. & A. Debrunner (2001). Grammatik des neutestamentlichen Griechisch. Bearbeitet von Friedrich Rehkopf (18th edition). Göttingen: Vandenhoek & Ruprecht. Metzger, B. M. (1971). A Textual Commentary on the Greek New Testament. New York: United Bible Societies. Mink, G. (1993). “Eine umfassende Genealogie der neutestamentlichen Überlieferung.“ New Testament Studies, 39, 481–499. —— (2000). “Editing and Genealogical Studies: the New Testament”. Literary and Linguistic Computing, 15, 51–56. —— (2002). “Kohärenzbasierte Genealogische Methode – Worum geht es?” . —— (2003). “Was verändert sich in der Textkritik durch die Beachtung genealogischer Kohärenz?”. In W. Weren & D.-A. Koch (Eds.), New Developments in Textual Criticism: New Testament, Early-Christian and Jewish Literature (pp. 39–68). STAR 8. Assen: Royal van Gorcum. Salemans, B. J. P. (2000). Building Stemmata with the Computer in a Cladistic, NeoLachmannian, Way. Nijmegen: Nijmegen University Press. Spencer, M., K. Wachtel, & C. J. Howe (2002). “The Greek Vorlage of the Syra Harclensis: A Comparative Study on Method in Exploring Textual Genealogy”. TC: A Journal of Biblical Textual Criticism 7. .



Kinds of variants in the manuscript tradition of the Greek New Testament Klaus Wachtel Westfälische Wilhelms-Universität, Münster

.

Introduction

One problem which editors of the Greek New Testament have to face is the large number of manuscripts which transmitted the text from antiquity to the age of printing. For the first installment of the Münster Editio Critica Maior (ECM)1 we had to deal with a total of 553 complete or fragmentary manuscripts. For the Gospels, this number is even three to four times larger. A solution to this problem is offered by an outstanding feature of rich manuscript traditions of antique texts in general: most of the manuscripts are very much alike. Before we decided which manuscripts to collate in full for the ECM, all of them had been tested by probe collations in 98 units of variation in the seven Catholic Letters.2 We could exclude more than two thirds of the manuscripts available, because with few exceptions they witness to the late medieval Byzantine text, the Greek Vulgate or Koine. The manuscript basis of the ECM as well as that of Gerd Mink’s study in the present volume is a selection of 164 complete or fragmentary manuscripts. The Byzantine text itself is represented in the edition by seven manuscripts which share the majority readings at the 98 test passages with a rate of nearly or actually 100 percent.3 I am going to deal first with the variants which distinguish the late Byzantine from earlier textforms and, secondly, with variations which occur within the Byzantine tradition. The first part aims at a revised assessment of the late Byzantine textform, the second at a better documentation of the Byzantine textform itself: one which would not simply have to rely on an admittedly arbitrary rate of agreements with the majority text, which we set at 90% for the Catholic Letters.



Klaus Wachtel

It may be useful to define the terms ‘Majority Text’ and ‘Byzantine Text’ at this point. ‘Majority Text’ is a merely quantitative term while ‘Byzantine Text’ is a historical and text-critical term. The Majority Text consists of the majority readings and passages that have been transmitted without substantial variation. A majority reading is a variant attested to by the majority of extant Greek manuscripts. As a rule, such a reading is attested to by the oldest and best as well as by the largest number of medieval manuscripts. Normally, there is little reason to doubt that it is also the original reading. Yet the Byzantine Text is characterised by those majority readings which differ from the supposedly original text and additionally by those readings which result from a division of the main stream into two and, sometimes, three branches. The branching in the main stream of the manuscript tradition allows the Byzantine and Majority texts to be clearly distinguished. Moreover, it allows us to understand why, for example, two manuscripts that agree with the Majority Text at a rate of about 90%, can still differ from each other at 20% of the variants in question. It should be possible to make use of these differences to come to grips with the problem of an adequate representation of the so-called Byzantine subgroups. I shall return to this point later. But for now some remarks about the features of the 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 the first place there are those variants which enhance the pragmatism of the text. Insertions of particles and pronouns which join sentences or improve their inner structure occur very often. They make it easier to read the text and appear to be meant to avoid any misunderstanding. Forms, word-order, and the wording itself are changed to the same effect. Although there are cases of Byzantine omissions, this textform is considerably longer than older textforms. In many places it seems that a striving for clarity and completeness motivated the rise and dissemination of Byzantine readings. Moreover, we find many harmonisations with the closer or wider context. The wording of the synoptic Gospels is assimilated and quotations from the Old or New Testaments are rendered more precisely. Nevertheless, although there is abundant variation, it is rare for such interpolations to change the meaning of the text considerably. As a rule they only emphasise what the text says in older text forms as well. Are these variants intentional changes? Might they even be regarded as traces 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 in the past and is still held by some.5 I think we can derive persuasive arguments against this theory from the nature of the variants found in the Byzantine text as compared with older text-forms: First, the changes are inconsistent. There are many places where there is no smoothing to improve the coherence of the text. I am thinking of some difficult passages in Paul, for example. On the other hand, one sometimes finds an obviously intentional change close to a very similar instance which has been left untouched. This may be demonstrated by the following example. In the second chapter of Luke, the infant Jesus is hailed as the coming salvation by Simeon, a man inspired by the Spirit. According to the oldest and best manuscripts, “his father and mother marvelled at what was said about him” (Lk 2:33). The majority of the Greek as well as Latin, Syriac and Coptic manuscripts have “Joseph” instead of “his father”. For obvious reasons, one may think – wasn’t Jesus born of a virgin? Did he not say that his father is in heaven? Might this variant not be classified as an “orthodox corruption of scripture”, a term used by Bart Ehrman in his celebrated study of the same title (Ehrman 1993)? Similarly, it is said in Lk 2:43 that Jesus’ parents did not know that he was with the teachers in the temple, whereas the majority of witnesses read that “Joseph and his mother” did not know where he was. But between these passages there is another where Joseph and Mary are called “his parents” (Lk 2:41) almost unanimously, and at 2:48 Mary says to Jesus, according to all but a few witnesses, “Your father and I have been looking for you anxiously”. This inconsistency, which can be observed in many more instances, excludes orthodoxy as the conscious motive behind the variants that distinguish the Byzantine from older text-forms. If we study the kinds of variants found in our oldest and best manuscripts, we will observe that the characteristics of supposedly typically Byzantine readings occur in these, too, and moreover in each and every manuscript. This means that such readings are not typical of the Byzantine text but of manuscript traditions in general. A scribe occasionally does not write what he reads in the exemplar, but what he understands the text to mean instead. Hence we frequently find what may be called scilicet variants. They normally result in an interchange of synonyms. The replacement of “his father” with “Joseph” in the above example is a variant of this kind, too. The scribe may very well have been subliminally puzzled by the wording of the text he was copying, and his interpretation crept into the text. The oldest Greek manuscript witnessing to this variant is from the 5th century, but there are many clearly secondary majority readings found in the earliest witnesses. Defenders of the general originality of the Majority Text have tried to strengthen their case by





Klaus Wachtel

listing single majority readings attested by early papyri. However, these readings demonstrate neither the early provenance of the Majority Text nor its originality. Rather, they demonstrate only the early genesis of many of the majority readings and the susceptibility of scribes to what Eduard Schwartz called “half-conscious trivialisation”.6 Yet on the other hand, there are many atypical Byzantine readings, i.e. majority readings which by no means smooth out the text but instead render it more difficult or seemingly distorted by obvious scribal errors. A majority reading from the letter of James serves as an example. Near the end of the letter (4:13f.), the author exhorts those striving for worldly gain: “Come now, you who say, today or tomorrow we will go to that town and spend a year there and trade and get gain, you who do not know what will be tomorrow, what your life will be like, for you are [ἔστε] a vapour appearing for a little while ...”.7 The majority reading here is “for there will be [ἔσται] a vapour for a little while”, or, referring to life, “it will be [ἔσται] a vapour for a little while”. This is an obvious scribal slip with an equally obvious reason: interchange of the diphthong -αιand 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 that the distinctive readings of the Koine text accumulated in stages (cp. Wachtel 1995: 180–198). It is probable that the last, decisive step to standardisation was the change to minuscule production in the ninth century. It was for this purpose that archetypes of many manuscript traditions of Greek literature were produced. For the Greek New Testament, this was the text used and to be used in liturgy and theological study. There can be little doubt that the Byzantine text is distinguished from the original text by many readings. There seems to be a large gap between the socalled Byzantine texttype and textforms preserved in manuscripts from the 3rd to 4th centuries if we focus on the variants attested to by the mass of younger manuscripts against the noble few. In recent times, editors of the Greek New Testament have striven for a reconstruction of the original text according to those old and trustworthy witnesses and have shunned Erasmus’ Textus Receptus, which made the Byzantine text dominate printed editions of the Greek New Testament for three centuries. But if we take the text of the Nestle-Aland edition (NA 27), which is most commonly used for scholarly purposes today, as a reasonably good approximation to the original text, and calculate the distance of each manuscript from this text by counting agreements with and differences from it, we find that pure Byzantine manuscripts come closer to the ‘original text’ than many older documents and their descendants. This has motivated a new interest in the Byzantine text. As editors of the Greek NT, we take it more

Kinds of variants in the Greek New Testament

seriously in all cases that do not show clear signs of scribal error. It is indeed worthwhile to ask which “lines of textual tradition ... fed into the tenth and eleventh centuries”, as did Frederik Wisse in developing the Claremont Profile Method for classifying and evaluating manuscript evidence (Wisse 1982: 5).

. Group readings of Koine manuscripts Due to the large number of extant manuscripts and the fact that most of them witness to the Byzantine text, a New Testament editor’s first task is the determination of three classes of manuscripts: 1. those witnessing to non-Byzantine texts, which probably represent older text-forms; 2. those witnessing to the Byzantine text, of which as many as possible will be eliminated from the list of witnesses that should be included in the edition; 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 question is, which manuscripts are suited to serve as representatives of the Byzantine text in a more discriminating and not merely quantitative manner. I shall use test collations of 1,785 manuscripts for this. These collations were done at 153 short passages in the Gospel of John as a first step towards a future critical edition of this writing. The results were entered into a database. I shall use these results to mark out Byzantine groups for the purpose of finding which manuscripts are typical of the main varieties of the Byzantine text. Such groups may provisionally be defined as sharing more readings with each other than with the majority text. Not all kinds of readings are suited for this evaluation. Above all it is necessary to distinguish between variants and mere readings. A reading is the generic term for the wording of a passage by which a manuscript is distinguished from one or more or from all other manuscripts. A variant is defined as one of at least two readings of the same textual unit which is grammatically correct and logically possible. A reading which does not fulfil these criteria is an error. This means that the manuscript will be treated as witnessing defectively to the variant rendered. Like errors, orthographically or morphologically equivalent forms of the same variant are not classed as textual differences.





Klaus Wachtel

Groups of manuscripts as shown below in the extract from the table of manuscript groupings were determined by the following procedures and qualifications: 1. For each pair of manuscripts which share at least one reading apart from the majority text, the total numbers of agreements including and excluding the majority readings were calculated. This was done by a program which compared their patterns of variants at the 153 test passages. 2. Two manuscripts qualified as members of a group if the percentage of their mutual agreements was greater than the percentage of the readings that one or both of them shared with the majority. 3. The degrees of agreement reached by a manuscript as compared to others were classified. Only those manuscripts which showed at least a third-rate degree of agreement with the base manuscript were included in a group.8 4. If the condition under point 2 was met, but the percentage of agreements apart from the majority text was less than 50%, the respective manuscripts were regarded as weakly related. Manuscripts with such rates of agreement are found in many groups defined by the condition under point 2. This sifting procedure resulted in 1,125 groupings like those shown in the extract. Most of them consist of between two and ten manuscripts, and in the realm of the Koine it is not uncommon for some of the manuscripts to agree at a rate of 100 percent. Such manuscripts are very likely to have been copied from a common exemplar, which may have survived as one of the group members. The biggest clear-cut Byzantine group was named Kr by Hermann von Soden, who discovered it (Soden 1911: (I.2) 757–765). The “K” means Koine, the “r” recension. It is a diligently made manuscript edition which shows the degree of perfection that could be achieved by scribes. There are 40 codices containing the gospel of John which witness to this edition without any deviation at the 153 test passages. The group is distinguished from the majority by four readings that, while supported by Kr , are usually supported by considerably more manuscripts besides. The lowest number of witnesses for one of the Kr readings is 272, the highest 367. This indicates that the editor(s) made deliberate decisions where they knew of variants. There are four group readings of Kr at the test passages, but they are not distinctive readings in the proper sense of the word, as they are shared by manuscripts belonging to other groups. This is the rule with manuscripts which have such a large share of majority readings. Thus it is clear that the starting point for an inquiry into the genealogical background of Kr must be the pattern of variants distinguishing the group from

Kinds of variants in the Greek New Testament

Table 1. Extract from the table of manuscript groupings ms MT 4 8

10

11

13

14

cms dg n

90.8% 973 2 97.4% 65 2 774 1 1168 3 1514 1 90.8% 342 1 895 2 1091 2 1194 2 1517 2 2676 3 248 2 96.1% 1212 2 570 1 1207 1 200 2 944 2 905 2 1444 2 188 2 1179 3 1351 2 2682 2 29 3 148 3 75.7% 543 2 828 3 788 3 346 2 98.7% 140 1 1343 1 2224 1 2522 1 95 1 405 1 123 2 208 2 1080 2 1191 2 1225 1 1585 1

1 8 2 1 3 48 1 1 1 1 1 1 1 3 3 2 3 1 3 1 1 14 131 3 2 1 1 1 1 6 6 6 6 6 20 10 6 10 10 11 9

agr1

agr2

92% (129/141) 98% (150/153) 98% (150/153) 98% (150/153) 98% (150/153) 100% (12/12) 99% (151/153) 98% (150/153) 98% (150/153) 96% (145/151) 96% (142/148) 93% (142/152) 99% (152/153) 99% (151/152) 99% (150/151) 99% (150/152) 99% (150/152) 99% (149/151) 99% (148/150) 98% (150/153) 98% (150/153) 98% (47/48) 97% (31/32) 97% (147/152) 96% (129/134) 93% (119/128) 91% (138/152) 89% (125/140) 89% (73/82) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (150/150) 100% (106/106) 99% (152/153) 99% (152/153) 99% (152/153) 99% (152/153) 99% (152/153) 99% (152/153)

67% (2/3) 100% (1/1) 100% (1/1) 67% (2/3) 100% (2/2) 100% (1/1) 100% (14/14) 92% (12/13) 92% (12/13) 77% (10/13) 91% (10/11) 100% (9/9) 100% (5/5) 100% (5/5) 100% (6/6) 100% (6/6) 100% (5/5) 100% (6/6) 100% (4/4) 100% (5/5) 100% (4/4) 100% (1/1) 100% (1/1) 100% (4/4) 100% (5/5) 80% (20/25) 79% (26/33) 89% (23/26) 87% (13/15) 100% (2/2) 100% (2/2) 100% (2/2) 100% (2/2) 100% (2/2) 100% (1/1) 100% (2/2) 100% (2/2) 100% (2/2) 100% (1/1) 100% (2/2) 100% (1/1)

ms: manuscript MT: percentage of agreements with the majority text cms: compared manuscript dg: degree of agreement reached by the compared manuscript n: number of manuscripts for which the same degree of agreement is reached by the compared manuscript agr1: percentage and number of agreements proportional to the number of shared test passages including majority readings agr2: percentage and number of agreements proportional to the number of shared test passages excluding majority readings Examples For manuscript 4 (ms, first line) there is only one manuscript which agrees more frequently with it than with the majority text: manuscript 973 (cms). The figures in agr1 show that the two manuscripts agree at a rate of 92% or in 129 of the 141 test passages they share including the majority readings. According to agr2, 4 and 973 agree at a rate of 67% or in 2 of the 3 test passages they share apart from the majority readings. According to the entry in dg, this is only the second best degree of agreement reached by 973. The entry in n shows that 4 is the only manuscript which 973 agrees with at the same degree.





Klaus Wachtel

Table 1. (continued) ms MT

cms

2509 584 1058 15 93.5% 2562 1439 1163 53 902 2374 2502 18 97.4% 35 55 128 201 479 480 645 696 769 789 867 928 955 1023 1072 1117 1147 1339 1401 1493 1496 1550 1560 1584

dg n

agr1

agr2

2 2 2 1 2 1 3 3 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

99% (152/153) 99% (151/152) 99% (104/105) 99% (151/153) 98% (149/152) 98% (145/148) 95% (146/153) 95% (146/153) 95% (143/151) 94% (141/150) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153) 100% (153/153)

100% (1/1) 100% (2/2) 100% (2/2) 100% (8/8) 90% (9/10) 90% (9/10) 75% (6/8) 86% (6/7) 100% (4/4) 100% (2/2) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4) 100% (4/4)

10 10 15 1 1 1 1 1 1 2 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40 40

This means that 4 and 973 are only weakly related. But as there is no other manuscript which comes closer to 4, we might still start investigating which non-majority readings connect the two manuscripts, if we were interested in the relations of ms. 4 with the varieties of the Byzantine text. We might then look at the manuscripts to which 4 is related by 973 (i.e. the closer relatives of 973 not shown in the present extract). The last manuscript in this extract, 18, which agrees with the majority text at a rate of 97,4% is shown to have many relatives which share more readings with it than with the majority. The first compared manuscript, 35, is one out of 40 which agree at the same rate of 100% with ms. 18, including or excluding the majority readings. Manuscripts 4 and 18 represent extremes. Most groupings consist of between two and ten manuscripts.

the majority text (“MT” in Table 1). The total number of Byzantine groups sharing at least one reading with Kr apart from the majority is 391. 57 of them are connected by patterns of variants which include the complete Kr pattern. The total number of manuscripts involved is 127. The following table shows the patterns coming closest to that of Kr , being represented by minuscule 18.9

Kinds of variants in the Greek New Testament

Table 2. Ms.

% MT

Group Pattern

18 83

97.4 96.7

141 147

96.7 96.1

155

96.1

167

95.4

27:3,

32:2,

189

93.5

12:3,

32:2, 39:4

246 386

97.4 96.7

32:2, 32:2,

394

96.7

32:2,

547

96.6

32:2,

553

96.7

32:2,

685

96.7

32:2,

55:5, 69:3, 55:5, 76:2, 55:5,

689

96.1

32:2,

55:5,

763

96.7

32:2,

781

96.7

797

95.4

32:2, 54:8, 32:2, 32:2B, 54:7,

16:1/2B, 26:3,

32:2, 32:2,

55:5, 55:5,

32:2, 32:2,

55:5, 55:5, 69:3, 55:5, 99,2, 55:5, 60:3, 66:4, 55:5, 69:3, 89:4, 55:5, 55:5, 96,3, 55:5,

32:2,

9:2, 12:3,

18:2,

100:4, 100:4, 127:9, 100:4, 100:4B, 100:4, 151:3 100:4B,

150:4 150:4 150:4 150:4 150:4 150:4

100:4B, 132:3, 134:3, 100:4B, 100:4,

150:4

100:4B, 118:3, 100:4,

150:4

100:4,

150:4

100:4B, 100:4B,

150:4, 152:5 150:4

55:5, 76:2, 55:5,

100:4B,

150:4

100:4B,

150:4

55:5, 69:1/2D

100:4,

150:4

150:4 150:4

150:4

The first line shows that ms. 18, a minuscule from the 14th century which represents von Soden’s Kr in a particularly pure form, agrees with the majority text at a rate of 97,4% of the test passages. The group pattern consists of variants at the test passages that distinguish the group from the majority (e.g. 32:2 is test passage 32, variant 2). Exactly the same pattern is shared by 41 manuscripts, 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 the majority at one more test passage (127:9). The mss. which follow in the table all include the Kr pattern, but all differ from the majority at one or more additional test passages. (Note that in ms. 246 the difference is but a subvariant of 100:4, indicated by the “B”.) 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 than 92% of the test passages. We have witnesses more remote from the majority text for each single reading of the pattern, but not for the pattern itself. This again indicates that the text of Kr was carefully edited. The editors wanted the new text as much as possible to be in concordance with those in official use, thus introducing a new standard. Our chance to trace the genealogy of Kr and the groups containing its characteristic pattern is not very good. As a rule the readings of the patterns including the one of Kr are attested to by relatively strong minorities of manuscripts and were occasionally picked up by scribes. But our test collations are not a suitable basis for genealogical analysis anyway. This is because the system of test passages cannot serve as a model for the entire text. They were selected in the first place for the purpose of distinguishing witnesses of the Byzantine text from those of other textforms. Yet the present study shows that we can go one step further. The patterns of variants at the test passages allow a more discriminating selection of Koine manuscripts, which should be subjected to full collations and then to genealogical methods as shown by Gerd Mink in the present volume.

. Conclusions 1. There are 1484 manuscripts containing the Gospel of John, which agree with the majority at more than 90 percent of the test passages. The remaining 301 manuscripts are certainly not all candidates for the apparatus of a future Editio Maior of the Gospel of John, because many of the differences from the majority may turn out to be distortions of majority readings. This has to be worked out. 2. The Koine group Kr was identified by means of a distinctive pattern of variants at four test passages. This means that the large number of manuscripts which deviate from the majority in between one to all four of the Kr variants will be well represented by a single Kr manuscript.

Kinds of variants in the Greek New Testament

3. As a clearly defined group Kr can serve as a starting point for further clarification of the group structure of the Koine manuscripts. One question will be whether manuscripts which include the full Kr pattern are more likely to be predecessors or descendants of Kr . Another promising investigation will deal with those Koine manuscripts which share no reading with Kr at the test passages. 4. One basic conclusion may be drawn as to the usefulness of test collation for determining manuscripts for further study: it is demonstrated by the fact that the outlines of the structure of the tradition can be drawn on that basis.

Notes . ECM IV 1997/2003. . Text und Textwert 1989–1999. . Cp. ECM IV, 1: 11*–12*; ECM IV, 2: B8–B9. . This chapter contains statements and conclusions which are explained on a larger scale in Wachtel (1995). . Cp. e.g. Westcott/Hort (1882: 132–139); Soden (1911: (I.2) 707–713); Metzger (1992: 212). . Schwartz (1909: CXLVI) (“halbbewusste Trivialisierung”). . Translation quoted from the Revised Standard Version (2 1971). . A first-rate degree of agreement means that the manuscript being compared does not reach a higher percentage of agreements with any other manuscript. If it has a second-rate degree of agreement, there is at least one manuscript which comes closer to it than the one which the second-rate degree applies to. . [32:2 – 4,1 – ‘Jesus’ instead of ‘the Lord’. Context is “Now when the Lord knew that the Pharisees had heard that Jesus was making and baptizing more disciples than John, although Jesus himself did not baptize...”. It is quite likely that original ‘Jesus’ was replaced with ‘the Lord’ here to avoid the clumsy repetition of ‘Jesus’, the more so because ‘the Lord’ was increasingly used in reference to Jesus. There is a relatively strong fraction of 367 witnesses which share the reading of Kr , some of them ranging among the oldest and best. [55:5 – 5,44 – replacement of ‘from one another’ by ‘from men’ in the sentence “How can you believe, who receive glory from one another and do not seek the glory that comes from the only God.” The reading of Kr makes for a more obvious opposition to ‘the only God’. The total number of witnesses is 381 here. [100:4 – 7,53–8,11 – This passage refers to the story of the adulteress, who is brought to Jesus by the scribes and Pharisees, and who finds forgiveness with him. There is overwhelming evidence for the non-Johannine origin of the pericope (cp. Metzger 1994: 187–189), which is found in the majority of witnesses. Kr has it but obelises it, which is an obvious trace of editorial work. The total number of witnesses for this obelization is 272. [150:4 – 10,39 – “...again they tried to arrest him.” The variant of





Klaus Wachtel

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

References ECM IV (1997/2000). Novum Testamentum Graecum, Editio critica maior, IV: Die Katholischen Briefe, hg. von B. Aland, K. Aland †, G. Mink, K. Wachtel. Part 1: Text, Part 2: Begleitende Materialien. 1. Lieferung 1997: Der Jakobusbrief, 2. Lieferung 2000: Die Petrusbriefe, 3. Lieferung 2003: Der Erste Johannesbrief. Stuttgart. Ehrman, B. D. (1993). The Orthodox Corruption of Scripture: The Effect of Early Christological Controversies on the Text of the New Testament. New York/Oxford. Metzger, B. M. (1992). The Text of the New Testament, its Transmission, Corruption and Restoration. Oxford 1964 (3 1992). Metzger, B. M. (1994). A Textual Commentary on the Greek Text of the New Testament. Stuttgart 1971 (2 1994). NA27. Novum Testamentum Graece, post Eberhard et Erwin Nestle editione vicesima septima revisa communiter ediderunt B. et K. Aland, J. Karavidopoulos, C. M. Martini, B. M. Metzger, A. Wikgren. Stuttgart 1993. Schwartz, E. (1909). “Prolegomena zu Eusebs Kirchengeschichte”. Eusebius Werke II, 3. Teil: Einleitungen, Übersichten und Register (GCS 9,3) Leipzig. Soden, H. Frhr. v. (1911). Die Schriften des Neuen Testaments in ihrer ältesten erreichbaren Textgestalt. Bd. I, 1–3 Untersuchungen, Göttingen 21911; Bd. II Text und Apparat. Göttingen 1913. Text und Textwert (1989–1999). Text und Textwert der griechischen Handschriften des Neuen Testaments, ed. by Kurt Aland et al. Vol. 1: Die Katholischen Briefe (1987), vol. 2: Die Paulinischen Briefe (1991), vol. 3: Die Apostelgeschichte (1993), vol. 4: Die synoptischen Evangelien (1998/1999). Berlin. Wachtel, K. (1995). Der Byzantinische Text der Katholischen Briefe: Eine Untersuchung zur Entstehung der Koine des Neuen Testaments. Berlin. Westcott, B. F. & F. J. A. Hort (1882). The New Testament in the Original Greek. Bd. I Text, Bd. II Introduction. Cambridge/London 1881 and 1882. Wisse, F. (1982). The Profile Method for the Classification and Evaluation of Manuscript Evidence as Applied to the Continuous Greek Text of the Gospel of Luke. Studies and Documents 44. Grand Rapids, Michigan.

How shock waves revealed successive contamination A cardiogram of early sixteenth-century printed Dutch Bibles A. A. den Hollander Universiteit van Amsterdam / Vrije Universiteit Amsterdam

. Introduction1 One of the complex problems a philologist has to deal with is a text tradition that is so entangled, that its transmission cannot be assessed in a straightforward way. Such a text tradition is generally considered to be ‘contaminated’: some or all of its members present a text not derived from one single exemplar, but composed from several predecessors. Until recently scholars had no aids in trying to handle this problem.2 In their article in the first volume of Studies of Stemmatology, however, E. Wattel and M. J. P. van Mulken offered an instrument, which could deal at least partially with the phenomenon of contamination (Wattel & Van Mulken 1996). In their contribution they distinguished three types of contamination: 1. simultaneous: several exemplars being used at the same time; 2. 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 the text, etc.; 3. incidental: one exemplar being used for the text, and other exemplars used to verify or improve the text through local interventions. The instrument Wattel and Van Mulken offered helps to trace the second type of contamination: successive contamination. In addition, they suggest a procedure for handling stemmatological problems caused by this phenomenon. Suc-

 A. A. den Hollander

cessive contamination makes it impossible to draw one stemma which correctly represents the genealogical relationships throughout the entire text. Stemmas can only be drawn for parts of the text and will only represent the relationships in those particular parts. It is, therefore, advisable to split up the text into sections, for each of which one valid stemma can be drawn. For making sound text divisions in a (successively) contaminated text, Wattel and Van Mulken suggest using a so-called cardiogram of the text tradition. In this article I intend to show how a cardiogram has actually been of great help in tracing successive contamination in the text tradition of the Dutch bibles printed between 1522 and 1545. On the basis of a cardiogram the text tradition could be split up into parts, each of which corresponded with logical units: one or more bible books. It turned out that the bible text in these successive 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 (manuscripts/editions). In a text tradition with unambiguous relationships, the similarity (or distance) between two witnesses will remain more or less stable. In case of successive contamination the similarity (or distance) between two witnesses will change at the point of change in relationship3 – probably dramatically. In order to find out whether and at what points in a text such changes in distances between witnesses occur, Wattel developed a distance distribution function, which records the distances between each pair of witnesses at every instance (marked by a variant) of a text tradition. The starting point in the production process of a cardiogram is the list of variants. This list of variants should have a clear data structure and should contain at least the following elements: 1. heading line – presenting the total number of lines and all the witnesses (manuscripts or editions) involved. Each witness is represented by a unique siglum; and each next line stating the following elements in a fixed order: 2. location – of the variant, e.g. number of chapter, verse or line, etc.; 3. formula – mathematical representation of the relation between witnesses; 4. readings – 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 1 2.1 2.2 4

abcdefg abcd/efg ab/cd/efg abcde/fg a/bcdefg

(heading line) reading 1 / reading 2 reading 1 / reading 2 / reading 3 reading 1 / reading 2 reading 1 / reading 2

In this example the text has 6 lines, with one variant in line 1, two variants in line 2 and again one variant in line 4. Additional information may be added in other columns in the list of variants, such as a typology of variants and weight factors. A typology of variants assigns each variant to a certain type, e.g. the type “inversion of words”. Such a typology makes it possible to test which types of variants are kinship-revealing in a given text tradition, and which are not and should therefore be left out. Therefore, no decision concerning possible genealogical relevance of a certain kind 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 factor may be assigned to the various types of formulas, in this case concerning typology of variations, not of variants. Type 2 variations, for example, show “two competitive variant readings, which are present in precisely two true groups of text versions”;4 a true group being a group containing two or more members. Type 2 variations, which are considered to be the most important materials for the stemma, can be given a much higher weight factor than the other variation types. Again, the whole process of weighing may be changed at any time and is reversible as well. No final decision concerning possible genealogical relevance of a certain type of variation must be taken during the process of building the list of variants. After the list of variants has been established, the similarity between each pair of witnesses will be measured throughout the entire list of variants. This step will result in a list for each pair, stating whether the two do or do not share a reading. Two witnesses sharing the same reading will have a positive score (similar), two witnesses not sharing the same reading a negative score (not similar).5 When adding up all the scores of a given pair over a large number of formulas results in a positive score, it can be concluded that these witnesses correspond in most cases. When the result is a negative score, it means that the two witnesses have different readings in most formulas. Wattel defined a distribution formula, which makes it possible to visualize the ‘similarity-score’ of a pair of witnesses as a function over the entire text. The graph produced by this function fluctuates somewhere between its extreme values +100% (complete similarity) and –100% (complete dissimilarity).6 This

 A. A. den Hollander

F D C B2

25

47 –2

–92

–44 –99

–39

37 –88

97

100 78

26

89 –39

23

67 –39

27

89 –39

24

89 –39

B1 A4 A3 A2 A1 total 0

100

200

300

400

500

600

700

800

900

1000 1100 1200 1300 1400 1500

Figure 1. Similarity of all mss of Charroi de Nîmes with ms B1. Horizontally: line numbers of the text; vertically: sigla of the mss; on the right: maximum and minimum values of the ‘similarity-score’; on the left: the average values.

‘similarity-graph’ clearly shows how each of the other witnesses relates to a given witness and whether any dramatic changes in relationship occur. A relationship is stable when the similarity remains stable through the entire text. A stable relationship will be visualised by a flat graph without many fluctuations. The graphs in Figure 1 show the relationships of eight manuscripts (A1–4, B2, C, D and F) of the manuscript tradition of Charroi de Nîmes with manuscript B1 of the same tradition.7 The graphs clearly show that the relationships between B1 and B2 as well as between B1 and D are stable; B1 and B2 are constantly very similar (maximum +100, minimum +78, average +97, without many fluctuations), and B1 and D very dissimilar (maximum –44, minimum –99, average –92, without many fluctuations). In this article, however, we are not interested in stable relationships. We want to know for each pair of witnesses where major changes in their relationships occur. We are, in other words, not interested in the flat parts of the ‘similarity-graphs’, indicating stable relationships, but in those parts where the graph rises or falls sharply.8 These rises and falls can be visualised in another graph – a so-called shock wave. This shock wave shows for each pair of wit-

How shock waves revealed successive contamination 

F D C B2 B1 A4 A3 A2 A1 total 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 Nîmes. Horizontally the line numbers of the text, vertically the sigla of the manuscripts.

nesses where the ‘similarity-graph’ is flat and where it fluctuates. In the case of a sharp fall or rise, the ‘shock wave’ has a high value, in the case of a stable relationship it has a low value. The minimum value of the shock wave is zero (completely flat ‘similarity-graph’), and its maximum is one (maximum fall or rise). Counting up the values from the shock waves of a given witness with all the other witnesses of a text tradition results in the ultimate shock wave of a witness. Figure 1, e.g., shows the shock wave of manuscript B1. It represents the average value of all the shock waves of pairs of manuscripts of the tradition of Charroi de Nîmes containing manuscript B1. The points where this average value graph sharply rises or falls still indicate important changes in relationships of manuscript B. They seem to occur in the vicinity of line 500, 950, and 1050. It is possible to compute shock waves of all the members of a text tradition. Figure 2 shows the shock waves of all the manuscripts of the tradition of Charroi de Nîmes. The graph at the base line gives the average value of all the shock waves.9 This graph is the shock wave of the entire text tradition of Charroi de Nîmes.

 A. A. den Hollander

Figure 2 gives the heart beat – the cardiogram – of the text tradition. The points where the shock wave of the entire text tradition peaks indicate changes in relationship in the text tradition. When all the manuscripts, or at least a substantial number, peak at the same point this could well be an indication of successive contamination.10

. Early sixteenth-century printed Dutch bibles During the period 1522-1545 some eighty Dutch bible editions were published.11 Six of them had a reissue during that period, one as many as three reissues. Seventeen of these eighty-nine editions were complete bibles. Five editions contained the text of the Old Testament, or parts of it, only, and seventy-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 of the bible text in the various editions. The similarities and differences in reading between the editions were taken up into a list of variants. Separate lists were made for the Old and the New Testament. The list of variants for the Old Testament counted 5099 variants, and for the New Testament 4573 variants. With the help of the instruments Wattel had developed two separate cardiograms were made for the Old Testament and for the New Testament. Figure 3 gives the cardiogram of the Old Testament, consisting of the shock waves of thirteen editions of (parts of) the Old Testament, leaving out the reissues; at the base line is given the shock wave of the text tradition of the Old Testament in early sixteenth-century printed Dutch bibles.12 Several observations can be made after a first glance at the cardiogram. First, observing the shock waves of the thirteen editions it is obvious that the graphs are quite flat and do not show a lot of fluctuation. There are a number of local symptoms, such as JvL1526 en DP1527 peaking sharply at line 780 – and some others peaking less at the same spot. The graphs also show a few more general phenomena, such as the fluctuations at line 213, line 1145, and line 1300. Also the shock wave of the entire tradition at the base line is quite flat, and without a great deal of fluctuation, let alone any number of significant peaks. Except 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 point where the heart beat of the text tradition at the base line fluctuated visibly, a change of relationship did indeed take place. Even points with minimal fluctuation still marked an actual change in relationship somewhere in the text tradition.13 With the aid of the heart beat the text tradition could be split up

How shock waves revealed successive contamination  I II

III II

IV

V II

II VI

VII VI

VIII IV

CvR 1525 SM 1544 JvL 1527 WV 1542 JvL 1542 HPvM 1541 HvL 1538 JvL 1535 HPvM 1535 JvL 1534 WV 1533 JvL 1532 WV 1532 WV 1528 [1531] WV 1528 DP 1527 JvL 1526 HvR 1525 total 0

180

360

540

720

900

1080

1260

Figure 3. The heart beat of the Old Testament. Horizontally the line numbers (based on the list of variants), vertically the abbreviations of the various editions.14

into thirteen separate parts.15 Within these thirteen parts all the relationships remained stable. The thirteen parts all correspond with coherent text units from the Old Testament – one or more bible books16 – and for all thirteen parts separate stemmas were drawn:

 A. A. den Hollander

1. Genesis, Exodus, Leviticus, Numbers, Deuteronomy 2. Joshua, Judges, Ruth, 1+2 Samuel, 1+2 Kings, 1+2 Chronicles, Ezra, Nehemiah, Esther (canonical part) 3. Esther (apocryphal part) 4. Job 5. Psalms 6. Proverbs, Ecclesiastes, Song of Solomon 7. Isaiah 8. Jeremiah, Lamentations, Baruch, Ezekiel, Daniel, Hosea, Joel, Amos, Obadiah 9. Jonah 10. Micah, Nahum, Habakkuk, Zephaniah, Haggai, Zechariah, Malachi 11. 3+4 Esdras 12. 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 always resulted in change of relationship, somewhere in the text tradition. Therefore, with each transition a new stemma was required to represent the genealogical relationships in the next part. Not all thirteen stemmas, however, were entirely different, since two or more non-successive stemmas were sometimes identical. A total number of eight different stemmas was required to represent the genealogical relationships in the entire text tradition of the Old Testament in the early sixteenth-century printed Dutch bibles.17 The process of dividing the text tradition into sections on the basis of the cardiogram drew our attention to some striking features of the text tradition of the Old Testament. Why, for example, was there a change in relationships between the books of Isaiah and Jonah? These books belong to a coherent part of the Old Testament, major and minor Prophets, respectively, in which changes in relationship were not be expected. A closer look into the part of the Prophets through study of the list of variants, and the ‘similarity graphs’ in the relevant section of the text tradition, revealed the cause of the change in relationship: various sources were used for the books of the Prophets in the 1526 Bible of the Antwerp printer Liesvelt. Jacob van Liesvelt was the first one to publish a complete bible in the Dutch language.18 His aim was to provide the Netherlands with a Dutch translation of the Luther Bible. In 1526, however, Luther had not completed his bible translation in full. He had translated the entire New Testament a few years earlier,

How shock waves revealed successive contamination 

but at that time substantial parts of the Old Testament remained to be done. Luthers translation of the Prophets was not finished until the year 1534. In 1526 Luthers translations of two of the minor Prophets had been issued separately: Jonah and Habakkuk.19 Jacob van Liesvelt must have been able to lay hands on a copy of the edition of Jonah almost at once. Only a few months later, this Luther edition appeared in a Dutch translation in the Liesvelt Bible. For the other books of the Prophets, except for Isaiah, Liesvelt simply reproduced the existing Dutch translation from the 1525 edition of the Old Testament of the Antwerp printer Hans van Ruremunde (HvR1525), which explains the sudden change in relationship.20 A cardiogram was also computed for the New Testament, see Figure 4. On 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 or more bible books.21 1. 2. 3. 4. 5. 6. 7. 8. 9.

Matthew, Mark, Luke, John Acts Romans 1+2 Corinthians, Galatians, Ephesians, Philippians, Colossians, 1+2 Thessalonians 1+2 Timothy, Titus, Philemon Hebrews James, 1+2 Peter, 1+2+3 John Jude Revelation

A total number of eight different stemmas was necessary to represent all the various relationships for the nine sections.22 The same stemma was valid for parts (2) and (9). In this case, it was not very surprising that the same stemma was valid for these two parts. The text of most of the early printed sixteenthcentury editions of the New Testament had been derived from other existing Dutch translations. Some of the editions of the New Testament were printed and published in parts, usually one part containing the Gospels, another part all the Letters, and a third part the two remaining books Acts and Revelation. These parts were independently distributed and were, therefore, separatedly used as sources for new editions. Suppose that the text of an edition of the New Testament A went back to another existing Dutch translation. Suppose also that for the text of Acts and Revelation one of the above-mentioned separate editions was used, which gave a completely different text from the text in the existing edition. In that

 A. A. den Hollander (B)

II = IX III

I

IV

V

VII VII IX = II

VI

MC 1539 AvB 1525 JvL 1523.2 JvL 1523.1 CL 1524 JvG 1525 JvL 1526 MHvH 1527 HPvM 1543 JvG 1524 JvL 1522 JvL 1523 JvG 1526.2 AvB 1523 AP 1525 AvB 1524 SM 1545 AvB 1524 WV 1529 CvR 1525 HvR 1525 CvR

1528

WV

1528

JvG 1526.1 CvR

1526

MK/GvdH

1525

DP

1523

HF

1525

total 0

40

80

120

160

200

240

280

320

360

400

Figure 4. The heart beat of the New Testament. Horizontally the line numbers (based on the list of variants), vertically the abbreviations of the various editions.23

case the shock wave of this New Testament A would peak between just before and directly after Acts as well as just before Revelation, indicating changes in relationship.24 On the basis of this shock wave the text would have to be split

How shock waves revealed successive contamination 

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

. Conclusions 1. The cardiogram has proved to be a useful instrument for getting, at one glance, an overall picture of the genealogical relationships within the text tradition of the early sixteenth-century printed Dutch bibles. It should be stressed that this overview could be obtained on the basis of just the list of variants. 2. The cardiogram made it visible at a glance how stable the genealogical relationships within the entire text tradition were. The shock waves also clearly revealed these relationships for each of the witnesses. It should be noted, however, that the text tradition of the early sixteenth-century printed Dutch bibles is a very ‘clean’ one. Incidental contamination, for example, can obscure a clear view.25 3. By means of the cardiogram it was also possible to point out those places in the text where changes in relationship might have taken place. Further research with the benefit of the list of variants and the ‘similarity-graphs’ did in fact reveal successive contamination. As a result, the text tradition was split up into coherent parts, and for each of these a separate stemma was drawn. 4. It turned out that the coherent parts were not only genealogically coherent text units, but that these parts also yielded a profile of the actual used sources. The fact, for example, that the single book of Jonah was separated from the other books of the minor Prophets as a non-coherent part raised the question what actual source had been used for the book of Jonah. The search for a potential source of Jonah led to the conclusion that its source had been a separately issued book of Jonah. This meant that other coherent text parts, established after division on the basis of the cardiogram, might also reflect (the size of) the actual used sources. This turned out to be true, as was shown above for separately issued parts of the New Testament. The same conclusions may hold for handwritten texts. Suppose that the cardiogram of a manuscript tradition divides up a text into parts. Since we know that manuscripts were sometimes split up into quires for reproduction, our text division on the basis of the cardiogram might very well result in text parts which 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 size of the exemplar.

Notes . This article uses results of my earlier study (Den Hollander 1997, esp. Chapter 3: “De teksttraditie”: 127–242). . In 1957 Paul Maas uttered “Gegen die Kontamination ist kein Kraut gewachsen” (Maas 1957: 31). . See for an example Wattel and Van Mulken (1996: 107). . See for this definition and a definition of other types of variations Salemans (2000: 23– 25). See also the contributions of Salemans and Wattel in this book. . See Wattel and Van Mulken (1996: 110–111) to learn how the scores could be computed. . The shock waves show the actual extreme values of each graph, as well as the average score. . This example was derived from Wattel and Van Mulken (1996: 116). . In order to level out the graph, the maximum rise or fall between two successive measuring points is about 30%. Increase of similarity, for example, from 5% to 95% requires a path of four successive measuring points minimal. The horizontal line consists of 300 measuring points, each measuring point corresponding with 5 textlines. In this way local information is suppressed in order to correct incidental influences. . That the shock wave of manuscript D hardly fluctuates is due to the fact that D hardly ever 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 (such as the list of variants). . Bible editions in this context are complete bibles, individual issues of the Old and New Testaments, individually issued books of the bible and fragments. I exclude individual editions of the Psalms and editions of the lessons from the ‘epistles and gospels’, since these two genres constitute separate text traditions. The bibles printed in this period form a well-defined group, and can be considered as an indepedent text tradition. . This cardiogram already gives an impression of which editions are related and which are not. Related editions will have a similar shock wave. . The small peak at line 175 (between Numbers and Deuteronomy), however, did not refer to 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 issue: HvR = Hans van Ruremunde; JvL = Jacob van Liesvelt; DP = Doen Pietersoen; WV = Willem Vorsterman; HPvM = Henrick Peetersen van Middelburch; HvL = Hansken van Liesvelt; SM = Steven Mierdmans; CvR = Christoffel van Ruremunde.

How shock waves revealed successive contamination . The parts covered lines (1) 0–123; (2) 219–509; (3) 515–519; (4) 525–566; (5) 572–717; (6) 723–781; (7) 787–851; (8) 857–1044; (9) 1050–1053; (10) 1059–1124; (11) 1130–1159; (12) 1165–1295; (13) 1301–1335. . Separate cardiograms can be made for each of the thirteen parts, which give a more detailed look into the interrelations; see Appendix A for the heart beat of part one of the Old Testament. . 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. 6 for 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, 392–395. . For the book of Isaiah Liesvelt once more made use of a different source. This time he published a Dutch translation of the Latin text of Isaiah from Oecolampadius (1525) Commentary on the book of Isaiah. This Latin text was not a Vulgate text, but Oecolampadius’ own Latin translation from the Hebrew. See Den Hollander (1997: 185). . The parts covered lines (1) 0–109; (2) 115–142; (3) 148–163; (4) 169–260; (5) 266–294; (6) 300–312; (7) 318–362; (8) 368; (9) 374–395. . See for the stemmas Den Hollander (1997: 99–211). . The total number of editions had been reduced through clustering, see Den Hollander (1997: 155–157). Only the representatives of these clusters have been presented in the cardiogram. The abbreviations consist of the abbreviated name of the printer and the year of issue: HF = Hiero Fuchs; DP = Doen Pietersoen; MK/GvdH = Merten de Keyser/Govaert van der Haeghen; CvR = Christoffel van Ruremunde; JvG = Jan van Ghelen; WV = Willem Vorsterman; HvR = Hans van Ruremunde; AvB = Adriaen van Berghen; SM = Steven Mierdmans; AP = Albert Pafraet; JvL = Jacob van Liesvelt; HPvM = Henrick Peetersen van Middelburch; MHvH = Michiel Hillen van Hoochstraten; CL = Cornelis Lettersnijder; MC = Matthias Crom. . Presuming that Revelation is the final book. Otherwise the shock wave would also peak directly after Revelation, dividing the text in five parts. . See, for example, the cardiogram of the text tradition of Perceval (Figure 5) in Wattel and Van Mulken (1996: 119).

References Hollander, A. A. den (1997). De Nederlandse Bijbelvertalingen/Dutch Translations of the Bible 1522–1545. Nieuwkoop: De Graaf. Hollander, A. A. den (1999). “Dat Oude ende dat Nieuwe Testament. Jacob van Liesvelt en de nieuwe markt voor bijbels in de zestiende eeuw”. Jaarboek voor Nederlandse Boekgeschiedenis, 105–122. Leiden: Nederlandse Boekhistorische Vereniging. Luthers (D. Martin) Werke. Kritische Gesamtausgabe. Die Deutsche Bibel. 12 vols. (1906– 1961). Weimar.





A. A. den Hollander

Maas, P. (1957). Textkritik (3rd. rev. edition). Leipzig: Teubner. Salemans, 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. & M. J. P. van Mulken (1996). “Shock Waves in Text Traditions. Cardiograms of the Medieval Literature”. In P. van Reenen & M. J. P. van Mulken (Eds.), Studies in Stemmatology (pp. 105–121). Amsterdam: John Benjamins.

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

WV 1542 JvL 1542 HPvM 1541 HvL 1538 JvL 1535 HPvM 1535 JvL 1534 WV 1533 JvL 1532 WV 1532 WV 1528 [1531] WV 1528 DP 1527 JvL 1526 HvR 1525 total 0

20

40

60

80

100

120

140

160

180

200

220

The manuscript tradition of the Cligés of Chrétien de Troyes A stemmatological approach Margot van Mulken Katholieke Universiteit Nijmegen

.

Introduction

In order to verify whether the proliferation of intricate relationships between manuscripts is not typical of the Perceval of Chrétien de Troyes, but at least typical of other works of the same author, I decided to examine the relationships between de manuscripts of the Cligés and to subject the manuscripts to a similar treatment as the procedure for the manuscripts of the Perceval (Van Mulken 1993). Like the Perceval, the Cligés is a complex tradition. To quote Foerster: So erklärt sich denn nur durch diesen merkwürdigen Zustand der Überlieferung die dann nicht mehr auffällige Tatsache, daß der Cligéstext jedesmal eine solch große Zahl von Kritikern gefunden hat, während andere Kristiantexte, [..] so gut wie ganz von denselben Herren Kritikern gemieden worden sind: denn hier handelt es sich um wirkliche Besserungen, und nicht um das Vorziehen einer unter zwei an sich passablen Lesarten; und es ist dazu nicht nur eine wirklich gründliche Kenntnis der Sprache, sondern auch wirklicher Scharfsinn unbedingt nötig. (Foerster 1910: lxxv)

The MS tradition of the Cligés consists of 7 more or less complete manuscripts (the fragments will be left out of consideration). In Table 1 all the manuscripts are listed, together with an approximate date attribution (based on Nixon 1993) and localization (based on Gregory & Luttrell 1993; Van Mulken 1993, 1999).



Margot van Mulken

Table 1. MS tradition Cligés Manuscript

Siglum

Date Attribution

Localisation

Tours, BM 942 Paris, BN, fr., 794 Paris, BN, fr., 1450 Paris, BN, fr., 12560

T A B C

12/13 13, 2nd q 13, 2nd q 13, mid

Paris, BN, fr., 1420 Paris, BN, fr., 1374 Paris BN, fr., 375

R S P

13, mid 13, 3rd q 13/14

Anjou Champagne North East Hand 1: East of Ile de France; Hand 2: West, Anglo-Norman Ile de France Yonne Arras

q = quarter, mid = middle

On the basis of textual differences and similarities a list of 370 variants has been established. Criteria for admittance into the list were derived from Den Hollander (1997). However, the following variants have not been included: 1. 2. 3. 4. 5.

Variants concerning a change in word order (see Van Mulken 1993) Lacunae or discrepancies in metre or prosody Variants concerning pronouns, prepositions, particles (Salemans 2000) Variants concerning prefixes (Salemans 2000) Variants concerning tense or mood (Salemans 2000)

It is important to first determine the relationships within the tradition without considering the direction of the variants (i.e. the origin of the readings), since in a first attempt the number of subjective interventions should be reduced as much as possible (in accordance with the Three Level Method, also known as Dees Method; Van Mulken 1993: 45–71). The list of variants, adapted to the guidelines published in Wattel and Van Mulken (1995), served as input to generate a table of quadruples1 and a first draft of a stemma (Figure 1). Of course, the reliability of this stemma depends on the evaluation of the philologist: a draft of a tree can always be drawn on the basis of quadruples, but the reliability rate will not always be high enough. All depends on the extent of noise the philologist is willing to accept, and the degree of inconsistency he/she is ready to allow in order to build a stemma. Some philologists insist upon the reduction of all noise, and are therefore willing to reduce the number of variants; other philologists are ready to allow for a reasonable amount of noise, since some variants may be due to parallelism (Salemans 2000; Van Mulken 1993: 29–32). In this particular case, I estimated that the first draft is far from reliable: only 103 out of the 370 variants (28%) are in concordance with the

The manuscript tradition of the Cligés of Chrétien de Troyes A

C

R

B

T

P

S

Figure 1. Draft structure Cligés, 72% of variants to be rejected

tree: a rather too large degree of inconsistency. We cannot but conclude that the Cligés is not a closed, but an open manuscript tradition. Some reputable philologists working on the Cligés were also disappointed with their trees. Compare the following laments: So ist denn schließlich der Stammbaum stellenweise recht unsicher – es müßten eigentlich deren mehrere aufgestellt werden je nach den verschiedenen Teilen. (Foerster 1910: lxxiii) Dieses schwankende Handschriftenverhältnis [..] führt dazu, daß man an vielen Stellen schwankend wird und mit der gewöhnlichen Formel des Stammbaumes nicht auskommen zu können glaubt. (Foerster 1910: lxxv) [..] Il (le copiste du groupe y) a copié un ms. contaminé qui offre dans son texte des leçons tantôt de S, tantôt de P, tantôt de B. C’est l’explication à adopter, si l’on imagine difficilement l’archétype y allant consulter trois mss. (Micha 1966: 114) Il ressort déjà que le ms. B n’est pas le seul oscillant, puisque S a des rapports assez suivis avec y. (Micha 1966: 112)

In such a case a philologist will conclude that the tradition is contaminated. So did these philologists. [Le ms. d’Annonay] confirme l’existence de ces groupes intermédiaires imputables à une tradition nécessairement contaminée. (Micha 1966: 121) Einzelne Schreiber, von ihren Auftraggebern mit den Vorlagen versehn, [hatten] zufällig beide Auflagen vor sich und [folgten] ihnen nach Gutdünken, falls nicht größere Lücken der Vorlage aus einer anderen, der 2. Redakzion angehörigen Hs. ausgefüllt werden mußten [...]. (Foerster 1910: lxxiv)

. Contamination If there is question of contamination, the premise is always that more than one 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 are always two copies of the same text in one location. However, there are two possible concepts of contamination. 1. Simultaneous contamination presupposes that the copyist used several sources simultaneously to produce a new text. When this is the case, it is only logical that the copyist did so intentionally: he compared readings and made choices. The role of the copyist in the reproduction process, therefore, was very important: the choice of a particular reading is a wellconsidered and deliberate choice, since the copyist judged that determining the ‘correct’ reading was hyper-important. 2. With successive contamination, the copyist refers to different sources one after another, and in this case, it is improbable that he does so consciously. The copyist is unaware of the fact that he is using more than one text as exemplar; he is unconscious of the differences between the versions. The explanation for the fact that he could have had several exemplars at his disposal lies in the transcription process: group production was not unfamiliar in the Middle Ages, when several copyists collaborated to produce a new text, and when copies were produced by quire. After transcription, all the quires were gathered and stitched. Since copyists did not distinguish between old and new quires, it became possible that newly transcribed quires were gathered together with quires of the original exemplar. In addition, it was possible that several exemplars of a text were present in a library already at the beginning of the transcription. In such a case, successive contamination could occur when a copyist, unaware of the differences between the exemplars, switched exemplars in the transcription phase. In three manuscripts of the Cligés tradition (P, C and S) paleographers have distinguished the presence of more than one hand (Stewart & Luttrell 1993). They are all manuscripts dating from the Later Middle Ages. So at least for these manuscripts, collaboration of copyists has been established. A combination of both types of contamination is of course possible, but it remains difficult to prove the existence of simultaneous contamination. The problem with successive contamination is only slightly smaller: the shift in relationships often occurred at a stage prior to the extant manuscript, when (one of) the ancestors of the extant manuscript may have been subject to successive contamination. In such a case, contamination can only be proved indirectly. What remains evident is that successive contamination always betrays itself by the occurrence of a sudden change in relationships.

The manuscript tradition of the Cligés of Chrétien de Troyes

The concept of contamination (and the type the philologist decides it belongs to) depends heavily on the view of text one ascribes to a copyist who writes in the vernacular. In the case of the Cligés, I find it highly improbable, also in view of the large number of singular readings in every extant copy in this tradition, that the copyists were so preoccupied by the exact ‘correct’ reading of a theme that they would have wanted to refer to several exemplars.2 The phenomenon of ‘mouvance’ and the liberal conception of text were generally accepted for the production of vernacular texts, in contrast to other genres (Zumthor 1972). The consequence of this presupposition, as Foerster himself suggests, is that several passages in the Cligés require the drawing of different trees. Successive contamination is an option that needs to be taken seriously in order to solve the contamination in the Cligés.

. The quire separator Thanks to the work of Evert Wattel (Free University Amsterdam), it is now possible to determine quite exactly at what places in the tradition a shift in relationships occurs, since a ‘sudden’ change in relationship can be detected by the presence of contradicting quadruples (see Note 1) (Wattel & Van Mulken 1993; Wattel this volume; Den Hollander this volume). In the case of a new affinity, manuscripts will regroup into new mutual relations. Between shifts in relationships, steady relations can be detected by the presence of consistent quadruples. According to this quire detector, the Cligés presents two important shifts in relationships. 1. The first consistent part covers the beginning of the tradition up to verse 1250 (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 which are responsible for the first shift: they change places in the region of verse 1250. It is striking that this shift actually occurs at a stage where a change of quires could have taken place: the length of the first part coincides with the number of verses in a quire consisting of 16 pages (i.e. 8 folios) with two columns of 38/40 verses each (i.e. 1216–1280 verses). This unit is among the most common





Margot van Mulken A

C

R

S

P

B

Figure 2. Structure I Cligés vv. 1-1250 A

C

R

B

T

P

S

Figure 3. Structure II Cligés vv. 1250-4800 A

C

R

T B

S

P

Figure 4. Structure III Cligés vv. 4800-end

quire partitions in the Middle Ages and the extant manuscript S has the same distribution. In the region of verse 4800 manuscript P changes places with a nontransmitted intermediary. This second change corresponds more or less with a quire of 24 pages (i.e. 12 folios) with 3 columns of 59/60 verses each (i.e. 4248–4320 verses).6 This is a less current format, but it is precisely the format of manuscript P in the Cligés 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 responsible for the multiple stemmata. This can only be established by careful codicological

The manuscript tradition of the Cligés of Chrétien de Troyes

Table 2. 1567 Einz li est touz li sens foiz Si que pres an est amuiz Mss ap

Einçois li est li sens failliz Si que pres qu il n’est amuiz Mss bcrsT

research into the extant manuscripts – and it remains possible that it was the ancestors of these manuscripts which were subject to quire shifts and not the manuscripts themselves. After having determined these consecutive structures, I recalculated the consistency within the trees and it appeared that now, in stead of 28%, 51% of the 370 variants could be preserved. This still implies a refutation of 49% of the variants, which is, of course, quite considerable. Every variant that has to be dismissed should therefore be carefully examined and evaluated in the light of parallelism. We shall not discuss all the contradictory readings, but many of the important conflicting variants concern cases of ‘rich rhyme’. For example Table 2. We see here that manuscripts A and P have a rich rhyme ‘foiz: amuiz’. According to Méla this must be an original reading (“[L]’accord AP justifie la correction [de C] pour une rime plus riche”) (Méla 1994: 136). However, the proliferation of rich rhymes is generally thought to be a posterior trend. In this respect, a remark of Foerster is relevant: “[...] daher auch die reichen Reime der späteren, schlechteren Hss.” (Foerster 1910: lxxiii). What is clear is that cases of rich rhyme are likely to have been subject to parallelism, and that they should therefore not be included in the variant list. In spite of the large number of variants that have to be dismissed, I consider the three structures as relevant and corresponding to actual underlying relation shifts.

. Orientation The consequence of the assessment of three underlying structures is that the other phases in the stemmatological process, according to the Three Level Method, are also to be multiplied by three. And this implies that for each structure the orientation and intermediate nodes must be determined. In earlier days, the orientation phase served as a basis to reconstruct the original text of the author. Nowadays, such an attempt is generally considered unrealistic and too far-fetched. The orientation phase is, however, still relevant in the study of the text production process: by orienting the structures we ob-



 Margot van Mulken

Table 3. 0313 tuit li baron les esgardoient por ce que biax et genz les voient car li vaslet molt lor pleisoient ne cuident pas que il ne soient mss ab

et li baron forment se taisent que li vallet trestuit lor plesent por ce que biaus et genz les voient ne cuident pas que il ne soient mss cprs

Table 4. 1027 et s’il n’aimme ne n’a amé donc ai ge en la mer semé mss ab

et s’il n’aime ne n’a amé donc ai ge en l’areine semé mss crT (p, s)

tain a better insight into the direction of the pedigree. In order to do this, I tried to determine the original reading of some of the more important variants. Here, I will discuss some of the variants that constitute the arguments for the final orientation. In Table 3 we see that mss. A and B share a reading that does not reflect a possible original reading (the intersection of verses 0313 and 0315, producing a repeated rhyme on -oient), whereas the other manuscripts present a more poignant rhyme. The reverse is true for the next variant in Table 4, where A and B share a probably original rhyme, whereas the other manuscripts have a reading that does not refer to the preceding passage of the ‘secret of the sea’ in verses 548-564 (cf. Méla 1994: 104). The orientation of the first structure must therefore not be looked for in the group AB nor in the group CPRST, and this implies that I located the orientation on the edge between manuscripts AB and CPRS(T). Since T only commences a few verses before the start of Structure II, because its first quire is lacking, it is difficult to determine the position of T in this structure. I decided to leave T out of the first Structure. In the second part, we find the readings as presented in Table 5. The versions in manuscripts APT, though not entirely identical, are similar enough to group them together confronting the other group BCR. The version in the first group is far more lively and detailed and can be considered archetypal. The modification of ‘iert toz jorz estable’ to ‘est si veritable’ (see Table 6) is perhaps semantically cognate, but the fact that four words were modified indicates that a case of parallelism is improbable.

The manuscript tradition of the Cligés of Chrétien de Troyes

Table 5. 1744 car de bien ferir se travaille el plus espés de la bataille vet ensi ferir un gloton que ne li valut un boton ne li escuz ne li haubers ne li valut un cendal pers ms a (ms s: pareisin) (mss. pT: k’en terre ne l’emporte envers)

va un chevalier envair si le fiert par si grant air que mort jus des arçons l’abat si qu’il ne se plaint ne debat (br) (c)

Table 6. 3118 cest amors ne fu pas resnable mes la moie iert toz jorz estable mss. ab

ceste amor ne fu pas renable mais la moie est si veritable mss bprsT

Table 7. 5725 onques meis si male golée ne pois tu doner au monde mss acrs

onques meis si male golée ne pot la mors haper au monde mss bp

In part III, in verses 5725-7 (see Table 7), we see that manuscripts BP present a reading that cannot possibly be original, since it occurs in a passage where Death is directly addressed, and the third person singular (‘pot la mors’) is a sudden stylistic change of perspective. The wordplay with ‘mors’ in the sense of ‘death’ and ‘bite’ in the context of ‘golée’ and ‘haper’ must be a refection. In the second and third parts of the MS tradition, in Structures II and III, the orientation must be located on the edge between AS and the other manuscripts.

. Conclusion We see, then, that the orientation, for each structure, must be looked for in the neighbourhood of manuscript A, the Guiot copy. This is not really surprising, since Micha and Foerster came to the same conclusion with regard to the general 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 S in the case of the Cligés. He motivates his choice as follows: Einmal ist es die Hs. S die Abschrift eines recht (auch des Französischen) unkundigen Schreibers nach einer guten Vorlage, die an manchen Stellen wieder so schlechte und wertlose Lesarten bietet, daß man nicht begreift, wie sie in den sonst so guten Text kommen konnten. [. . . ] Hier wird (meiner Wertschätzung der Hs. entsprechend) meist an S unter allen Umständen gegen die andern Hss. festgehalten. (Foerster 1910: lxxv–vi)

Micha, on the contrary, continued to consider the Guiot copy as the best manuscript, in the tradition of Bédier, and used it as manuscrit de base for his edition of 1957. Similar arguments led Lutrell and Gregory to edit the Guiot copy in 1993, just as Poirion decided to edit Guiot in 1993. It is true that manuscript A is a remarkable manuscript, but in Van Mulken (2002) I demonstrated that a picardian predecessor must be inferred in the case of the Cligés. The version of the Cligés as laid down in Guiot’s copy is coloured by distinct picardisms, although the champenois provenance of the manuscript remains undisputed: (one of) the ancestor(s) of Guiot must have been a picardian manuscript. But we also know that Chrétien’s dialect was originally a champenois dialect! This implies that the version of Cligés in manuscript A first travelled to Picardy, and then returned to Champagne to be transcribed in the Guiot copy. In other words: even the best manuscript in a tradition, on the grounds of quality and age, may have had a considerable history before it became the best manuscript. The dialectological argument confirms the fact that manuscripts in the vernacular had a complicate production history, as demonstrated by the stemmatological analysis of the Cligés, where the use of the quire separator throws an important light on the contemporary view on vernacular texts in the Middle Ages. Not only the Perceval, but also the Cligés proves to be a ‘rich’ tradition.

Notes . A quadruple is the minimal unit in a Type-2 variant: it covers two similarities and a difference. A variant such as ‘abrs/cp’ can be broken down into: ab/cp; ar/cp; as/cp; br/cp; bs/cp; rs/cp. The advantage of quadruples is that no more than two Type-2 variations exist that can be in contradiction with the quadruple: ‘ab/cp’ can only be in contradiction with ‘ac/bp’ and ‘ap/bc’. If a manuscript tradition is closed and linear, this implies that for each quadruple the score for the competing quadruples will be zero or next to zero (depending on the

The manuscript tradition of the Cligés of Chrétien de Troyes

degree of inconsistency admitted by the philologist). Thanks to the table of quadruples, the philologists 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 Chrétien (Foerster 1910: lxxiv), Micha thinks several copyists had more than one exemplar at their disposal (Micha 1966: 114). . All verse references follow CFMA Micha 1957. The following important variants are in accordance with the stemma structure of the first part: ab/cprs 0022; 0176; 0290; 0297; 0313; 0331; 0430; 1028; abp/crs 0044; 0054; 0327; 0432; 0631; 0905; 0973; abps/cr 0527; 0724; 0806; 1083 . The following variants are in accordance with Structure II: as/bcprT 1190; 1380; 1386; 1409; 1487; 1535; 1572; 1612; 1640; 2368; 2628.2; 2666; 2800; 2987; 3063; 3086; 3118; 3162; 3362; 3555; 3602; 3743; 3948; 4133; 4380; 4446; 4695; aps/bcrT 1344; 1590; 1613; 1633; 1638; 2360; 2512; 2628; 2945; 3105; 3420; 3478; 3545; 3631; 3651; 3657; 3776; 3824; 4400; 4476; 4607; apsT/bcr 1151; 1201; 1244; 1250; 1285; 1330; 1348; 1403; 1521; 1544; 1666; 1719; 1722; 1741; 1744; 1745; 1754; 1797; 2247; 2394; 2405; 2428; 2588; 2601; 2679; 2681; 2703; 2726; 2836; 2919; 2921; 3026; 3253; 3430; abpsT/cr 1296; 2402; 3106; 3237; 4082.2; 1232; 1244; 1322; 1544; 1602; 1719; 2375; 2394; 2428; 3086.2; 3577; 3707.2 . The following important variants are in concordance with Structure III: as/bcprT 5309; 5560; 5773; 6040; asT/bcpr –; cr/abpsT 5693; 5926; 6084; bp/cprsT 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 transition of an underlying structure into another cannot be detected, due to lack of variants at the precise border of a structure, it is very well possible that a different combination of underlying quires must be presupposed.

References Dees, A. (1988). “Analyse par l’ordinateur de la tradition manuscrite du Cligés de Chrétien de Troyes”. In D. Kremer (Ed.), Actes du XVIIIe Congrès International de Linguistique et de Philologie Romanes (pp. 62–75). Tübingen: Niemeyer. Gregory, S. & C. Luttrell (1993). “The Manuscripts of Cligés”. In Keith Busby, Terry Nixon, Alison Stones, & Lori Walters (Eds.), Les Manuscrits de Chrétien de Troyes: The Manuscripts of Chrétien de Troyes (pp. 41–48). Amsterdam: Rodopi. Hollander, A. A. den (1997). De Nederlandse Bijbelvertalingen 1522–1545. Nieuwkoop: De Graaf. Micha, A. (1966). La tradition manuscrite des romans de Chrétien de Troyes. Genève: Droz.



 Margot van Mulken

Mulken, M. van (1993). The Manuscript Tradition of the Perceval of Chrétien de Troyes, a stemmatological and dialectological approach. Ph.D. Dissertation: Amsterdam. Mulken, M. van (1999). “Les changements de parenté dans le Cligés de Chrétien de Troyes.” Atti dei Convegni Lincei, 151, 105–114. Roma: Accademia Nazionale dei Lincei. Mulken, M. van (2002). “Le manuscript de Guiot et les influences picardes dans le Cligés de Chrétien de Troyes”. In H. Jacobs & L. Wetzels (Eds.), Liber Amicorum Bernard Bichakjian. Nijmegen: University Press. Nixon, T. (1993). “Romance Connections and the Manuscripts of Chrétien de Troyes”. In Keith Busby, Terry Nixon, Alison Stones, & Lori Walters (Eds.), Les Manuscrits de Chrétien de Troyes: The Manuscripts of Chrétien de Troyes (pp. 17–26). Amsterdam: Rodopi. Salemans, B. (2000). Building Stemmas with the Computer in a Cladistic, Neo-Lachmannian Way. Ph.D. Thesis Katholieke Universiteit Nijmegen: Nijmegen University Press. Wattel, E. & M. van Mulken (1996). Shock Waves in Text Traditions – Cardiograms of the Medieval Literature. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 105–121). Amsterdam: Benjamins. Zumthor, P. (1972). Essai de poétique médiévale. Paris: Seuil.

Editions Foerster, W. (1910). Kristian von Troyes. Cligés, Textausgabe mit Variantenauswahl, Einleitung, Anmerkungen und vollständigem Glossar. Halle: Niemeyer3 . Luttrell, C. & S. Gregory (1993). Cligés – Chrétien de Troyes. Bury St. Edmunds: St. Edmundsbury Press. Méla, Ch. (1994). Chrétien de Troyes: Cligés, Edition critique du manuscrit B.N. fr. 12560. Lettres Gothiques. Livre de Poche. Micha, A. (1957). Les Romans de Chrétien de Troyes, Cligés, édités d’après la copie de Guiot (Bibl. nat., fr. 794). Les Classiques Français du Moyen Age. Paris: Champion. Poirion, D. (1993). Œuvres Complètes de Chrétien de Troyes. Paris: Gallimard.

P II

Textual variation

Genealogy by chance! On the significance of accidental variation (parallelisms) Ulrich Schmid Kirchliche Hochschule Bethel, Bielefeld / Vrije Universiteit, Amsterdam

.

Introduction

Relationships between witnesses of a given text tradition are usually defined by the variant readings they share. Structures like ‘groups’ or ‘clusters’ or ‘families’ appear and ideally an overall stemma of the tradition can be established by means of the kinship-revealing process of comparing the variant readings that are displayed by the tradition. Such a process can be executed in different ways. No matter which way is chosen, usually one presupposition is accepted by all: Agreement in reading is agreement in ancestry. Although we may consider some variant readings or even some types of variant readings as not very impressive with regard to their kinship-revealing nature, we usually do not exclude them from the body of variant readings which is registered, counted, and processed. This is all the more true for variant readings that are not mere ‘orthographicals’ or ‘nonsense readings’,1 but sensible readings providing a readable and intelligible text. If variant readings of the latter type are to be noted and found in more than one witness, we must not exclude them from our data processing. Quite to the contrary. This type of evidence is the backbone of every genealogical or stemmatological work: Agreement in reading is agreement in ancestry. Yet, the problem of accidental variation, i.e. identical variant readings created purely by chance, has not escaped the eye of textual scholars. Most recently B. J. P Salemans, using the designation parallelism to describe the phenomenon of accidental variation has treated the issue (Salemans 2000: 64–71). He formulates the “text-genealogical basic rule: A genealogical

 Ulrich Schmid

(or relationship-revealing) variant is a textual difference that fits well and inconspicuously in a text version” (Salemans 2000: 64). The rule is intended to leave the genealogically relevant information exclusively with readings that do not attract the attention of a copyist nor hint at his or her orthographical, interpunctional, dialectal, etc. proclivities. Additionally Salemans describes various forms of parallelisms, which include the exchange of synonyms on the level of individual words, exchange of tenses, numbers or cases on the level of individual lexems, and the “absence of small, (highly) frequently used words”.2 In his view textual phenomena of those types should not be used for genealogical purposes. Of course, it is one thing to theoretically assess the liability of certain types of variant readings to accidental variation, and it is quite another thing to actually observe real parallelisms. Therefore, it is very interesting to note that Salemans, from within his own deductive framework, is even able to observe these kind of phenomena after processing the textual information collected from a body of fourteen versions of the late medieval vernacular text Lanceloet van Denemerken. The way he proceeds is based upon complete (electronic) transcriptions of all fourteen versions. Then a computer program, designed to apply the text-genealogical rules and criteria as developed deductively, filters out all the information that is considered to be genealogically unfit according to these criteria. The remaining body of information is then used to build a stemma. A large part of Salemans’ study is devoted to the evaluation of his textgenealogical rules and criteria by comparing the results based upon the various types of unfiltered variant readings with the ‘true’ stemma based upon the filtered set of data. From within his own framework Salemans is able to vindicate most of his ‘prejudices’ against certain types of variant readings that he has discarded from the onset. Regarding the omission/interpolation of small words, e.g., Salemans is now even more confident in defining that type of words: “containing four or fewer phonemes and belonging to other word categories than nouns and verbs” (Salemans 2000: 294). Nevertheless, he does not consider his filters and their evaluation to be ‘universal’. Other text traditions may require adaptations and refinements (Salemans 2000: 297). Undoubtedly Salemans has made a good case for critically and systematically reviewing the types of variant readings that are being used in genealogical studies. Moreover, he certainly pinned down the ones that are predominantly suspect of accidental variation. However, his case is built by means of data deliberately manipulated according to certain ‘prejudices’ set out before. From a scientific viewpoint this is of course perfectly legitimate, even necessary. But it should be interesting to have a look at the issue of parallelism based upon data that 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 specific genealogical purposes, because it was considered as allowing genealogically safe conclusions. Upon reviewing those data part of them should be unmistakingly identifiable as purely accidental. Then we should be able to (a) assess the types of variant readings that happened to be accidental in a certain constellation of witnesses, and (b) relate the subset of parallelistic readings to the entire body of data that had been selected for genealogical purposes among those witnesses. If it turned out that quantity and quality of the demonstrably parallelistic readings does actually outweigh the testimony of the non-parallelistic readings, we should have an empirically based case for Genealogy by Chance! Although it seemed extremely unlikely that anyone should come across such a case, thanks to the infallible fallibility of human nature I am able to provide one. The case is built upon the fact that in 1919 a scholar published a list of readings drawn from a 9th-century Gospel Harmony manuscript, which was intended to link the basically Vulgate Harmony text under scrutiny to a particular allegedly considerably older textual tradition. The text that was used, however, was not that of the 9th century historical artefact but that of its 19th-century edition. Upon comparison of the edition with the real manuscript a considerable number of errors showed up, which were partly incorporated in the list of readings mentioned, thus linking the 19th-century counterfeit, but not the real thing, to the older textual tradition. In what follows I will elaborate the case in three steps. First of all, the scholarly debate in which the mentioned Harmony text is situated will be reconstructed. Secondly, the parallelistic readings will be identified and discussed. Finally, some of the implications of this case will be pointed out.

. Medieval Latin Gospel Harmony manuscripts – Reconstructing the scholarly debate Gospel Harmonies are texts which present the materials of the four separate Gospels of the New Testament as one single coherent narrative. Gospel Harmonies are as old as the second half of the 2nd century, when a certain Tatian composed his famous Diatessaron,3 of which not a single direct witness survives. The oldest extant Gospel Harmony is in fact preserved in a famous Latin Vulgate manuscript, the so-called Codex Fuldensis.4 Codex Fuldensis contains the entire New Testament, with a Gospel Harmony replacing the usual four canonical Gospels. Victor, Bishop of Capua,5 commissioned the manuscript, finally approving it on April 12, 547 C.E.6 In the 8th century the famous Anglo-

 Ulrich Schmid

Saxon missionary St. Boniface acquired the manuscript and donated it to the then newly founded monastery of Fulda, whence it derives its name. Today Codex Fuldensis still resides in Fulda, now in the Landesbibliothek. The oldest copies of the Fuldensis Harmony date from the 9th century. Today, we have got a total of four of these 9th century manuscripts, and the question of how closely they are related to Codex Fuldensis is a matter of scholarly debate. Apart from Codex Fuldensis, two of the four Harmony manuscripts from the 9th century have been edited. The editors were Germanists who studied two of the oldest texts in vernacular German. One of the two Harmony manuscripts, the 9th century Codex Sangallensis,7 actually contained one of the oldest texts in Old High German, namely a translation of the Gospel Harmony. Technically speaking St. Gallen, Stiftsbibliothek Cod. 56 is a Latin-Old High German bilingual manuscript, presenting on every page a column with the Latin Harmony text on the left and an Old High German version on the right. The second manuscript, the so-called Codex Cassellanus,8 was edited because of 19th century research into the sources of the Heliand.9 The Heliand is an Old Saxon poem, again one of the oldest texts in that Germanic idiom, retelling the Gospel stories in the form of a Harmony. Although the outline of the Heliand is basically the same as found in Codex Fuldensis, yet for some of the textual details C. W. M. Grein, the editor of Codex Cassellanus, felt that the latter was closer to the Heliand than the former. From the Germanist’s point of view it is absolutely crucial to identify as precisely as possible the sources of the two vernacular texts, because this may lead to a better understanding of the translator’s choices. Moreover, should it be possible to locate these sources in time and space, it will help to allocate the specific blend of the vernacular dialect as displayed by the mentioned texts to a certain area. This, of course, is primary evidence for compiling a map of the historic development of a language. It comes as no surprise, therefore, that Germanists were very pleased to view Codex Sangallensis as a direct copy of Codex Fuldensis, as it attributed a place of origin to the Sangallensis’ Old High German blend. On the other hand, with respect to the intermediary Codex Cassellanus, which separates the old Codex Fuldensis from the Heliand, the link between the Heliand and Fulda as its place of origin seems not so strong. Anyway, most Germanists believed that Codex Fuldensis was the ultimate source of all of their vernacular harmonised texts, either directly – in the case of the bilingual Codex Sangallensis – or, perhaps, indirectly as with respect to the Heliand. These same three Harmony manuscripts, Codex Fuldensis, Codex Sangallensis, and Codex Cassellanus, are now studied from a different angle. As already noted, the oldest and most famous Gospel Harmony, Tatian’s Diatessaron, has

Genealogy by chance!

not been preserved in the original. Scholars who were interested in reconstructing the lost text studied Gospel Harmonies that had been preserved for possible distant echoes from this text, which must once have been very popular. Diatessaronic scholarship developed the theory of a very ancient Old Latin, i.e. pre-Vulgate, translation of Tatian’s Harmony, which was composed in either Syriac or Greek. This Old Latin Harmony was then textually adapted to the Vulgate Textform, which is exactly what Codex Fuldensis represents. Apart from this, however, the Old Latin or less vulgatized Latin version of Tatian’s Diatessaron is thought to have exerted its influence on other parts of the Harmony tradition in Western Europe. Contrary to what most Germanists believe, Diatessaronic scholars suspected not Codex Fuldensis to be the ultimate source of the Western Harmony tradition but the Old Latin Harmony.10 Consequently, they looked for evidence for their guess, by screening bilingual Codex Sangallensis, Latin Codex Cassellanus, and even the Heliand against Codex Fuldensis. The idea was that every difference between Codex Fuldensis and the later texts weakens the position of Codex Fuldensis as the ultimate source of the Western Harmony tradition. Their idea was corroborated by the deviations from Codex Fuldensis, which could often be paralleled from known Old Latin text traditions of the separate Gospels and/or remote witnesses such as Syriac or Arabic texts or Church writers’ testimonies of the first centuries. These deviations then suggested that some source other than the vulgatized Codex Fuldensis must have been operative, namely the ancient Old Latin translation of Tatian’s Diatessaron. After a time of reorientation on the Germanists’ side, when they tried to come to terms with the new situation of a dethroned Codex Fuldensis, the pendulum swung back. The Germanist J. Rathofer devoted detailed studies to the relation between Codex Sangallensis and Codex Fuldensis and concluded that the former was a direct copy of the latter.11 Once again, the Germanists are happy to have their most interesting Old High German text located in time and space to the mid 9th-century monastery of Fulda. Rathofer’s successful demonstration was based on reconsidering the manuscripts anew, with special focus on features that had not been readily available in the 19th-century editions of Codex Fuldensis, Codex Sangallensis and Codex Cassellanus, namely palaeographical data, layout phenomena and marginal addenda. Incidentally, this demonstrates that kinship-revealing features are also, even decisively, to be found outside the narrowly defined textual variants of a manuscript tradition. Anyway, Rathofer’s move away from the editions and back to the manuscripts clears the stage for my demonstration of accidental variation resulting in identical 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 manuscripts from different times (Vogels 1919). Part of his study was devoted to the presenting of examples of textual differences between the oldest, still almost purely Vulgate manuscript Codex Fuldensis and the younger manuscripts, partly or considerably differing from Codex Fuldensis. For his study Vogels used the editions of the manuscripts, when available, among them Grein’s edition of Codex Cassellanus and, of course, Ranke’s famous edition of Codex Fuldensis. On two pages Vogels presented a list of 44 readings where 9th-century Codex Cassellanus deviates from 6th-century Codex Fuldensis (Vogels 1919: 128–129). The significance of this list for Vogels and other Diatessaron scholars lay in the fact that all of these deviations can be paralleled from either Old Latin witnesses and/or remote witnesses such as Greek, Syriac, Arabic and Armenian texts. It was exactly this type of evidence that caused Diatessaronic scholarship to postulate influence, even descent, from an ancient source representing a less vulgatized text than Codex Fuldensis. Now, when Rathofer, as explained above, made his move back to the manuscripts, he noted in passing that especially Grein’s edition of Codex Cassellanus contains a large number of mistakes; Rathofer counted 360 mistakes for the Harmony text, some of which even entered Vogels’ list! (Rathofer 1972: 348) This goes right to the heart of the problem. After checking Vogels’ 44 deviations of Codex Cassellanus from Codex Fuldensis against Grein’s edition and a microfilm of Codex Cassellanus, I detected that 24 of them are based on errors in Grein’s edition! Vogels didnot make mistakes; he simply recorded the differences between two 19th-century editions and assumed that they accurately represented the 6th and 9th-century manuscripts. This highlights first of all the usually neglected fact that modern editors of ancient manuscripts are basically less or more accurate copyists of the manuscripts they edit. In that sense they not only contribute to the study of a manuscript tradition, but with their errors they are part of the manuscript tradition itself. This simple insight is usually obscured by the fact that modern editors’ ‘manuscripts’ are extant in several hundred identical copies. In a broader perspective, however, Grein’s errors as unsuspectingly reproduced by Vogels severely question the received perception of manuscript genealogies. First of all, there can be little doubt that Vogels’ conclusions, based on the 44 readings, are highly questionable, for he thought they ruled out the possibility that Codex Cassellanus was a direct copy

Genealogy by chance!

of Codex Fuldensis (Vogels 1919: 126). Then, by consequence, the fact that more than 50% (24 out of 44) of those readings are not to be found in Codex Cassellanus, but only in its 19th-century edition should cast severe doubts on what the edition actually pretends to be, namely an edition – i.e. a direct copy – of Codex Cassellanus. Secondly, and even more devastatingly, the 44 readings were not chosen randomly just because they were supposed deviations from Codex Fuldensis. Quite to the contrary, they were carefully selected as genealogically significant deviations from Codex Fuldensis, because they could be paralleled from Old Latin and even more remote Syriac, Arabic, or Armenian witnesses. They should serve to make the point that some influence other than the standard Vulgate text as displayed by Codex Fuldensis was operative within the Latin Harmony tradition. The repercussion of this assumption is devastating, since, whatever may have influenced the copying process of Gospel harmonies in the early Middle Ages, it left exactly the same traces in a 19th-century copy of a Gospel Harmony, i.e. in Grein’s edition of Codex Cassellanus. It is essential to realise that Vogels’ findings remain valid, because the 44 readings are obviously shared by the remote witnesses he adduced, even though only 20 of them actually belong to the actual manuscript he was aiming at. But what to think of the other 24 readings that were produced within the 19th-century copy of Codex Cassellanus? All of them can also be paralleled from the same type of remote witnesses. If it is assumed that the 20 readings belonging to the actual manuscript are due to the influence of a remote Old Latin text, by the same token this has to be assumed for the additional 24 readings of the 19th century edition as well. This second assumption, however, is simply beyond imagination. I can think of no condition under which a modern scholarly editor would consciously modify the text he or she is editing in such a way without mentioning anything about it. If the idea of external influence is to be upheld, unconscious or subconscious ways of infiltration must be looked for. Should we suppose that Grein was influenced by some sort of ’local’ text he grew up with or was familiar with? What could such a text have looked like in mid-nineteenth century Germany? If he were a good Protestant, Grein certainly would have been exposed to Luther’s translation of the Bible in Sunday services and confirmation classes. How likely is it that such a vernacular upbringing would influence the editing of a Latin Gospel Harmony? Even if one were a Roman Catholic in those days, which was probably the only way to grow up with or be surrounded by a Latin Bible text, the text should have been that of the Clementine Vulgate, not any Old Latin type of text. In short: it is highly unlikely that unconscious or subconscious influence from a text contemporary to 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 been completely coincidental, and would thus not allow any scientific conclusion. I may, therefore, conclude that the deviations between the text of the 9th-century manuscript and its 19th century edition were not purposely created, nor in any likelihood due to influence from a ‘local’ text the editor was familiar with. Instead, they were created ad hoc by the 19th-century editor while transcribing the manuscript he edited, by committing the usual blunders that every copyist of substantial portions of text is guilty of. Then, by consequence, these deviations between the manuscript and its edition hit any of their ‘parallels’ as adduced by Vogels purely by chance! Since the proportion of accidental hits is more than 50% (24 out of 44), there is nothing significant left with respect to the rest. In other words: The 20 readings which are not based on errors in the edition might have hit their parallels by chance, too. This brings the whole case full circle. It must be kept in mind that this demonstration was performed on material not intentionally selected for the purpose. Quite to the contrary, another scholar made the selection in full confidence of the kinship-revealing quality of the chosen variant readings. With all the parallels that he adduced he aimed at a genealogically ‘safe’ conclusion with respect to a certain constellation of witnesses. Yet, unexpectedly and unintentionally, he not only demonstrated that his conclusion was genealogically unsafe, but once and for all that the genealogical method in itself is intrinsically vulnerable. Not only is there empirical evidence that genealogically relevant information could have been created by chance. In the case under discussion the factual proportion of accidental variation did actually blow up the entire genealogical conclusion based on the evidence presented. Thus, Vogels’ genealogy regarding Codex Cassellanus is completely arbitrary. So let us now consider the type of evidence the study of Codex Cassellanus and its 19th-century edition does provide us with. . A short empirical survey of scribal errors and accidental variation The following discussion is based on the evidence gathered from comparing a microfilm of a 9th-century Gospel Harmony manuscript (Codex Cassellanus) with its 19th-century edition. I do not aim to present an exhaustive collection of scribal errors or of accidental variation (parallelisms); I would not even dare to think of compiling such lists. I am interested in a rough grouping of the phenomena 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 nine randomly chosen pages from Grein’s edition of Codex Cassellanus12 against the microfilm of the manuscript. This was done in order to gain a general insight into the overall distribution of errors as well as the types of errors evidenced. Secondly, I checked the 44 variant readings given by Vogels in his list of supposed deviations between Codex Fuldensis and Codex Cassellanus against the microfilm of the manuscript. .. Spot checks on Grein’s overall error rate The lemma (italics) gives Grein’s text. The reading of the manuscript is in bold type. a. Grein, p. 130 Lk 1:4 visum est mihi] visum est et mihi Jo 1:3/4 factum est nihil, quod factum est. In ipso vita erat] factum est nihil • Quod factum est. in ipso vita erat Although the manuscript’s system of punctuation is different from modern systems, in this very prominent case Grein choose to ignore the capitalised “Q(uod)”and the lower case “i(n)” given by the manuscript, as well as the raised stop after “nihil” and the normal stop after ‘est’. I will not go into other punctuation questions. But in this particular case Grein’s edition definitely alters the meaning of the manuscript’s reading. Jo 1:4 Lk 1:5 1:6

et vita erat et lux hominum] et vita erat lux hominum Herodis regis] Herodis recis*; ‘regis’ corrected (scribe?) sine querela] sine quærela

b. Grein, p. 131. No error detected c. Grein, p. 132 Lk 1:45 1:60 1:62 1:63 1:63 1:64

dicta sunt ei] dicta sunt*; ‘ei’ supplied supralinear by scribe sed vocabitur] sed vobitur (sic) patri eis] patri eius pugilarem] pugillarem et nomen] est nomen dominum] deum

d. Grein, p. 190. No error detected (Grein choose to represent the numerals XCVIIII [Mt 18: 12–13] and X [Lk 15:8] given by the manuscript with ‘nonaginta novem’ and ‘decem’)



 Ulrich Schmid

e. Grein, p. 191 mercenarii] mercennarii13 mercenariis] mercennariis in misericordia] misericordia mercenariis] mercennariis in manum] in manu si peccaverit] attendite vobis si peccaverit lucratis eris] lucratis es*; (e)ri(s) supplied supralinear (scribe?) Mt 18:17 si autem aecclesiam] si autem et aecclesiam

Lk

15:17 15:19 15:20 15:21 15:22 17:3 17:3

f. Grein, p. 192 Mt 18:21 quotiens peccavit] quotiens peccavit in me g. Grein, p. 259 Jo

20:19 dum esset sero] cum esset sero

h. Grein, p. 260 (Jo 20:26 octo] VIII) Jo 20:29 quia audisti me] quia vidisti me Jo 21:11 et cum tanta] et cum tanti i. Grein, p. 261 Jo

21:17 Simon Johannis amas me] + contristatus est petrus quia dixit ei tertio amas me 21:22 quid ad te] + tu me sequere

Our spot check reveals a total of 25 errors in nine pages (excluding the different representation of the numerals), involving about 40 words. Extrapolating from that figure to the 131.5 pages of Grein’s edition adds up to a total of c. 365 errors involving about 584 words (c. 2.8 errors per page). This figure is virtually identical with the number of errors recorded by Rathofer. All sorts of errors occur: apart from (presumingly) printing errors (e.g., Lk 1:62.632 ) and unrecorded corrections (Lk 1:5.45: 17:3) there are orthographicals (e.g., Lk 1:6.631 ; Lk 15:17.19.21), punctuation (Jo 1:3/4), and a homoioteleuton error (Jo 21:17). Moreover, we find two omissions of the highly frequently used conjunction et (Lk 1:4; Mt 18:17), two additions of small words (Jo 1:4; Lk 15:20), three substitutions of words (Lk 1:64; Jo 20:19.29),14 and three changes of word forms (Lk 15:22; Jo 20:19; 21:11). This is in perfect agreement with the types of readings that Salemans records as especially liable to parallelism. However, we find another three readings (out of a total of 25, i.e. 12%) which

Genealogy by chance!

are as far as I can judge not a type recorded by Salemans, namely the omission or addition of more than one word (Lk 17:3; Mt 18:21; Jo 21:22).15 Mt 18:21 includes the omission of a prepositional object (two words); Lk 17:3 (two words) and Jo 21:22 (three words) include the addition of entire sentences. Within that context we may even add the homoioteleuton error (Jo 21,17), which caused the omission of nine words.16 .. Grein’s errors spotted from Vogels’ list (Vogels 1919: 128–129) a. It must be remembered that Vogel’s list was intended to register differences between Codex Cassellanus and Codex Fuldensis which could be paralleled from other remote witnesses; that is to say that the text of Codex Fuldensis is the point of reference. In order to simplify comparison I reproduce Vogels’ listings according to his conventions, e.g., ∞ aut l. neque (Grein, p. 151) means: Codex Cassellanus (according to Grein’s edition p. 151) has aut where Codex Fuldensis has neque. However, in every single case listed below the 9th-century artefact Codex Cassellanus does not deviate from Codex Fuldensis; only the errors in Grein’s edition make it deviate. Consequently, all these readings are purely parallelistic. Mt 4:4 6:25 15:6 17:5 22:31 24:14 24:21 28:7 Mc 3:17 8:19 Lk 2:26 13:18 14:3 14:12 16:27 22:37 Jo 1:19 4:6 4:30 11:50

dixit] + ei (Grein, p. 141) ∞ aut l. neque (Grein, p. 151) ∞ et l. aut (Grein, p. 179) om. ipsum audite (Grein, p. 187) ∞ domino l. deo (Grein, p. 210) om. hoc (Grein, p. 227) ∞ saeculi l. mundi (Grein, p. 227) om. ecce 1o (Grein, p. 256) ∞ nomen l. nomina (Grein, p. 146) om. plenos (Grein, p. 185) ∞ christum dominum l. chr. domini (Grein, p. 135) ∞ regn. caeli l. regn. dei (Grein, p. 171) om. dicens (Grein, p. 199) om. neque fratres tuos (Grein, p. 200) om. enim (Grein, p. 197) om. hoc (Grein, p. 238) om. ad eum (Grein, p. 140) om. sic (Grein, p. 181) ∞ et exierunt l. exierunt (Grein, p. 182) om. homo (Grein, p. 220)



 Ulrich Schmid

12:28 ∞ glorifica. glorificavi. glorificabo l. clarif. (Grein, p. 222) 14:6 et nemo l. nemo (Grein, p. 237)

b. The following readings need to be specified: Mt 20:1 om. enim (Grein, p. 198)

Grein failed to note that ’enim’ was supplied supralinear by the scribe himself. Jo 14:1 inc: et ait discipulis suis (Grein, p. 236)

Grein failed to note that the words were added by a different post 9th-century hand at the bottom of the page, partly exceeding the usual layout of 28 lines per page. When we now consider the character of the 24 errors from Grein’s edition that Vogels paraleled in his list, we find that orthographicals, punctuation and nonsense readings are not included. In agreement with Salemans’ criteria of what is likely to be due to parallelism we find nine omissions of one word (Mt 20:1; 24:14; 28:7; Mc 8:19; Lk 14:3; 16:27; 22:37; Jo 4:6; 11:50), though one omission involves a noun (Jo 11:50); one omission of three words is (most likely) due to homoioteleuton (Lk 14:12). Moreover, we have got three additions of one word (Mt 4:4; Jo 4:30; 14:6); five substitutions of one word (Mt 6:25; 15:6; 22:31; 24:21; Jo 12:28), of which in one case three different forms of the same word have been substituted by the pertinent forms of a potential synonym (cf. Jo 12:28); and three changes of word forms (Mc 3:17; Lk 2:26; 13:18). And yet, three of the real parallelisms, i.e. 12,5% of the total of 24 variant readings, are not accounted for by Salemans. Two of them involve the omissions of two words (Mt 17:5; Jo 1:19), one of which is actually the omission of a complete sentence. This is in perfect line with what we already discovered in our first sample (see above). In addition, a new type of variant reading that is liable to parallelism occurred, i.e. the addition of a whole sentence consisting of four words (Joh 14:1). To sum up: It is true that almost 90% of the real parallelisms presented here are more or less ‘typically’ covered by Salemans. Nevertheless, the remaining more than 10% send most disturbing signals. Not only comparatively simple omissions and additions of single words, but omissions of compound expressions and sentences, even the addition of an entire sentence can hit a ’parallel’ purely by chance. Thus the issue of parallelism is considerably more complicated than hitherto thought of.

Genealogy by chance! 

. Concluding observations I should like to conclude this paper by adding two observations, one on what might have contributed to the devastating results of Vogels’ genealogy, and one on using typologies of variant readings with respect to issues of parallelism. 1. To my mind it is a most remarkable fact that Vogels’ list of 44 variant readings from Codex Cassellanus, for which he gave parallels from remote witnesses provides us with a full empirical case of Genealogy by Chance. It is important to keep in mind that our findings did not simply raise doubts as to whether or not Vogels’ overall genealogical conclusions might still be valid, because some of his readings could be proven to be accidental hits. The case is far more serious. Since the proportion of accidental hits is more than 50%, Vogels’ entire case crumbles to pieces. This proves that the comparison he performed must have been completely inadequate. What can have contributed to this devastating result? 1a. First of all, the entire Harmony text consists of about 40.000 words.17 Upon comparison of two Harmony manuscripts Vogels produced out of them a list of 44 variant readings involving 63 words that he paralleled from remote witnesses. These proportions could simply be too low, since they involve only 0.16% of the actual text. 1b. Secondly, Vogels comparison base consists of witnesses of different genres and languages, and varying degrees of preservation.18 This alone should qualify as a big question mark as to whether Vogels comparison base can indeed be accepted as forming one single coherent text tradition at all. A closer look at the witnesses which were used to form the comparison base reveals that they are, taken individually, not consistent in providing parallels for all of the 44 variant readings. Only as a group consisting of about 30 members do they testify to all 44 of them. However, if we split up the group, we find that the highest number for a single witness is 13 out of 44; this witness is Old Latin Gospel manuscript b. If we combine the testimony of the four individual Syriac witnesses (sycs.c.pal.p ), we reach 14 out of 44. However, the usual frequency for individual witnesses is 6 to 10 out of 44 variant readings. We may conclude, then, that there is a considerable lack of coherence among the witnesses that form the comparison base. 1c. Thirdly, the witnesses that were used to form the comparison base were selected because of their historical remoteness with respect to the two Vulgate Gospel Harmony manuscripts under review. Moreover, their remoteness was 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. Although not explicitly stated by Vogels, this combination of remoteness of witnesses and alleged rareness of readings could be understood as some sort of antidote against the problems of (a) low numbers and (b) incoherent comparison base. The implicit reasoning could have been: readings that are only rarely testified to by scattered testimonies must be ‘survivors’ of an older Textform, which successfully resisted the assimilation process to the now dominant Textform. Thus, they are safe, even if they are few in number and incoherently present in a rather diverse array of witnesses. Vogels’ actual case of the 44 readings from Codex Cassellanus suggests a different interpretation: shared readings which are only few in number and testified to by historically remote witnesses appear to be more likely than not accidental hits or ‘false positives’. The reasons for such an impression need to be further explored, which is beyond the scope of this paper. However, my guess is that it has to do with the sheer quantity of the text tradition involved. The New Testament Gospels display probably the largest text tradition ever produced in the Western Hemisphere, with tens of thousands of mostly unexplored manuscripts in Latin alone. Some amount of coincidental variation (accidental hits, parallelism) seems to be inevitable. However, one may wonder whether a comparison base which is selected in a different way would provide better results. It seems reasonable to assume that with witnesses, of which the historical relations (in time and place) other than shared variant readings provide a link with the two Harmonies, even a small fraction of such readings might prove to be significant. 2. There can be little doubt that the vast majority of the parallelisms discovered in our sample are accounted for by Salemans. There are, however, some that are not accounted for, i.e. omission/addition of more than one word and most notably of entire sentences. Should we now simply ban those types of variant readings like the other ones Salemans already excluded in order to stay on the safe side in genealogical studies? The problem with such an approach is threefold. 2a. First of all, the fact that variant readings which are assigned to certain types can be shown to be parallelistic does not mean that the reverse is true as well, namely that all or even most of the variant readings belonging to those types are indeed parallelistic. Quite to the contrary, more often than expected even in Salemans’ own analysis some highly suspect types of variant readings seem to fall in neatly with the ‘true’ stemma.19 In other words: Their assembled testimony may not effectively obstruct the stemma based on the filtered set of data.

Genealogy by chance!

2b. Secondly, to the extent that the present study adds new types of readings to the pool of possibly parallelistic readings, severe doubt is cast upon the remaining set of data. With only 24 readings we were able to identify additional cases of parallelistic readings which Salemans did not account for. A larger sample might have provided even more evidence and could have added different types. This in turn enhances the possibility that we might not be able to find a single type of variant readings that is entirely free from liability to parallelism. 2c. Thirdly, let us assume that there is indeed no completely safe line to draw between those types of variant readings that are prone to parallelism and those that are never exposed to the threat. The challenge we are facing is that genealogical study must either surrender or intelligently use data that in all likelihood will never be completely weeded of potentially parallelistic inclinations. One solution could be to determine statistical correlations between individual subsets of variant readings and their liability to parallelism as observed in a given text tradition in order to weigh them accordingly. A typology of variants like the one proposed by Salemans is certainly helpful to identify appropriate subsets as well as the usual suspects, i.e. those types of variants that seem to be more likely exposed to parallelism than others. But it should not be used to eliminate data without proper evaluation of their statistical inclination towards parallelism in a given text tradition. Otherwise we should be left with individual subsets of data in which the differing parallelistic inclinations are completely unaccounted for.

Notes . ‘Nonsense readings’ are readings that leave the text unintelligible at a certain passage. . Salemans (2000: 70), see also the list on p. 67–68, n. 44. . Cf. Petersen (1994). . Ranke (1848). Descriptions of the manuscript can be found in Scherer (1905: 6–12) and Fischer (1963: 519–600, esp. 545–557). . A town in Northern Italy in the region of Campania (medieval: liburia). . Fischer (1963: 546). . Cf. the recent edition, Masser (1994). Descriptions of the manuscript are given in Masser’s edition as well as in an older edition by Sievers (1892). An extensive study of the manuscript (including photographs) was presented by Masser (1991). . A recent description is given in Kahlfuss (1994: 37–39).



 Ulrich Schmid . The manuscript was edited as an appendix to Grein (1869: 127–262). . The first scholar to issue that theory was Th. Zahn, see Zahn (1881: 300–303), and Zahn (1894: 87–107). . Rathofer (1971, 1972, 1973). . These include the first and the last three pages, and three pages from the center; i.e. pp. 130–132, 190–192, 259–261, out of a total of 131.5 pages. . On p. 217 (Jo 10:12) 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 graphic representations of the two words in Codex Cassellanus. The ms like most early medieval mss does not distinguish between the vowel ‘u’and the semi-vowel ‘u’ = ‘v’; the back stroke of the ‘a’ is very upright with a rounded foot towards the right, thus closing the gap to the next letter. Therefore, the two letter-combinations ‘au-’ and ‘ui’ are basically both dominated by three vertical strokes. As far as I can see, Salemans does not place much emphasise on graphic similarities between words as likely causes for accidental variation. . Salemans discusses the “addition (or interpolation) and omission of complete verses” (Salemans 2000: 101–102, 294–295). This is very likely due to the fact that the text he researched 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 Ex Quatuor. Genealogie und Rezeption einer Lateinischen Evangelienharmonieüberlieferung. The Harmony text in Patrologia Latina 68 contains 40770 words. The edition is based on a manuscript from the 13th century. Spot checks revealed that the later versions are slightly expanded (400–1000 words) compared to the old version from the 6th to 9th centuries. . The witnesses adduced include Gospel Harmonies, a Harmony commentary, a few Church father testimonies, and Gospel manuscripts, out of which some are fragmentary in at least five different languages! . See, e.g., Salemans (2000: 257), “Evaluation of the Word Category Conjunctions”. Although there are examples of parallelisms with the medieval Dutch conjunction “ende”, this does obviously not hold true for other conjunctions: “We can offer many examples in which a conjunction is added at the beginning of a verse... However, these interpolations all agree with our stemma... The conclusion at the end of this section must be stated cautiously, because we did not find clear, convincing examples to show that conjunctions are parallelistic. Yet I am still convinced that they are parallelistic, in view of the ease with which they have been interpolated in the Lanseloet texts” (pp. 257–258).

References Fischer, B. (1963). “Bibelausgaben des frühen Mittelalters” [La Bibbia nell’alto Medioevo]. Settimane di studio del centro italiano di studi sull’alto 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, Landesbibliothek und Murrhardsche Bibliothek der Stadt Kassel. H.-J. Kahlfuss (Ed.), Bd. 1,1: Manuscripta Theologica. Die Handschriften in Folio (bearbeitet von K. Wiedemann). Wiesbaden. Masser, A. (1991). In Nachrichten der Akademie der Wissenschaften in Göttingen. I. Philologisch-Historische Klasse. Jahrgang 1991, Nr. 3. Göttingen. Masser, A. (1994). Die lateinisch-althochdeutsche Tatianbiblingue Stiftsbibliothek St. Gallen Cod. 56. SAHD 25. Göttingen. Petersen, W. L. (1994). Tatian’s Diatessaron. It’s Creation, Dissemination, Significance, and History in Scholarship. VigChr.S. 25. Leiden. Ranke, E. (1848). Codex Fuldensis – Novum Testamentum Latine Interprete Hieronymo ex manuscripto Victoris Capuani edidit.... E. Ranke (Ed.). Marburgi et Lipsiae. Rathofer, J. (1971). “Zur Heimatfrage des althochdeutschen Tatian. Das Votum der Handschriften”. Annali (instituto universitario orientale, sezione germanica), 14, 7–104. Rathofer, J. (1972). “‘Tatian’ und Fulda. Die St. Galler Handschrift und der Victor-Codex”. In K.-H. Schirmer & B. Sowinski (Eds.), Zeiten und Formen in Sprache und Dichtung. Festschrift für Fritz Tschirch zum 70. Geburtstag (pp. 337–356). Köln/Wien. Rathofer, J. (1973). “Die Einwirkung des Fuldischen Evangelientextes auf den althochdeutschen ‘Tatian’. Abkehr von der Methode der Diatessaronforschung”. In A. Önnerfors et al. (Eds.), Literatur und Sprache im europäischen Mittelalter. Festschrift für Karl Langosch zum 70. Geburtstag (pp. 256–308). Darmstadt. Salemans, B. J. P. (2000). Building Stemmas with the Computer in a Cladistic, NeoLachmannian, Way. The Case of Fourteen Text versions of Lanselot van Denemerken. (Doct. Thesis). Nijmegen. Scherer, C. (1905). “Die Codices Bonifatiani in der Landesbibliothek zu Fulda”. Festgabe zum Bonifatius-Jubiläum 1905. Fulda. Sievers, E. (1892). Tatian, lateinisch und altdeutsch. BADLD 5. Paderborn 18922 . Vogels, H. J. (1919). Beiträge zur Geschichte des Diatessaron im Abendland. NTA 8.1. Münster. Zahn, Th. (1881). Tatian’s Diatessaron. FGNK 1. Erlangen. Zahn, Th. (1894). “Zur Geschichte von Tatians Diatessaron im Abendland”. NKZ, 5, 85–120.

Constructing initial binary trees in stemmatology E. Wattel Vrije Universiteit Amsterdam

.

Introduction

The main purpose of stemmatology is to construct a pedigree of actual and putative texts, in order to describe the relations between the manuscripts of a historical text tradition. The expectation is that such a pedigree, which is usually called a stemma, will help to obtain insight into the transmission process which led to this text tradition. Questions like ‘Is one manuscript a direct ancestor of another one’, and ‘Which reading in case of variants should be considered original?’ can sometimes be solved if a reliable stemma is available. Since the only information available is usually in the texts themselves, we need a method to extract the genealogical information from the comparison of texts. To enable the use of modern computers the coding of this information should be well defined and easily available for automatic processing. Therefore we will use the database conventions of e.g. Wattel and Mulken (1996a), which is described in the next section. The genealogical information, which is coded in version formulas, is usually internally inconsistent, and so many investigators direct their efforts towards explaining away version formulas which disturb favoured parts of a stemma under construction. Most of the arguments used to delete a version formula are ad hoc and a posteriori, and are not suited for an objective construction method. A much better approach to the problem was investigated by Salemans (2000). In this work a version formula has to meet eleven admission criteria, based upon seven basic principles, to be accepted for stemma construction. If this leads to a complete stemma, we have certainly found a reliable, objective solution, but it is possible that some inconsistencies remain, or that the

 E. Wattel

number of version formulas which meet the requirements is insufficient to construct a stemma. An alternative approach is to keep in mind that no information in the version formulas is completely and absolutely reliable. So the best we can hope for is a stemma with only a small amount of inconsistency. In order to estimate these inconsistencies we assign to each version formula a positive formula weight. After the construction of a tree we can add the weights of all the formulas which contradict the tree, and so we can estimate the performance of a tree. A tree is better when the estimated inconsistency is a smaller number. Now a stemma can be evaluated with respect to the values of the version formulas, and the stemma with optimal value can be chosen. Since the Salemans (2000) admission criteria are suitable for assessing the structure and the contents of variation, formulas meeting those criteria will get a high formula weight, but formulas previously rejected can still be kept, albeit with a much lower weight. In this way we hope that good information will not be choked by bad information, but we can still use downgraded indications, in cases where nothing else is available. At this point it is necessary to assess variation formulas. We can distinguish two types of observations which influence the value of the formula: i.

Linguistic. These observations consider the contents of the versions. Here we take grammatical, semantical, logical and aesthetic aspects of the versions in account. ii. Computational. Here we judge how suitable the structure of a version formula is for the construction of a stemma. The most prominent of those criteria is the type 2, which indicates that a formula contains precisely two concurrent readings, both shared by at least two manuscripts. In order to separate the data collection phase from the data processing, it is strongly advised to discard the information mentioned in (ii) when the value of a version formula is determined in the data base. During the construction of a stemma from the data base the information in (i) should have no effects which are not coded into the value of the formula. In this way one investigator can supervise the data acquisition and a different expert can monitor the data processing. In this contribution we will concentrate on the computational problems, under the assumption that the linguistic considerations have already determined 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 a set 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 declare the tree with the optimal value to be the stemma. However, such a procedure is not feasible. The reason for this failure is called computational complexity, which means that the number of possible trees is very large, even for a small number of available manuscripts. In order to get a credible final solution we start out from a reasonable tree, and evaluate the sum of the version formulas which contradict it. We try to improve this tree by a search algorithm which proceeds along the following lines: We construct a small set of related trees, and evaluate the sum of the contradicting formulas for each of these trees. If the evaluated sum is smaller, we obtain a better tree. If a better tree is found, we discard the first tree, and continue with the best tree encountered. This process is repeated until no further improvement is obtained. An algorithm of this kind is called a local search method. In stemma construction this local search is quite applicable, but again the theory of computational complexity implies that success cannot be guaranteed. One possibility is that the search goes on for ever, effecting smaller and smaller improvements all the time. Another possibility is that the search never as much as approaches the optimum over all possible trees, and that the result is a fake which is far from optimal. We should especially avoid the second possibility. Since expectations for the search algorithms cannot be very high, a great deal of effort should be put into the construction of the initial tree. Since our material is not collected in order to supply just any arbitrary tree, but specifically to yield a stemma reflecting the transmission, we hope that the true stemma will be clearly distinguishable from the background noise of inconsistencies. The main aim of this paper is establishing a method to construct an initial stemma which is hopefully close enough to the optimal stemma to function as a starting point for any type of local search. From the above it should be clear that the program described here is part of a larger set. Obviously, other programs should deal with stepwise refinement and edge contraction (see for those programs Wattel & Mulken 1996b). One of the set is a program which can spot version formulas which are in contradiction with their context, and therefore must be erroneous or contaminated. The method used is an implementation of the ideas of Dees on this subject. It is also possible to look for a stable section of the text by means of the shock wave method described in Wattel and Mulken (1996a). Although the program described here is meant to yield an initial tree, which should be refined by other programs, with the algorithm of this present contribution it is possible

 E. Wattel

to construct an unordered tree structure straight from a collection of version formulas. In order to get a complete stemma, we have to define an ordering for the manuscripts, and this can be done by adding an extra manuscript in the version formulas, which is only mentioned when the original version for the formula can be determined. The position of this manuscript in the final result will be at the top of the tree, which then defines the direction of all the edges. In the present contribution we have tried to implement the experiences which have been obtained in the last ten years. In this respect the work of Salemans (2000) was essential. The main difference between Salemans’ approach and ours is that he refuses to use version formulas which are not completely reliable, while in our approach the reliability of the formulas is accounted for in the weighing schemes. In the near future we hope to redesign the stepwise refinement and contraction programs, in order to improve their performance and to get statistical information on the probability of error of the final stemma. In this paper Section 2 will be devoted to the structure of the data base. The subject of Section 3 are lacunary version formulas, i.e. formulas in which some manuscripts of the text tradition are missing. In Section 4 we will give a small introduction to the theory of complexity and explain the computational difficulties of exhaustive search methods. This section is the main motivation for the binary tree construction, but can be skipped if the reader is not interested in the theory. In Section 5 we will give a construction algorithm for initial trees, in an iterative way, in which formula weights are adapted in the process if the corresponding version differences turn out to be either unreliable or not suited for tree construction.

. Material Following e.g. Wattel and Mulken (1996a), version differences can be coded in a data base consisting of a single header line and a collection of version formulas, as in Table 1. In our convention the heading line consists of two fields: firstly an integer indicating the maximum number of verses in the text, and secondly a list of codes, indicating the available manuscripts. In this example the total number of lines is 5, and there are seven available manuscripts: ‘a ‘b ‘c ‘d ‘e ‘f and ‘g. The verse numbers should not be taken too strictly, they are used to find corresponding verses in different manuscripts,

Constructing initial binary trees in stemmatology 

Table 1. A simple data base 5 1.1 1.6 3 3 5

‘a ‘b ‘c ‘d ‘e ‘f ‘g ‘a ‘c ‘d / ‘b ‘e ‘a ‘b ‘d ‘g / ‘c ‘e ‘a ‘d / ‘b ‘c / ‘f ‘g ‘a ‘b / ‘c ‘e ‘f / ‘g ‘a ‘f / ‘b / ‘c / ‘d / ‘e / ‘g

7.5 0.5 5 0.75

(heading line) (first formula) (second formula) (third formula) (fourth formula) (fifth formula)

and in e.g. Wattel and Mulken (1996a) verse numbers are used to indicate the natural order of the episodes and verses in the texts. All the lines apart from the heading line in the data base contain coded information from variant readings in the manuscripts. The version formulas in the data base consist of two or three fields. The first field indicates the position of the variation, usually a line number or a verse number where a set of comparable variant readings can be found. The important part in such a variant formula is the second field, which shows which manuscripts share each of the various readings. Manuscripts sharing a reading are listed in groups, and these groups are separated by slashes ( / ). The third field may contain a number indicating the relative importance of the formula. If a version formula reveals more about the relationship of the manuscripts than usual, this field should contain a number which is larger than one, if the version formula indicates differences which can easily arise from other sources than the used examplar, this field should contain a small number. If the third field does not exist, the importance is supposed to be 1, which means that the information in the formula is moderately revealing. Fields are separated by tabs. In order to keep the data base readable, it is possible to add comments and version indications as fourth, fifth, etc. fields in a database line. Usually the fourth field contains the various readings themselves. The program uses the coded information of the second field. Verse numbers in the version formulas need not be unique. One verse may contain several version differences. In the example, the third and fourth version differences both occur in verse number 3. Also shown in the example is that we can use fractional verse numbers in stead of integer ones to indicate where in the verse the difference occurs. When one version difference occurs at the beginning of a verse and another at the end, we use e.g. 1.1 and 1.6 to indicate that both formulas were extracted from verse 1 but one from the beginning and one from the end. The lines with verse numbers 1.1 and 1.6 indicate precisely two different readings 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 are by far the most suitable for constructingstemmas. Especially when all the manuscripts are represented in an unquestionable type 2 formula, the group of manuscripts is divided into two subgroups. We call such a partition of all the material into exactly two groups a dichotomy. All the lines in a tree represent a dichotomy of such a kind. If we have a number n original manuscripts, then we need n – 3 of those dichotomies to construct a tree. The requirements for this set of n – 3 variants are that they are different, unquestionable, complete, and compatible. The ideal case in which such a collection can immediately be selected from the data base is like hitting the jackpot. 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 these readings is found in only one manuscript. A formula is type 3 when there are more than two different readings, but all except one of those readings are found in only one manuscript. The formula for verse 5 of the example is type 3. A version formula is type 4 when there are more than two different readings, and at least two of those readings are shared by more than one manuscript. The formula with verse number 1.6 and both formulas for verse number 3 of the example are version formulas of type 4. It must be observed that formulas in which no two manuscripts share a reading are useless for tree construction, and are therefore omitted from the data base.

. Lacunary version formulas Seven manuscripts are mentioned in the heading line of the example but only the fifth formula contains all seven manuscripts, in none of the other version formulas do all the manuscripts occur. Such formulas are called lacunary. There can be several reasons why a manuscript is not mentioned, e.g. part of a page may be damaged, or a page may be missing. In such a case it is better not to force the manuscript into the formula. This type of lacuna is totally different from a revealing text omission; such a text omission occurs when some words, a phrase, or even a couple of verses are missing from the text on an undamaged page. These omissions have to be put in the data base as a separate variant reading. In the example of Table 2 we have an ordinary version difference in manuscripts ‘a, ‘b, ‘c and ‘d, in which the words “lanterns” and “lamps” interchange. Manuscripts ‘f and ‘g have a revealing omission, because the words

Constructing initial binary trees in stemmatology

Table 2. Omissions and lacuna Manuscript versions ‘a ‘b ‘c ‘d ‘e ‘f ‘g 3

Carrying torches and lanterns Carrying torches and lamps Carrying torches and lamps Carrying torches and lanterns Carrying @@@@@@@@@ Carrying lanterns Carrying lanterns ‘a ‘d / ‘b ‘c / ‘f ‘g

(version 1) (version 2) (version 2) (version 1) (lacunary) (omission) (omission) (version formula)

“torches and” are not in their texts, although the lines appear to be undamaged. Manuscript ‘e is unreadable, and hence it is lacunary at this verse. In the version formula we have three established different readings: version 1 – ‘a ‘d, version 2 – ‘b ‘c and omission version – ‘f ‘g. Manuscript ‘e must not be included in the version formula because it may belong to any of the other groups. It goes without saying that processing this formula should have no effect on the relations between ‘e and any of the other manuscripts. We will formulate this as a general rule: Manuscripts which are not mentioned in a lacunary formula should have no influence on the result of the computation when such a formula is processed.

This rule has as a consequence that e.g. the distance between two manuscripts should not be computed as the sums of all the formulas in which these manuscripts do not share a reading. If the distance between two manuscripts were to be computed as the sums of all the formulas in which they differ, then a heavily damaged manuscript would get positioned close to all the other texts. A damaged manuscript is then easily considered central, and old, therefore reliable, and thus its merits could be overestimated. The lacunary version rule is meant among other things to prevent this type of bias. Moreover, the rule and its consequences are obligatory if we want to use lacunary formulas. It is clear that we cannot avoid lacunary information if we want to make stable stemmas. Imagine what the consequences would be of not allowing lacunary information, if a highly damaged manuscript is found after the stemma was established: should we throw away all the information which was valid until the appearance of the extra manuscript which made most version formulas lacunary? If lacunary information is allowed, even type 2 version formulas will no longer separate the set of manuscripts into two distinct sets. It is not clear where





E. Wattel ‘@

‘@

‘ˆ5

‘ˆ5

‘ˆ3 ‘a

‘ˆ4 ‘f

‘ˆ2 ‘ˆ1

‘d

‘ˆ3 ‘g

‘ˆ2 ‘a

‘ˆ4 ‘f

‘ˆ1 ‘d ‘b

‘g

‘c

‘b ‘c 3 ‘a ‘d / ‘b ‘c / ‘f ‘g (version formula)

Figure 1. One formula with two different stemmas

the lacking manuscript should go. This could be a major objection against the use of lacunary formulas for those philologists which restrict themselves to only type 2 information. However, unapplicable and lacunary information is so common in text traditions that lacunary formulas cannot be ignored without loosing a large part of the available information. So even the authors who base their information processing on cladistic methodology (cf. Salemans 2000) cannot maintain their objections against lacunary type 2 version formulas under all circumstances. The reason for the restriction to type 2 information can be seen in the example of Figure 1. In this figure there are two different stemmas, with six available manuscripts ‘a, ‘b, ‘c, ‘d, ‘f, ‘g and a putative archetype ‘@. In order to obtain a structure in which every internal node joins exactly three edges we have five intermediate nodes ‘ˆ1, ‘ˆ2, ‘ˆ3, ‘ˆ4, ‘ˆ5. Suppose, moreover, that the formula ‘a ‘d / ‘b ‘c / ‘f ‘g is compared to both stemmas. If the case is that the version in ‘a and ‘d is the original one, it is clear that the deviation in ‘f and ‘g must have occurred for the first time in node ‘ˆ4, and the deviation in ‘b and ‘c must have occurred in node ‘ˆ1. Also, if we have established the deviation in node ‘ˆ1, then all the descendants of ‘ˆ1 share this reading, and the same observation holds for ‘ˆ4. But in case the version of ‘f and ‘g is the original one, then the formula introduces two different deviations in one branch of the stemma in the left tree, but two independent deviations in the right tree. In the left tree the deviation of ‘a and ‘d occurred in ‘ˆ3, but not all decendants of ‘ˆ3 share its readings. In the right tree the situation is even worse. Either of the deviations could have occurred

Constructing initial binary trees in stemmatology

for the first time in ‘ˆ3, but it could also be that this intermediate still has the original version. These ambiguities are not solved when the original versions and the stemma are well established, and so several authors claim that a formula of this type is not fit for tree construction. Contrarily, type 2 formulas do not introduce difficulties of this type. In the left tree we can have formulas ‘a ‘b ‘c ‘d / ‘f ‘g, ‘a ‘f ‘g / ‘b ‘c ‘d and ‘a ‘d ‘f ‘g / ‘b ‘c. In the right tree we have ‘a ‘b ‘c ‘d / ‘f ‘g, ‘a ‘d / ‘b ‘c ‘f ‘g and ‘a ‘d ‘f ‘g / ‘b ‘c. In all these cases it is possible to assign a unique node to the deviation whenever the original version is established, and all the descendants of that node share the deviation. In the left tree this node will be either ‘ˆ3 or ‘ˆ4 for the first formula; it will be ‘ˆ2 for the second formula; and ‘ˆ1 for the third formula. In the right tree this node will be again either ‘ˆ3 or ‘ˆ4 for the first formula; it will be ‘ˆ2 for the second formula; and ‘ˆ1 for the third formula. Salemans 2000 follows this line of thought in the stemma constuction and he uses type 2 information only. His argument is that we can be sure that a clean formula will introduce a well defined node which may be held responsible for the deviation. The sheer existence of such a node is necessary, even if we cannot (yet) decide which version is original. If the processing faithfully observes the lacunary formula rule, then it is not a big intervention to split up a formula into its constituent parts. In the case the original formula ‘a ‘d / ‘b ‘c / ‘f ‘g is split into three lacunary subformulas: (i) ‘a ‘d / ‘b ‘c, (ii) ‘a ‘d / ‘f ‘g and (iii) ‘b ‘c / ‘f ‘g, which are all type 2 formulas. The two stemmas of Figure 1 will be reduced to a substemma for each of those subformulas. This is done by deleting manuscrips which are not in the formulas and contracting the internal nodes which do not distinguish between the other manuscripts.The result can be seen in Figure 2. In these substemmas

‘@

‘@

‘@

‘@

‘@

‘@

‘ˆ5

‘ˆ5

‘5 = ‘3

‘5

‘5

‘3 = ‘4 = ‘5 ‘a ‘2 ‘1 ‘d ‘b ‘c ‘a ‘d / ‘b ‘c

‘3 = ‘2 ‘4 ‘a ‘d

‘f ‘g

‘a ‘d / ‘f ‘g

‘3 = ‘2 = ‘1 ‘4 ‘b ‘c

‘f ‘g

‘b ‘c / ‘f ‘g

‘2 ‘a ‘d

‘1

‘3 = ‘2 ‘4

‘3 = ‘1 ‘4

‘b ‘c

‘a ‘d

‘b ‘c

‘a ‘d / ‘b ‘c

‘f ‘g

‘a ‘d / ‘f ‘g

Figure 2. Three substemmas constructed from each of the two stemmas

‘f ‘g

‘b ‘c / ‘f ‘g



 E. Wattel

there is always a node which can be held responsible for the deviation if one of the versions of the subformula is non-archetypical. For instance, in the leftmost substemma node ‘1 must be held responsible for the deviation, and the reading shared by ‘a and ‘d could be the original, while the reading of ‘b and ‘c is probably not archetypical. If we compare this observation to an analysis of the same lacunary subformula and the fourth stemma of this figure, we get totally different results. In the fourth substemma node ‘1 is responsible if the version in ‘b and ‘c is not original, and node ‘2 is responsible for a deviation in the manuscripts ‘a and ‘d. So either of the concurrent readings could be archetypical. The other substemmas have the same structure as the fourth. We must conclude that the use of restricted type 2 formulas can be helpful to construct stemmas, and that the position where a deviation occurred can be determined, if there is prior knowledge as to which version is archetypical. The example above shows that it is safe to extract type 2 formulas from formulas with more alternatives. Even when the number of alternatives is more than three, and the number of shared readings is higher, we still encounter no additional difficulties in contracting consistent stemmas and each contracted substemma has at least one node which can be held responsible for some deviation. It is a well known problem that the number of genuine type 2 formulas is seldom sufficient for the construction of a stemma, or even an unrooted unoriented tree. Several authors, among them Dearing, Dees, and Dekker, have tried to construct additional type 2 formulas by combining the available version information. Salemans (2000) shows that the methods of Dearing can yield combinatorially erroneous results. The methods of Dees and Dekker are intrinsically correct, but since we know that version information hardly ever has absolute value, even careful combination of information usually increases the error probability. Therefore we are strongly in favour of constructing the necessary type 2 formulas only by splitting larger formulas, into all possible lacunary type 2 subformulas, and adhering strictly to the lacunarity rule that manuscripts not mentioned must have no influence on the computation.

. Computational complexity When all the version formulas have been collected and the weights (based on the linguistic contents of those formulas) have been determined, it should be a

Constructing initial binary trees in stemmatology

matter of computation to find the optimal stemma. This task is not as easy as it sounds by far, even when computers can be used for the evaluation of trees and stemmas. The first step is to decide along which lines the combination of a version formula and a tree should receive a value. We can illustrate this as follows. Given a stemma or a tree, e.g. the stemma of Figure 1 left, and a formula, e.g. 3

‘a ‘d / ‘b ‘c / ‘f ‘g 1

how should this combination be valued? There are several possibilities for this valuation. We could say that there is complete agreement and therefore count 1, the original value of the formula. On the other hand, we could say that there is agreement for each of the three type 2 subformulas, and therefore count it as 3. A third possibility is to count all the mimimal type 2 subformulas, which also happen to yield 3. Those minimal type 2 formulas always have the form ‘w ‘x / ‘y ‘z. Because there are always exactly four manuscripts involved, those formulas are called quadruples (or quartets, cf. Bryant & Steel). If we make the same evaluation for the stemma of Figure 1 left and the formula 3

‘a ‘d ‘b ‘c / ‘f ‘g 1

the result is 1 again if we take only overall agreement into consideration; the result is also 1 if we consider type 2 subformula evaluation, and now as much as 6 if we count quadruple agreement. The number of possibilities becomes even larger if we also start counting formulas which do not agree with the tree. Consider e.g. the formula 3

‘a ‘d ‘b / ‘c ‘f ‘g 1

If we consider quadruple evaluation we have three possibilities to choose an agreeing quadruple, and six possibilities to choose a disagreeing quadruple, and so the value would be – 3. For the full formula valuation we could argue that this formula is in total disagreement, and so evaluation will yield the value – 1, and the same is the case for the subformula valuation. Then there is a problem when large trees are matched with large formulas and only one single manuscript is in the wrong set. The effect would be disagreement with the resulting value of – 1 for this small deviation, which is rather exaggerated. A sophisticated solution for curbing this effect is to see what happens when one manuscript is removed from the formula. If we get agreement, we count a positive fraction of the formula weight. We could subtract a fraction of the weight for each of the manuscripts which have to be omitted before 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. It is 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 complete tree in one sweep, but to compare a formula to each dichotomy which can be 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 is positive; if a disagreeing quadruple can be selected from the formula, we have a negative result. If we have neither agreement nor a disagreeing quadruple we call the combination consistent. Sometimes this means that the formula is in agreement with some other dichotomy of the tree, sometimes we have disagreeing quadruples elsewhere in the tree. In Wattel and Mulken (1996b) the disagreeing formulas receive very negative values, while consistent dichotomies receive small positive values. When we combine the different possibilities, there are at least twenty different valuation and optimisation schemes, some of which are treated in Bryant and Steel. This present paper will also make clear that the comparision of possible stemmas is not easy even when the optimisation scheme has been decided upon. The reason is a consequence of the stupendous number of possible trees that have to be considered in the computations, and although this number more or less depends on the type of valuation, almost all computation schemes suffer from the same unmanageability problem. There is a compartment of information theory called computational complexity, which estimates the number of computational steps that have to betaken to solve some problem, as a function of the input of the problem. The stemmatological optimisation problems score rather badly in this respect. The input length of a file is about the product of the number of formulas times the number of available manuscripts. Since the number of formulas does not cause problems, we will concentrate on the number of manuscripts to estimate the complexity. To make this point clear: if the number of available manuscripts is four, there are 32 possible unoriented trees, and 394 possible stemmas. If we increase the number of manuscripts to five, the result is 262 different trees and 4336 different stemmas. So one single manuscript increases the possibilities with a factor of 12.3 resp. 16.5. The number of trees for just eight manuscripts is as high as 3.7 million, and the number of stemmas is 72 million. Supposing that we have several hundreds of formulas, and that we can evaluate a tree or a stemma in a millisecond, it will then take an hour to evaluate all the trees, and a day to evaluate all the stemmas. Should we find just one more manuscript and valuate the same number of formulas, the number of trees and stemmas

Constructing initial binary trees in stemmatology

increases with a factor over thirty, and it will take more than a day to evaluate all the trees and more than a month to evaluate all the stemmas. In the same way every additional manuscript will increase the number of trees with a large factor. The increase of processor speed does not give us much help. At the moment the processor speed of modern computers doubles about every two years. This means that it will take another ten years of processor speed improvement before the full evaluation of all the trees in the case of nine manuscripts is reduced to one hour. It follows that full evaluation of all the stemmas is not feasible if the number of manuscripts mounts to fifteen, like in the “Lanceloet” (cf. Salemans 2000) or the “Perceval” (cf. Maas). However, the really bad news is that there is no algorithm which is certain to yield the optimal tree for a fixed set of formulas without considering all the possible trees. The optimal weighted tree construction problem is not the only one of this kind, several other computational problems in science and business applications share this characteristic. The best-known problem of the type is the travelling salesman problem: find the shortest round trip for a travelling salesman who is going to visit his clients, when all the distances between pairs of clients are given. This class of problem is called the class of Nondeterministic Polynomial Optimization problems, or NPO for short. In fact, we should say that they are optimisation reformulations of the Nondeterministic Polynomial problems (NP for short), where the question is: ‘Is there a solution (not) exceeding given limits’. This class of NP problems has been studied extensively over the last fourty years, and for none of its members a fast algorithmic solution has been found up to now. Moreover, almost all members of this class NP are equally hard to program. They share the characteristic that a fast algorithm which would solve it under all circumstances would also guarantee a fast algorithmic solution for every other member of the class NP. These problems are called Nondeterministic Polynomial Complete, or NPC. So in a way, should one solve one of those problems, one would solve them all. Since no fast solution has ever been found for any member of the NPC class problem, it is assumed that all these problems are intrinsically difficult. It is clear that the optimisation variants cannot be easier than the corresponding problems which only ask for a solution within given limits. From these observations it follows without saying that we should proceed with extreme caution. In constructing a stemma we are not just looking for the best 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 background noise. We proceed in two steps. First we construct a tentative initial stemma from the material, and then we try to improve the overall structure by means of a local search method. Such a local search method make small alterations and adaptions to an initial structure, in the hope of achieving improvements. The absolute optimum will be found if the initial stucture was not too far away from it, and could be transformed into the optimal solution by a set of local search improvement steps. Clearly, we will never obtain a 100% certainty, but we do the best we can by starting with a carefully designed method for the making of the initial guess. In this paper we do not go into the details of stepwise improvement. Wattel and Mulken (1996b) gives a possible algorithmic solution to that problem in the case of a given formula weight and some fixed valuation design. In the next section of this present paper we will improve on the initial guess of Wattel and Mulken (1996b) with an initial tree building algorithm which adapts the formula 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 manuscripts. If a version in a formula is indicated as original, we put an extra fictive manuscript ‘@ (indicating the possible archetype) into that version of the formula. The aim is to construct an unoriented tree, in which every internal node is a fictive manuscript which is directly connected to exactly three other nodes. The sigla and the putative manuscript ‘@ are the terminal nodes of this tree. Trees of this kind are called fully resolved (cf. Bryant & Steel). In order to have a consistent notation, we will call indications in the version formulas of existing and fictive manuscripts symbols. The construction proceeds in the following way. We work in different stages, and every stage consists of two passes. In the first pass of each stage we look for the two most closely related symbols, and join them together to make a new symbol. The second pass consists of remaking the set of version formulas, in such a way that the two closely related symbols are deleted in favour of the single new symbol. The weight of the version formulas is then adapted according to the suitability of the pairing of those two manuscripts. In the process the fictive distances between the new symbol and each of the two old symbols is accumulated, as well as the weights of the contradicting quadruples. After this stage is finished, we start the next stage, in which the number of symbols is

Constructing initial binary trees in stemmatology 

decreased by one. When only three symbols are left, those three are joined to make the top symbol of the tree, and again we can make a distance indication for those three last symbols and the top. The unoriented tree is now made by starting out from the top symbol as top node. Next, we take the three symbols which were joined to make the top; we make them into three new nodes, and connect each of them to the top with an edge. We use the computed distance as an indication for the fictive ‘length’ of these edges. As long as the tree contains constructed symbols we can add two nodes for its two constituent symbols, and connect them with edges with length indications in accordance with the computed distances. This rather simple algorithm immediately yields an unoriented fully resolved tree. The method is quite common, cf. e.g. Wattel and Mulken (1996b), and the main sophistication should go into the decision which two symbols are closest at a given stage. Errors made in this decision tend to cumulate over the following stages, and propagate into the final tree. Therefore we have to make deliberate design choices to evaluate the algorithmic consequences of the weights of different formulas, the weights of pairs inside a formula, and the comparison of the pair weights at the end of the stage to decide which pair should be squeezed into one new symbol. These are the decisions of the first stage. The design choices of the second stage concern the new formula weights after the pair has been squeezed, and the computations of distances and/or the weights of the quadruples contradicting formation of the pair. We will proceed this section along these steps. . First pass – 1 – algorithmic structure valuation of version formulas We have already observed that linguistic criteria should be used to decide the data base value of the version formulas. Salemans (2000) uses exclusively type 2 information, except when a manuscript could be an intermediate. In that case he also admits type 1 information. These decisions are in line with the usual practice in cladistics. We are well aware of the importance of type 2 information, but we consider it too restrictive to abandon the other information. Therefore we make the following design choices: i.

If a version formula contains precisely two version groups with more than one manuscript each, we consider this a type 2; to stress its importance, the weight of such a formula is multiplied by 4. ii. If a version formula contains precisely one version group with more than one 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 by 0.25. iii. Since this algorithm concentrates on pairs of manuscripts, it seems reasonable to multiply the weight of the formula by the number of pairs that it contains. In that way the importance of the formula is directly related to its influence on the pairing process. . First pass – 2 – pair weights inside a version formula It is clear that in a certain version formula pairs sharing a group should count positive, and pairs the members of which are in different groups should count negative. In a formula we will make all positive contributions add to the weight established in the previous paragraph. In order to have the total contribution of the formula 0, the absolute value of the negative contributions should also add to this total weight. For the positive contributions we will take into account the observation of Maas that a group with precisely two manuscripts is a reliable witness for a pairing construction. Therefore pairs in groups with a small number of manuscripts should be favoured above pairs in groups with a large number of manuscripts. Because the total process will be iterated, this is not contradictory to the observation of Dees that type 2 formulas with large groups of manuscripts should count more heavily, because they are entered into the process more often. This situation can be easily established if we take: iv. The total weight of the formula is distributed over those groups with more than one member according to the numerical squares ng 2 of the number ng of members of a group g (see example below). Then the positive contribution 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

‘a / ‘b ‘c / ‘d ‘e ‘f / ‘g ‘h ‘i ‘j / ‘k ‘l ‘m ‘n ‘o / ‘p ‘q ‘r ‘s ‘t ‘u / ‘v 270

has seven different readings, and a final weight 270 after the multiplications of the previous steps. Now this weight has to be distributed over the groups in the proportions 0 / 4 / 9 / 16 / 25 / 36 / 0, so the group ‘b ‘c gets value 904 ·270 = 12, group ‘d ‘e ‘f gets 27, etc. Now the only pair in group ‘b ‘c gets the full 12, the three pairs in ‘d ‘e ‘f get 9 each, the six pairs in ‘g ‘h ‘i ‘j get 8 each, the ten pairs in ‘k ‘l ‘m ‘n ‘o get 7.5 each, and the fifteen pairs in ‘p ‘q ‘r ‘s ‘t ‘u get 7.2 each. In fact, a pair in a group of two gets almost double the weight a pair in a large group gets.

Constructing initial binary trees in stemmatology

In order to distribute the formula weight over the pairs in different groups to get the negative contributions, we distinguish two cases: (a) one of the symbols is a group in itself, i.e. it is a single reading, or (b) both symbols are from groups with at least two members. We observe that it is a bad idea to join the two symbols in case (b), because that would introduce a contradicting quadruple in the tree. In case (a) we have at worst a single deviating reading which will vanishes in the next stage, which is no big deal. So pairs of type (b) are much more important, and we will give them a factor of for example 16 in comparison with pairs of type (a). When we look in the example above, we have 41 pairs of type (a) and 155 pairs of type (b), and so the weight factor is to be divided by 41 + 16 · 155 = 2521. So the pairs of type (a) ‘a ‘b, ‘a ‘v, ‘b ‘v etc. get 270 a score of – 2521 = –0.1071. The pairs of type (b), for example ‘b ‘d, ‘e ‘g etc. get a 270 score of – 2521 ·16 = –1.7136. We should again formulate this as a design choice, and obtain: v. Pairs whose members are in two different non-single groups receive a negative value which is 16 times as large as pairs with one or two members in single reading groups. . First pass – 3 – deciding upon the closest pair of symbols By the end of this pass, we have collected a sum Sp of all the positive contributions by pairs, a sum Np of the absolute values of negative contributions by pairs. This value Np can be split up into two parts, Qp accounting for quadruple contributions, and Rp representing the contributions from single readings. Now we can make some design choices to determine the pair that is to be squeezed into one symbol at this stage. Possibilities are: (a) minimal absolute value of Qp , so the number of contradictions introduced is as small as possible, (b) maximal value of Sp – Np , which chooses the pair with maximal support, (c) minimum of Qp divided by Sp , i.e. the amount of contradictions is minimal in proportion to all contributions. Our favoured choice is: (d) the negative contribution is proportionally minimal: N

vi. The pair for which Sp +pNp is minimal will be taken together to form a single new symbol, to replace the two symbols of this minimal pair. . Second pass – 1 – production of new version formulas Since in the previous pass it was decided which pair of symbols should be replaced 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. This requires two main decisions: (a) how to replace, and (b) how to adjust the weights. These two steps are closely interrelated, and they should be expressed for every type of occurrence of the pair in the version formulas. We dismiss all the weight adjustments made in the previous pass. The input is the original data base or the data base constructed in the previous stage. When output is generated, it is put into the new database. We start with copying the header, omitting the two symbols of the pair and adding the new single symbol. The design choices (vii) through (xi) express the necessary decisions made for all of those different occurrences: vii. If neither member of the pair occurs in the version formula, it is just copied to the output. If only one member of the pair is in the formula, that member is replaced by the new symbol, andthe version formula is then just copied to the output with the original weight. viii. The two symbols of the pair are members of the same group. In that case replace the two symbols by the new symbol. If the new formula still has groups with more than one manuscript, output the line with the original weight. ix. If one of the members of the pair is a single reading, just omit it and replace the other member of the pair in that formula with the new symbol. If there is more than one group left, copy the new version formula to the output with the original weight; if not, omit the formula. x. If both members of the pair are single readings, just omit them. If there is more than one group left, copy the rest of the version formula to the output, with the original weight. xi. If the members of the pair are in different groups, and each of those groups contains more than one manuscript, we have contradicting quadruples. We then multiply the weight by a factor 0.4, and output the new version formula twice: (a) once omitting the first member of the pair and replacing the second member by the new symbol, and (b) once omitting the second member of the pair and replacing the first member by the new symbol. In this way, the total weight of the original formula loses a fifth of its influence in the next stages. This is a reasonable compensation for the contradictory nature of this formula which has been discovered at this stage.

Constructing initial binary trees in stemmatology 

. Second pass – 2 – computing edge distances This step and the next step are not important for the tree construction itself, but they will help to understand the trees. We will again use proportions, so we will register contributions to the distance as well as weight contributions for the formulas. We concentrate on each member of the pair, and use the original data base, or the database constructed in the previous stage, just as in second pass step 1. xii. If only one member of the pair is in a formula, we skip this formula for distance computation. xiii. If both members are single readings, the distance contribution is half the formula weight, and the weight contribution is the full formula weight. xiv. If both members are in different groups and each of those groups contains more than one manuscript, the distance contribution is also half the formula weight, and the weight contribution the full formula weight. xv. If one member is a single reading and the other member is not, the weight contribution as well as the distance contribution for this member is the full formula weight. If the other member is a single reading, and this member is not, the distance contribution is 0, and the weight contribution is the full formula weight. xvi. If both members are in the same group, the distance contribution is also 0, and the weight contribution is the full formula weight for both members of the pair. xvii. The distance indication of the edge connecting a member of the pair with the node of the new symbol is the quotient of the distance contributions and the weight contributions (the easiest way to express this value is as a percentage). . Second pass – 3 – quadruple contradiction weights In the same way as in the previous step we can compute a quadruple contribution percentage. Again we start with the original data base or the result of the previous stage, and compute a quadruple contribution as well as a weight contribution. In this step not the symbols themselves but the pairs are important. It should also be stressed that the edge which is involved most in the quadruple contradictions is the edge which emerges from the new symbol in the direction of 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 contribution should be the formula weight. If the two are members of two different groups, each with more than one manuscript, we have a contradicting quadruple and the quadruple contribution equals the formula weight for this version formula. xix. The quadruple indication of the edge emerging from the new symbol in the direction of the top is the quotient of the quadruple contributions and the weight contributions, and the easiest way to express this value is as a percentage. . Final stage The algorithm has to stop when there are only three manuscripts left, in which situation we can look for the closest pair, and compute distance contributions. It has already been mentioned that those three symbols should be connected to a top node. In this case there is no quadruple indication, but the edges which connect the three symbols to the top will have received a quadruple indication in the previous stages for the non-terminal symbols. If the data base gets empty at a certain stage, or if all version formulas are trivial, it makes no sense to continue the algorithm any further. The pairs could be squeezed in a random way to get a fully resolved tree, and all the distance indications should be put at 0. Quadruple indications should be considered undefined. If this happens, the internal nodes of the tree which still exist at that stage could be identified with the top node if we use an internal contraction method afterwards (cf. Wattel & Mulken 1996b).

. Conclusion Although the program described here is meant to yield an initial tree, it can also be used to construct an unordered tree structure straight from the collection of version formulas. It is even possible to construct a fully oriented stemma with it, if we indicate the unquestioned archetypical versions in the formulas with a sigla ‘@, 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 and yields consistent results most of the time. It does not contain a contraction method and so we need a contraction program in cases where it is expected that some of the manuscripts are intermediates, and when the stemma is not fully resolved.

Constructing initial binary trees in stemmatology 

The results of the program itself are usually consistent with the cluster structure of the set of manuscripts. There is a need for a new program implementing stepwise refinement according to the latest ideas, which could give statistical information about the error probability of the final result.

References Bryant, D. & M. Steel (1999). “Fast algorithms constructing optimal trees for quartets”. In Proceedings of the Tenth Annual ACM-SIAM Symposium on Discrete Algorithms (pp. 147–155). Maryland: ACM/SIAM. Dearing, V. A. (1974). Principles and practice of textual analysis. Berkely, Los Angeles & London. Dees, A. (1977). Over stambomen van handschriften. Forum der Letteren 1977, 63–73. Dekker, M. C. H. (1986). Reconstruction methods for deviation trees. Doctoral thesis Mathematics and Informatics, Vrije Universiteit Amsterdam. Hollander, A. A. den (1997). De Nederlandse Bijbelvertalingen 1522–1545. Ph.D. Thesis Vrije Universiteit Amsterdam. Nieuwkoop: De Graaf. Hollander, A. A. den (this volume). “How shock waves revealed successive contamination : A cardiogram of early sixteenth century printed Dutch Bibles”. Maas, P. (1957). Textkritik. Verbesserte und vermehrte Ausgabe. Leipzig. Salemans, B. P. J. (1996). “Cladistics or the resurrection of the method of Lachmann”. In P. T. van Reenen & M. P. van Mulken (Eds.), Studies in Stemmatology (pp. 3–70). Amsterdam & Philadelphia: Benjamins. Salemans, B. P. J. (2000). Building stemmas with the computer in a cladistics neo-lachmanan way. Ph.D. Thesis Katholieke Universiteit Nijmegen. Nijmegen University Press. Wattel, E. & M. P. van Mulken (1996a). “Shock waves in text traditions”. In P. T. van Reenen & M. P. van Mulken (Eds.), Studies in Stemmatology (pp. 105–122). Amsterdam & Philadelphia: Benjamins. Wattel, E. & M. P. van Mulken (1996b). “Weighted formal support of a pedigree”. In P. T. van Reenen & M. P. van Mulken (Eds.), Studies in stemmatology (pp. 135–168). Amsterdam & Philadelphia: Benjamins.

Trouble in the trees!* Variant selection and tree construction illustrated by the texts of Targum Judges Willem F. Smelik University College London

.

Textual history

The early textual history of Targum Jonathan to the Prophets is a matter of conjecture. Not a single complete manuscript predating the early twelfth century ce has survived. Our earliest evidence consists of some isolated quotations in classical rabbinic literature (second to eight centuries ce)1 and magic bowls, produced between the third and sixth centuries ce.2 The reasons for this lack of hard evidence are miscellaneous. The practice of burying and depositing outworn and faulty manuscripts, which would not actively be destroyed because they contain the Sacred Scriptures, may partly account for this situation, which has no doubt been aggravated by the vicissitudes of migration and persecution. By the time when Aramaic had been superseded by Arabic in the Islamic countries and Indo-European languages in the West, the Targums had largely lost their originally important place within the curriculum of rabbinic students and the liturgy of the synagogue. Obviously, this development would not help preservation of extant manuscripts, all the more so since there may have been rather few manuscripts in the first place. Because of the status of the Targum as oral Torah, it is not impossible that the textual transmission of the Targums remained predominantly oral during the first millennium ce.3 Any construction of the textual history will thus have to make do with little evidence. By common consent, dialect, contents, and the rabbinic regulations for the liturgical use of Bible translations favour an origin of the Targums to the Torah and the Prophets in or soon after the second century ce – although

 Willem F. Smelik

the issue of dialect is still controversial.4 Because of exegetical parallels between Targum Jonathan and later Amoraic literature, we may safely assume that the Targum was subject to changes and supplements in the following three centuries, even though the extent of these changes has still not been explored in a satisfactory way. A stemma for the available textual witnesses would, ironically, only cover the part of the textual history when the heyday of this genre was long past, and the fluidity of targumic traditions had given way to a fairly stable textual tradition. Only 24 (nearly) complete manuscripts of the Targum to Judges are extant nowadays, a number similar to that to any of the prophetic books; however, hundreds of manuscripts have the Targum to the lectionary readings, the so-called haft.aroth.5 These manuscripts have been produced in such widely divergent places and periods as twelfth-century Italy, fourteenth-century Germany, fourteenth–fifteenth-century Aleppo, fifteenth-century Yemen and sixteenth-century Spain. Their relationships have hardly been explored6 beyond the fairly obvious family connections of the Babylonian, Yemenite, Sefardi and Ashkenazi manuscripts. One of the long-standing questions concerns the position of the Yemenite manuscripts, and especially the manuscript in two volumes which Alexander Sperber selected as the basis for his commonly used edition of Targum Onqelos and Targum Jonathan, ms Or. 2210-11 (British Library, London).7 This manuscript was previously thought to be unique among the Yemenite witnesses, but it receives substantial support from some of the fragmentary Babylonian witnesses, widely held to represent the oldest text type for historical, textual and grammatical reasons.8 It is plausible that his manuscript of choice was a crock of gold. It should be noted that he gave preference to a grammatically correct text,9 so that it may not be entirely coincidental that Dalman had given preference to this manuscript in his grammar of Jewish-Palestinian Aramaic.10 In recent years some stemmatologists have focussed on the issue of variant selection for the purpose of tree construction. Some scholars, notably Salemans,11 argue against a quantitative approach which allows any variant reading to have its influence. They suggest that a rigorous selection of variants should precede any attempt to build a stemma. Other scholars would disagree, arguing that all variant readings should be taken into account, and that selection should occur in the final stages of analysis. A middle of the road approach allows most variant readings to have their influence, but would attribute lower values to circumspect variants, and higher ones to more reliable variants.

Trouble in the trees! 

No generalizations should be applied to any given textual tradition without consideration of its idiosyncrasies.12 Codicological studies of medieval Hebrew literature have shown that scribes assumed considerable freedom in revising, adapting and improving the text of their exemplars; not infrequently, the authors even appealed to scribes to correct mistakes and even, on at least one occasion, to add useful material.13 Because draft versions of certain books sometimes circulated before the final version was ‘published’, the texts often existed in multiple, co-existing versions. For that reason it will be useful to observe, justify, and verify the reliability of several types of variant readings for the Targum to the Prophets to avoid the pitfalls of either including or excluding too much information. To establish these categories, variations introduced in dual copies of a single exemplar, or in the copy of a known exemplar are of great value. By study of these variants we hope to disclose patterns and tendencies in the process of textual transmission, and to identify purely random variations that did not carry any value in the eyes of the scribe him- or herself. Especially in dual copies of a single exemplar we may identify the kind of variation that the scribe would not deem very important, and that I will call ‘random’ variation. A genealogy of the textual witnesses for the Jewish Aramaic Bible translations is problematic because our copies reveal traces of infuences from other sources than their exemplars: other copies which the scribe consulted, the Hebrew text which was usually transmitted within the same source, and quite possibly memorized passages.14 These influences thwart a linear type of genealogy, where each copy is produced from a single exemplar. All the same, linear relationships between the texts remain important and the extent to which we are able to retrieve such relationships should concern us here. In a previous study I have briefly explored a handful of criteria for the selection of variant readings; in this article I will supplement and evaluate these criteria. In tandem these two studies highlight random variation in our textual witnesses in order to establish an empirical, rather than theoretical, basis for the selection.15 Surprisingly, no such criteria have ever been set for targumic literature.16 This article will supplement these earlier observations on variant selection; in addition, it will evaluate the value of the classification. The evaluation will assume the form of interpreting alternative trees and shock waves (both of which have kindly been produced by Evert Wattel of the Free University in Amsterdam).17 The core of this study thus consists of observations on the phenomenon of random variation in manuscript reproduction. Random variations are those variants 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 agreements between manuscripts would not reflect their genealogy. To identify such coincidental variation I will draw extensively on two Spanish manuscripts and the first 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 information as could possibly be provided by these types of variant readings, even those regarded with the utmost reservation, will be discussed and related to the various levels of stemmatological studies.18 The Spanish manuscripts used here were written by the Jewish convert Alfonso de Zamora in 1532 and 1533.19 The exemplar of both apographs is unknown, but may well be the lost ms that had been purchased for the Complutensian Polyglot,20 published between 1518 and 1520, as he wrote them in Alcalá de Henares, where he had been a professor of Hebrew since 1512 and where the manuscripts had been stored after publication of the Polyglot. Both copies contain the Aramaic text of Targum Judges in the left column, and its Latin translation in the right column. The First and Second Rabbinic Bible, edited by Felix da Prato and Jacob ben H ayyim respectively, and published by Daniel Bomberg in 1517 and 1525,21 . represent a parent and daughter text; the first edition, presumably based on Codex Solger, now housed in Nürnberg, served as the basis of the second one; however, Jacob ben H . ayyim evidently consulted one or more manuscripts in addition to his exemplar.22

. Random variation and tree construction To single out variant readings that do not unequivocally carry genealogical value, I will explore the phenomenon of ‘random variation’ in manuscript reproduction. While isolating these phenomena, I will pay attention to the levels of inquiry at which they can still have their say, because they add information that we cannot afford to lose. The following fields of variants were distinguished: a. b. c. d. e.

Orthography Error Separation and contraction Vocalization Abbreviation

f. g. h. i. j.

Plus Minus Substitution Semantic shift Preposition and copulative

Trouble in the trees!

k. Relative particle l. Verbal morphology m. Number

n. Person o. Gender p. Status

Inevitably, the categorization is heuristic, and the categories themselves are blurred.23 This particular classification is devised to test the reliability of certain variants, whose reliability has been called into question in previous studies, which tend to single out certain verbal and nominal variants.24 These latter variants, numbered f to i, are deliberatedly contrasted with the remaining categories that are deemed to produce much less reliable results, such as orthography and morphology. The following observations will only produce a tentative result in order to answer two questions: (1) Which variants can be fruitfully employed to build a provisional stemma? And (2) to what extent do all variant readings shed light on particular relations between textual witnesses? Since the first five categories have been discussed in greater detail elsewhere,25 the following discussion will focus on the remaining categories; however, I will summarize my findings for the first ones. . Orthography The vast majority of variant readings between the two copies of Alfonso de Zamora concerns the use of vowel letters (70%). The variation shows that these variants do not bear on his copies’ relationship to their exemplar, hence they obscure rather than illuminate genealogical relationships. This conclusion should not be applied to all textual witnesses, however, since the First Great Rabbinic Bible and its successor, the Second Great Rabbinic Bible, almost always agree in their use of vowel letters. From these observations two conclusions can be inferred: spelling variants should not be admitted as evidence for the initial construction of the network between all textual witnesses; all the same, they should not be discarded altogether, because they may corroborate our observations about certain individual relationships between textual witnesses. . Errors Scribal errors represent unintentional changes that are quite common in manuscript reproduction,26 and may therefore have been introduced by scribes independently of each other. Common scribal errors like the omission of words based on similar beginning or ending of clauses (homoioarcton, homoioteleu-



 Willem F. Smelik

ton) and the confusion of similar letters occur in almost every manuscript. These errors may have been transmitted down a family line, when they have been faithfully transmitted by copyists. In such cases they reveal a genealogical relationship between these manuscripts, as, for example, between the closely related Yemenite manuscripts Or. 2371 and 1471, housed in the British Library, which share four unique cases of haplography,27 as well as small minuses unique to them elsewhere.28 Nonetheless, common errors can never be taken to represent genealogical information at face value for two reasons.29 Scribal slips are often typical, and susceptible to repetition; indeed, examples of shared, but no doubt unrelated, errors are found in several verses. Moreover, errors were susceptible to correction. Phonetic errors and Hebraisms could easily be reverted into the original reading in a later copy,30 and minuses could be restored on the basis of the parallel transmitted Hebrew text. Only if there is circumstantial evidence to support the assumption of a real, genealogical connection between the mss sharing a scribal error, would we take this type of evidence into consideration. . Vocalization The use of variant vocalization is fraught with difficulties. The systems of vocalization differ, the scribe did not always vocalize the text himself,31 or, as seems likely in case of Alfonso de Zamora, he invented the vocalization himself. . Separation and contraction The separation and contraction of prefixed morphemes such as and resulted in a variation that appears to reflect scribal conventions, and sometimes typographical ad hoc considerations, rather than genealogical information. . Abbreviation Abbreviations reflect scribal conventions and typographical considerations, hence they vary by text. They are only useful if a copyist made a mistake when he filled out an abbreviation in his exemplar, or when the versions differ in their readings. Some of the variants agree with each other by pure chance, because copyists filled out abbreviated words.32

Trouble in the trees!

. Grammatical properties Many variant readings concern a change of grammatical properties, like gender, number, status and morphology. Some of these variant readings are inspired by a zeal for grammatical improvement, for instance the introduction of the feminine third person plural suffix, , which, in earlier manuscripts, seems not to have been distinguished from the masculine form, .33 As grammatical ‘improvements’, if indeed that is what they are, they may have been introduced independently by different scribes at different times. Such examples would not carry genealogical information, but rather would distort our picture of the network between the manuscripts and editions. Table 1 provides many examples of improvements on the part of Jacob ben H . ayyim, who edited the Second Rabbinic Bible for Daniel Bomberg. On the other hand, these data will sometimes reveal a dialectal kinship between groups of texts, as with the gender of the word , which is invariably masculine in the Eastern texts, but feminine in the Western ones. Another frequent variation concerns the number of nouns after numerals; consider the following variation, in Table 2 below.34 Some of these texts, of which we either know or suspect a close relationship, nicely fit together, but there are many irregularities to be noted as well. On the assumption that the Babylonian and Yemenite manuscripts represent the oldest text-type, the singular state of the noun is original, and the lack of variation except for one position in Eb66 is significant. The most likely explanation for the variation in this table is that copyists occasionally felt that a plural form was necessary; however, they may have made their ‘corrections’ almost unconsciously, as there are blatant inconsistencies even within a single chapter. One should also take into account that the syntagma is followed by a plural participle in 18.11, 16 and 17; some versions correct this plural form into the singular but once again inconsistently.35 As a consequence, these ‘corrections’ may well have been introduced independently from each other, representing a polygenetic, irrelevant type of variant readings, as some variations do indeed suggest. Thus the disagreements cannot be relied upon, whereas the agreements appear to be reliable. To deal with these readings, it seems advisable to create a separate class of readings, so that the validity of this class, including many occurrences not included in Table 2, can be assessed in comparison with other classes. Rather than curbing their possible blurring effect by using value factors in computational approaches, or by their immediate identification as noise, I would argue in 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, as the occurrences are not frequent enough to isolate them. The source of these variant readings is equally difficult to pinpoint. The variant for

Trouble in the trees!

Table 2.

38

vs.

Siglum

3.31

7.6

7.7

7.8

7.16

p Eb1 x y w q63 Eb66 K B O S W M F P T d E a C A J Q N Rb1 Rb2

s – s s s s s s s s s pc pc s s s s s pc pe s s s pa pa pa

s s s s s s s s s ? s pc pc s s s s s s s s s s s s s

s s s s s s s s s pa pa pa pa s s s pc p39 s s s s s pa pa pa

s s s s s s s s s s s s s s s s s s s s s s s s s s

s s s s s s s s s s s s s s s s s s s s pa pa s s s s

Variation place 7.19 8.4 18.11 s s s s s s s s s s s s s s s s s s s s pa pe s s s s

s – s s s s s pa s s s s s s s s s s s s s pa s pe pe pe

s s s s s s s s s s s s s pa pa pa pa pa pa pa pa pa pa pa pa pa

18.16

18.17

20.47

s s s s s s pa s pa pa pe pa pa pa pa pa pa pa pa pa s pa pa pa pa pa

s s s s s s s s s s s s s s s s s s s s s s s s s s

s s s s s s s s s s s s s s s s s s s s s s pa s pa s

in 11.19, for instance, may well reflect influence of Deut. 2.27 on either the Hebrew or the Aramaic text, or on a later corrector working from another exemplar with a different text. The variant for in 11.8 is less likely to reflect intertextual influence; here, however, an error is not inconceivable. Both readings, however, contradict the Masoretic Text (mt) and appear in old Eastern manuscripts, which is a factor to be considered, as it is an old rule of thumb that readings contradicting the mt are important for the textual criticism of the Hebrew Bible.40



 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 with these lexemes on a semantic level. Such equivalents are called substitutions. They obviously represent purposeful choices; in the sample texts, our single example occurs in 8.24. More frequent is the use of an equivalent which partly overlaps with the alternative reading; a phenomenon labelled as a semantic shift.41 Together, the examples occurring in the Second Great Rabbinic Bible are listed in Table 3. I will include in these categories verbs, nouns and adjectives only;42 conjunctions, particles and prepositions will be discussed separately. Which substitutions are susceptible to multiple introduction? Some substitutions reflect a surprise about the choice of translation equivalents, when a stereotyped equivalent was not adopted; at these points variant readings occur that prefer the more usual equivalent. In fact, the Masorah to Targum Onqelos seems to be designed at least in part to report exceptional translations, so as to prevent alterations resulting from hyper-criticism. Other substitutions are related to Hebrew variant readings;43 these are likely to be introduced by more than a single scribe, and thus not reliable as unique mutations in the textual history. Similarly, it stands to reason that semantic shifts or near-synonyms which reflect dialectical developments, such as for in 1.6,44 cannot be accepted as reliable evidence at face value; obsolete words may well have been replaced by later copyists (substitution by synonyms).45 But the extent of such practices should be investigated and verified before assuming this hypothesis to be true. Readings introduced on the basis of a Hebrew text can easily be pointed out in the textual history of Targum Judges; whether semantic shifts due to dialectical processes occur is far less clear.46 A fair number of substitutions and semantic shifts consist of theological and exegetical changes, which have been inspired by non-textual considera-

Trouble in the trees! 

tions;47 yet they are often not so obvious that they could have been introduced independently by several scribes without one knowing about the other. The change recorded for Rb2 in 8.24 is far from obvious, even though it reflects an exegetical concern that may have been shared by others as well. Other examples are related to established exegetical discussions, and may involve a series of substitutions or semantic shifts; a good example is a set of semantic shifts in chs. 17 and 18.48 The more complicated such an operation of alteration is, the less likely it becomes that several scribes carried out the same activity. There is a real chance of contamination here, for differences such as these would not go unnoticed. They would often be scribbled in the margins of manuscripts, either as an alternative reading or as a correction. Whether the editor of Rb2, for instance, noticed the example listed above himself or found them in the margins of his (master) exemplar, is unclear; the source for these particular changes cannot readily be identified. These readings result in contamination, because they do not reflect the Vorlage of Rb1, yet they do not seem to have originated with the editor without recourse to alternative texts. For our purposes, however, the real issue is whether the reading is susceptible to independent introduction by more than one scribe. ‘Simultaneous contamination’, which occured when a scribe consulted other sources in addition to his exemplar,49 is not a problem to eradicate, because it is stemmatologically relevant; it is a phenomenon to detect and preserve. Random variation is the real culprit. In sum, most of the semantic shifts and substitutions should be evaluated on a case by case basis. Those reflecting Hebrew variant readings should not be taken into account; all others should be tested. The presence of polygenetic readings is possible in this category as a whole, so one must interpret cautiously. Unfortunately, the relatively small number of such readings does not allow a separate test to be carried out on them. . Pluses and minuses Pluses and minuses are not invariably valuable, so that different types of pluses and minuses should be distinguished. The majority of minuses reflect typical scribal errors, especially haplography,50 and a good many pluses may reflect the influence of the Hebrew text. The following examples in Rb2 (Table 4) illustrate how limited the value of such variants may be. Three of these examples in Rb2 reflect the influence of the Hebrew text (7.20; 7.22; 11.35), although the first one is shared by all other witnesses and may 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 (11.29), another a correction of an error in the exemplar (11.31), and the sixth represents a Sefardi (?) influence (contamination, 11.39). None of these readings reflects a trustworthy genealogical relationship. The minuses and pluses in the copies of Alfonso de Zamora, printed in Table 5, do not fare very well either. Alfonso de Zamora strove to achieve a faithful copy of his exemplar,51 as his corrections show (marked by ‘mg ’). Any minus has usually been restored in the margin by Alfonso de Zamora himself, while the equivalent of the minus in the parallel Latin translation confirms that it really is a scribal error.52 Nonetheless, many of these minuses are shared at random by other mss, which points to independently occurring, but identical, errors. The evaluation of the translation plays a prominent part in the creation of variant readings. When later tradents, readers or correctors evaluated the adequacy of the translation, and felt it did not accurately reflect the Hebrew, they introduced changes or noted alternatives. The phenomenon of double translations allows us to detect such later alterations of the text, when a copyist combined two alternative translations,53 apparently on the assumption that one of them, perhaps written in the margin, had been omitted by accident – or perhaps he was unsure which one to select. In most cases, to be sure, these alternatives were not combined into an erroneous double translation,54 and the only trace of such evaluations is a plus or minus in some of the textual witnesses, that do not appear to be mere errors.55 A particularly fine example is found in 2.1, where the (majority) translation for Hebrew reads ;56 some copyists or glossators apparently believed the Hebrew construction should be preserved literally, hence in a handful of Yemenite manuscripts, the more ‘literal’ translation replaced the longer reading,57 whereas in other manuscripts both translations were jumbled together, .58 A further source of minuses and pluses consists of actual Hebrew variant readings, that have found their way into the Aramaic text. Whether these pluses

Trouble in the trees! 

Table 5. Pluses and minuses in mss 7542 and M-1

belonged to the ‘original’ text of Targum, or have been introduced into the text at a later stage, often remains obscure.59 In fact, copyists often would alter their readings, adjusting their text to a Hebrew variant in their bilingual exemplar.60 Interestingly, a number of such instances has been found in manuscripts of poor grammatical quality. This also illustrates that no matter how poor a manuscript may be, it should not too readily be omitted from an edition, even if related manuscripts in a better condition exist. Finally, some pluses may have resulted from standardized language or lacunary exemplars. In 6.36, the plus to is indicative of a standard combination (K has the plus at this point), so that this reading does not necessarily have any genealogical value. Other variant readings may have resulted from difficulties in reading the exemplar.

 Willem F. Smelik

It is hard to assess the actual degree of parallelism – that is, accidental agreement in change – in a fair but consistent way. The combined reading in 2.1, which was discussed above, is likely to reflect accidental agreement between the texts that attest it, rather than a genealogical relationship. Several scribes may have included in their main text a reading that was noted in the margin of their exemplars (namely, ). On the other hand, the two other readings ( and ) do reveal a genealogical relationship, and we may have a fair idea about the original reading as well as later developments, so that these instances can often still be fruitfully employed. In other words, part of the ‘formula’ is reliable, and part of it is not.61 For the purpose of evaluation, a ‘safe’ file without these instances should be preferred; the information thus left out of consideration can always be brought to bear upon our understanding of the genealogy in detailed refinement. Other cases to be excluded are the minuses that reflect typical scribal errors, and the pluses that involve stereotyped phrasing, will be categorized separately. . Prepositions, and In the sample selection of the two copies produced by Alfonso de Zamora, only a single difference in their use of prepositions occurs, which is due to scribal instead of ; interestingly, the Antwerp polyglot, which error (M reading is based on M’s exemplar, has the same mistake). Rb2 has a different preposition than its exemplar Rb1 in six instances (see Table 6). Of these six, no less than five agree with the Hebrew (Masoretic) Text, which suggests that the parent text casted a long shadow over its translations (even if some of these Hebraisms may have been suggested by the consultation of other Aramaic versions, they ultimately appear to be Hebraisms). Apart from such reversals to Hebrew language, prepositions may also have been changed for idiolect changes or idiomatic adaptations. The only instance that is not a Hebraism Table 6. Prepositions in Rb2

Trouble in the trees!

Table 7. The particles in Rb2

(7.16) consists of a grammatical improvement. None of these examples represents a reliable source of information for building a stemma; therefore, their impact on tree construction should be considered separately. The plus or minus of the copulative should be studied in conjunction with Hebrew textual criticism, which falls outside the scope of the present study.71 The variation in using reflects inner-Aramaic variation, which is not so prone to Hebrew influence. On the other hand, could as easily be introduced as omitted in many instances, as Table 7 of such changes in Rb2 might suggest, and it seems wisest to refrain from the use of these readings in building the initial stemmas. . Additional types of variants The present survey has not been exhaustive, but has focussed on the most frequently found and relevant phenomena. For example, auxiliary verbs, transposition and paratextual elements have not been included. Auxiliary verbs do not occur at random in targumic literature, and their presence or absence, which is rejected as inflectional parallelism by Salemans,72 should, accordingly, be included in the variant selection. Transpositions do not occur often in this particular textual tradition, for obvious reasons: primarily, because these errors would have been easily detected in comparison with the Hebrew text, which the Targum either physically or functionally accompanied. The Aramaic translation tends to mirror the Hebrew word order faithfully;73 for this reason, inversion of verses is rare as well.74 Paratextual elements, however, shed a far more interesting light on the genealogy of manuscripts. The Toseftot, ‘additions’, were often added in the margin of texts, but as an alternative translation they were soon incorporated in the body of the text, here juxtaposed to the original translation, there ousting that ‘original’ reading. As such, their genealogical value and textual position



 Willem F. Smelik

may vary between exemplar and copy, especially because the marginal readings may not have been written by the same scribe who signed for the body of the text.75

. Evaluation . Methodology It is one thing to argue for categories of variants, but another to test them. Even if we try to keep our assumptions in check by comparing the variation between texts whose relation is known to us, it seems worthwhile to evaluate the results even further. Two tools have been used in the following evaluation: similarity graphs and shock waves. Shock waves are used to detect possible non-linear influences between the (sub)groups of textual witnesses, usually called ‘contamination’. Successive contamination occurs when a copyist (or editor) switched from one to another exemplar at some position in the text; up to this position, his copy follows exemplar X, thereafter exemplar Y. To detect such shifts in relations, Wattel and Van Mulken developed the instrument of shock waves, which peak wherever a change of affinity between textual witnesses occurs.76 Similarity graphs show the relations between the witnesses on the basis of each category, and it is possible to construct several images of these witness relations to see how these images vary among themselves, and where irregularities may point to important information about either my classifications or the data. The images, or genealogical trees, represent the network of relations between the textual witnesses. Their construction follows the procedures as developed by Evert Wattel, and they can be based both on similarity scores between witnesses and on dichotomies of pairs of witnesses, the latter by quadruple calculations. Thus it is possible to create a set of images for each category. To see the wood for the trees, only a few tell-tale graphs based on the same method will be reproduced here.77 Finally, the known relations between manuscripts and editions as outlined above will be brought to bear upon these trees; wherever the trees are in conflict with these data, the construction must have been wrong, and since the procedures have been tested with great care, it is likely that my selection of readings was unreliable. Some of the categories produced a sufficient amount of variant readings to be processed separately, but most did not. Therefore they were combined into the following groups:

Trouble in the trees! 

1. Scribal conventions: orthography; separation and contraction (or word boundaries). 2. Scribal mistakes: errors, including erroneous changes in the suffixed pronouns.78 3. Grammatical properties (nouns and verbs): number, gender, state. 4. Variation in other wordgroups and inflection: preposition, copulative, conjunction; tense and conjugation. 5. Nouns and verbs: plus, minus, semantic shift and substitution. 6. 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 case evaluation). These groups essentially contrast nominal and verbal variations with those in other wordgroups, inflection, errors and spelling. The shock waves for all these groups did not display a significant shift of affinity, not even on the borderlines of categories that were combined into a single file. This result justifies the combinative approach, no matter how pragmatic it is. It also suggests that there is no case of successive contamination among any of the witnesses involved. Nonetheless, there are some interesting observations to be made on these shock waves; detailed comments will follow below wherever relevant for a category and separately, in an assessment of the extent to which the tradition is contaminated. The first test that these trees and shock waves pass is based on the relations between texts which are known to us: the two copies made by Alfonso de Zamora, the Nürnberger Codex and its daughter-editions, and the two affiliated Yemenite mss bearing the sigla ‘w’ and ‘y’ in Sperber’s edition. The graphs in which the direct relations between these witnesses have been distorted, discredit the data on which they were based. Wherever there is little hard evidence, we must resort to other ways to assess our results. A heuristic way is to evaluate the cohesion of different images which are based on a single category. In some cases, the trees for one category do not agree among themselves, whereas in other categories the trees are much more homogeneous, suggesting that these latter categories are more reliable. . Category 1: Spelling The first category contains all orthographic variant readings, and this file was expected to produce the poorest result. Even though the picture yielded by the readings 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 0

15

30

45

60

75

90

105 120 135 150 165 180 195 210 225 240 255 270

Figure 1. Shock waves for Category 2

categories,79 it is essentially flawed. External evidence, as well as careful comparison with the other categories, proves that a number of witnesses have ended up in the wrong position. The Antwerp polyglot, o, based on the same exemplar as W and M, has not been grouped together with these texts. The sigla Eb1, Eb91 and Eb4 (all Babylonian manuscripts) have been separated from

Trouble in the trees!

the mainstream of Eastern manuscripts, to which they undoubtedly belong. Likewise, but less marked, Eb66 and q2 have been separated from this branch. In almost all other trees, they are grouped together. Similarly, S has not been grouped together with W M O, as in four out of six categories. Likewise, the close connection between a A B J and Q has been distorted. Although some of the information agrees with that presented by the remaining categories, the overall picture is not reliable. It should be remarked that G and s4, two of the haft.aroth, contain too little evidence to produce reliable 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 and stable relationships in between; see Figure 1. In many cases, the peaks occur where an obvious error has been corrected in subsequent copies, either of the scribe’s own accord or by following a marginal correction. A point in case is provided by some errors in Codex Solger, where the First and Second Rabbinic Bible follow its marginal corrections. Such corrected errors blur the picture, as the shock waves seem to illustrate, which confirms the point that errors should only 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 Sefardi mss, M, W, O, S; see Figure 2. Whereas we know that M and W have been copied by the same scribe, either from the same exemplar, or from each other, nothing in this tree suggests as such. As a result of editorial activity, a number of errors in N and Rb1 have been corrected in subsequent editions; indeed, these editions have been torn apart in the present picture. The quadruple method achieved a somewhat better result, but separated M and W, besides other errors such as the separation of w, y even in a category where I had expected them to be brought together (see Section 2.2). All in all, this category seems to yield the most unreliable results of all. . Category 3: Grammatical properties The third category, which includes all grammatical properties like number, gender and state, achieves far more stable results. The Sefardi group,80 the Solger group, the late Ashkenazi group, and the Eastern subgroups agree more or less with the remaining categories. As for the shock waves, none of the peaks displayed in the bottom line was remarkable, and minor peaks appear to reflect either minor divisions or



 Willem F. Smelik ˆ34 42 J

4

37

ˆ32

4

ˆ27

A ˆ26 9

ˆ25

O 3

2 ˆ24

ˆ22

ˆ9

ˆ21

ˆ20

W

4

ˆ3

6

5

Rb6

0

F

2 Eb1

ˆ18 15

2 ˆ14

ˆ1 18

2

0

p

M

@

1

2

2

ˆ2

s4

K 4

ˆ12

3

3

2 Rb2

Rb3 ˆ13

3

6

3

2

2

1

3 ˆ5

ˆ4 0

ˆ7

ˆ19

11

a

Q

y

3

ˆ17

E

Eb4

G

w 3

3

6

1

4

4

2

5

7

N

Rb1

T

S

3

ˆ23

3

3

14

3

3

ˆ16

16

4

11

C

8

ˆ28

B

D

7

ˆ29

P

33

ˆ31 41

1

ˆ30

ˆ33 11

23

5

5

5

1 q63

d

ˆ11

ˆ10

2

2

o 4

Eb91

Eb66

2 ˆ8

4

ˆ6 4

q2

1 x

Figure 2. Initial tree for Category 2

noise. An example is the singular (10.11) in a Yemenite subgroup (x y w q63), which agrees with a single Ashkenazi manuscript B, and with the Rabbinic Bibles Rb1 and Rb2, over against the plural in all other witnesses. The reading of Rb1 can be explained as a result of an abbreviation in N, its exemplar, filled out independently; Rb2 here follows Rb1. These witnesses agree

Trouble in the trees! 

with the Hebrew text in having a singular; the singular in B is another example of independent adaptation to the Hebrew text.81 The results are flawed according to the pairing method (based on similarity scores) which separates the Eastern subgroups from one another (w y x q63 and q2 Eb1 p Eb91 Eb66). In a graph produced according to the quadruple method, this dichotomy has been removed; as the agreement with graphs for the following categories shows, this must be considered an improvement.82 The difference between the two results based on the same data is remarkable in itself, and suggests that the data still suffer from noise. Apart from some floating nodes which are not too problematic, or vague positions where the data are limited (e.g. Eb4, 11%; Eb91 13%; G 16%), there is some difficulty in grouping the Ashkenazi texts, example, a J Q; A,B and the Bomberg group N Rb1 Rb2 Rb3 Rb6 and E K and d. The grouping does not appear to be correct in its entirety in comparison to the images based on the following 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 points out unexpected alliances that usually represent polygenesis. For example, the for in some witnesses which we would not norreading, in 11.21, mally consider to represent a group; this variant renders the text smoother than it was. Hence it appears to be likely that this variant reading was introduced more than once.83 Another example concerns the phrase (7.22), in which four Yemenite mss read , which is more an idiomatic use of language. In the same verse, the geographic indication has been changed into in most witnesses, except for a hotchpotch of eight mss that may well have copied the Hebrew wording in their Aramaic text.84 While the graphs for this category offer a slight improvement in the Eastern grouping, as well as in their cohesion, a few odd positions are noteworthy: q2 (32% formulas) and Eb91 (15%) are dislocated, and so is T (95%). The quadruple method provides a far more reliable result, here and elsewhere, as q2 and Eb91 here join the Eastern group. This reflects the real situation better, since these witnesses always join the main group, and never disagree with p; therefore, the pairing method based on similarity scores, distorts the stemma here. However, the quadruples’ graph does not give a proper picture for Eb1 (33%) and d, the incunable of Leiria; the incunable is linked to the Sefardi witnesses, in contrast to the pictures yielded for the previous and the two following categories.85 Although we know this text was produced in Portugal,

 Willem F. Smelik ˆ34 7

43

31 a

ˆ33

B 11

32

ˆ32

J 24

9 ˆ31

A 12

8 ˆ30

ˆ28

8 ˆ19

ˆ18 6

6

ˆ24 8

5 ˆ5

ˆ14

d

3

7 10 K

1

1 0

C

ˆ23

Rb6

D 2

5 ˆ6

O 3

Rb3

ˆ1

N

10

ˆ9

7 3

15

11

15 S

ˆ13

5

Rb1

ˆ21

ˆ26

Q

14

ˆ25

10

11

ˆ29 14

7 7

ˆ27 12

8

G 4

4

ˆ2

W

M

1

2

q2

Rb2

o 0 Eb91

12

7

E

ˆ22

13

5 ˆ20

P

5

7

ˆ17

F

5

7 Eb1

ˆ16 4

5

ˆ15 4 ˆ10 ˆ7

p

q63

ˆ11 1

1 Eb66

ˆ8 1

2

ˆ4 22

4

2 T

2

3 2

ˆ12 7

ˆ3

w 1

0 @

x 1 y

1 s4

Eb4

Figure 3. Initial tree for Category 4

hence a Sefardi link would not have come as a surprise, the text is closer to the Nürnberg–Bomberg group than to the Sefardi textual tradition. The position of this edition requires further study. The Antwerp Polyglot, based on the exemplar from which M and W derive, was disjoined from the Sefardi group in the pairing tree and linked closer to O and S following the quadruple method.

Trouble in the trees! 

In brief, the graphs seem to confirm that the results in this category are not entirely reliable. . Category 5: Substitution, plus, minus, semantic shift More convincing are the results for the fifth and the sixth category, but the differences between their graphs beg the question how to explain them. The Eastern group of manuscripts and the Sefardi and Ashkenazi subgroups are obvious in each of the graphs, but there is still some lack of clarity in the pairing graph because w y x q63 was too far separated from the remainder, p Eb1 Eb66 Eb91 q2; this was resolved in the quadruple approach. Nonetheless, the graphs are inconsistent in their Eastern subdivisions; whereas the pairing method has two main branches, which it separates, the quadruple approach resulted in one main branch starting with w y x q63 and the remaining Eastern manuscripts as several sprouts. Between the various graphs, some witnesses are floating. The fragmentary character 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. In any case, s4 is attached to the first group according to the quadruple approach, and to the second one according to the pairing method. This picture is different from 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 been created independently in more than one witness for other reasons than the categorical ones are omitted. Substitutions and semantic shift are susceptible to secondary changes, as explained in Section 2.7, and any of these which is likely to have been introduced repeatedly by copyists has been omitted here. The case for omitting minuses as evidence has been made above (2.8); including pluses here obviously implies that I have made a decision in some cases that a reading is unlikely to be authentic. I have omitted all pluses and minuses that could somehow be related to a correction in the Aramaic which is based on the Hebrew original; all other instances were evaluated one by one. It is difficult to argue at this stage of my studies whether the refinement in this category produces an essential improvement over the former one, although there are some indications that it does. Both the pairing method and the quadruple approach present the Eastern manuscripts as one group with two subgroups, w y x q63 and p Eb1 Eb66 q2 s4; apart from the floating manuscript

 Willem F. Smelik ˆ34 43

15

A

29

ˆ33 29

B 17

ˆ32 23

26 ˆ28

ˆ25

ˆ24

ˆ20

6

ˆ22 7

6

3

2 w

M

2 Rb2

W

6

Eb4

7

ˆ15

ˆ4 6

q63

T

5

0 p

ˆ14

2

ˆ9

Rb6

0

P

G

7

ˆ12

ˆ2

N

1

6

11

0

2

Rb3 2

1

ˆ16

ˆ3

5

o

ˆ17

9

5

0

C

ˆ18

ˆ13

S

Rb1

6

ˆ7

d 6

2

ˆ8

ˆ21 8

10

10

O

2

14

ˆ19

6

9 ˆ10

22

ˆ26 7

F

7

6

9

10

K

ˆ23

32

11 ˆ27 8

6

ˆ29

ˆ30

a

Q

14

13

J

12

12

6

ˆ31

4

3 y

ˆ6 3 @

Eb91

E

D

6 ˆ11

x 1

3

ˆ5 4

3

0

Eb1

ˆ1

1

0 q2

s4

1 Eb66

Figure 4. Initial tree for Category 6

Eb91 (16%), there is no real difference between these graphs. The cohesion of the network images has thus been improved. The floater s4 (12%) is now firmly placed within the second group. Eb4 (13%), admittedly, is floating around as it has been doing ‘consistently’. Overall, the picture is clearer than that of the pre-

Trouble in the trees!

vious category, but whether we have resolved the tree or eliminated dissension remains to be argued. A and B are more closely linked together than in the former category, over against a J Q in the same branch of a later, Ashkenazi type of text. This change confirms how influential the presence of data can be, because all the formulas in this category are present in the former as well. This influence may be explained in two ways: either the small amount of data used here distorts the picture, alternatively, it can be argued that the use of unreliable data in the former category has undesirable effects. . Contamination as a characteristic The cardiograms offer one surprising phenomenon: the categories which were deemed most reliable have the most unstable shock waves, while the utterly unreliable first category has less pointed shock waves; compare Figures 1, 5 and 6. To some extent, we should not attribute too much importance to this phenomenon, because a lower number of readings increases the pictorial impact of the waves. However, there is more to this phenomenon than representation alone. The shock waves for Categories 3 and 4 were also more stable than those for Category 5 (and 6). To put these images in perspective, we may briefly consider the highly stable pictures achieved for some traditions and the less stable pictures for others in previous studies.86 There is sufficient reason to assume that the textual tradition of Targum Judges is essentially, perhaps sui generis,87 contaminated; I believe that this degree of contamination bears on the stability of the shock waves. Further study of contaminated and non-contaminated traditions should verify or falsify this assumption. The contamination is not of the successive type, where a copyist (or editor) switched from one to another exemplar at some position in the text. Rather, simultaneous and incidental contamination, where a copyist consults another exemplar either throughout the process of copying, or incidentally, have considerably influenced the text relations.88 The phenomenon of parallelism should be added to this picture: in many positions, a ‘self-evident’ correction may have occured to several scribes who were quite unaware of one another. Such activity results in agreements that do not reflect any real relation between the texts.89 The authoritative status of the Hebrew parent text of the translation is the main reason for this high level of contamination (hereafter used as a shorthand for contamination and parallelism). The Aramaic text was never intended to be used 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

Figure 5. Shock waves for Category 1

and with which it was often transmitted. When the Hebrew and Aramaic alternate by a single verse in a manuscript, it stands to reason that the scribe may still have had a Hebrew word in mind when he was writing the Aramaic text. Subsequent correctors have been quite active, too. We have already had occasion to refer to readings where one group of texts reads A, another B, and a

Trouble in the trees! 

third one simply took them together, A B (see n. 53 above). Such conflations are typical of a tradition that was subjected to continuous gauging under the influence of the original text, variant versions, and the authoritative rabbinic tradition. For that reason this textual tradition is contaminated to the bones; would not the purest file, then, produce unstable (but reliable) relationships, whereas, from a genealogical point of view, fringe elements such as errors and spelling may produce a more stable (and yet less reliable) result? It is perhaps useful to point out that instances of contamination are true reflections of how texts were related; it is the complexity they create that we are wary of, but the readings themselves, and the relations they reflect, should not be dismissed lightly. While accidental agreements are genealogically irrelevant, contamination is not.

. Epilogue To construe an image of the witness relations of a textual tradition, we must not proceed blindly and include either all readings, or a selection that has not been accounted for by philological considerations and empirical analysis. Even though for some text traditions all variant readings seem to produce reliable results, as the high level of consistency which August den Hollander observed in the readings of Dutch Bible translations illustrates,90 it goes without saying that such a consistency cannot be assumed by definition, but must always be demonstrated. Philological insights inform the classification and selection of variants. All variant readings are important, but some are more important than others, and the degree of difference should be worked out in detail for each textual tradition. The Aramaic Bible translation of the Prophets presents its own case. For the genealogy of this particular tradition, there is no watertight case to be made from variations of orthography, vocalization, separation and contraction, and grammatical properties like gender, number, status and morphology; these aspects vary unpredictably within (phases of) the language. Variants, then, are to be expected against the grain of a tree. Codicological arguments account for additional categories. Errors and abbreviations in manuscripts and editions are not unassailable, as the comparison between closely related witnesses has demonstrated; hence, such variations are unreliable source materials for tree construction. The genealogical value of lexemic variations of nouns, adjectives and verbs proved 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

Figure 6. Shock waves for Category 5

lexemic change in the Targum, such changes may easily have been reproduced independently by scribes working in different places and periods. Dialectical developments may well account for a number of semantic shifts between textual witnesses, and so do exegetical considerations; these cases should not be generalized but have to be evaluated on an individual basis.91

Trouble in the trees! 

Determining categories on the basis of general, philological and codicological considerations is one thing, to keep all these premises in check another. The main thrust of this study is the evaluation of all categories of variation in two steps. First, it was suggested that such variants between multiple copies of a single exemplar, or of copies produced from established exemplars, would provide the necessary hard evidence. Needless to add, such information may well be supplemented by, for example, codicological data. Then the categories were tested by a series of computations that were aiming at falsification of our assumptions, not at an optimal tree. The variants between manuscripts and editions of Targum Judges, which are valuable as a first model for the Targum of the Prophets as a whole,92 eventually confirmed the general point that constructing a tree requires rigorous selection of readings. The categories of error and orthography are notorious rogues. Their graphs disagreed with the others and were incoherent. The next two categories of grammatical properties, prepositions, copula and relativa, produce ambiguous results: their trees are quite close to the trees produced for the last two categories, but there is some evidence that they distort known relations between textual witnesses. Their graphs (including those not reproduced here) lack the cohesion of those for the last two categories. Hence, it seems wisest to refrain from using their input in the initial tree construction, while their input should not be disregarded in a case by case evaluation. The last two categories, of semantic shift, substitution, plus and minus, produce the best results in terms of hard evidence and cohesion. Further testing will be necessary to evaluate the differences which occur at this level, especially between the fifth and sixth category; the danger of refining one’s statistical source materials too much should never be underestimated. At this juncture the question should be raised whether the problem of contamination and parallelism has been overcome. This question can not be answered with either ‘yes’ or ‘no’; the procedures followed here usually aim at an optimal tree, rather than a historically correct representation of text relations. Yes, the final picture could be confirmed by some hard evidence, but no, not all details of the picture could be confirmed by lack of evidence. The differences between the graphs for the fifth and sixth category do not warrant overconfidence. A case should be made for the preservation of all readings, since none of the categories analyzed above proved to be absolutely irrelevant. Many of the categories listed above may, in possibly varying degree, have been introduced by scribes independently from each other. This polygenetic nature of these categories renders them unsuitable for general genealogical purposes, but it

 Willem F. Smelik

should be stressed that such readings remain valuable. Corrections of number in nouns and verbs, for example, seem to have flown easily from a scribe’s pen, hence rendering such readings less useful for stemmatological purposes. At the same time, however, most of such readings will not have been changed from one copy to the other. As a result they are useful for the detailed study of subgroups and the correlation of findings with the actual readings themselves. Spelling may reveal the provenance of a witness. Shared errors may reveal intermediacy in some cases. Even abbreviations, as demonstrated, may be of value here. The course of individual readings can be illuminated from the most unexpected perspectives. Moreover, the readings themselves may be authentic. Our images of text relations should be taken as a heuristic device to understand the course of variant readings.93 Parallelism, incidental and simultaneous contamination lead to a variety of historical relations, which these images cannot capture. Only to some extent will peaks in the cardiograms reveal positions, where such crossovers occur. The contaminating readings may, however, preserve the better text, even in a witness that is considered to be of less generic value for the lost original. As a result, the value of trees for individual readings should not be over-estimated; in particular not in a highly contaminated tradition. If trees cannot predict the authenticity of readings,94 it is perhaps less appropriate to use such trees for the selection of manuscripts in a critical edition, as has recently been proposed by Houtman for Targum Isaiah.95 Are we losing relevant readings if we select those manuscripts which are the best representatives of a (sub)group? We should consider the possibility that poor manuscripts may have unique readings to offer, even though, or because they are not the best representatives of their (sub)group. Slightly odd and contaminated manuscripts are very valuable for the readings they preserve! For the Targum of the Prophets, some poor copies, riddled with errors, spring to mind, all of which have some unique readings, but none of which represents the Ashkenazi texts very well.96 Weitzman, in an important study of the Peshitta, the Syriac Bible translation, has recently highlighted the importance of so-called ‘poles’ in a manuscript map;97 this aspect should also be taken into account, especially when the number of complete manuscripts is rather small, as it is here. Of course, such selection procedures are inevitable when the number of manuscripts becomes too large for comprehensive collation; even then, however, these considerations should play a prominent part. Finally, it is interesting briefly to compare the results achieved here with the excellent discussion of characteristics that Salemans has provided, and which has inspired the present investigation.98 Although I have emphasized the

Trouble in the trees! 

phenomenon of hybridization, based on the specific conditions of targumic literature, and have taken a different view of the importance of word order and auxiliary verbs,99 the results appear to confirm his insistence on the use of variant lexemes consisting of nouns and verbs only.100 This is a remarkable result, and shows that philological observations may benefit from taking generalized principles into account, based on a different tradition and a different language. Such principles, however, should always be evaluated by the philologist, because 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 Universiteit Amsterdam, 13 October 2000. I would like to thank Piet van Reenen for his kind invitation to participate in this stimulating workshop. This article is also published in AS 1.2 (London: Continuum Press 2003), 247–287. . See A. Berliner, Targum Onkelos (Berlin: Gorzelanczyk 1884), II, p. 112; P. Churgin, Targum Jonathan to the Prophets (New Haven: Yale University Press, 1907 [= 1927]). . See S. A. Kaufman, ‘A Unique Magic Bowl from Nippur’, JNES 32 (1973), pp. 170–174; C. Müller-Kessler, ‘The Earliest Evidence for Targum Onqelos from Mesopotamia’, JAB 3 (2001), pp. 181–198. . 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: Yale University Press, 2000); M. Jaffee, Torah in the Mouth: Writing and Oral Tradition in Palestinian Judaism 200 bce–400 ce (Oxford: Oxford University Press, 2001); W. F. Smelik, ‘Orality, Manuscript Reproduction, and the Targums’, in A. den Hollander, U. Schmid and W. 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 Bible translations, see W. F. Smelik, ‘The Rabbinic Reception of Early Bible Translations as Holy Writings and Oral Torah’, JAB 1 (1999), pp. 249–272. . See Müller-Kessler, ‘The Earliest Evidence for Targum Onqelos’. . Based on an extensive survey of library catalogues and sporadic examination in situ, I have compiled a comprehensive but eclectic short-list of targumic manuscripts in which the information of the Institute of Microfilmed Hebrew Manuscripts has later been taken into account. I would like to thank B. Richler for his invaluable assistance for providing the essential catalogue card copies. Short-lists for the books of Psalms, Judges, Isaiah and Samuel have been produced, but it is hoped that the entire list will be revised in co-operation with David Kroeze and Dineke Houtman (Kampen, Netherlands). . With the exception of Targum Hosea, no stemma has ever been produced of the extant manuscripts of Targum Jonathan; J. Ferrer i Costa, El targum d’Osees en tradició iemenita (Col.lecció de Tesis Doctorals Microfitxades, 869; Ph.D. dissertation Universitat de Barcelona, 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, ‘Textual Tradition of Targum Jonathan to Isaiah’, in J. Targarona Borrás and A. Sáenz-Badillos (Eds.), Jewish Studies at the Turn of the Twentieth Century (2 vols.; Leiden: E. J. Brill, 1999), I, pp. 145–153. The situation is slightly better for the targums of Job, Lamentations and Canticles; F. J. Fernández Vallina, ‘El Targum de Job’ (unpublished Ph.D. dissertation Universidad de Madrid, 1981); A. van der Heide, The Yemenite Tradition of the Targum of Lamentations: Critical Text and Analysis of the Variant Readings (Leiden: E. J. Brill, 1981); C. Alonso Fontela, ‘El Targum al Cantar de los Cantares (Edición Crítica)’ (unpublished Ph.D. dissertation Universidad de Madrid, 1987); D. M. Stec, The Text of the Targum of Job: Introduction and Critical Edition (AGJU, 20; Leiden: E. J. Brill, 1994); D. Shepherd, ‘Before Bomberg: The Case of the Targum of Job in the Rabbinic Bible and the Solger Codex’, Bib 79 (1998), pp. 360–379. . A. Sperber, The Bible in Aramaic (5 vols.; Leiden: E. J. Brill, 1959–1973). . Followed by the Yemenite mss, and finally the Western witnesses. I hope to elaborate on these findings, which I presented at the third meeting of the International Organisation for Targum Studies (IOTS), Oslo, July 1998. An eventual pedigree of the available textual evidence will be based on almost half of the book of Judges, evenly distributed over beginning, middle and end (1.1–4.1, 6.28–8.32, 10.10–14.8, 17.12–18.24, and 20.47–21.25; 307 out of the 618 verses, or 49,7%, made up by the book of Judges), in correspondence with accepted practice. The support for ms Or. 2210-11, hereafter ‘p’, should be explored in greater detail and set in relief with the Western textual evidence. A brief additional exploration of the evidence in the books of Joshua to Kings suggested the possibility that some Yemenite manuscripts, other than Sperber’s 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 jüdisch-palästinischen Aramäisch (Darmstadt: Wissenschaftliche Buchgesellschaft, 1905 [repr. 1905]), p. xvi. . B. J. P. Salemans, Building Stemmas with the Computer in a Cladistic, Neo-Lachmannian, Way: The Case of Fourteen Text Versions of Lanseloet van Denemarken (Nijmegen: Nijmegen University Press, 2000); see also Salemans, ‘Cladistics or the Resurrection of the Method of Lachmann’, in P. van Reenen and M. J. P. van Mulken (Eds.), Studies in Stemmatology (Amsterdam: John Benjamins, 1996), pp. 3–70; K.-H. Uthemann, ‘Which Variants are Useful in Discovering the Deep Structure of the Manuscript Tradition of a Text? Contra a So-called Essentially Quantitative Approach’, pp. 249–261 of the same volume. . For example, variations of word order, or omission of clauses and verses, are unreliable in most targumic texts as a result of the usually parallel transmitted Hebrew parent text which would alert many scribes and correctors to conspicuous differences, and allow them to revert these changes. Thus the fifth text-genealogical rule formulated by Salemans, Building Stemmas, pp. 81–85, which attributes genealogical information to differences of word order, cannot be applied to the targums without modification, because in this tradition the rule interferes with his first principle, that a variant ‘fits. . . inconspicuously in a text version’ (p. 64). The modification does not falsify Salemans’ principles. For a fuller discussion, see below, under ‘Substitution and Semantic Shift’ and ‘Minus and Plus’, Sections 2.7 and 2.8 respectively.

Trouble in the trees!  . See M. Beit-Arié, ‘Transmission of Texts by Scribes and Copyists: Unconscious and Critical Interferences’, in P. S. Alexander and A. Samely (Eds.), Artefact and Text: The Re-Creation of Jewish Literature in Medieval Hebrew Manuscripts. Proceedings of a Conference held in the University of Manchester 28–30 April 1992 (=BJRL 75/3 [1993]; Manchester: John Rylands University Library, 1994), pp. 33–51. . See Smelik, ‘Orality, Manuscript Reproduction’, pp. 76–80. . See W. F. Smelik, ‘How to Grow a Tree: Computerised Stemmatology and Variant Selection in Targum Studies’, in J. Cook (Ed.), Bible and Computer, AIBI-6 (Leiden: E. J. Brill, 2002), pp. 613–644. ‘Empirical’ is a goal rather than an achieved result in the strict sense of the word, because of the inevitability of some working assumptions about the texts, and the lack of hard data. . Houtman, ‘Textual Tradition’, p. 148, does not mention any criteria beyond the standard exclusion of spelling and scribal error, nor does she make mention of the number and type of readings on which her study is based. . These two fundamental aspects of stemmatology, tree construction and shock waves, will be explained below, Section 3.0. . For a distinction between various levels, cf. P. van Reenen and L. Schøsler, ‘From Variant to Pedigree in the Charroi de Nîmes: A Typology of Variants’, in Van Reenen and Van Mulken, Studies in Stemmatology, pp. 263–304. . The copy of the Prophets and the Writings he produced on behalf of the University of Salamanca, earned him the sum of 12 ducats as recorded in the archives of the university (see F. Marcos Rodríguez, ‘Los manuscritos pretridentinos hispanos de ciencias sagradas en la Biblioteca Universitaria de Salamanca’, Repertorio de Historia de las Ciencias Eclesisticas en Espaa 2 [1971], pp. 261–507). De Zamora was baptised in 1506, after he had begun working for Cisneros, but the codicological aspects of his manuscripts show he was a trained scribe in the Jewish-Sefardi tradition (so that the inclusion of his manuscript in evaluating the process of manuscript reproduction in the Jewish tradition appears to be valid). Christian patronage is revealed only by certain peculiarities: the manuscripts are to be read from left to right, following the direction of the parallel, literal Latin translation, and occasionally Aramaic words are hyphenated which is rare in Jewish Hebrew manuscripts. Comparison with other Sefardi manuscripts shows that De Zamora was faithful to his exemplar, but to facilitate singling out non-literal additions to the base text translation, he sometimes inserted the marker ‘Tosefta’ in the running text. Whether both mss were copied from this exemplar, or one was copied from the other, is uncertain; see M. Taradach and J. Ferrer, Un targum de Qohéleth: Ms. M-2 de Salamanca. Editio princeps. Texte araméen, traduction et commentaire critique (Le monde de la Bible, 37; Genève: Labor et Fides, 1998), who argue that the differences point to a common Vorlage, rather than one being copied from the other. This unqualified assessment prompts the question of how many errors and accidental (?) differences should exist between two manuscripts to disconnect the umbilical cord between them. The intermediacy of manuscripts should be argued along other ways, in particular by examining the presence or absence of unique readings (type-1 variation; see M. van Mulken, ‘The Manuscript Tradition of the Perceval of Chrétrien de Troyes: A Stemmatological and Dialectological Approach’, unpublished Ph.D. dissertation, Amsterdam 1993, p. 50; Salemans, Building Stemmas, pp. 25–27, 155–156; A.

 Willem F. Smelik

den Hollander, De Nederlandse Bijbelvertalingen 1522–1545 [Nieuwkoop: De Graaf, 1997], p. 152 n. 74). One should note that there is not much difference for our purposes whether both manuscripts are apographs of an exemplar now lost, or whether one has been copied from the other. . The work for this Polyglot was launched by Cardinal Francisco Jiménez de Cisneros, the archbishop of Toledo, in 1502 and led to publication in 1518, although it went into circulation only after Pope Leo X’s approval in 1520. In 1504 Alfonso de Zamora (c. 1474–1544) was hired to produce a Latin translation of the Targums of the Latter Prophets and the Writings. Due to vehement opposition, among others by the new archbishop of Toledo, Juan de Tavera, neither the Targums nor their Latin translations were ever to be included in the Polyglot, with the exception of Targum Onqelos. Fortunately, however, most of the manuscripts were to be preserved in the biblioteca de San Ildefonse in Alcalá for consultation. Later, they would 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. 146– 224; J. S. Penkower, ‘The Chapter Divisions in the 1525 Rabbinic Bible’, VT 48 (1998), pp. 350–374; idem, ‘Verse Divisions in the Hebrew Bible’, VT 50 (2000), pp. 379–393 (383–384). . This conclusion was presented at the sixth meeting of the Association Internationale Biblique et Informatique, Stellenbosch, July 2000; to be published in a forthcoming study, ‘Targum Judges in the Great Rabbinic Bibles’. . Some pluses and minuses are errors, other variant readings combine several characteristics. . See esp. Salemans, Building Stemmas. . Smelik, ‘How to Grow a Tree’. . The category of scribal errors can easily be confused with the concept of ‘error’ in some stemmatological models, in which ‘error’ represents a non-original reading. . In Targum Judges 2.1, 14; 11.18; 12.4. .

in 2.10;

in 2.15. There are also some minuses shared by other mss.

. See the first text-genealogical rule as formulated by Salemans, Building Stemmas, pp. 64–71. . For phonetic errors, see Smelik, ‘How to Grow a Tree’; for Hebraisms, see for example in N and Rb1, corrected into in Rb2. the Hebraism in 8.28, . See Smelik, ‘How to Grow a Tree’; B. Narkiss, ‘The Relation between the Author, Scribe, Massorator and Illustrator in Medieval Manuscripts’, in J. Glénisson and C. Sirat (Eds.), La paléographie hébraïque médiévale (Colloques internationaux du CNRS, 547; Paris: Éditions du Centre national de la recherche scientifique, 1974), pp. 79–86; M. Beit-Arié, ‘The Worms Mah.zor – MS Jerusalem, Jewish National and University Library Heb. 40 781/1: Würzburg? (Germany), 1272’, in idem, The Making of the Medieval Hebrew Book: Studies in Palaeography and Codicology (Jerusalem: Magnes Press, 1993), pp. 152–180 (162). . For an example, see below, Section 3.3.

Trouble in the trees!  . A frequent phenomenon concerns the expression of the genitive by a construct chain or by the intermediate use of the particle . Changes from the emphatic to the absolute state or vice versa occur frequently as well. . See G. Dalman, Grammatik des jüdisch-palästinischen Aramäisch (Darmstadt: Wissenschaftliche Buchgesellschaft, 1960), p. 191 (§38.5). . 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 error here). . 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 absolute; c 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 of person should sometimes be glossed over (a step which should not be obscured for obvious reasons), when variant readings belong to two different categories. In 11.24, for example, ; one ms reads , another . Obviously, the all witnesses except for two mss read variant readings attest to the same verb, over against all other witnesses, but they do not reflect the same person. . See W. F. Smelik, ‘Translation and Commentary in One: The Interplay of Pluses and Substitutions in the Targum of the Prophets’, JSJ 29 (1998), pp. 245–260; idem, ‘Concordance and Consistency: Translation Studies and Targum Jonathan’, JJS 49 (1998), pp. 286–305. . The distinction between adjectives and nouns is often difficult to draw. Cf. Salemans, Building Stemmas, pp. 85–89 (87 n. 71). occurs in some Hebrew mss, and it is supported by the Peshit.ta, . In 2.9 the reading Vulgate and in Targum by mss A M O S E D B Wmg (as well as the Antwerp Polyglot); the . other mss support . Contrast, however, E. van Staalduine-Sulman, The Targum of Samuel (Leiden: E. J. Brill, 2002), pp. 158–159. See also C. A. Dray, “Is Subtlety in Translation the Reason for the Targumic Use of various Verbs of Fleeing”, AS 2 (2004), pp. 25–35. . M. O. Wise, ‘Accidents and Accidence: A Scribal View of Linguistic Dating of the Aramaic Scrolls from Qumran’, in T. Muraoka (Ed.), Studies in Qumran Aramaic (AbrNS, 3; Leuven: Peeters, 1992), pp. 124–167. For a stemmatological perspective, see Salemans, Building Stemmas, pp. 70, 236. or vice versa is not related to dialectical processes; nonetheless, . The change of to this type of variation appears to be rather unreliable. . In this connection it should be recalled that, according to Uthemann, intentional changes should not be taken into account when studying the genealogy of a manuscript tradition; Uthemann, ‘Which Variants are Useful?’, p. 257. Intentions may have been operative independent from the exemplar, and therefore may have influenced several scribes independently of each other. However, it is possible that the terminology obscures more than it reveals. Whatever 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 witnesses without the confusion of analogue, but independent, developments, and hybridization. That is exactly what Uthemann strives to achieve, so that the confusion hinted at here may derive from different labels rather than fundamentally opposed approaches. . W. F. Smelik, Targum of Judges (OTS, 36; Leiden: E. J. Brill, 1995), pp. 594–596. . For a definition of the various types of ‘contamination’, see E. Wattel and M. J. P. van Mulken, ‘Shock Waves in Text Traditions’, in P. van Reenen and M. J. P. van Mulken (Eds.), Studies in Stemmatology (Amsterdam: John Benjamins, 1996), pp. 105–121 (105–106). . Such errors may, when shared by manuscripts or editions, point to a common ancestor that introduced the error, but they are, generally speaking, not a reliable indication of genealogical relationship. . L. Díez Merino, ‘Fidelity and Editorial Work in the Complutensian Targum Tradition’, in J. A. Emerton (Ed.), Congress Volume, Leuven 1989 (VTSup, 43; Leiden: E. J. Brill, 1991), pp. 360–382. . The Salamanca ms contains more uncorrected minuses that are apparently unintentional than the Madrid ms. In W, eight uncorrected minuses occur, in 1.1, 17, 25; 3.7, 16, 25; 6.32; 8.23; in M, only three, in 3.21; 6.34; 18.1. . The following examples reflect the combination of two alternative translations: in 6.35 in mss A J; in 8.10 the reading in ms B; in 8.18 the reading the reading in ms A; in 12.9 the reading attested by mss N Rb1 Rb2; in 11.22 the reading in the Leiria edition (d). See also Smelik, ‘Orality, Manuscript Reproduction’, p. 77. . There are of course double translations which seem to be original. . For example, in 2.19 the Hebrew does not qualify the Israelite ‘practices’ as ‘evil’, as does the Targum with the plus ; the absence of this plus in some mss (T Q) and Rb1 may simply reflect an adjustment toward mt. The interpretation of geographic indications, for in example, has sometimes been supplemented with a transliteration; see, for instance, Rb2 and edition d in 7.22. This also applies to minuses; the reading for in 7.24 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 Rb3 Rb6. . In mss x y w q63; this is one of the occasions, where Sperber’s base manuscript may be more reliable than the other Yemenite manuscripts he used. This reading may first have been in the margin. an alternative one, noted as . In mss Eb66 A d J. Theologically motivated changes are often more obvious. A marked example is to be found in 11.23-24 in a ms in Jena, where a neutral statement about a non-Israelite deity was deemed to be improper, and has been recast into the standard denunciation of foreign deities. . See, for example, the plus

in 14.2, reflecting Hebrew .

. See Smelik, Targum of Judges, p. 643 n. 15. . An analysis of contamination did not show a remarkable indication for pollution at this point. The shock waves at this juncture were not marked by peaks.

Trouble in the trees!  . The standard translation is , although some witnesses do occasionally read throughout Targum Jonathan to the Prophets. Even Rb2 usually has the first reading; it deviates in Josh. 5.12; Judg. 6.38 and 21.4; contrast 1 Sam. 5.3, 4; 11.11; 18.10; 20.27; 30.17; 31.8; 2 Kgs. 8.15; Jer. 20.3; Jon. 4.7. . Although G has the same preposition as Rb2, it is prefixed to a different noun, a discussion of the translation, see Smelik, Targum of Judges, pp. 544–546.

. For

. 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 is erroneous 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 is a 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 K Q N Rb2. . The following variations occur: 8.16, N Rb1 / Rb2; 10.10, N Rb1 / Rb2 ; 11.9, N Rb1 / Rb2 ; 12.2, N Rb1 / Rb2 ; 12.4, N Rb1 / Rb2 ; 13.2, N Rb1 / Rb2 ; 13.14, N Rb1 / Rb2. . Salemans, Building Stemmas, pp. 70, 252–256; characteristic 4b. . See Smelik, ‘Targum of Judges’, Ch. 3. . Salemans, Building Stemmas, characteristics 8 and 10. . Different positions apply to the insertion of the marker , a paratextual element, marking a part of the translation that has been added later on, or an alternative translation added by the scribe, editor or glossator. These Toseftot themselves yield interesting information, as their position in the running text, or margin, differs, as does their contents. On the allocation of glosses, see Smelik, Targum of Judges, pp. 162–179. . E. Wattel and M. J. P. van Mulken, ‘Weighted Formal Support of a Pedigree’, in Van Reenen and Van Mulken (Eds.), Studies in Stemmatology, pp. 135–167; idem, ‘Shock Waves in Text Traditions’, pp. 105–121; E. Wattel, ‘Clustering in Stemmatological Trees: How to Handle a Large Number of Versions’, in Van Reenen and Van Mulken (Eds.), Studies in Stemmatology, pp. 123–134; Den Hollander, ‘How Shock Waves Revealed Successive Contamination’, pp. 1–2. . A total of 12 graphs per category has been produced by Evert Wattel, and all of these have been taken into account. Wherever percentages are provided, the witnesses involved are lacunary or even fragmentary, and the percentage indicates the number of formulas in which 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, Rb6, are convincing.

 Willem F. Smelik . Due to a technical error, the readings of ‘o’ have been largely omitted from this category (only 6% included). nonetheless, this was enough for the quadruple method which grouped it correctly together with W and M. . The plural appears to be more original, as it reflects the collective meaning of the Hebrew; this is based on the considerations that this reading (a) has a greater distance to mt while (b) it better reflects the translation strategies of the targum. Another peak appears to reflect a correct split of the Sefardi mss from the remaining witnesses. That it results in a peak is not incongruous with the theory. . It should be noted that in the fifth category this Yemenite subgroup was still separated from 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 have difficulties in positing K; according to the quadruple method, K is close to the Nürnberg – Bomberg group. . Contrast the shock waves in A. den Hollander, ‘How Shock Waves Revealed Successive Contamination: A Cardiogram of Early Sixteenth-Century Printed Dutch Bibles’, forthcoming, with Wattel and Van Mulken, ‘Shock Waves’, p. 119. For brevity of argument, the shock waves for categories three and four have been omitted. . For more details, see n. 3 above. . See now Den Hollander, ‘How Shock Waves Revealed Successive Contamination’, pp. 1–2. . See esp. Salemans, Building Stemmas, pp. 67–71. . Den Hollander, Nederlandse Bijbelvertalingen 1522–1545, p. 136. . Of course, taking similar changes into account; ‘individual’ here does not mean ‘atomistic’. . Onqelos may carry its own characteristics in view of its even more careful editing, the Palestinian versions, the higher number of texts, and the inclusion of the whole text in the liturgy. . Cf. M. P. Weitzman, The Syriac Version of the Old Testament: An Introduction (Cambridge: Cambridge University Press, 1999), pp. 316–317. . To some extent, the focus on the original reading in critical editions should be called into question. Why should later readings be less interesting? . Houtman, ‘Textual Tradition’. . ms 11 in Göttweig, Austria (A); ms El.f.6 in Jena, Germany (J); ms Or. Fol. 1–4 in Berlin (B); ms 26879 in London (a). For more details on A, see Smelik, ‘Orality, Manuscript Reproduction’. . Weitzman, The Syriac Version, pp. 316–322. . Salemans, Building Stemmas, passim.

Trouble in the trees!  . See n. 12 and Section 2.10 above. . Thus excluding morphological, inflectional and orthographical variants of verb and nouns (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 B D J K N Q C A G F P T

Add. 26879, British Library, London Or. Fol. 1–4, Staatsbibliothek, Berlin parm. 3188, Biblioteca Palatina, Parma El.f.6, Universitätsbibliothek, Jena Reuchlin 3, Karlsruhe Cod. Solger 3.20 , Stadtbib. Nürnberg hébreu 18, Bibliothèque Nationale, Paris BH III, Biblioteca Civica Berio, Genoa 11, Stift, Göttweig Heb. A 10, Kaufman Coll., Budapest Urbinates–Vaticani 1, Vatican Laud. Or. 326, Bodleian, Oxford Or. 72, Biblioteca Angelica, Roma

Ash. Ash. Ash. Ash. Ash. (?) Ash. Ash. Ash. Ash. Ash. Ash. Ash. Ash.

complete complete complete complete complete complete complete complete complete haft.arah complete complete complete

E

hébreu 75, Bibliothèque Nationale, Paris

Or.

complete

W M O S

1, Biblioteca Antigua, Salamanca 7542, Biblioteca Nacional, Madrid Opp. Add. 40 75, Bodleian, Oxford Kennicott 5 (2329), Bodleian, Oxford

Sef. Sef. Sef. Sef.

complete complete complete complete

p y w x q63/2 q2 s4

Or. 2210, British Library, London Or. 2371, British Library, London Or. 1471, British Library, London Or. qu 578, Staatsbibliothek, Berlin 63/2, R. Y. Kapah., Jerusalem 2, R. Y. Kapah., Jerusalem Sassoon 1154

Yem. Yem. Yem. Yem. Yem. Yem. Yem.

complete complete complete complete complete incomplete haft.arah

Eb1 Eb4 Eb66 Eb91

229, JTS library, New York 505, JTS library, New York Cambridge, UK & JTS, New York H . olon, Y. L. Nah.um

Bab. Bab. Bab. Bab.

incomplete incomplete nearly complete incomplete

 Willem F. Smelik

Table 8. (continued) siglum

description

provenance type

Editions (chronological order) d Rb1 Rb2 Rb3 Rb4 o Rb5 Rb6

Former Prophets, Leiria 1494 1st Rabb. Bible, Bomberg, 1516–1517 2nd Rabb. Bible, Bomberg, 1524–1525 3rd Rabb. Bible, Bomberg, 1548 4th Rabb. Bible, Bomberg, 1568 Antwerp Polyglot, 1569–1573 5th Rabb. Bible, Bragadin, 1617–1619 6th Rabb. Bible, König, 1618–1619

Sef. Ash. Ash. Ash. Ash. ‘Sef.’ Ash. Ash.

complete complete complete complete complete complete complete complete

Ash. = Ashkenazi; Bab. = Babylonian; Or. = Oriental; Sef. = Sefardi; Yem. = Yemenite

Scribal variations When are they genealogically relevant – and when are they to be considered as instances of ‘mouvance’?1 Lene Schøsler University of Copenhagen

. Introduction The intention of this paper is to look for linguistic criteria for a distinction between genealogically relevant and genealogically irrelevant scribal variations. I have tried to do so before, when working on the mss. of Narcisse and of the Charroi de Nîmes (Schøsler 1988, 1989; Van Reenen & Schøsler 1996). I then proposed to distinguish three, later four, levels of linguistic variation (see Van Reenen & Schøsler 1996, Section 3: Local variants): 1. 2. 3. 4.

differences in spelling and phonology, differences in morphology and syntax, content differences in related passages, and content differences resulting in unrelated passages.

The genealogical relevance of each level differs: level 1 may present dialectally, hence possibly genealogically, relevant information about rhyme and assonance (1996: 279). Like level 1, level 2 may contribute to the identification of the dialect of both the exemplar and a copy, and may thus be genealogically relevant (1996: 279). Level 3 may offer dialectally determined lexical variation, i.e. possibly genealogically relevant information. More importantly, it may permit the establishment of the so-called “type-2 oppositions”, which are genealogically relevant lexical variations opposing at least two mss. against two other mss. in exactly two groups,2 e.g. AB versus CD (1996: 280). A distinction is made between passages showing at least some resemblance (level 3), and pas-

 Lene Schøsler

sages which are completely independent (level 4). At level 3 and 4, we may find important, genealogically relevant “type-2 oppositions” (1996: 281). I have realised – as have most of my colleagues working in the field of stemmatology – that it is extremely difficult to differentiate between, on the one hand, variations revealing the absence of a well-established linguistic norm or literary authorized version and, on the other hand, variations revealing different manuscript traditions. Until recently, I did not see any way to cast more light on this problem. A possible way to a better understanding of medieval variation could be to study two copies of one exemplar made by the same scribe, but such cases are rare. However, they do exist, and Keith Busby has recently discovered an interesting one (1993a, 1993b): mss. T and V of Chrétien de Troye’s Perceval. I find his argumentation absolutely convincing, and I accept his results concerning the identity of the hand of the mss. T and V of Perceval. According to Busby (1993a: 54), it is not possible to show that T is a copy of V or vice versa; it is more likely that they are copies of the same exemplar.3 In the following, I will examine these two copies as illustrations of what I will assume to be genealogically irrelevant variations. In Section 1 I shall study the variations between the two mss. at each level, as mentioned above. In Section 2 I shall compare the types of variation with those found for comparable mss. among the nine mss. of the Charroi de Nîmes. None of these mss. were copied by the same hand. In Section 3 I will conclude the investigation by considering whether the study of the Perceval copies can lead us to a further understanding of which types of variation are genealogically relevant and which are not. The two Perceval mss. are from the thirteenth century, and their geographical provenance is Northern, possibly from Oise. For more details on the two mss., see Busby (1993a & 1993b) and Van Mulken (1993).

.

Variations between mss. T and V

The two mss. have been studied on microfilm, the quality of which is sometimes rather bad. Therefore, it has not been possible for me to make a complete inventory 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 two mss. are indicated in Busby (1993a: 51); they amount to 3098 verses. Busby counts over 350 points of difference, with the exclusion of purely orthographical ones: on average one every 8.75 lines (Busby 1993a: 55).

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 reveal 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. The level of dialectal adaptation is almost the same in the two mss. – as should be expected for the same scribe: T has a dialectal coefficient of 68, V has a dialectal coefficient of 64.56, the maximum being 100.4 Even so, spelling predilections differ from one ms. to the other, as already observed by Van Mulken: T prefers the spellings vos, nos, totes; V prefers vous, nous, toutes (Van Mulken 1993: 219). Additional examples are: T: tor (5901), Escalibor (5902); V: tour (5901), Escalibour (5902).

Most frequently, the spelling is identical: doute:toute (5931-2). In the following section, I will examine some different spelling variations. The dialectally relevant distinction, or non-distinction, between an et en is discussed in Van Mulken (1993). We find hesitation in both mss.: T: anqui (6049), espans:tans (6261-2), desfendre:prendre (6745-6), descent:bauchent (7261-2), V: encui (6049), espens:tens (6261-2), desfendre:prandre (6745-6), descent:bauchant (7261-2).

The northern spelling ch, corresponding to the central spelling c, seems more frequent in T than in V, the mss. showing both the central and the northern spelling: T: che que (5933, 5943 ...), por che (5976), chité (6145), comenche (6216), atache (6340), charme (6340), V: ce que (5933, 5943 ...), por ce (5976), cité (6145), comence (6216), attace (6340), carme (6340).

The mss. hesitate between the spellings s and z. I get the impression that V has a predilection for s: T: mandez:demandez (5957-8), faz:solas (5971-2), avez (5978), V: mandes:demandez (5957-8), fas:solas (5971-2), aves (5978).

A frequent spelling variation without any phonological basis is that between u and l. I have the impression that T prefers the vowel whereas V prefers the consonant:

 Lene Schøsler

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 (1993a: 56), which implies variations of the type qu’ – que, e.g. 6607: T: jusqu’anuit – V: jusque anuit. Spelling variations between T and V are very frequent. We know that there is no important temporal difference between the two copies.5 The ‘input’ dialect being common to both copies, and the ‘output’ dialects being common as well, the numerous spelling variations reflect the absence of a norm – even of a strict personal norm. What is permanent about our copyist is that he keeps the same attitude towards his act of copying (see Schøsler 1995: 246): he is not mirror-copying in one ms. and freely adapting in the other. In both mss. he respects the original rhyme and adapts the spelling inside the verse to his own dialectal habits. Compare Van Mulken (1993: 168): The behaviour of manuscripts t and v once again illustrates the kind of scribal conservatism with regard to rhyme, known as diglossia, which we encountered in manuscript u for OR words: in rhyme position, manuscripts t and v occasionally use the ‘ou’ spelling, though there are no linguistic circumstances which would have obliged the scribes to modify their usual spelling. These partial mirror-copyists, too, seem to have preserved the spellings of their models in rhyme position, whereas they felt free to alter the spelling within the rest of the text.”

. Variations at the level of morphology and syntax (level 2) I will first present paradigmatic variation and, afterwards, discuss syntactic variation. The best known dialectally relevant morphological variations are variations in the declension system, differing forms of gender marking, and variation in the pronominal forms. Conjugation variations are also found. Our two mss. show variations in all these cases:

Declension Both Chrétien’s champenois dialect and the northern dialect of the 13th century are conservative with respect to declension. Accordingly, we find a regular system with additional -s-marking in the singular nominative forms of some feminine and masculine nouns, although this is more frequent and more consistent in feminine than in masculine nouns. A few examples:

Scribal variations

MASC maistre(s) (6072), traitre(s) (7559); but not, although it would be possible, in the following masculine nouns: sire, hom, prestre, frere (see e.g. 6302, 6305, 6321, 6415, 6454, 6804, 9113 ...) FEM raisons (7182, 8020, 8878), mer(s) (7590), cor(s) (8897), ...

Gender In the northern dialects, the feminine singular definite article may take the form le instead of the central form la. Throughout the two mss. the two forms alternate, with la as the dominant form: T: la espee (5903), la traïson (6095), le cort (6107), le color (6138), le pointe (6376), la crois (6496), la voie (6620), la palme (7020) ... V: le espee (5903), le traïson (6095), la cort (6107), le color (6138), le pointe (6376), le crois (6496), le voie (6620), le palme (7020) ....

Compare also Busby (1993a: 58). 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 in verse 8330: tele.

Pronominal paradigms There are a few differing forms, but they are too infrequent to permit any conclusions. Alternations are found among personal pronouns: T: jou (6492), on (6152); V: je (6492), l’en (6152); possessive pronouns: T: soe – V: soie (6416); relative pronouns: T: qui – V: cui (6415, 6951), T: celui a cui – V: celui cui (8938). Conjugation Besides a few straightforward copying errors, such as T: mirois (2nd person singular) instead of V: miroit (3rd person singular) (6678), we find only few cases of differing forms like T: averiez – V: avriez; T: feïsse – V: fesise (8358); T: estoit – V: ert (9146); T: aiue – V: aide (6466); T: lairai – V: laisserai (8418). If morphological variation is rather rare, syntactic variation – especially variation in word order – is less infrequent, as will be shown below. Declension As mentioned above, the declension system of both ‘input’ and ‘output’ dialects is conservative. We find very few ‘errors’ in the use of the nominative forms, with one frequent exception: terms of address are generally found in the accusative, instead of the expected nominative forms (for declension of terms of address, see Schøsler 1984). Examples are: vassal (T & V: 6880, 7014,



 Lene Schøsler

7030 ...), chevalier (T & V: 7283, 8421 in the rhyme ...), Gauvain (T & V: 6140, V: 7094, 8902). There are a few unexpected accusative forms instead of nominative forms in the subject function: fief (V: 7393), chevalier (T: 9012); and an accusative form of the subject complement (T & V: 7013, in rhyme position, V: 7394), especially following the verb sembler: 7188, 7324 (in the rhyme). There is one strange nominative form instead of the expected accusative form in 7337-8: ... que il l’en ront li senestres tot en ront (corrected into the accusative form by Roach and by Busby in their editions: le senestre).

Tense A well known feature of older literature is the alternation of narrative tenses, especially between the historic present and the preterite (see Schøsler 1973, 1994). Busby (1993a: 58) mentions several cases, e.g.: sai – sot (6035); vint – vient (6036 ...); fu – est (6260 ...); fist – fait (9229). There are a few other unpredictable cases of variation, e.g.: T: covenoit – V: convient (6026); (7984) poez – porrez; (9138) a – ara; and also one case of change of person, i.e. second sing. – first plural: (8213) mejerois – mengerons. We have seen syntactic phenomena here which show a certain degree of variation. On the other hand, it is also highly interesting to find syntactic phenomena that are stable. Historians of Old French have often looked in vain for factors and parameters determining certain variations. Curiously enough, what is often considered as inexplicable variation or as instability due to ongoing linguistic changes (like the use of the determiners, the position of the adjectives, the use of the pronominal subject, the choice of person in address, etc.) seems to be stable for our copyist. I will first consider the structure of the Noun Phrase, and afterwards the Verb Phrase. Noun Phrase The use or omission of determiners is largely stable (except for the pair on:l’en), with some alternation between the possessive and the definite article, e.g. (6197): T: tote sa paine – V: toute la paine. The position of the adjective does not show variation, both mss. have either anteposition, as in 8083: sa lie chiere, or postposition, as in 6530: un palefroi noiret petit. There is no blurring of the distinction of the two demonstrative paradigms: cil versus cist.

Scribal variations 

Verb Phrase The expression or omission of a pronominal subject is largely stable in the two mss., with some fluctuation concerning the type of pronoun: personal, relative or demonstrative (see Busby 1993a: 57): qu’il – qui (7449 ...), il – cil and cele – ele (168, 8446), etc. There is hardly any variation, in the form of address, between the 2nd person singular and plural, which have, however, been observed to fluctuate in an intricate way within narrative texts; see Foulet (1967: 198ff.). The only change I have found is clearly provoked by a change of the rhyme: (7419-20) T: en vos fier:garder – V: en toi:foi.6 The use of the subjunctive is stable; I have found only four cases of difference between T and V: 6041: T: ert (imperfect, indicative) – V: fust (imperfect, subjunctive) and 6921: T: soit (present, subjunctive) – V: est (present, indicative); (8423) T: avez (present, indicative) – V: eüssiez (imperfect, subjunctive); (8457) T: deüst estre (imperfect, subjunctive) – V: devoit estre (imperfect, indicative). The first two types of context are known to show fluctuation of mood, as the governing verb is one expressing uncertainty, e.g. 6040-1: De che que mesire Gavains ert/fust el chastel ne savoit mot; 6920-1: Or quit je que cis chevaliers soit/est mors ...; see Foulet (1967: 208). The last two examples show the well known fluctuation of mood in connection with hypothesis. Concord of the past participle with the direct object of a compound tense is a difficult matter in Old French: it is hard to understand the fluctuation in concord versus non concord. However, our scribe appears to know some sort of a system, because he is largely consistent about it; see e.g. 5957, 6242. I have found hardly any variations in the valency patterns of the verbs. In one case (6198) there is a difference in the preposition introducing a prepositional 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 infinitive clause as subject: Que ne seroit pas vostre biens T: de / V: a demorer en cest rivage. Fluctuation in the use of prepositions introducing infinitive clauses is in fact not infrequent; see Van Reenen and Schøsler (1993). In Medieval French, word order is largely free. One might therefore expect variation in word order, such as (S)OV – (S)VO – OV(S). In our two mss. variation in word order is nevertheless limited and mainly concerns adverbial phrases and pronouns (see, however, v. 6560, below). This could be due to the limitations of the octosyllabic rhyming verse. At subsequent levels of variation, i.e. levels 3 and 4, though, these limitations do not seem to play any role, so influence from the metre should not be overestimated. A few typical cases show the types of variation found:7

 Lene Schøsler (5970) T: Et tenez vos m’ent a vilaine – V: Et tenez m’ent vos a vilaine. V S C V C S (6146) T: Se destorner vos en pleüst – V: Se vos destorner en pleüst. Inf C C Inf (6298) T: Doivent estre hui en peneance – V: Doivent hui estre en penitance. Inf Adv Adv Inf (6560) T: Quant mesire G. vint la – V: Mesire G. quant vint la. Conj S V S Conj V (7404) T: Que je le cheval n’en eüsse – V: Le cheval que je ne l’eüsse. Conj S O O Conj S (8028) T: Que sachiez bien je ne porroie – V: Que bien sachiez je ne porroie. V Adv Adv V

Compared to the frequency of spelling variation, morphological and syntactic variation is limited. This presents a contrast with with the state of affairs at level 3, 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) concern variations at the level of content, i.e. lexical variation, in related passages. Our mss. offer a precious source for identification of synonyms or near synonyms of Old French. We find synonyms for words belonging to all word classes: nouns, adjectives, pronouns, articles, conjunctions, prepositions, verbs and adverbs. First, I shall quote some illustrative examples of what I consider to be synonyms. Afterwards, I shall proceed to near synonyms and words or expressions having related meanings. Finally, I shall consider ellipses and variations concerning more than one word. I first give the form found in T, then the one found in V. .. Synonyms Nouns: (6298) peneance – penitance; (7144) ambleüre – aleüre; (7372) nacele – bargele; (8868) onor – grant los; (9136, 9145) chaceor – coreor;

Adjectives: (8464) trestot bien – tout le bien;

Pronouns: See Section 1.2 above; (6086) il – cil; (7346) il – on;

Scribal variations 

Articles: See Section 1.2 above; (6197) tote sa paine – toute la paine;

Conjunctions: (5988) et – mais, (6521) ou – quant; (6609) que – car; (7095) que que – coi que; (8401) mais que – fors que; (8910) ains que – ançois que;

Prepositions: (6036) venir el – al chastel; (6081) venir a – en la tor; (7069) desor – desus; (7266) en le – ens le;

Verbs: (5999) voist avant un pas – face avant un pas; (6627) alever – eslever; (7166) n’atoche – ne toche; (9160) lasser – pener;

Adverbs: (5946) onques – ainc; (5964) pas – mie; (6204) hors – fors; (6493, 6767...) issi – ensi; (6788) neporoec – neporquant; (7537) molt – tant; (8077) tant – molt; (8215) amont – çasus; (8462) buen – buer; (8480) si hautes – molt hautes.

My list is not exhaustive, but I believe the relative frequency of synonyms to be representative. Particles appear to be more easily interchangeable synonyms than other words, with the exceptions discussed in Section 1.2. Nouns, adjectives and verbs have a more specific lexical meaning, which apparently makes it difficult to have full synonyms, while near synonyms or related expressions are more frequently found, see Sections 2 and 3 below. .. Near synonyms It is of course impossible to draw a clear line between ‘synonyms’ and ‘near synonyms’; however, the following examples illustrate the latter type: (6306) la gloire de Dieu – la gloire del Ciel; (6467) veve dame – veve amie; (5948) malvais – coart; (6305) sains hom – bons hom; (8270) preus – grans; merveillous – coragous; forsené – molt pené. .. Words from the same semantic field or with related basic meanings Still more differing are cases where words have distinct meanings, but still belong to the same or a related semantic field; or, if more than one word is concerned, the basic meanings are at least related:

 Lene Schøsler

Variation between different kinds of trees: (6676) carme – orme; between different, but related activities: (6697) pensez – volez; (7160) pot – sot; (8944) rasamblai – redonai; (8490) paser – aler; (8852) vendrons – serons; (8898) la verité en savez toute – la verité t’ai dite toute; (9102) establie – aramie. Even more differing are: (8274) aparole – acole. ..

Ellipses and variation concerning more than one word (a hemistich or a whole verse) In many cases, the copyist simply replaces a word by its synonym or by a related word; he may skip a word, or insert another, or he may even modify the whole verse, normally without seriously changing its meaning. The following examples illustrate major modifications: (8310) T & V: jel vos dirai T: sanz detriier – V: bien volentiers; (8454) T: Lasses! por coi somes – V: Or mais por coi somes; (8473) T: ... celi et dist: Bele, or me dites – V: ... pucele, fait il, or me dites; (7406) T: au chevalier faillir / V: tollir ne(l) doi; (8522) T: si s’est 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 simple errors: (6631) T: a nul sens – a nul tens; (8344) T: dont l’en laisserai je issir – V: et je l’en laisserai issir; (8416) T: Gavains i vient, si le salue – V: Il vient vers li, si le salue; (8418) T: car ci ne vos lairai je mie – V: ci ne vous laisserai je mie; (8423) T: Certes bataille avez assez – V: vous eussiez bataille assez; (6176) T: ou morir ou languir set ans – V: morir ou languir bien set ans; (8344) T: dont l’en laisserai je issir – V: et je l’en laisserai issir; (8348) T: ne vos anuit – V: qui qu’il anuit.

Sometimes the scribe has simply forgotten one or several words, as is apparent from the metre: (5967) T: en fui – V: en is lacking; (5997) tant ne redout – T: ne is lacking; (6000) T: damoise – V: damoisele; (6189) se vos la lance – V: has skipped one of the two la-syllables; (7599) cil rendroit as dames lor terres – T: as dames is lacking.

Scribal variations 

Or the verse contains too many words, as in V (7249) T: furent vestues les pluisors – V: furent vestues richement les pluisors. In one case only (7419-20) do we find a change of word order necessitating a change of rhyme (see also Busby 1993a: 58): ms. T Me porrai je O V S De mon cheval C

en vos C en foi C

fier – Inf garder? Inf

ms. V Porrai me je V O S De mon cheval C

fier en toi Inf C garder a foi? Inf C

Variations of the type examined in this paragraph often provide the basis for oppositions which are accepted as genealogically relevant. If supported by other mss., they may provide type-2 oppositions that can be used for establishing a stemma. I was very surprised to see the liberty of one and the same scribe copying the same exemplar, and I fear that many of the variations that we think genealogically relevant could instead be independent, free innovations, made spontaneously by the scribes – i.e. cases of “mouvance” (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 exemplar, made by the same person, we should not expect to find any variations that are unrelated in content. In fact, we find numerous confirmations of relatedness that have not yet been mentioned here; e.g. common readings of a different sort, opposing T and V to other mss., such as common erroneous succession of verses (6184,86,85; 6230,33,34; 6496,99,6500; 6616,15; 6646,51,52; 6664,67 ...). We also find common, but not significant, errors of content – due to the similarity of characters, as in 8406: li coroit soz l’auberc le sanc, corrected by Roach and Busby to: li coroit sur/sor le hauberc blanc, and a common number indication: .vii.c. in stead of .v.c. in 7566, corrected by Roach and Busby. However, we find at least three cases of order of verses where the mss. do not agree: V presents the order 6851,52 – T has 6852,51; later, T has the order: 7200,1,2,3,4,5,6,6a,6b,7 – whereas V has: 7200,1,2,3,4,6a,6b,7 – without verse 7205-6. T has theorder : 9092,3,4,6,5 – V: 9092,3,4,5,6. Moreover, we find an additional 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. A very curious case is found in 8424, where the two mss. have almost opposite meanings, but one of these must be an error: T: se mes amis ne fust lassez – V:

 Lene Schøsler

se mes anemis ne fust grevez. Other cases of opposite meanings are found in the following examples: (8476) T: ainz que je n’aie / V: perde vostre grace; (5975) T: n’autre folie n’i pensai – V: n’onques folie n’i pensai; (7354) T: desor la penne – V: desoz la penne.

There are cases, where the basic meaning is not really opposite, but only very vaguely related: (8266) T: li notoniers dont vos ai dit – 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 – V: Je ne sai pas ou plus bas soit / l’iaue est trop parfonde, ce dolt.

In these cases, we no longer have identity of meaning, whatever may be the reason. There is no reason to think that a change of relationship in the mss. took place (see Van Mulken 1993: 219), and my impression that variations increase in frequency after folio 12 (verse 8000) is probably due to a sudden new inspiration of our scribe as he reached that part of the text. The conclusion to be drawn from the investigation of this level is that what could be considered as genuine unrelated variations are indeed very rare.

. Variations between the mss. A2, A3 and A4 of the Charroi de Nîmes In this section I shall briefly compare the types of variation found in mss. T and V of the Perceval to those found for comparable mss. of the Charroi de Nîmes, which differs in genre: it is an assonanced, decasyllabic, epic text, much shorter than Perceval (some 1500 verses). This text has survived in nine mss. of which none were copied by the same hand and none were copied from any of the other extant manuscripts. Three mss. were copied at approximately the same moment (1300), from the same exemplar, i.e. from the same ‘input’ dialect, which is from the south-eastern part of northern French, into the same ‘output’ dialect, that of Nièvre-Allier. (Ms. A1 has the original ‘input’ dialect: HauteMarne, slightly north-east of Nièvre-Allier.) The mss. have almost the same score of dialectally marked linguistic features: 73, 76, 79 out of a maximum of 100. The mss. are largely similar and might have been copied in the same atelier, according to Tyssens (1967), to whom I refer for further information upon these 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 of individual liberty in copying in the case of one scribe working several times in succession on the same exemplar, as compared to the liberty of several scribes working 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 Charroi is very similar to what has been found for Perceval. Throughout the mss., we find variations like the ones quoted below: Variation between en and an: (1198) A2: Angleterre – A3,4: Engleterre; (1112) A2,4: emdementiers – A3: andemantiers; (507,528,545,601 ...) A2,3: hennor – A4: anor. Between ein and ain: (1112) A2: einsi – A3,4: ainsi. Between ai and é: (687) A2,3: aidier – A4: edier; (60) A2: ferai – A4: feré; (340) A2,3: dirai – A4: diré, etc. (generally, A4 has a predilection for é). For more details of vowel variations, see Schøsler (1995). In the consonant system, there is much hesitation between the spellings c and qu: (542) A2,4: coronne – A3: quorone; (948): A4: car – A2,3: quar; (1176) A3: cuens – A2,4: quens; (1386) A2,3: corent – A4: queurent. There is also much variation as to simple or double marking of identical consonants, oral as well as nasal; there is alternation between the spellings l and u, etc. In short, we get a comparable variation pattern between these independent mss. 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 in the Perceval mss. The declension system of the three mss. is less conservative than the system we found in T and V. In particular, the scribe of A4 has great difficulties in trying to master the two-case system. We may even find internal ‘contradictions’ of case inside Noun Phrases, like in 1086: A2,3: li glorieus, li fi(l)z Sainte Marie (correct nominative forms) – A4: le glorieus, le fiuz Sainte Marie (mixture of nominative and accusative forms). We find the same alternation between different articles: (1439) A2: li cors (correct nominative form) – A3: lor cors (possessive, indeclinable article) – A4: le cors (erroneous accusative form). The position of the adjective is identical in the three mss.; so is the con-

 Lene Schøsler

cord, or absence of concord, in the past participle of compound tenses (a few conflicting examples are found, e.g. in 723). In contrast with the Perceval mss., there seems to be a slight blurring of the 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 neutralised demonstrative form ce. As for tense-variation, the mss. of the Charroi do often vary, but they normally have the same choice of tenses, except for a few cases: (188) A4: membre (present) – A2: membra (preterite); (237) A2,3: ai ocis (perfect) – A4: ocis (preterite); (707) A2: ç’a fet (perfect) / A3,4: ce fet (present) Aymes le Viell; (831) A2,4: done (present) – A3: dona (preterite); (851) A4: savez (present) – A2,3: saroiz (futur).

There are a few cases of differences in transitivity, e.g. between direct and indirect constructions, such as dire – dire de (908); tirer – tirer a (1332). Minor variations in word order and small words skipped in one or two mss., are like those found in T and V, e.g. (739): Et si te paines A2,4: de moi molt empirier – A3: molt de moi empirier. At this level, once again, we get a comparable variation pattern between these independent mss. and the two Perceval copies. . Variations at the level of content in related passages (level 3) There are many cases of synonyms, near synonyms or related constructions. I shall quote a few illustrative cases: Nouns: (488) A2,4: chatel – A3: regné; (753) A3: Rois Loys – A2,4: Loys, Sire; (880) A2,3: regne – A4: pais; (1057) A3: corneles – A2,4: toreles; c’est nom de (A2,3): pute gent – (A4): mescreant; (1188) A2,3: fié – A4: terre; (1200) A2,3: empire – A4: compeignie;

proper names show often variations: (609) A2,4: Guielin le franc – A3: Guielin l’enfant; (952, 957): A2,3: Ricordane – A4: Cordane; Tiacre/Fiacre (1136, 1138, 1155, 1186 ...); (1295): A3: Gilebert de Faloise sor Mer – A2,4: Gilebert de Faloise le Ber; (1364) A2,3: Raol d’Omacre – A4: Raoul de Marche;

Scribal variations 

Adjectives: (621) A2,3: tot – A4: trestot;

Pronouns: (370) ... bien vueil que tuit (A2,4) / vos (A3) l’oiez;

Articles: Variations between cel/tel – cele/tele are frequent – probably due to the difficulty in distinguishing the letters c and t, e.g. 210, 329;

Conjunctions: (1449) A2,4: com – A3: quant;

Prepositions: (191) A2: en ta cort – A4: a ta cort; (635) A2,3: sor – A4: sus (several occurrences of the alternation sus/sor); several occurrences of the varation between trusqu’au – jusqu’au, e.g. 1005, 1083;

Verbs: (161) A2: refis – A4: en fis; (425) A2: fet – A3,4: dit; (601) A2: si sorrist – A3,4: si s’en rist; (808) A2,3: s’aseons – A4: si seons; (848) A3,4: aresonez – A2: resonez; (910) A2,3: preismes – A4: veismes; (1240) A3,4: creverent – A4: coperent; (1333) A2: errache; A4: arache – A3: estache; (1359) A2,3: se prist a escrier – A4: comança a crier; (1366) A2,3: mener – A4: amener;

Adverbs: (461) A2,3: ausi – A4: ainsi; (574) A2: molt – A3,4: si; (604) A2,3: faintement – A4: faussement; (989) A3: com fierement – A2,4: comfetement.

In some cases, the variations between the mss. are more significant – so significant that it is difficult to speak of clear identity of content. These are in fact difficult border-cases. Yet we still find an underlying similarity of structure, which indicates that these variations are related and thus differ from those classified as level 4-variations; see below. Cases which are still related, although different in meaning, are e.g. (917): n’en iert mençonge (A2,4): oïe / (A3): dite; (1159) Par voz merciz, faites nos (A2,4): en doner / (A3): entendant. Simple errors include e.g. (794) A2: demander – A3,4: dementer; (1277) A2,4: delez – (A3) celez; (1365) A3,4: l’apostre – A2: la porte.

 Lene Schøsler

. Variations at the level of content in unrelated passages (level 4) The three A mss. have a common exemplar, a. None of the daughters of a are free adaptations, and no variations occur at level 4. As shown in e.g. Van Reenen and Schøsler (1996), unrelated (type-2) variations are altogether rare in our text; I here quote two of them in order to illustrate how different variations can be when they are unrelated. In both cases we have an opposition between the A and the B families on the one hand and the mss. C and D on the other hand: A1 A2 A3 A4 B1 B2 C D

v. 0029 AB/CD Li cuens Guillaume fu molt gentix et ber Li quens Guillaume fu molt gentis et ber lacuna Li quens Guillaume fu molt gentil et ber Li quens Guillaume fu molt gentiz et ber lacuna A son ostel descendi au degre En la grant place est descendus li ber

A1 A2 A3 A4 B1 B2 C D

v. 0161 AB/CD S’il le deffent, bien en doi avoir blasme S’il le deffent, bien en doi avoir blasme lacuna S’i le deffant, bien en doi avoir blame S’il s’en deffent, bien en doi avoir blasme S’il s’en defent, bien en doi avoir blasme S’il le desdist, prest sui que m’en combate S’o volt noier, pres suis d’a lui conbatre

To sum up: variations at level 4, i.e. at the level of content in unrelated passages, are not found between these mss. copied by different scribes, copying from one and the same exemplar, at approximately the same moment and at approximately the same place. . Conclusion Does the study of the Perceval copies, compared to the Charroi copies, bring us closer to an understanding of which types of variation are genealogically relevant and which are not? I will summarize what we have learnt from the study of the two sets of mss.:

Scribal variations 

Spelling variations (i.e. variations at level 1) and lexical variations (i.e. variations at level 3) are very frequent. Morphological and syntactic variations (i.e. variations at level 2) are much less frequent. As expected, unrelated variations of content (i.e. variations at level 4) are virtually non-existent. It has been established for both sets of mss. that the common dialect of the copies differs from the original dialect. Type and frequency of variations do not, however, differ significantly between the two sets. When mss. have made the same itinerary, at approximately the same moment, we do not find essential differences in variation between one copyist copying the same exemplar and different copyists copying one and the same exemplar. Absence of a norm makes individual variations so frequent that independent copyists and one and the same copyist working on the same text behave equally freely when copying. The conclusion to be drawn from this comparison, then, is that we have no linguistic clues for distinguishing a single person’s individual variation from different person’s variations – when copies were made at approximately the same moment and in the same dialect. Fundamental differences may arise in cases of different copying attitudes or techniques (mirror copying versus adaptation; transcription versus dictation), and when copies were made at different moments and in different dialects. But our cases, with apparently the same copying attitude and common ‘output’ dialects, do not permit me to postulate the existence of two essentially different linguistic situations. Only the identity or the difference of the hand itself can tell us whether the spelling variations stem from one or from different persons. On the other hand, it should not be forgotten, of course, that external evidence, dialectal evidence and chronological evidence are the first, important, points of genealogical investigations – but these points are of no relevance here: we are talking here about specific linguistic clues in stemmatology. Sometimes, one even gets the impression that scribes make spelling variation a point of honour. Let me quote the case of Yvain vv. 2025–2031 with regular alternation between antel and entel that Kajsa Meyer has kindly brought to my attention (quotation adapted here: see Meyer (1995: 69) for a diplomatic transcription of the Guiot-manuscript): (... anquel maniere?) antel que graindre estre ne puet entel que de vos ne se muet Mes cuers n’onques aillors nel truif antel qu’illors pansser ne puis entel que toz auos m’otroj

 Lene Schøsler

antel que plus uos aim que moj entel s’il uos plest a deliure.

The conclusions of the investigation into mss. T and V, and the comparison with the Charroi A group of manuscripts, are threefold: 1. Individual variation in spelling (level 1) and lexicon (level 3) is much larger than I had previously believed it to be. On the other hand, morphological and 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 their time, these findings will have implications for our understanding of the way people used their language in northern France in the 13th century.8 3. The unexpected liberty of variation, of ‘mouvance’, seen in one and the same person must have implications for our investigation of relations between mss. Philologists often base their arguments on lexical variations – but we have seen in our two mss. that lexical variation without any genealogical relevance is actually very frequent. Levels 2 and 4 (morphology, syntax and unrelated lexical variations) may offer better clues for the investigation of manuscript variations. But one of the important conclusions to draw from this investigation is that we will have to reconsider which variations are genealogically relevant and which are not. I believe that these conclusions support many of the claims made by the socalled “new philology”, see e.g. Schøsler (forthcoming) with references.

Notes . “Mouvance”: “le caractère de l’œuvre qui, comme telle, avant l’âge du livre, ressort d’une quasi-abstraction, les textes concrets qui la réalisent présentant, par le jeu des variantes et remaniements, comme une incessante vibration et une instabilité fondamentale” Zumthor (1972: 507) quoted apud Mulken (1993: 31). Cf. also Micha (1966: 69–70): “...nous sommes en présence d’une réfection libre constante ... nous sommes extrêmement éloignés d’une tradition mécanique du texte où le copiste reproduit consciencieusement ce qu’il a sous les yeux ...”. . 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 of the same exemplar; see however Busby (1993b, notes to 7261, 7338, and 7725) for indications that V could be a copy of T. See also Busby (2002), Chapter 2: “Varieties of Scribal Behaviour”. . See Dees et al. (1987).

Scribal variations  . See Busby, e.a. p. 54: T – “the third or fourth quarter of the thirteenth century” and V – “after 1250”. . The relevant verses are quoted in Section 1.3. . V = verb, S = subject, C = (different sorts of) complement, Inf = infinitive, Adv = adverb, Conj = conjunction, O = object. . It might be interesting to compare my findings with a few other texts copied several times by the same copyist. Keith Busby has kindly informed me of some fabliaux having been copied twice, e.g. La vieille Truande.

References I Primary sources Perceval – T, Paris, Bibliothèque Nationale, f.fr. 12576; V, Paris, Bibliothèque Nationale, nouv. acq.fr. 6614. Charroi de Nîmes – A2, Paris, Bibliothèque Nationale, f.fr. 1449; A3, Paris, Bibliothèque Nationale, f.fr. 368; A4, Milano, Biblioteca Trivulziana 1025. Busby, Keith (1993b). Le Roman de Perceval ou le Conte du Graal. Edition critique d’après tous les manuscrits. Tübingen: Niemeyer. Hanchard, J. (1955). Le Charroi de Nîmes, chanson de geste du XIIe siècle; édition du manuscrit de Boulogne-sur-Mer. Louvain. Lange-Kowal, Ernst-Erwin (1934). Das altfranzösische Epos vom Charroi de Nîmes: Handschrift D herausgegeben mit sprachwissenschaftlichem Kommentar und Glossar. Jena. Meyer, Kajsa (1995). La copie de Guiot, fol. 79v-105r du manuscrit f.fr. 794 de la Bibliothèque Nationale, “li chevaliers au lyeon” de Crestien de Troyes. Éditée par Kajsa Meyer. Amsterdam-Atlanta: Rodopi. McMillan, Duncan (1972). Le Charroi de Nîmes, Editée d’après la rédaction AB, avec introduction, notes et glossaire. Paris: Klincksieck. Roach, William (1959). Chrétien de Troyes: Le Roman de Perceval ou Le Conte du Graal. Paris: Droz. Walker, B. J. (1955). The Boulogne text of the Charroi de Nîmes. University of London.

II Philological and linguistic studies Busby, Keith (1993a). “The Scribe of MMS T and V of Chrétien’s Perceval and its Continuations”. In K. Busby et al. (Eds.), Les manuscrits de/The manuscripts of Chrétien de Troyes (pp. 49–65). Amsterdam: Rodopi. Busby, Keith (2002). Codex and Context: Reading Old French Verse Narrative in Manuscript I–II. Amsterdam: Rodopi. Comfort, T. E. (1954). The Charroi de Nîmes: Old French Chanson de geste edited from the Manuscript of Boulogne-sur-Mer. University of Illinois.

 Lene Schøsler

Dees, Anthonij et al. (1980). Atlas des formes et des constructions des chartes françaises du 13e siècle. Beiheft zur Zeitschrift für romanische Philologie, Band 178. Tübingen: Max Niemeyer. Dees, Anthonij et al. (1987). Atlas des formes linguistiques des textes littéraires de l’ancien français. Beiheft zur Zeitschrift für romanische Philologie, Band 212. Tübingen: Max Niemeyer. Foulet, Lucien (1967). Petite Syntaxe de l’Ancien Français. Paris: Champion. Jodogne, Omer (1956). Le manuscrit de Boulogne du “Charroi de Nîmes”. Publicaciones de la Facultad de Filosofía y Letras II, (17), 301–326. Zaragoza. Micha, Alexandre (1966). La tradition manuscrite des romans de Chrétien de Troyes. Genève: Droz. Mulken, Margot van (1993). The Manuscript Tradition of the Perceval of Chrétien de Troyes. A Stemmatological and Dialectological Approach. (Ph.D. Thesis.) Amsterdam. Reenen, Pieter van & Lene Schøsler (1993). “Les indices d’infinitif complément d’objet en ancien français”. Actas do XIX Congreso Internacional de Lingüística e Filoloxía Románicas, Vol. V, 523–545. La Coruña. Reenen, Pieter van & Lene Schøsler (1996). “From Variant to Pedigree. A Typology of Variants”. In P. van Reenen & Margot van Mulken (Eds.), Studies in Stemmatology (pp. 263–304). Amsterdam – Philadelphia: Benjamins. Salemans, B. J. P. (1990). “Text Genealogical Remarks on Lachmann, Bédier, Greg and Dearing”. Leuvense Bijdragen, 79(4), 427–468. Louvain. Schøsler, Lene (1973). Les temps du passé dans Aucassin et Nicolete. L’emploi du passé simple, du passé composé, de l’imparfait et du présent “historique” de l’indicatif. Etudes romanes de l’Université d’Odense, vol. 5. Odense. Schøsler, Lene (1984). La déclinaison bicasuelle de l’ancien français, son rôle dans la syntaxe de la phrase, les causes de sa disparition. Etudes romanes de l’Université d’Odense, vol. 19. Odense. Schøsler, Lene (1988). “La constellation de Narcisse. Distribution spatiales et temporelles, constellations des manuscrits”. Etudes de variation linguistique offertes à Anthonij Dees à l’occasion de son 60me anniversaire, 247–263. Amsterdam. Schøsler, Lene (1989). “Problèmes de stemmatologie, illustrés par le cas de Narcisse”. Jaarboek 1988–1989, 167–174. Amsterdam. Schøsler, Lene (1994). “Did ‘Aktionsart’ ever dominate Verbal Aspect in Old French?” In Carl Bache, Hans Basbøll, & Carl Erik Lindberg (Eds.), Tense, Aspect and Action. Empirical and Theoretical Contributions to Language Typology (pp. 165–184). Berlin – New York: Mouton de Gruyter. Schøsler, Lene (1995). “New Methods in Textual Criticism: the Case of the Charroi de Nîmes”. In J. Fisiak (Eds.), Medieval Dialectology (pp. 225–276). Berlin – New York: Mouton de Gruyter. Schøsler, Lene (forthcoming). “The Copyist at Work. How did he Work? What are the Consequences for Linguistic Research and for Editorial Policy?” Tyssens, Madeleine (1967). La Geste de Guillaume d’Orange dans les manuscrits cycliques. Paris: Les Belles Lettres. Zumthor, Paul (1972). Essai de poétique médiévale. Paris: Seuil.

The effects of weighting kinds of variants Matthew Spencer*, Linne R. Mooney, Adrian C. Barbrook*, Barbara Bordalejo**, Christopher J. Howe*, and Peter Robinson** University of Cambridge* / University of Maine, Orono / De Montfort University, Leicester**

.

Introduction

Manuscript production by manual copying introduced many different kinds of variants into texts, from changes in spelling and punctuation to the insertion or deletion of whole lines or sections. Are all these kinds of variants equally reliable for stemma reconstruction, and if not, how should we deal with differences in reliability? It will be helpful to distinguish two different classes of variants: those that are biased with respect to the true stemma, and those that are unbiased but relatively uninformative. Both classes have analogues in the reconstruction of phylogenies (evolutionary trees depicting the lines of descent of organisms) by evolutionary biologists. For example, the external appearance of many organisms is often more affected by the conditions under which they live than by their lines of descent. It has been known for more than a century that such external characters are of little use in establishing evolutionary relationships. For example, in Chapter XIV of The Origin of Species (Darwin 1859), “No one regards the external similarity of a mouse to a shrew, of a dugong to a whale, of a whale to a fish, as of any importance”. Spelling and punctuation variants are good examples of variants that are likely to be biased, as they may be strongly influenced by the dialect spoken by the scribe. Such variants are usually excluded from stemmatic analyses (Robinson 1997: 72–74). Unbiased but relatively uninformative variants are usually those that show a high frequency of change. Most stemmatic analyses are based on the principle that we should 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. If a variant arose independently many times, distributed over changing groups of manuscripts, it can tell us little about the relationships among manuscripts because its distribution is unlikely to be simply related to the true stemma. On the other hand, a very improbable variant gives strong evidence that all the manuscripts in which it occurs are related by descent. In evolutionary biology, Darwin stressed the importance of relatively constant characters over characters showing large amounts of variation (Darwin 1859: Ch. XIV). Here, we assume that biased classes of variants have been removed, and examine the circumstances under which unbiased but uninformative variants may cause problems for stemma reconstruction. We use Lydgate’s Kings of England as an example. Many of our methods are derived from evolutionary biology. We cite relevant papers in the biological literature, but we translate biological terms into their analogues in stemmatology (for example, ‘taxon’ becomes ‘manuscript’, and ‘phylogenetic tree’ becomes ‘stemma’).

. The data Lydgate’s Kings of England consists of 105 lines in most versions. 35 manuscripts and three early printed editions survive. We transcribed all 38 surviving copies and manually coded them so as to distinguish eleven kinds of variants (Table 1. The table gives the weight for a change from the base text to each kind. For change from one kind of variant to another, we used the maximum of weights for each kind of variant (for example, if one manuscript had a variant portion of a line and the other had that portion omitted, we gave the difference between these two a weight of 50). Missing data due to damage were always given weight 0). Each coded text contained 860 single-letter symbols. We describe our methods and results in more detail elsewhere (Mooney et al. 2001). Determining the values that should be used to weight different kinds of changes is problematic. If there is a priori information on the frequency or importance of different kinds of change, one could use this information to assign weights. In most cases, such information will not be available, at least until after the stemma has been produced. It may be possible to optimize the initial guesses (Ronquist 1995: 82), but this will be difficult if there are many different kinds of variants. Our aim was to determine whether weighting different kinds of variants is likely to be important for a given text. We therefore compared stemmata constructed using two plausible sets of weights: uniform weights,

The effects of weighting kinds of variants 

Table 1. Non-uniform weights used for each kind of variant. Kind of variant

Weight

Stanza omitted Line changed completely Variant portion of line, changes meaning Word change affecting rhyme Line or portion of line omitted Proper noun variant, changes meaning Word variant, changes meaning Major word added/omitted, changes meaning Two (or more) words in reverse order Word variant without change in meaning Minor word added/omitted, without change in meaning

90 90 50 40 30 25 20 15 15 10 5

and our subjective ideas of the importance of different kinds of changes, as given in Table 1. We gave high weights to kinds of variants that we thought unlikely to occur independently more than once (such as a completely changed line) and low weights to kinds of variants that we thought could easily have arisen independently several times in the tradition (such as the omission of a minor word without changing the meaning of a sentence). We stress that these weights are subjective, in that other scholars might assign different values or different rank orders to these kinds of variants. However, we analyzed several other weighting schemes, including random weights, with similar results which we do not report here. We are therefore confident that our conclusions are not much affected by the choice of weighting scheme. We return in the Discussion to the problem of objective assignment of weights. One reason why weightings might make no difference is that if all variants are transmitted via the same stemma, the same stemma should be recovered whatever weights are used for different kinds of variants (although sampling error in texts of finite length may introduce some disagreement). To test this idea, we also compared stemmata constructed using each of the four commonest kinds of variants in the Kings of England data: major words added/omitted; minor words added/omitted; word variants changing meaning; and word variants not changing meaning. We did not examine rare kinds of variants because there were too few data to be reliable.

 Matthew Spencer et al.

. Tree reconstruction and comparison We used neighbour-joining to reconstruct stemmata from matrices of pairwise distances between manuscripts. Neighbour-joining (Saitou & Nei 1987: 406– 425) is a simple clustering algorithm that sequentially separates pairs of manuscripts from an initially unresolved stemma, such that at each step the sum of least-squares branch length estimates is minimized. For biological data, both simulations (e.g. Saitou & Imanishi 1989: 524; Tateno et al. 1994: 273) and analyses of real data with well-supported phylogenies (Wiens 2000: 624–625) have shown neighbour-joining to be among the most successful tree reconstruction methods. For uniform weights, we estimated pairwise distances as the proportion of non-missing locations at which two manuscripts had different coding symbols. For non-uniform weights, differences between coding symbols were weighted as in Table 1. We gave high weights to kinds of changes we thought unlikely to have occurred independently many times, and low weights to those that we thought could easily have arisen independently in different parts of a manuscript tradition. Although the observed number of differences between a pair of manuscripts ignores the possibility of multiple changes at single locations, such events would have been rare in the Kings of England tradition because the maximum pairwise uniformly weighted distance was 0.21. With distances this small, there will be few cases of several changes at a single location. When this is not the case, one should attempt to estimate the actual number of differences (Spencer & Howe 2001). We used three methods to check for contamination: split decomposition (Huson 1998: 68–73); Mantel correlations (Upton & Fingleton 1985) between distance matrices from adjacent sections of text; and the construction of separate stemmata from separate sections of text (Robinson 1997: 75–79). None of these methods suggested extensive contamination, so a branching stemma is probably a reasonable representation of this text tradition. All our stemmata are unrooted: they indicate the topology of the stemma but not the location of the archetype. Stemmata could subsequently be rooted by judgements about the originality of readings, palaeographic evidence or internal manuscript information on exemplars. To help illustrate the main features of the stemmata, we have grouped manuscripts into categories (A, B and C) suggested by manual stemmatic analysis (Mooney et al. 2001). We used two different methods to compare stemmata: partition distances and consensus stemmata. The partition distance (Penny & Hendy 1985: 76) between a pair of stemmata is a measure of their topological similarity. Removing

The effects of weighting kinds of variants  1

3

2

1

4

5

2

3

4

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 is common to both of a pair of stemmata if, in both stemmata, there is an edge whose removal divides the stemma into the same two sets of manuscripts. The partition distance between two stemmata is simply the number of edges found in one but not both of the stemmata (Fig. 1. Each of the bold edges is present in only one of the two stemmata. For example, the bold edge in the upper stemma divides the manuscripts into the sets {1,2},{3,4,5}. There is no single edge in the lower stemma whose removal splits the data into these two sets. All other edges are present in both stemmata). We can scale this distance by the number of edges that could possibly be found in one of two stemmata (2 (N – 3) for a bifurcating stemma with N extant manuscripts), to give a distance measure between 0 and 1. The probability distribution of partition distance is known for random stemmata with small numbers of extant manuscripts, and can otherwise be obtained by simulation. Thus we can determine the probability of observing a given partition distance under the hypothesis of no relationship other than chance between a pair of stemmata. A consensus stemma includes an edge only if it is present in some defined proportion of the set of stemmata being compared. Here, we use strict consensus, in which an edge is only included if it is present in every stemma, because we want to indicate where disagreements occur, although several other consensus rules are possible (Page & Holmes 1998: 34–35). Where the stemmata disagree, the consensus stemma will have a polytomy rather than a bifurcation. We do not present branch lengths for consensus stemmata, as their purpose here is simply to show areas of disagreement between stemmata constructed using different weightings.

 Matthew Spencer et al.

We used PAUP* (Swofford 2001) for neighbour-joining, calculating partition distances and preparing a consensus stemma. We calculated distance matrices and generated input files for PAUP* using Matlab 5.3 (The Mathworks, Inc., Natick, MA).

. Results Figure 2 is a stemma obtained by neighbour-joining on uniformly-weighted distances. The different fonts represent manuscript groups obtained by manual stemmatic analysis (Mooney et al. 2001): A bold, B italic, C underlined, others normal 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 (our representations also differ in orientation, but this has no meaning). Both had a strong C group (manuscripts in underlined font) clearly separated from the A (bold font) and B (italic font) groups (apart from the manuscripts Bodley 48 and Bodley 686, which our manual stemmatic analysis 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 branch lengths were different in the two stemmata. For example, de Worde and R. Wyer were further away from other manuscripts when the weights in Table 1 were used than when uniform weights were used. This happens when different manuscripts have different proportions of each kind of variants, either because each manuscript represents a small sample or because there are systematic differences. However, many small groups of manuscripts appeared in both stemmata (e.g. {R. Wyer, de Worde, Pynson} and {TCC 601 1, TCC 601 2}). Table 2 gives the distribution of scaled pairwise partition distances estimated from 1000 random bifurcating stemmata on a set of 38 extant manuscripts (only the values given in this table occurred). The observed scaled partition distance between stemmata using uniform weights and weights in Table 1 was 0.57. Table 2 shows that a distance this small between two randomlygenerated bifurcating stemmata for a set of 38 taxa would be very unlikely by chance. Polytomies in the strict consensus stemma (Fig. 4) show where the differences occur. We remain uncertain about the relative positions of a large set of manuscripts in the A and B groups. We were also unable to resolve the position of the C group relative to Ipswich, Pynson, de Worde and R. Wyer, but the C group (other than Bodley 48 and Bodley 686) was clearly separated from A and B. Several smaller groupings (e.g. {Harley 2251, BL Ad 34360, Digby 186}

The effects of weighting kinds of variants  R.Wyer de Worde

Jesus 56 Leiden Ipswich Harley 372 Lansd. 699 Rawl.C. 48 Caius 249 Lansd. 210

Peterborough Pynson

Harley 7333 Egerton 1995 Linc. Lat 129 TCC 601 2 TCC 601 1 Fairfax 16 BL Ad 310 42 Nottingham Bodley 48 Bodley 686 Dublin 516 Bodley 912 Rawl.C. 316 Buhler 17

Galba E. VIII Harley 2261 Titus D.XX Lambeth 306 Harley 2169

Digby 186 Lansd. 762 Harley 2251 BL Ad 34360

CUL Ad6686 Stowe 69

0.01 distance units

Ashmole 59

Figure 2. Stemmata obtained by neighbour-joining on uniformly-weighted distances.

and {Rawl.C.48, Jesus 56, Leiden, Harley 372, Lansdowne 699}) also occurred under both weightings. Table 3 gives scaled partition distances between each of the four stemmata drawn from the four commonest kinds of variants. These distances were higher

 Matthew Spencer et al.

Ashmole 59

Dublin 516 CUL Ad6686 Harley 2169 Bodley 686 Bodley 912 Buhler 17 Linc.Lat 129 Rawl.C. 316 Nottingham Fairfax 16 Digby 186 BL Ad 34360 BL Ad 31042 Bodley 48 Harley 2251 Titus D.XX Lansd. 762 Egerton 1995 Harley 2261 Harley 7333 Galba E.VIII

Peterborough Lansd.210

Stowe 69 Ipswich

TCC 601 1 TCC 601 2

Rawl.C. 48 Pynson

Caius 249 Leiden

Lambeth 306 Jesus 56 Harley 372 Lansd. 699

0.1 distance units

de Worde R.Wyer

Figure 3. Stemmata obtained by neighbour-joining on distances weighted as in Table 1. Ms groups indicated as in Figure 2. Branch lengths to scale (in arbitrary units), but the scale is different from Figure 2.

The effects of weighting kinds of variants  R.Wyer

de Worde

Rawl.C. 48

Pynson

Jesus 56 Leiden

Ipswich Caius 249 Harley 2251

Lansd. 210

Ashmole 59 Bodley 912 Fairfax 16 Linc.Lat 129 CUL Ad6686 Galba E.VIII Titus D.XX Harley 2169 Harley 2261 Lansd. 762 Stowe 69 Lambeth 306 Nottingham Peterborough

BL Ad 34360 Digby 186 Harley 7333 Egerton 1995 TCC 601 2 TCC 601 1 Buhler 17 Rawl.C. 316 Dublin 516 Bodley 686

Harley 372

Lansd. 699

Bodley 48 BL Ad 31042

Figure 4. Strict consensus stemma for neighbour-joining stemmata using uniform weights or weights as in Table 1. Ms groups: as in Figure 2. Polytomies: points where different weightings lead to different topologies. Branch lengths not to scale.

than those between the two different weightings of the whole data set, probably because they represent quite small sample sizes. However, they were still unlikely to have arisen by chance (cf. Table 2).

Table 2. The distribution of scaled pairwise partition distances estimated from 1000 random bifurcating stemmata on a set of 38 extant manuscripts. Scaled partition distance

Probability

0.86 0.89 0.91 0.94 0.97 1.00

6.01 ×10–6 1.06 ×10–4 1.88 ×10–3 0.02 0.18 0.79

 Matthew Spencer et al.

Table 3. Scaled partition distances between stemmata drawn using neighbour-joining using the four commonest kinds of variants: 1 – major words added/deleted, 2 – minor ditto, 3 – word variants changing meaning, 4 – ditto not changing meaning.

1 2 3 4

1

2

3

4

– 0.86 0.86 0.83

– 0.80 0.89

– 0.86

–

. Discussion The differences between stemmata for the Kings of England constructed using different weightings were small (Fig. 4, Table 2). This was probably because all variants were fairly rare, and all were transmitted along the same true stemma. All the uniformly-weighted pairwise distances between manuscripts were small, so even the most frequent kinds of change would only rarely show reversals and coincident variation. The pairwise scaled partition distances among stemmata drawn using single kinds of variants were smaller than expected by chance alone (Table 3), supporting the idea of a single true stemma for all these variants. If it is not generally important to give different weights to different kinds of variants, the task of stemma construction may be simplified. For example, automatic coding systems would not need to distinguish different kinds of variants, and it will be easier to develop explicit mathematical models for copying errors. However, there will be cases in which giving different weights to different kinds of variants might lead to major differences between stemmata. If we were unable to identify biased variants beforehand, different weightings could lead to completely different stemmata. Furthermore, if there were many differences between manuscripts, reversals and coincident variation might be common enough that the most frequent kinds of variants were unreliable indicators of the true stemma. A worst-case scenario, combining both of these problems, is the analysis of a large manuscript tradition extending over a very long time period. Changes in language over time could introduce nonrandom patterns of variation unrelated to descent, for example in the substitution of one word for another or in conventional word order. Similarly, the large number of copying events could allow frequent reversals and coincident variants. Because such cases clearly exist, the choice of weights for different kinds of variants merits more attention than it has received so far. Other situations in

The effects of weighting kinds of variants 

which the choice of weights might affect the stemma include contaminated traditions in which different kinds of variants were systematically transmitted along different pathways, and traditions with so few variants that sampling error becomes important. A survey of the effects of weighting in different text traditions would be the best way to test these ideas. There are several unresolved problems with the analyses we have described here. The manual coding we used is subjective and error-prone. Automatic coding could potentially distinguish between some kinds of variants (Robinson 1989: 103–104), but has not yet been employed on a large scale. The choice of weights is of more concern. We arrived at the weights in Table 1 after extensive discussion about the importance of different kinds of change and the likelihood of scribal errors. We have no way of knowing whether another set of weights might not be much better (although we did analyse some other sets of weights, including random weights, with similar results). Non-textual evidence such as a full or partial stemma produced independently of textual variants (for example on the basis of scribal statements of the exemplar used when producing a manuscript: Robinson & O’Hara 1996: 126) will always be needed to distinguish between biased and unbiased kinds of variants, if each kind appears to support a different stemma. In order to establish objective weights based on the probabilities of different kinds of change, we need a stemma, but we need weights in order to obtain that stemma in the first place. There are two possible solutions: 1. Examine the frequencies of different kinds of variants on a stemma which has already been established by other means. We would then have to assume that the frequencies of these kinds of variants are the same in other manuscript traditions. However, a stemma (or a part of a stemma) could only be established independently of any kind of information on variants when there is non-textual evidence for relationships (for example, a statement in the manuscript of the exemplar used). Such cases are rare, although not completely unknown (O’Hara & Robinson 1993: 55–56). 2. A related problem is the determination of reliable and unreliable locations in a text. One would like to give high weights to reliable locations and low weights to unreliable locations, but it is difficult to know in advance which locations are reliable. It has been suggested that one can construct an initial stemma using some guesses at weights (uniform weights might be a reasonable guess), and weight locations according to their degree of concordance with this stemma. A new stemma is then constructed and the process repeated until the topology of the stemma stops changing. Al-

 Matthew Spencer et al.

though this approach sometimes improves the tree reconstruction process (Farris 1969: 374–385; Fitch & Ye 1991: 147–154), it has been criticized on the grounds that it is best to avoid such circularities whenever possible (Lee 1999: 726–728). With either method, one still has to make a subjective decision about the form of the relationship between frequencies of change and weights, although simulations can suggest forms that are likely to perform well (Farris 1969: 377–380). In conclusion, our stemmata for Lydgate’s Kings of England are unlikely to be substantially affected by the differing usefulness of different kinds of variants (other than spelling and punctuation variants, which we did not code). However, we have identified situations in which one might need to weight some kinds of variants more highly than others in order to obtain a reliable stemma. Determining 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 of Favored Races in the Struggle for Life. London: Murray. Farris, J. S. (1969). “A Successive Approximations Approach to Character Weighting.” Systematic Zoology, 18, 374–385. Fitch, W. M. & J. Ye (1991). “Weighted Parsimony: Does It Work?” In M. M. Miyamoto & J. Cracraft (Eds.), Phylogenetic Analysis of DNA Sequences (pp. 147–154). Oxford: Oxford University Press. Huson, D. H. (1998). “Splitstree: Analyzing and Visualizing Evolutionary Data.” Bioinformatics, 14(1), 68–73. Lee, M. S. Y. (1999). “Circularity, Evolution, Systematics ... And Circularity.” Journal of Evolutionary Biology, 12, 724–734. Mooney, L. R., A. C. Barbrook, C. J. Howe, & M. Spencer (2001). “Stemmatic analysis of Lydgate’s “Kings of England”: a test case for the application of software developed for evolutionary biology to manuscript stemmatics”. Revue d’Histoire des Textes, 31, 275– 297. O’Hara, R. & P. Robinson (1993). “Computer-Assisted Methods of Stemmatic Analysis.” In N. Blake & P. Robinson (Eds.), The Canterbury Tales Project Occasional Papers, Vol. 1 (pp. 53–74). London: Office for Humanities Communication Publications.

The effects of weighting kinds of variants 

Page, R. D. M. & E. C. Holmes (1998). Molecular Evolution: A Phylogenetic Approach. Oxford: Blackwell Science. Penny, D. & M. D. Hendy (1985). “The Use of Tree Comparison Metrics.” Systematic Zoology, 34(1), 75–82. Robinson, P. (1997). “A Stemmatic Analysis of the Fifteenth-Century Witnesses to the Wife of Bath’s Prologue.” In N. Blake & P. Robinson (Eds.), The Canterbury Tales Project: Occasional Papers, Vol. II (pp. 69–132). London: Office for Humanities Communication Publications. Robinson, P. M. W. (1989). “The Collation and Textual Criticism of Icelandic Manuscripts. (1): Collation.” Literary and Linguistic Computing, 4(2), 99–105. Robinson, P. M. W. & R. J. O’Hara (1996). “Cladistic Analysis of an Old Norse Manuscript Tradition.” In S. Hockey & N. Ide (Eds.), Research in Humanities Computing 4 (pp. 115-137). Oxford: Oxford University Press. Ronquist, F. (1995). “Reconstructing the History of Host-Parasite Associations Using Generalised Parsimony.” Cladistics, 11, 73–89. Saitou, N. & T. Imanishi (1989). “Relative Efficiencies of the Fitch-Margoliash, Maximum-Parsimony, Maximum-Likelihood, Minimum-Evolution, and Neighbor-Joining Methods of Phylogenetic Tree Construction in Obtaining the Correct Tree.” Molecular Biology and Evolution, 6(5), 514–525. Saitou, N. & M. Nei (1987). “The Neighbor-Joining Method: A New Method for Reconstructing Phylogenetic Trees.” Molecular Biology and Evolution, 4(4), 406–425. Spencer, M. & C. J. Howe (2001). “Estimating distances between manuscripts based on copying errors”. Literary & Linguistic Computing, 16(4), 467–484. Swofford, D. L. (2001). Paup*. Phylogenetic Analysis Using Parsimony (*and Other Methods). Vers. 4.0b8. Computer software. Sinauer Associates. Tateno, Y., N. Takezaki, & M. Nei (1994). “Relative Efficiencies of the Maximum-Likelihood, Neighbor-Joining, and Maximum-Parsimony Methods When Substitution Rate Varies with Site.” Molecular Biology and Evolution, 11(2), 261–277. Upton, G. & B. Fingleton (1985). Spatial Data Analysis by Example. Vol. I. Point pattern and quantitative data. Chichester: John Wiley & Sons. Wiens, J. J. (2000). “Reconstructing Phylogenies from Allozyme Data: Comparing Method Performance with Congruence.” Biological Journal of the Linnean Society, 70, 613–632.

Cluster analysis and the Three Level Method in the study of the Gospels in Slavonic Dina Mironova Saint-Petersburg State University

.

Introduction

. Historical background The Slavonic alphabet was designed in the middle of the IX century in Byzantium at the court of patriarch Photius. The originator was an outstanding scholar of his time, Constantine the Philosopher. The script in the course of time received the name Glagolitic. Fifty years later another script appeared which was called Cyrillic. The new script was based on the Greek alphabet supplemented with a few letters to denote specific sounds of Slavonic. Old Church Slavonic was one of the last literary languages in Europe which came into being together with the new alphabet. Along with the alphabet the Slavs received their first Bible texts. A two-century investigation of the tradition has formed the following view upon 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 can only speak about a number of manuscripts, which date back to the IX–XII centuries. These manuscripts have features that can be traced back to the original translations and bear signs of a considerable revision of the end of the IX – beginning of the X centuries. We shall refer to these manuscripts as the Old Text Type. A text type, as E. Colwell (1969: 45, 10–11) defines it, is “the largest identifiable group of related New Testament manuscripts”. A group of manuscripts form one text type if they agree in a number of variants against other groups and if there is agreement of the group in the majority of variants.

 Dina Mironova

2. Slavonic literature was flourishing during the reign of tsar Simeon (893– 927) in Bulgaria. The activity of translators in this period was characterized by mass lexical revision, mostly the revision of Christian terminology. This group of manuscripts is known under the name of the Preslav Text Type (after a large cultural center where many scribes worked). The Preslav Text Type has more extant witnesses than the Old one. 3. At the beginning of the XIII century St. Savva of Serbia introduced the Typicon of Jerusalem in the Serbian Chilander Monastery on Mount Athos. It was meant to serve as the basis for monastery needs. A. Alekseev calls this the New Liturgical Tetraevangelion (Evangelije 1998), because the term Tetraevangelion had already been used for Liturgy before. The New Liturgical Tetraevangelion was based on the continuous text of the Old Text Type with some additions from the Bulgarian revisions of the X century.1 4. In the first half of the XIV century two new recensions of the Gospels emerged in the Slavic Monasteries of Mount Athos. Like the New Liturgical Tetraevangelion these were Tetraevangelia with a rather stable text. They rest upon the Christian terminology of the original translation and have no linguistic features of the Preslav Text Type. They are known as the Athonite Text Types. The second of these comprises the largest number of sources. It achieved predominance in the XIV–XV centuries, and in the XVI century laid the foundation for printed editions. Today, with slight linguistic modifications, it is read during the Orthodox Liturgy. 5. A new translation of the whole New Testament, called the Chudov New Testament, was made in the middle of the XIV century under the supervision of Moscow Metropolitan Alexios (c. 1293–1378). It didn’t circulate in many sources, being forced out by the Athonite Text Types, and is preserved in just a few extant witnesses. We shall refer to this text type as to the Chudov New Testament Text Type. This translation shares many features of the Athonite Text Types. The recensions of the Slavonic Gospels are so closely connected that they form a kind of textual continuum, and the borders between them are vague. There is no complete data about all the Old Church Slavonic New Testament manuscripts. A. Alekseev refers to more than 8000 XVI–XVII century manuscripts and about 500 XI–XV century Gospel manuscripts, though this number doesn’t embrace all sources (Alekseev 1999: 132).

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 factors. Firstly, the contaminated nature of the tradition puts a ban on considering relations between single witnesses, and therefore hinders employing genealogical methods. Readings that can be considered truly genealogically significant are rare and cannot form the basis for the classification.3 The only possible way to investigate the tradition is to consider relations between groups of witnesses with the same text, but never relations between single witnesses.4 The arrangement of the tradition into groups with the same text is done on the basis of variation units (passages in the text where witnesses have different readings). Secondly, the tradition has numerous extant witnesses. Their number is estimated at more than ten thousand and prevents us from making full collations. The collation of only 500 older manuscripts would take years. A compromise way out is collating corresponding parts of every manuscript, each part being long enough to provide a sufficient number and variety of variation units.5 A reliable classification of a contaminated tradition can be obtained if it is based on all variation units found in the investigated piece of text. It is highly undesirable to choose only the most significant variation units, especially because we have too little information to dispense with part of it. The following two arguments prove the necessity to use all available material. Firstly, the manuscripts in each group are very closely connected, and it is often difficult to point out readings which characterise it. Therefore it is almost impossible to detect subgroups within one text type if not all variation units are considered. The more variation units we have, the more likely we should be to find all available groupings, large and small. Secondly, when we lack information, some late manuscripts with old text could mistakenly be positioned into the group of late recensions, as their old text was obscured by new syntactical and grammatical features in the process of copying.6 Fortunately, the time-consuming processing of all the variants is considerably facilitated by computer, which saves time and make it possible to work with any number of manuscripts. In the field of Old Church Slavonic New Testament Studies the method of A. Alekseev, based on cluster analysis, has been used since 1980. In 1999 E. Wattel tested the method he designed for stemmatological needs on some of the Slavonic Gospel manuscripts, and the results in general coincided with the previously made classifications. It became clear that the stemmatological idea of building a non-oriented graph at the initial stage of stemma construction goes 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. Alekseev and E. Wattel on the basis of the results of both classifications. The research comprises 531 manuscripts containing the Gospel according to Matthew, passage 14.14–14.34. This piece of text has about 300 words and provides 545 variation units. We shall first describe the current project on studying the Old Church Slavonic New Testament, and then, outline the essence of the methods of A. Alekseev and E. Wattel will be outlined. After that we shall present the classifications provided by both methods and discuss their similarities and differences. According to the obtained results it should be possible to evaluate the advantages of each formal method for Old Church Slavonic New Testament studies 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 a project in the St. Petersburg branch of the Bible Society in Russia. The project aimed at preparing the editions of Old Church Slavonic Gospels based on all available sources and was launched in December 1994 by the Committee of the United Bible Societies. Before that, as early as January 1993, a large number of linguists had started the collations in the framework of the Slavonic Bible Foundation. The further work on the data and the preparation of the edition was 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 Old Text Type and a number of representatives for every other recension.7 It was for the first time that all 30 witnesses chosen for the critical apparatus of the Slavonic Gospels were cited throughout the whole text for each and every instance where they had a different reading. The basis for the edition is the XI century Marianus Tetraevangelion (Mr). The edition shows the textual history of the Slavonic Gospels as reflected in the extant manuscripts and discovered during the research.

. The method of A. Alekseev The assumption underlying the research is that the tradition of the Slavonic Gospels is controlled (Evangelije 1998; Alekseev 1985), which means that the

The study of the Gospels in Slavonic 

scribe could use more than one copy of the same Gospel for his work. The text was constantly corrected in order to be kept as close to the original as possible. Hence the tradition’s high level of stability, which makes it difficult for the contemporary scholar to trace back textuological connections among the extant witnesses. It is not possible, in any of the text types, to find a variant which occurs in all the manuscripts belonging to that particular text type, and not in other text types. One can only speak of variants which occur in most manuscripts of a given text type. The same is true for the Greek New Testament, and A. Alekseev has borrowed some ideas expressed by the American New 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 (1969: 63–83) shows that it is not relevant to apply a genealogical method to the text type, because a text type of the Greek New Testament tradition is a highly contaminated group of manuscripts. A genealogical method implies that every manuscript has one parent, whereas one extant witness of one text type can have more than one parent. If we regard contaminated sources as nodes of a tree, most pairs of nodes will be connected by only one edge (see Figure 1). There is no way to orient this stemma in such a way that every node has no more than one parent. m1

m2

m7 m3

m6 m4 m5

Figure 1. A stemma for manuscripts of the same text-type.

 Dina Mironova

E. Colwell (1969: 15–19) believes that the “effort to restore the text of a Text type” is misleading. For every text type E. Colwell suggests finding average, typical, most characteristic representatives, which combine in themselves most typical textual peculiarities of the given text type, and presenting their evidence in an apparatus. If families (the smallest and most intimately related groups) can be pointed out within a text type, one representative of a family should also be a valuable addition to the critical apparatus. In order to determine text types within the Greek New Testament Tradition E. Colwell chose a quantitative method. . Similarity score E. Colwell and E. Tune showed the inconsistency of using Textus Receptus as a basis for the collation of Greek New Testament manuscripts. They suggested comparing manuscripts by pairs and thus forming a list of variation units. After the variation units have been found, the number of readings in common for every pair of manuscripts is counted. Then they calculate the ratio of variants in common for every pair and the variation units relevant for this pair. This figure is converted into a percentage. Working with percentages enables them to compare complete and lacunary manuscripts on an equal basis. If Ep is the number of variants manuscripts ‘a’ and ‘b’ have in common, and Vp is the number of variation units relevant for the pair, then the similarity score (Sp ) for the pair ‘ab’ is: Sp = (Ep / Vp ) × 100%

This is one of the coefficients often used in cluster analysis. It requires the least number of calculations, and in our case it is difficult to justify the usage of more elaborate coefficients. The percentage of readings in common for each pair is put down in a square matrix on the intersection of lines and columns, lines and columns standing for manuscripts. In such a matrix E. Colwell calls a group a number of manuscripts which have a higher percent of similarity with each other than with other sources. . Data presentation The contribution of A. Alekseev was that he added one more step to the method to facilitate the analysis. He proposed rearranging the matrix so that the manuscripts with more closely related text would be positioned together, and clusters of the same text type would be visible.

The study of the Gospels in Slavonic 

at least one member of the given pair already belongs to a cluster yes

no

do both mss already belong to some clusters

a new cluster is formed no

the unassigned ms is joined to the same cluster as its pair

do they belong to the same cluster? no

two clusters are joined together

yes

yes

nothing changes in the clusters

Figure 2. Kuznestova’s algorithm for grouping manuscripts in clusters.

The program was designed by E. Kuznetsova (see Alekseev & Kuznestova 1987). The similarity score is measured by Colwell’s coefficient mentioned above. The objects are joined to the cluster by a single-link clustering criterion (for the terminology cf. Galloway 1979), which requires a single strong resemblance of an object to any member of the group. The program looks for a pair of manuscripts with the next highest similarity score. The algorithm is shown in Figure 2. The clustering stops when all the objects are clustered together. The sequence of the manuscripts within the final cluster reflects the order in which they were clustered. After getting an ordered matrix we look for borders between clusters. The matrix is designed in such a way that manuscripts within one cluster and the clusters themselves are ordered by the decrease of the similarity score. The border between clusters is the rise of the percent. Let us consider the example in Figure 3. The similarity score between ‘a’ and ‘b’ is 90%; ‘c’ has an 87% and 88% correspondence with each of them respectively. The correspondence of ‘k’ with them is 79%, 81% and 85%. The next two mss, ‘l’ and ‘m’ 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 didn’t join any cluster: f, g, h, i

Figure 3. The final matrix of clusters. Manuscripts are ordered by the decrease of similarity score.

with ‘k’. Thus, the similarity score goes from 90% down to 85% and then increases up to 90% among the mss ‘k’, ‘l’ and ‘m’, staying lower among the latter three and ‘a’, ‘b’, ‘c’ (85%–79%). The first cluster we can point out is, therefore, ‘abc’. The second cluster comprises mss ‘k’, ‘l’, ‘m’, ‘n’, ‘o’. The percentage of similarity goes down to 80% and then again up to 87%. This is where the new cluster ‘xyzpq’ starts. When we compare it to the previous clusters, we should note that here the similarity between the first two objects is much higher than among the rest. The fourth cluster is ‘uv’, with ‘f ’ probably forming part of it.

The study of the Gospels in Slavonic 

. Disadvantages of the classification In the chosen type of clustering the objects which didn’t join any cluster are positioned at the bottom of the table, ‘at the periphery’. Thus, manuscripts ‘g’, ‘h’, and ‘i’ do not form part of the ‘uv’ cluster, but are united with all the clusters at the last stage. A disadvantage of such an algorithm is that the periphery of the table is a mixture of manuscripts mostly belonging to the periphery of different clusters. In Figure 3 we can see that the similarity score of ‘g’ with the cluster ‘xyzpq’ (73%–70%) is higher than with its neighbours ‘u’, ‘v’, ‘f ’, ‘h’ (68%–64%). Manuscript ‘h’ has a higher degree of similarity with the cluster ‘klmno’ (72%–70%). And finally, manuscript ‘i’ has relatively the same similarity score with all clusters. This drawback, however, doesn’t affect the result. It has already been mentioned that the borders between clusters of the Old Church Slavonic New Testament are vague. Due to contamination different clusters have a periphery in common, and in the table manuscripts ‘g’, ‘h’, ‘i’ can be said to form the periphery of all the clusters above. If a manuscript is positioned at the bottom of the table by mistake, it is easy to reposition it correctly if needed (as in the case of manuscripts ‘g’ and ‘h’, if the variation units prove their belonging to certain clusters). . Stages of classification The method provides a possibility to work with any number of manuscripts, but the first matrix is not the final one: the work is done in two stages. At the first stage in the first matrix we look for groups with identical witnesses, and substitute one representative for each of these groups. The philologist decides which representative to choose. At the second stage we produce a new matrix, where one representative stands for each group of identical manuscripts. In this matrix we determine borders between clusters. The current project covered 1100 available manuscripts. Most Athonite manuscripts could be detected before the stage of computer classification. Most of the witnesses of older recensions and the older witnesses of the Athonite Text Type (altogether 531 manuscripts for the Gospel according to Matthew) were chosen for the computer classification. The method of A. Alekseev yielded the following clusters: 321 manuscripts with the Athonite Text Type, 49 with the Old Text Type, 96 with the Preslav Text Type, 4 with the New Liturgical Tetraevangelion, and 3 with the Chudov New Testament Text Type. The remaining 58 manuscripts are at the periphery and should be studied by purely philological analysis. Thus we have systematised the witnesses and significantly reduced

 Dina Mironova

the number of those needing further investigation. It should be specified here that the classification algorithm is but a small part of a big project. The complete philological part is done by scholars, as well as the final evaluation and checking of the program’s output.

. The method of E. Wattel The method of A. Alekseev remained the only one which could be used for a contaminated tradition with large number of sources. The application of the method of E. Wattel to the Slavonic material has shown that it is also possible to turn to the tools of stemmatology. It has also demonstrated the large scale of employment for the method. Since Wattel’s concept is in detail explained in Wattel (this volume), we shall dwell only on the items which are significant for the 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, developed by A. Dees (1975). A. Dees suggests three stages on the way of building a genealogical classification for the sources: first define the deep structure, then the intermediate structure, and only afterwards the oriented stemma. A deep structure reflects the internal relations among the manuscripts. It is a fully resolved tree structure – a tree in which every extant witness is represented as a terminal node, and every other node in the tree represents a fictive (lost) manuscript with precisely three connections to other nodes. An intermediate structure is a contracted deep structure. E. Wattel devised a set of algorithms to build a deep structure and to contract it (Wattel e.a. 1996, this volume). We shall see that the initial tree or deep structure he constructs is a different representation of the same ideas which lay behind A. Alekseev’s matrix with clusters. For the further discussion we need to connect the concepts of variation unit and version formula. A variation unit deals with the opposition of variants (e.g. variant ‘1’ vs. variant ‘2’), whereas a version formula implies opposition of one or more manuscripts to other manuscripts. For example, if we have variation unit ‘X’ with two variants, ‘1’ and ‘2’, and manuscripts ‘a’ and ‘b’ share variant ‘1’, while manuscripts ‘c’ and ‘d’ share variant 2, then the version formula for this variation unit will be manuscript ‘a’ and manuscript ‘b’ vs. manuscript ‘c’ and manuscript ‘d’.

The study of the Gospels in Slavonic 

. Formula weight Whereas A. Alekseev refrains from weighting variation units, E. Wattel distinguishes two weight constituents in every variation unit (every version formula): linguistic and computational. Linguistic weight is supplied if needed in the input data by the linguist himself. It reflects the hierarchy of importance for different types of variation units. Computational weight is a characteristic of the version formula (not of a variation unit, because the information about the computational weight is derived from the combination of manuscripts in the version formulas). It reflects “how suitable the structure of a version formula is for the construction of a stemma” (Wattel this volume). In the original database linguistic weight for every version formula is multiplied by a computational coefficient depending on the type of the formula. This figure is added up by all positive (a pair of manuscripts sharing the same variant) and all negative (a pair of manuscripts having different variants) contributions of every pair of manuscripts in the given version formula. Thus the original linguistic weight obtains a computational constituent. After every formula has acquired weight this weight is distributed in the formula between pairs, and every pair gets weight in every formula. This weight is positive if both members of the pair have the same variant in the formula, and negative if they have different variants. The weight of a pair of manuscripts with the same variant in a version formula also depends on the number of manuscripts sharing this variant in this version formula, and the weight of a pair of manuscripts with different variants in a version formula depends on the overall number of pairs in this version formula. Having obtained the weight for every pair of manuscripts in every formula we can calculate the sum of positive and negative weight contributions for 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 calculates the distance between the objects (manuscripts) pairwise, and A. Alekseev calculates the similarity score, and each step in structuring the data is the search for the next pair of manuscripts with the smallest distance (highest similarity score). Let us dwell upon the function of E. Wattel. If Np is the sum of negative weight contributions for a pair of manuscripts ‘a’ and ‘b’, and Sp is a sum of positive weight contributions for ‘a’ and ‘b’, 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 relevant for the pair) and Np corresponds to Ep (units of variation in which both manuscripts of the pair have the same reading) in the similarity score used by A. Alekseev, and while A. Alekseev is always looking for a pair with the highest similarity score (Sp ), E. Wattel is looking for a pair with the minimal distance (Dp ). Tests with the Slavonic Gospels have shown that both formulas generally give the same results, and the pair with the highest similarity score has the smallest distance. The divergence of the algorithms starts after the first step has been made, and the first pair has been found. As was described above, the algorithm of E. Kuznetsova extracts both members of the pair from the database, and keeps on looking for the next closest pair. The algorithm of E. Wattel substitutes one new symbol for both members of the pair in every version formula of the database, adjusts the weights of all formulas, and only then starts looking for the next pair with minimal distance (the procedure is described in detail in Wattel this volume). . Stages of classification and data presentation The outcome is presented as a stemma (the procedure of drawing a stemma can be found in Wattel e.a. 1996). The algorithm stops when there are only three manuscripts left, if the database turns out to be empty at a certain stage, or if all version formulas are trivial. In the latter two cases “the pairs could be squeezed in a random way to get a fully resolved tree” (Wattel this volume). Just like the first matrix of A. Alekseev is not final, and is used to detect identical manuscripts and substitute them by one representative, so the pilot stemma of E. Wattel is the basic tree, which can be refined. On the way to obtaining the final classification, A. Alekseev repeats the clustering algorithm, while E. Wattel designes a new algorithm of stepwise refinement of the initial tree. Every step consists of detaching a branch from the pilot tree, attaching it elsewhere in the tree, and checking whether the new tree is a better reflection of the information from the version formulas. So far we have been dealing with trees, where terminal nodes represented extant witnesses, and intermediate nodes hypothetical witnesses. Every intermediate node had precisely three connections to other nodes. The latter restriction was caused by the need to reduce the computational complexity. In the final stage of classification E. Wattel designs a tree without this type of restriction. After the best initial tree has been obtained, the contraction phase begins (the procedure is described in Wattel e.a. 1996: 137–138, 157–161). It means that some internal nodes have a number of

The study of the Gospels in Slavonic 

connections larger than three, and some extant witnesses become intermediate. This is a type of tree presented in Appendix 2 (below) which is the final stemma 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 our view on the presentation of results. Remembering that the stemma of E. Wattel doesn’t show such relations as in Figure 1, we must refrain from drawing it for a contaminated tradition. Nevertheless, if we regard the stemma from a different position, and consider it only as a representation of several clusters, then we can compare it with the matrix of clusters on equal grounds. In doing so we should look for correspondence between clusters in the matrix and subtrees in the stemma. We cannot interpret the edges connecting manuscripts in the stemma, because they are meaningless for a text type, but we can interpret every group of nodes which were joined into the tree at a certain stage. To do that we need to introduce one definition. We shall call a subtree of a hypothetical node n (or a subtree n) a graph with the node n and all the edges incident to n with all their incident nodes, which have a number smaller than n (we shall consider any end node or other node corresponding to an extant witness smaller than any other node). Let us look at the final stemma in Appendix 2. The number of a hypothetical node equals the number of the step at which next manuscript joined the tree or two or more subtrees were joined together. Therefore, the lower the value of n, the smaller the distance between the end nodes of n. Subtree 37 is the stemma itself. It is made up of subtrees 36, 34, 29 and 23. Each of them is in its turn made up by two or more subtrees. We expect to find such subtrees where all or most end nodes are manuscripts belonging to one recension. Such a subtree can be compared to clusters in the matrix. Thus we can see that the given stemmatological method itself doesn’t contradict the principles of the study of the Old Church Slavonic Gospels, and can be applied to this tradition. The only real restriction to its employment is that not more than 250 manuscripts can be processed. We have to remember that the weight depends on the number of manuscripts sharing one variant, and the number of manuscripts in the version formula. Therefore if we process 250 manuscripts instead of e.g. 500 we reduce the reliability of weighting algorithm. . Disadvantages of the classification The main difference between the methods is in the possibility to trace mistakes. Should a mistake propagate into the final classification, it cannot be seen in the

 Dina Mironova

stemma, because the stemma doesn’t visualize information about the edge distances, and it could be seen in the matrix, because the matrix supplies similarity scores for all pairs. The latter presentation is verifiable. Meanwhile, if a mistake is made in the stemma, we do not know about it, unless another classification gives us a contradictory data. This is the drawback of the presentation of the method of Wattel.

. Results Let us consider the classification of 74 Old Church Slavonic New Testament witnesses (the list of witnesses is given in Appendix 1). This sample includes representatives of each recension, found during the processing of 531 manuscripts. The Old Text Type includes 17 oldest and most famous sources. The Preslav Text Type includes 21 older representatives, which are the core of the multiple group forming this recension. The New Liturgical Tetraevangelion is represented by four most characteristic witnesses. The Athonite Text Type includes the Ostrog Bible of 1581 (OB), the most authoritative printed edition, which strongly influenced manuscript production, and 27 representatives of the core, including several manuscripts, which form subgroups. A few almost identical manuscripts were taken to show how uniform the recension is. The Chudov New Testament Text Type comprises the Chudov New Testament itself and 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 results of the procedure of Alekseev (Appendix 3) and Wattel (Appendix 2) correspondingly. In Appendix 2 intermideate nodes 1–37 are virtual. The increase in number of these nodes corresponds to the decrease in the strength of ties between the manuscripts which it connects. We shall consider subtrees 23, 29, 34, 36, which are joined into the final tree in node 37. We start from node 23, which is the smallest number. Subtree 23 is formed by subtrees 22, 19, and 6. All the endnodes of subtrees 19 and 6 are Athonite manuscripts. Outside these two subtrees no more Athonite manuscripts are found. They number 28 in all. Endnode @A is the cluster 21, 286, OB, 60, 285, 296, 46, 64, 323, Q. These manuscripts are alike or almost alike. They form the first cluster in the table as well (numbers 1–28). It should be mentioned that the Athonite Text Type isn’t characterised by vivid distinctive textual 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

The study of the Gospels in Slavonic 

this text type. Examples of a few variants which characterize one or more text types are given in Figure 4. Peculiarities of the Athonite Text Type can be seen in lines 14.15a, 14.15b, 14.22a, 14.24b, 14.28a, 14.33a, 14.34 in Figure 4. Often the distinctive features of the Athonite Text Type overlap with those of other recensions, mostly the Old one. As we see here and shall observe again further on talking about other text types, in the majority of variation units more than one text types agree in a variant. Again we understand that only a combination of 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 are the same both in the table and in the tree. Both pictures have a group of 311, HH, II, 54. In the stemma one can see better that manuscripts 311 and HH stay more closely together. This group like other Athonite subgroups has but few characteristic readings. Specific of such subgroups is that all their members give the same reading in the majority of variation units. Cluster 1z, Ht, 42, 1p (let us call it 1z*) is also present in both pictures. Characteristic readings of this group for the most part coincide with the readings of the Athonite Text Type, and in certain cases those of the Old Text Type, including Os. In the table manuscript 185 joins this cluster, too. In the stemma it is positioned together with manuscripts 69, 352, 85, 355 (we shall call them 69*) and a subgroup a, 294, 113 (we shall call it a*). Judging from the figures in the table, manuscript 185 has a better position there than in the stemma. This needs to be verified in the text. As a matter of fact, this manuscript is lacunary. One pericope of the two considered was ommited. Therefore its affiliation with any subgroup is only approximate. The text shows that only in one variation unit (14.14–2) does manuscript 185 have a reading different from that of 1z*, this variation being an omission of the conjunction ‘and’, which often occurs spontaneously. In six cases 185 has a reading different from 69*, and in five cases a reading different from a*. In all other variation units found in the passage all these manuscripts have the same readings. These data are in favour of positioning 185 closer to 1z*. Now we pass on to group 69*. As one can see in the stemma, manuscripts 69, 85, 352, 355 do not form a separate group. They are clustered together with 185 and subtree 3 (subgroup a*). We have already discussed manuscript 185. Now we shall try to see whether the text will confirm stronger links within 69* compared to the links of 69* with other Athonite subgroups, first of all a*. In 12 variation units manuscripts 69, 85, 355, 352 are opposed to a* (in one of these cases we have a dichotomy 69, a* vs. 85, 355, 352). Only two cases position a* and 69* on the same side of a dichotomy. Accordingly, the output in the table is more trustworthy.

 Dina Mironova

i

Old Slavonic and both mean time. To distinguish the two synonyms in English we denoted them as time-1 and time-2. ii We give the second variant when an important witness has a variant different from the other representatives of the text type it belongs to. The sigla of the witnesses are given in brackets. In cases where there is no agreement about a variant among the witnesses of the given text type, both variants are presented without specification of the sources. iii spectre-1, spectre-2, spectre-3 are three synonyms for spectre in Old Slavonic. iv Old Slavonic grammar allows both prepositional and non-preposotional variants here without change of meaning.

Figure 4. Examples of variations, characterising different text types of the Old Church Slavonic New Testament, Mt. 14.14–14.34.

The study of the Gospels in Slavonic 

It is typical of all the Athonite subgroups that most of the time they share either the reading of the Athonite Text Type, or the reading of the Old Text Type, and rarely give a reading of their own. There is not a single occurrence of a reading in common for one of these subgroups and any other text type except for the Athonite and the Old ones. Coming to subtree 22 we see that its non-hypothetical nodes are the manuscripts from the Chudov New Testament Text Type. Note that in the table manuscripts Nkn and Pg have the similarity score of 96%, and their similarity to Cd is 91% and 89% respectively. This goes along with the stemma, where Nkn even happens to be antecedent to Pg (for which there is no obvious philological confirmation). This type of text shares a number of characteristic features of other recensions (cf. e.g. lines 14.15a, 14.15b, 14.33 in Figure 4). It also adds certain characteristic variants of its own, mostly lexical substitutions. The studied piece of text gave 10 such variants (e.g. lines 14.17a, 14.17b, 14.23a, 14.24 in Figure 4). It is interesting to compare the different positioning of this cluster in the table and in the stemma. In the stemma it is joined directly to the Athonite Text Type. This is quite logical, because these two texttypes are very close, as was mentioned in Section 1 above. In the matrix the cluster is positioned at the bottom, away from the rest. This can also be justified, because we are dealing with quite a special recension, a new translation of the New Testament. At the same time, in the matrix we can see that its percentage of similarity with the Athonite Text Type is slightly higher than with the other ones. Thereby we can see that the two different representations of A. Alekseev and E. Wattel supplement each other and show the relations between recensions at different angles. All the endnodes of subtree 29, which we are considering next, are manuscripts of the Preslav Text Type. There isn’t any Preslav manuscript which is not an end node of this subtree. In the table we also have only one cluster containing all the Preslav manuscripts of the selection (numbers 45–65). The Preslav Text Type has ten characteristic variants in this passage, most of which are lexical substitutions, and the majority of them are typical of this recension only, and not of any other (cf. e.g. lines 14.15a, 14.15b, 14.24, 14.26, 14.33, 14.34 in Figure 4). The readings do not speak in favour of pointing out subgroups among the representatives of this recension chosen for the sample. Therefore, although subtree 29 is made up by two subtrees, and these in turn consist of more subtrees, we shall regard them all as one group. We do not know whether the stemma gives a correct division into small groups, because the variations do not give information about them. Therefore we cannot rely on 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 representatives of the Old Text Type, both its core (manuscripts which were joined together at an early stage of the tree construction) and periphery (manuscripts which were joined together at a later stage of the tree construction). By the numbers of hypothetical nodes we can see that the manuscripts from the core in subtree 7 (Mr, Gl, Tp, Zg, Os) are rather close, and the manuscripts from the periphery in subtrees 25 and 30 (192, Ar, 122, As) are both far from the core (subtree 7) and from each other (25 is a rather high number for a virtual node), which doesn’t prevent them all from being part of the same subtree. Assemanus (As), for example, belongs to the core of the Old Text Type, but has so many individual features that computer classification often positions it at the periphery. Among the characteristic readings of this recension we can point to lines 14.15a, 14.15b, 14.22, 14.34 in Figure 4. Subtree 21 in the stemma also belongs to subtree 34, but four out of its six terminal nodes (1r, WW, 385, 387) in reality represent a different recension, the New Liturgical Tetraevangelion, and we shall deal with them later. In the matrix (Appendix 3) the Old Text Type forms two clusters: one in the middle (numbers 33–44), and a smaller one at the bottom (numbers 66– 69). They have more representatives, including also those which stay behind subtrees 7, 25 and 30 in the stemma (Appendix 2, see subtree 36).The first cluster is bigger than the core of the Old Text Type in the stemma: it also includes manuscripts As, 372, 122, 48, Mv, 225 (see nodes 32 and 8). The second cluster includes only 4 manuscripts (Kh, Ar, 192, Vr), and there are no more representatives of the Old Text Type outside these two clusters. This is important for the comparison of the two classifications, because in the stemma four manuscripts, which belong to the periphery of the Old Text Type, are not even to 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 the Old Text Type. Many witnesses are scattered in the stemma, and their position is a good illustration of A. Alekseev’s statement that the features characteristic of the Old Text Type are more difficult to describe than those of its descendants. This phenomenon is explained by the fact that the peculiarities of the archetypical text are more or less evenly distributed in all manuscripts, including later text recensions, whereas newly acquired features belong only to a part of the tradition (Evangelije 1998: 9). Such a special text type immediately reveals the differences between the methods, implicit in the difference between the similarity score and the distance function. We have already stated that the peripheral manuscripts often have a mixed text of more than one recension,

The study of the Gospels in Slavonic 

therefore they are not significant for the purpose of describing a recension and finding its best representatives. When the two classifications position peripheral manuscripts differently, it does not significantly influence the overall picture. It is a different matter when a core manuscript is mispositioned. Unfortunately this mistake occurs once in the stemma. Manuscript Mv is found at the periphery, in subtree 36, whereas its text proves that it is one of the core manuscripts of the Old Text Type, and it should be positioned together with the core witnesses. The New Liturgical Tetraevangelion (manuscripts 1r, WW, 285, 387) is based on the continuous form of the Old Text Type (see Section 1) therefore it is not surprising that these two recensions are so closely linked. The New Liturgical Tetraevangelion does have its typical readings (cf. e.g. lines 14.15b, 14.22, 14.28 in Figure 4), but so few that in the stemma it is mixed with the witnesses of the Old Text Type. In the table the New Liturgical Tetraevangelion forms a separate cluster (numbers 29–32), although it also comes very close to that of the Old Text. Manuscript Sav has very many specific readings of its own, and is never positioned in any group, although it belongs to the Old Text Type.

. Conclusion We have shown that a stemmatological method can be applied to the study of a contaminated tradition, if we regard the stemma from a specific angle. It is a useful consideration from the methodological viewpoint, for sometimes when a theory is not applicable to the material the methods employed by the theory can still be used. Since the tools for studying old manuscript traditions are limited, we shouldnot neglect this possibility. The main argument in favour of using all available tools for those who start a classification is that identical results mean that we are on the right path, while divergences will make the researcher check the data carefully (as was done with manuscript 185 above). The present research is the first attempt to apply a new method to the Slavonic Gospels. On the whole our conclusion is that the output of both methods is relatively the same. This happens because the input is the same, and because both methods use cluster analysis as a classification technique. Nevertheless, stemmatological orientation has a certain impact on the classification, causing differences in the output. After studying the differences we found the following.

 Dina Mironova

In some cases the text proves that the clustering was correct in the matrix and incorrect the stemma. This can be illustrated by the mispositioning of Mv in the stemma. There is no other evidence for this mispositioning. Moreover, it is not clear why it was mispositioned. But since the cluster analysis positioned it correctly, we must conclude that the variation material provided enough information. In other cases the clustering in the matrix is more general compared to the clustering in the tree, but the variants do not provide any evidence in support of the clustering of the tree. For example, in the stemma the Preslav Text Type is subdivided into subgroups which cannot be verified. When we examine the text in the witnesses of the suggested subgroups more closely, we do not find variants which oppose this subgroup to the rest. The explanation for this situation lies probably in the introduction of weight of a version formula. Firstly, the computational weight depends on the type of the formula (type 1, type 2, etc.).8 Since we cannot regard the stemma as a stemma, but deal only with the groups it provides, type 3 version formulas have the same value as type 2, and type 1 do not have to be considered at all. That is why the originally ascribed computational weight can entail mistakes in the end. Secondly, pair weight inside a version formula can lead to inaccuracies when not all the manuscripts are classified, because pair weight depends on the number of manuscripts sharing the same variant. These observations entail the following conclusions: a. the more precise method of E. Wattel can lead to a bigger divergence from reality in the case of a contaminated tradition (cf. the mispositioning of Mv, As, 185 in the stemma); b. it is not always possible to give a philological explanation for a more precise classification. We should remember that if there is no information in the text, this information cannot be obtained by calculations either. The fact that the results provided by both methods are almost identical allows us to choose between the two methods in the future. If a tradition under consideration contains less than 250 witnesses, it is desirable to use both methods to study it. If the tradition is bigger and is contaminated, then the method of A. Alekseev could be sufficient. If there is a possibility to check the result using the method of E. Wattel, this would be a valuable addition. Without the stemma we could have overlooked such cases as the closeness of manuscript 225 to the New Liturgical Tetraevangelion. The stemma was also a good visualisation of the relationship of the Chudov New Testament and the Athonite Text Types. But

The study of the Gospels in Slavonic 

in some cases it is important to confine oneself to rough results, because rough results are less likely to contradict the text.

Acknowledgements I would like to thank the Organizing Committee of the Stemmatology workshops for the financial assistance to attend the workshops, and Dr. Wattel for his 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 (continuous 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 there can be seen “genetic divergence, that is the divergence of readings, not merely in the order of their likeness, but in the sense in which they have actually arisen” (Greg 1927: 31–32). . Under the same text in a group of manuscripts here and further on we understand one recension or text type. . We use the terminology of E. Colwell, where a variation unit means “the length of the text wherein our manuscripts present at least two variant forms”. And “a variant (or variant reading) is one of the possible alternative readings which are found in a variation unit” (Colwell 1969: 97–100). . This was the case with two well-preserved 13th century Four Gospels with old text of the Gospel according to John (Evangelije 1998: 8). . The representatives preferred are: mss with the least lacunae, known in philology, with the least number of mechanical mistakes, with more correct grammar, with older text. . For the types of variants see Greg (1927: 18–21).

References Alekseev, A. (1986). “Opit tekstologicheskogo analiza slavianskogo Evangelia (Po spiskam bibliotek Bolgarii). (Textual analysis of Old Slavonic Gospels (the study of manuscripts from Bulgarian libraries).)” Starobulgarica, X, 3, 8–19. Alekseev, A. (1999). Tekstologija slavianskoj Biblii (Textual history of the Slavonic Bible). Köln. Weirmar. Wien.

 Dina Mironova

Alekseev, A. & E. Kuznetsova (1987). “EVM i problemi tekstologii drevneslavianskih tekstov. (Computer and the problems of textual criticism of the Old Slavonic texts)”. Lingvisticheskije zadachi i obrabotka dannikh na EVM, 111–120. Moscow. Colwell, E. (1969). Studies in Methodology in Textual Criticism of the New Testament. Leiden. Dees, A. (1975). “Sur une constellation de quatre manuscrits”. Mélanges de linguistique et de littérature offertes à Lein Geschiere, 1–9. Amsterdam. Evangelije ot Ioanna v slavianskoj traditsii. (The Gospel according to John in the Slavonic tradition.) (1998). St. Petersburg. Galloway, P. (1979). “Manuscript Filiation and Cluster Analysis: the Lai de l’Ombre Case”. La pratique des ordinateurs dans la critique des textes, 87–96. Paris. Garzaniti, M. (2001). Die altslawische Version der Evangelien. Köln. Weirmar. Wien. Langbroek, E., A. Roeleveld, & E. Wattel (2002). “Valentin and Namelos discover their parentage: Narrative elements in the family tree of an international medieval tale”. VU Amsterdam. Also this volume (Roeleveld et al.). Reenen, P. van & M. J. P. van Mulken (Eds.). (1996). Studies in stemmatology. Amsterdam: John Benjamins. Wattel, E. & M. J. P. van Mulken (1996). “Weighted Formal Support of a Pedigree”. In Van Reenen & Van Mulken (Eds.), 135–167. Wattel, E. (2002). “Constructing Initial Binary Trees in Stemmatology”. VU Amsterdam. This volume.

Appendix 1 The witnesses I. The Old Text Type Sigla Signature

Type of Book

Date

Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Concise Lectionary Concise Lectionary

C11 C11 earlyC12 1144 1056/57 C11

the core Zg Mr Tp Gl Os As Mv Ar

St.-Peterburg, RNB Glag. 1 (Zographensis) Moskva, RGB Grig. 6 (Marianus) Moskva, RGADA 381#1 (Typographensis) Moskva, GIM Sin. 404 (Galicianus) St.-Petersburg, RNB F.p.I.5 (Ostromir Evangelion) Vatican, Vat., Cod. Slav. (Assemanus) Belgrad, Narodnij Muzej 1538 and St.-Peterburg, RNB F.p.I.83 (Miroslav Evangelion) Moskva, RGB M.1666 (Archangel Evangelion)

Expanded Lectionary 1185 Concise Lectionary 1092

other manuscripts Sa Kh 122 Nkl

Moskva, RGADA 381#14, (Sava Codex) Odessa, OGNB 1/3 (Kokhno Gospel) St.-Peterburg, RNB Pog. 11 Dublin, Chester Beatty 23 (Nikolskoje Gospel)

Concise Lectionary Concise Lectionary Concise Lectionary Tetraevangelion

C11 C12 C12-13 C14

The study of the Gospels in Slavonic 

372 225 48 192 Vr

Moskva, GIM Uvar.480 Moskva, GIM Uvar.93 St.-Peterburg, RNB Q.I.924 Moskva, GIM Hlud. 16 Sofia, NBKM N19 (Vrachansko Gospel)

Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Concise Lectionary

C14 C15 C15 C14 C13

Tetraevangelion

c.1381

Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary

c. 1128 c. 1117 1164 1363

Tetraevangelion Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary

c.1356 C14 1362 1307 C14 1463 C14 C13 C14 C15

Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary

C15 C14 1409 C12-13 late C14

manuscript with features of different text-types Jv

Moskva, RGADA 381#2 (Javilovo Gospel)

II. The Preslav Text Type Ju Ms Dl B IA 293 376 366 43 118 K 418 75 257 Pv d 273 Tr 032 02 383

Moskva, GIM Sin. 1003 (Jury Evangelion) Moskva, GIM Sin. 1203 (Mstislav Evangelion) Moskva, RGB Rum. 103 (Dobrilovo Evangelion) Moskva, RGADA 381#8 London, the British Museum Add. 39627 (Evangelije Tsara Ivana-Alexandra) Moskva, RGB 304 III#1 Moskva, RGB Sof. 3 Moskva, GIM Sin. 740 St.-Peterburg, RNB F.p.I.9 St.-Peterburg, RNB Pog. 18 St.-Peterburg, RNB Sof.2 Moskva, RGB 256#106 St.-Peterburg, RNB Sof.5 Moskva, GIM Hlud. 170d St.-Peterburg, BAN 34.7.20 (Pivoavrov Evangelion) Moskva, NBMGU 2Bg45 Moskva, GIM Sin. 71 Moskva, GTGK 5348 St.-Peterburg, RNB F.p.I.64 Jaroslavl, JaIAMS 15690 (Spasskoje Evangelion) St.-Peterburg, RNB F.p.I.15

Expanded Lectionary mid C13 Expanded Lectionary C14

III. The Athonite Text Type Ob 21 286 60 285 296 46

A phototypic edition in 1988 (Ostrog Bible) Moskva, RGB 173#1 Moskva, RGB 304 I#46 St.-Peterburg, RNB Sof.27 Moskva, RGB 304 I#66 Moskva, RGB 299#538 St.-Peterburg, RNB Q.I.1198

Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion

1581 C15 1500 C16 C15 C15 C15

 Dina Mironova

64 323 Q HH 311 II 54 42 Ht 1p 1z 185 352 355 69 85 a 294 113 209 210

St.-Peterburg, RNB Sof.21 St.-Peterburg, BAN Celepi 40 Moskva, RGB 113#17 St.-Peterburg, RNB Q.p.I.4 Moskva, RGB 256#119 St.-Peterburg, RNB Q.p.I.2 Moskva, RGADA 201#8 St.-Peterburg, RNB F.p.I.12 Moskva, RGB 304 III#3 (Evangelije Hitrovo) Jaroslavl, JaIAMS 15569 St.-Peterburg, RNB F.p.I.109 Moskva, GIM Tchert. 81 St.-Peterburg, BAN 32.13.7 St.-Peterburg, BAN 32.13.27 Moskva, RGB Sof. 14 St.-Peterburg, RNB F.I.27 Moskva, NBMGU Vetk. 339 Moskva, RGB 299#706 St.-Peterburg, RNB Pog. 23 Moskva, GIM Uv. 745 Moskva, GIM Uv. 704

Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion Expanded Lectionary Tetraevangelion Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Expanded Lectionary Tetraevangelion Tetraevangelion Tetraevangelion Expanded Lectionary Expanded Lectionary

C16 C16 C15 C15 C15 C14 C15 C14 C14-15 C14 C14 C14 C16 C16 C16 C16 C15 C15-16 C15-16 C16 C16

Tetraevangelion Tetraevangelion Tetraevangelion Tetraevangelion

C13-14 C14 C14 C14

Tetraevangelion

C14

Tetraevangelion Tetraevangelion

C14. 1399

IV. The New Liturgical Tetraevangelion WW 385 1r 387

Athos, RMPA Slav. 2 Moskva, RGB 178#891 St.-Peterburg, RNB Q.p.I.44 St.-Peterburg, RNB Gilf. 18

V. The Chudov New Testament Text Type Cd

a phototypic edition in 1892 (the Chudov New Testament) Pg St.-Peterburg, RNB Pog. 21 Nkn Moskva, RGB 304 III#6 (Nikonov Evangelion)

The study of the Gospels in Slavonic 

Appendix 2 Stemma for 74 witnesses of the Old Church Slavonic New Testament using the method of E. Wattel

376 d

Tr

366

D1 383 032 257 02 12 273 Pv 118

48 Nk1

27

Kh

31

20

14

28

Mv

43 75 418 B1t 293 Ju Ms K

5

10

11

15

2

35

33

30

122 As

25

21

23

24

7

6

22

Cd

Nkm

13 8 372 Mr G1 Tp Zg Os Pg

lr WW 225 387 385

1

19

34

mm

1z Ht 42 1p

9

17

37

36

192 Ar

311 HH

II 54

@2 4

29

32

Vr

64 @A

3

185 69 352 85 355

a 294 113

 Dina Mironova

Appendix 3 Classification of 74 witnesses of the Old Church Slavonic New Testament using the method of A. Alexeev

The study of the Gospels in Slavonic 

Different kinds of tradition in Targum Jonathan to Isaiah Alberdina Houtman Theologische Universiteit, Kampen

. Prolegomena Targum Jonathan, an ancient Aramaic translation of the biblical books of the Prophets, has since the late 1980s been the subject of a large-scale research project at the Theological University of Kampen (the Netherlands).1 Within this project, as a matter of course, the question of the textual history of the text is also considered.2 This history is quite complicated. Internal and external evidence strongly suggests a long formative period before the text took its present shape, and the text as we have it now, contains at least two different kinds of traditions. The main text consists of a strictly edited explanatory translation which stays relatively close to the Hebrew source text, but at some points it is interspersed with small textual units of a quite different character. In this paper, I will examine whether both kinds of traditional material can be equally used for the construction of a stemma. First, I will give a short general introduction to the Targum genre, followed by some words on Targum Isaiah. Then the question of method will be dealt with. The pièce de résistance is the case of the so-called Tosefta Targums. It will be considered whether or not they can be used as kinship revealing variants.

.

Introduction3

In 586 before Common Era, the Babylonian king Nebuchadnezzar destroyed Jerusalem. In the event, several thousands of the inhabitants of the city-state of Judea 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 from family and compatriots, the Jews in exile had to find a new way of religious expression. While in spiritual matters they kept strictly to their own cultural heritage, in daily life they soon took over the vernacular of the Babylonian empire, which was Aramaic. Eventually even the Jews who remained in Palestine switched to the new lingua franca. As a result, by the 4th century BCE Aramaic had ousted Hebrew as the daily language of many Jews. Since religious legislation prescribed that the Bible could only be read in the Holy tongue, which is Hebrew, a habit developed of translating the weekly Scripture lesson into Aramaic after its liturgical reading in Hebrew. In the schools Aramaic Bible translations were used to teach the children Hebrew, and at home the translations were used for private preparation for the lesson in the synagogue. These Aramaic Bible translations are called Targums. In Palestine they were originally not standardized, witness the occurrence of several different translations of the Torah. At a certain point in time, however, one of these Palestinian Targums found its way to Babylon and was there edited and adapted to the local needs and promoted as the one official Targum. In the 5th to 6th century, this revised Targum probably found its way back to Palestine and was supplemented with old local traditions. These additions are generally called Tosefta Targums,4 Tosefta being the Aramaic word for “addition”. In fact, it must be said that about this last stage the opinions differ.5 The liturgical practice of Targum passed out of use in the Middle Ages when as a result of the Islamic conquests the vernacular of the oriental Jews became Arabic. Only in Yemen the tradition survives to this day. From the Targum on the Torah there remain some Palestinian versions beside the official Babylonian version. From the Targum on the Prophets however, only the standardized Babylonian version, which is called Targum Jonathan, is preserved. The mentioned additions to the official Targum may or may not be the remnants of an older Palestinian tradition.

. Targum Isaiah There is hardly a biblical book that has been more popular through the ages then the book of Isaiah. The rabbinical literature is replete with references to Isaiah, and from the oldest times the liturgical readings of the Prophets, the so-called haftarot, which accompanied the Torah reading in the synagogues, were to a large extent taken from Isaiah.6 It is, therefore, no wonder that a comparatively 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 of varying 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 hundreds of fragmentary manuscripts and collections of haftarot. Of the 26 known continuous manuscripts, 22 are such that they can be used for stemmatological research.7 Generally, Targum text witnesses can be classified into an Eastern and a Western tradition. The Eastern tradition is quite homogeneous. It consists of the oldest Babylonian tradition – of which unfortunately only very little remains – and the Yemenite tradition. On the level of consonant readings, these two strands are very close. The Western tradition is more heterogeneous, but broadly speaking it can be subdivided into an Ashkenazi8 tradition and a Sefardi9 tradition.

. Method The task at hand is to classify the extant text witnesses into a pedigree to enable the study of the historical development, something which is easier said than done with a large book that has survived in so many textual witnesses. We therefore have to look for a way to construct a reliable stemma without having to scrutinize all the textual material first. A possible way to do this is to carry out a sample survey.10 The sample has to meet two conditions. Firstly, it has to be representative of the whole manuscript. Secondly, the sample has to be unbiased. If we are interested in discovering the source of a certain manuscript, it is better to keep away from theologically hot items, because in those cases a copyist might be inclined to tamper with his sources. In the same manner, beginnings and endings of textual units have also proved to be the object of later rewriting more often than the core of the same units. For our research, the liturgical readings, the so-called haftarot, are the main textual units.11 A large percentage of the expansive variant traditions, which are generally called Tosefta Targums, are found in the introductions or the terminations of the haftarot,12 so we have to take a verse somewhere from the middle of the haftarot. Fortunately, the haftarot are fairly evenly distributed over the book of Isaiah. Only twelve out of the sixty-six chapters of the book do not contain haftarah readings. From those chapters a verse from the middle was taken. The total sample, therefore, included 66 verses, one from each chapter.

 Alberdina Houtman

The sample verses being selected, the next step was the selection of the textual witnesses to be included in the research. The continuous text tradition looked like a feasible start, using all the complete continuous manuscripts mentioned above, supplemented with some early printed editions. As all stemmatologists know very well, some tedious groundwork has to be done before one can start with any stemmatological construction, namely the collation of the witnesses. For this part of the process, I used the program Collate, which was designed and produced by Peter Robinson.13 This sophisticated program is a great help in the dull job of collecting variant readings, although one still has to transcribe the manuscripts first. The variant readings that were registered in this way were evaluated with regard to their relevance. Using the possibility of data abstraction, most orthographic variants can be filtered out automatically. Nevertheless, a number always succeed in slipping through the net. In those cases the formulas were adjusted manually. Variant readings that proved to be obvious errors were removed from the formulas. Not every variant is equally important. The formulas were therefore given a weight factor expressing their significance for the establishment of a stemma. I counted characteristic orthographic variations such as different spellings of loanwords for 0.5, minor syntactical changes for 1, substitutions, additions, and omissions (when evidently not due to reading errors) for 2. The reason for weighing is obvious. If the technique of the copyist consisted of reading a certain part of the text and then copying it from memory, it is very well possible that he changed the orthography unconsciously. It is even conceivable that he changed it consciously. If he knew the word in a different spelling and thought his exemplar inaccurate, he might be tempted to correct it. On the other hand, the deviating orthography may also have been part of the exemplar. So, if two or more witnesses share the same deviating orthography, it cannot be rejected altogether as evidence for relationship.14 The same holds true to a lesser degree for minor syntactical changes. Conversely, substitutions, additions, and omissions are in considerable measure kinship revealing. This kind of variation must therefore weigh more heavily in the construction of a stemma than the other kinds. However, although the choice for weighing is clear, the determination of the weight factor is not. It must be admitted that this factor is a contrived choice, because it is impossible to determine in exact measure how much more important one kind of variant is than the other. For the construction of the tree, I adopted the Three Step Method that was developed in the 70s by Antonij Dees (Dees 1975, 1976, 1977). According to this method, in the first step the witnesses are clustered into subfamilies on purely quantitative grounds. In the second step, witnesses that might have been

Different kinds of tradition in Targum Jonathan to Isaiah 

intermediary in the process of transmission are identified.15 These two steps produce the chain of relationships that underlies the genealogical tree. At this stage, the nature of the relationships between the different members is settled, but not as yet their direction. This must be determined at the third step, where the point of suspension, i.e. the root of the tree, is established on the basis of qualitative arguments. This last step is the most difficult one, and all possible means must be employed to arrive at a well-founded decision, such as: assessment of the origins of the variants, palaeographical and codicological data, and historical information. In recent years, the Three Step Method has profited greatly from the development of a computer program that takes care of the first two steps. This program was designed and developed by Evert Wattel of the Department for Mathematics and Information Science of the Free University of Amsterdam.16 Thanks to this program, one can now work with large amounts of data. The program has been tested extensively in several projects and has proved its worth.17 What remains for the philologist is the challenge of evaluating the output and interpreting it in the light of circumstantial data. Wattel processed the data concerned in different ways and produced the underlying structure of the stemma. The outcome confirmed on the whole what was known from previous research on other biblical books, i.e. a clear division between Eastern and Western textual witnesses and a subdivision of the Western witnesses into an Ashkenazi and a Sefardi branch. This grouping was used as the point of comparison for the present research.

. Tosefta Targums Whereas in general the text of Targum Jonathan is remarkably stable, the seemingly random occurrence of Tosefta Targums is an intriguing phenomenon. Although Targum Jonathan is more of a paraphrase than a translation, it still keeps close to the Hebrew original. If one knows the theological premises of the translators and the hermeneutic rules they used to interpret the biblical text, it is nearly always possible to reconstruct the process that led to the given paraphrase. At some places, however, the characteristic style is broken by digressions that are only tangentially related to the text. These digressions occur mostly within the so-called Tosefta Targums. The following example may give an impression of the nature of the different kinds of tradition (Diagram 1). The version of Targum Jonathan still clearly reflects the Hebrew text, albeit with some changes. Instead of the Hebrew rendering that says that it was in the

 Alberdina Houtman

Diagram 1. Isa 6:1 Hebrew Bible

Isa 6:1 Targum Jonathan

In the year that King Uzziah died,

In the year that King Uzziah was stricken with leprosy,

I beheld my Lord seated on a high and lofty throne; and the skirts of His robe filled the Temple. 1

Isa 6:1 Tosefta Targum1

In the year that King Uzziah died, that means the year in which he was stricken with leprosy. For there are four who in their lives are considered as dead while until now they hold out: one who is stricken with leprosy; one whose eyes are blinded; and who has no sons; and one who went down from the strength of his property. The prophet said the prophet, Isaiah said, I beheld the glory of the I beheld the glory of the Lord Lord seated on a throne, high seated on His throne, high and elevated and elevated in the in the highest heavens; highest heavens; and the Temple was and the Temple was filled with the filled with the splendor splendor of His glory. of His glory.

According to MS Hébreu 75, Bibliothèque Nationale, Paris

year that King Uzziah died, Targum Jonathan contends that it was the year in which he was stricken with leprosy.18 Secondly, Targum Jonathan paraphrases part of the text in order to keep off an inexpedient anthropomorphic understanding of God. Thirdly, the impression that Isaiah actually saw the Lord had to be avoided. This was considered theologically incorrect because in Exod. 33:20 it says, “man may not see Me and live.” In the Tosefta Targum an explanation is given for the translation of “dead” as “stricken with leprosy.” To this end, it cites a rabbinic tradition that is also known from other sources.19 This specific Tosefta Targum is inserted immediately after the translation of Targum Jonathan, without any label. In another manuscript the expansive version is given as the first translation, while the short version of Targum Jonathan is given afterwards, indicated as “another translation”.20 This is one of the strange features of the Tosefta Targums. They are scattered throughout the manuscript traditions, where they may be inserted in the text at the appropriate place, indicated as Tosefta Targum or the like, or not indicated at all. At other times they are written in the margin, either by the first or a second hand, or gathered together at the end of the main text. There also existed special collections of Tosefta Targums (Bernstein 1986: 151; Klein 1986: 410; Gleßmer 1995: 165).

Different kinds of tradition in Targum Jonathan to Isaiah 

Since the Tosefta Targums are not found in the oldest strands of the textual tradition, they apparently consist of extraneous material that was added to the common text of Targum Jonathan, hence the name Tosefta, which means addition. In some manuscripts they are indicated as Targum Jerushalmi, i.e. “Targum of Jerusalem”, which points to a Palestinian origin. However, both the contents of these additions,21 as well as their language,22 often reflect Babylonian influence. These facts apparently contradict each other. The occurrence of Toseftot is not restricted to a specific branch of the genealogical tree. They occur in all textual traditions, except in the Babylonian tradition (Kasher 1996: 60–62). There are, however, Toseftot that only occur in certain geographic areas (Kasher 1996: 60). Many of the Tosefta Targums occur within the haftarah readings, especially within the festive portions, and may thus attest to the instruction heard by the congregation in the synagogue (Kasher 1996: 16–18). Moreover, a significant part of them occurs at the beginnings and endings of the haftarah readings.23 Since beginnings and endings of textual units have been found to be the object of later rewriting more often than the core of the same units, this makes them unfit for stemmatological purposes (Den Hollander 1997: 138). On the other hand, being the most substantial variants in an otherwise quite stable text tradition, it is very tempting to use them. Willem Smelik, in his book The Targum of Judges, employed them for stemmatological purposes, and they seemed to confirm the results that were based on other criteria (Smelik 1995: 129–153). Therefore, having a stemma in hand that was based on an unbiased sample of the witnesses (Houtman 1999a), I decided to test the value of Tosefta Targums as kinship revealing variants. To that end, I re-checked the continuous text witnesses 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 unbiased sample, to set a benchmark for comparison. The first group contains the Eastern tradition. The second group contains broadly the Sefardi tradition (I will return to this). The third group contains the Ashkenazi tradition, and the last group, finally, consists of manuscripts which cannot be convincingly placed in any of the other categories.

 Alberdina Houtman

Diagram 2. Eastern f Babylonian fragments v Or 2211 P Hébreu 1325 e Gaster 673 z Or 1474 L Lutzki 239 Y Qafih. 5

“Sefardi” N Solger b First Rabbinic Bible g Second Rabbinic Bible c H. 116 k Kaufmann A13 H Hébreu 75 O Opp. Add 40 76 r Villa Amil o Antwerp Polyglot

Ashkenazi Irregular Q Hébreu 18 p Hébreu 96 K Codex Reuchlinianus B Or. Fol. 2 J El. F.6 d Add. 26,879 F Urbinas 1 C B.H. V A Göttweig 11 D Parma 3188

Eastern tradition The Eastern manuscripts, encompassing the Babylonian fragments and the Yemenite manuscripts, make up a strong cluster, which contains two recognizable sub-clusters, the internal connections of which are considerably stronger than 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 tradition. The group that I have called Sefardi for convenience, contains one exceptional sub-group. This consists of MS Solger, the First Rabbinic Bible, and the Second Rabbinic Bible. It definitely belongs to the Western text tradition, but scrutiny of the place of this sub-group within the tradition does not disclose a clear kinship to either the Ashkenazi or the Sefardi tradition. Within the group of the Sefardi witnesses, H.116 and MS Kaufmann A13 can be discerned as a strong subgroup. The other subgroup is less coherent.

Ashkenazi manuscripts The third group consists of the Ashkenazi manuscripts. These manuscripts are by far not as closely mutually related as the manuscripts within the other two groups. They are distinguished from the other groups more through their difference than through a mutual likeness. Nevertheless, some connections can be observed. There seem to be two sub-groups, of which the first consists of

Different kinds of tradition in Targum Jonathan to Isaiah 

Codex Reuchlinianus, MS Hébreu 18, MS El f.6, and MS Add. 26,879. The second sub-group has very loose internal links. It consists of MS Urbinas 1, MS B.H. V, MS Göttweig 11, and MS Parma 3188.

Irregular manuscripts Finally there are two manuscripts that do not fit into one of the groups defined, namely MS Hébreu 96, and MS Or. Fol. 2. It is difficult to account for the place of MS Hébreu 96. Although it was probably written in Spain in a Sefardi script, it seems textually closer to the Yemenite than to the Sefardi text tradition. MS Or Fol. 2 has the outward appearance of an Ashkenazi manuscript, but its consonantal text does not conform to any of the groups defined.24 We will now compare these findings with the results of the collation of the Tosefta Targums. Two things need to be mentioned in advance. Firstly, no Babylonian fragments available contain the verses concerned.25 Therefore, they will not be found in the table. Secondly, I have not included Tosefta Targums that are known from the marginal readings of the Codex Reuchlinianus only, because these unique readings were taken from different sources that have not yet been identified.26 Including this material would therefore not add to the text genealogical information. So, restricting ourselves to Tosefta Targums that are not unique to Codex Reuchlinianus, we find eight cases. One of them, a Tosefta Targum to Isa 10:32 occurs in two quite different recensions.27 I have collated these separately, so that there are nine cases to be considered. See Diagram 3. What do these data tell us? If we first consider the Yemenite tradition, we see that it has no Tosefta Targums integrated in the text. In ‘v’, ‘e’, ‘z’, and ‘L’ there is a Tosefta Targum to 10:32b, but it is not part of the original text. In ‘v’, ‘z’, and ‘L’ it is added in the margin, while in ‘e’ it occurs as an annex on a separate page. Apparently the Yemenite tradition, like the Babylonian tradition, did not include Tosefta Targums originally. It is a known fact that the publication of the Rabbinic Bibles in the 16th century strongly influenced the Yemenite tradition. We see, accordingly, that the added material of the Yemenite MSS coincides with the First Rabbinic Bible (‘b’). The Sefardi tradition is characterized by the Tosefta Targums on Isa 6:1, 49:15, 49:24-25 and 50:10-11. ‘c’ is exceptional, since it has no Tosefta Targums at all.28 Except for ‘H’, none of the Sefardi manuscripts has a Tosefta Targum to Isa 10:32. Within the Sefardi tradition, ‘r’ and ‘o’ have the same occurrence of Tosefta Targums. This is not surprising, because ‘r’ was probably the ex-

 Alberdina Houtman

Diagram 3.

+ present – absent m present in margin ∼ text not available or illegible

emplar of ‘o’, which is the Antwerp Polyglot (Díez Merino 1994: 83–85; Stec 1994: 7–16). The First and Second Rabbinic Bibles (‘b’ and ‘g’) and ‘N’, which is the exemplar of the First Rabbinic Bible (Houtman 1999b: 191–202), share the same version of Isa 10:32. As already indicated above on the basis of the unbiased sample research, the place of this small group within the Western tradition is problematic. The pattern of Tosefta Targums, being dissimilar to the patterns of both the Sefardi and the Ashkenazi group, confirms this finding. The Second Rabbinic Bible is sometimes said to be only a slightly revised edition of the First Rabbinic Bible. We see, however, that it includes Tosefta Targums that were not present in the first edition. This indicates that the editor of the Second Rabbinic Bible, Jacob ben H . ayyim, who came as a refugee from Spain to Italy, incorporated the traditions of his home country. Five out of the eight Ashkenazi manuscripts share the same version of Isa 10:32, of which four also contain the Tosefta Targum to Isa 10:33.29 Obviously these two form a pair.30 It is remarkable that MS Or Fol. 2 (‘B’), which has the outward appearance of an Ashkenazi manuscript, also has this typically Ashkenazi combination of Isa 10:32a combined with a Tosefta Targum to Isa 10:33. This evidence may push it more strongly towards the Ashkenazi group. ‘F’ and ‘C’ share the otherwise unique Tosefta Targums to Isa 66:1-2 and 66:23. Although, except for Isa 66:1-2 in ‘C’, they are not incorporated in the running text, but are appended at the end of the text by the same hand as the basic text, they may be considered an integrated part of this tradition. ‘p’ is exceptional in its unique combination of the “Sefardi” Tosefta Targum to Isa 6:1 and a marginal reading of the Tosefta Targum to Isa 10:33. This

Different kinds of tradition in Targum Jonathan to Isaiah 

unusual combination confirms the results based on the unbiased sample, that this is probably a contaminated tradition. So far for the data that confirm the earlier results. There are, however, also elements that contradict the general tendencies: –

– –

The Ashkenazi manuscripts ‘Q’, ‘K’, and ‘A’ are extraordinary, because their basic text contains no Tosefta Targums at all. The same holds for the Sefardi manuscript ‘c’. In an automated comparison, these MSS would probably have been grouped with the Eastern tradition on the basis of the (original) absence of Tosefta Targums. The “Ashkenazi” version of Isa 10:32, occurs unexpectedly in the Sefardi manuscript ‘H’. On the other hand, the decidedly Ashkenazi manuscript ‘D’ contains a Tosefta Targum to Isa 6:1, which seemed to be a Sefardi characteristic.

. Conclusions What can we conclude concerning the use of Tosefta Targums as relationship revealing variants? Because of the mentioned contradictions, it seems not advisable to include the Tosefta Targums in the automatic procedure for the production of the underlying structure of a stemma. On the other hand, they are too important to be neglected. If used with care, they can be of aid in the reconstruction of the textual history. For example, in the case of the Second Rabbinic Bible, it seems that the editor, who generally followed the text of the First Rabbinic Bible slavishly, made an exception for the Tosefta Targums that were known to him from his home country. In cases of complete conformity of the occurrence of Tosefta Targums, a close relationship is probable, like between ‘N’ and ‘b’, and ‘r’ and ‘o’, where it is a matter of direct dependence. In the case of ‘F’ and ‘C’, there is no direct linear dependence, but the relationship between these manuscripts is certainly closer than between the other Ashkenazi manuscripts, which are very heterogeneous. The strange combination of Tosefta Targums in ‘p’ confirms the results of the unbiased sample research that this must be a contaminated tradition. The name Tosefta means ‘addition’. Perhaps that is how we should use them, as additional evidence in the reconstruction of the textual history. Within the Three Step Method, it should therefore belong rather to the third step than to the first two steps.

 Alberdina Houtman

Acknowledgments The investigations were supported by the Foundation for Research in the Field of Humanities, which is subsidized by the Netherlands Organization for Scientific Research.

Notes . The project includes the production of a bilingual concordance as well as a series of monographs: Bilingual Concordance (1995–); 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 member W. F. Smelik: Smelik (2002), and his contribution in the present volume. . The following overview is largely based on Alexander (1998, esp. 247–250). . The designation “Tosefta Targum” is somewhat biased and therefore misleading. But since this designation is commonly used in scholarly literature on the subject, I decided to adopt the term. . See the useful overview of previous research into date and provenance of the Tosefta Targums in Smelik (1995: 77–85). . Although the details of the history and the content of the lectionaries are still under debate, the broad outline is clear. According to Wacholder (1966), 25% of the haftarot in the annual cycle were taken from Isaiah; while in the triennial cycle almost half of the haftarot are from Isaiah (he counts 187 different haftarot including the festival readings, of which 93 are taken from Isaiah). According to Ch. Perrot (Perrot 141–143) the book of Isaiah is read on 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 form of 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. M. Metzger (Metzger 1992: 181). . The division into chapters is a rather late Christian invention, and therefore irrelevant for our purposes. Bishop Stephan Langton introduced the division at the beginning of the 13th century in the Latin Bibles. Gradually this division also entered the Jewish Bibles. See Ginsburg (1897: 25–31). . See Kasher (1985: 76) and Kasher (1996: 17). A. Shinan noticed the same phenomenon for the weekly reading of the Torah, the sedarim. See Shinan (1987: 106–107) and Shinan (1992: 26–31). . Robinson (1994).

Different kinds of tradition in Targum Jonathan to Isaiah  . 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 some other MSS. . See e.g. Wattel (1996). . E.g. Van Mulken (1993); Den Hollander (1997); De Visser-van Terwisga (1995–1999); see also the contribution of A. Roeleveld, E. Langbroek, and E. Wattel in the present volume. . The Aramaic does not use the word “leprosy”, but the euphemistic expression “he was stricken with it”. This expression is generally used in connection with leprosy. The employed verb is related to the Hebrew word that was used in II Chron. 26:20, where the story of Uzziah’s leprosy is related. The story concludes there with the words “for the Lord had struck him”. The reason that the translator chose not to translate the Hebrew literally was that according to the rabbinic tradition this verse describes the beginning of Isaiah’s prophetic activities. Since it is known that Isaiah prophesied during the reign of Uzziah, he could not possibly have had his first prophetic vision on the day Uzziah died. This is an example of harmonization of Scripture, which is a regular principle in the Targums. . E.g. Babylonian Talmud, Nedarim 64b. . MS Hébreu 96, Bibliothèque Nationale, Paris. . Bacher (1874: 4); Epstein (1895: 50). . See Tal (1975: 191–200); Kasher (1996: 14–16). . See above, n. 12. . Here as elsewhere, this situation may be restricted to only part of the manuscript. For the Book of Judges, for instance, Smelik has shown a relationship of this manuscript with Codex Solger, MS El. f.6 and MS Hébreu 18. See Smelik (1995: 142–147). Also Kasher classifies it as an Ashkenazi manuscript. See Kasher (1996: 303). . So unfortunately, we cannot check Kasher’s assertion that this tradition does not contain Tosefta Targums for the case of Isaiah. . For the unique character of the glosses in the Codex Reuchlinianus, see Smelik (1995: 170–175). . Actually there are three recensions of which one occurs in the Codex Reuchlinianus only. See Kasher (1996: 151–155). . I.e. for Targum Isaiah. Some of the other books of the Prophets do contain Tosefta Targums. See Luzzatto (1844: 132–137); Sperber (IVb: 139–140). . Were we to include MS Or. Fol. 2 on the basis of its outward appearance as an Ashkenazi manuscript, 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. (1995– . . . ) Project Director Johannes C. de Moor, Chief Editors Willem F. Smelik and A. Houtman. 20 vols. to date. Leiden: Brill. Alexander, P. S. (1988). “Jewish Aramaic Translations of Hebrew Scriptures”. In M. J. Mulder & H. Sysling (Eds.), Mikra: Text, Translation, Reading and Interpretation of the Hebrew Bible in Ancient Judaism and Early Christianity (pp. 217–253). Assen: Van Gorcum; Philadelphia: Fortress. Bacher, W. (1874). “Kritische Untersuchungen zum Prophetentargum”. Zeitschrift der deutschen morgenländischen Gesellschaft, 28, 1–72. Bernstein, M. J. (1986). “A New Manuscript of Tosefta Targum”. In Proceedings of the Ninth World Congress of Jewish Studies, Jerusalem, August 4–12, 1985, 4 vols., I (pp. 151–158). Jerusalem: World Union of Jewish Studies. Dees, A. (1975). “Sur une constellation de quatre manuscrits”. In A. Dees et al. (Eds.), Mélanges de linguistique et de littérature offerts à Lein Geschiere par ses amis, collèges et élèves (pp. 1–9). Amsterdam: Rodopi. Dees, A. (1976). “Considérations théoriques sur la tradition manuscrite du lai de l’Ombre”. Neoplilologus, 60, 481–504. Dees, A. (1977). “Over stambomen en handschriften”. Forum der Letteren, 18, 63–78. Díez Merino, L. (1994). “Targum Manuscripts and Critical Editions”. In D. R. G. Beattie & M. J. McNamara (Eds.), The Aramaic Bible; Targums in their Historical Context (pp. 51–91). Sheffield: JSOT Press. Epstein, A. (1895). “Tosefta du Targoum Yerouschalmi”. Revue des Etudes Juives, 30, 45–51. Ginsburg, Ch. D. (1897). Introduction to the Massoretico-Critical Edition of the Hebrew Bible, photostatic reprint of edn. London 1897. New York: Ktav. Gleßmer, U. (1995). Einleitung in die Targume zum Pentateuch. Tübingen: Mohr. Hollander, A. A. den (1997). De Nederlandse bijbelvertalingen 1522–1545. Nieuwkoop: De Graaf. Houtman, A. (1999a). “Textual Tradition of Targum Jonathan to Isaiah”. In J. Targona Borrás & A. Sáenz-Badillos (Eds.), Jewish Studies at the Turn of the 20th Century, 2 vols., I (pp. 145–153). Leiden: Brill. Houtman, A. (1999b). “Targum Isaiah According to Felix Pratensis”. Journal for the Aramaic Bible, 1(2), 191–202. Houtman, A. (2000). “Planning a New Targum Edition: Look before You Leap”. Journal for the Aramaic Bible, 2(2), 213–231. Kasher, R. (1986). “The Aramaic Targumim and their Sitz im Leben”. In Proceedings of the Ninth World Congress of Jewish Studies, Jerusalem, August 4–12, 1985, 4 vols., I (pp. 75–85). Jerusalem: World Union of Jewish Studies. Kasher, R. (1996). Targumic Toseftot to the Prophets. Jerusalem: World Union of Jewish Studies. Klein, M. L. (1992). “Targumic Toseftot from the Cairo Genizah”. In D. Muños León (Ed.), Salvación en la Palabra: Targum–Derash–Berith: En memoria del profesor Alejandro Díez Macho (pp. 409–418). Madrid: Ediciones Cristiandad.

Different kinds of tradition in Targum Jonathan to Isaiah 

Luzzatto, S. D. (1844). “Nachträgliches über die Thargumim”. Wissenschaftliche Zeitschrift für jüdische Theologie, 5, 124–137. Metzger, M. (1992). The Text of the New Testament: Its Transmission, Corruption, and Restoration (3rd ed.). Oxford: At the Clarendon Press. Mulken, M. van (1993). The Manuscript Tradition of Chrétien de Troyes: A Stemmatological and Dialectological Approach. (Doct. Thesis Vrije Universiteit). Amsterdam. Perrot, Ch. “The Reading of the Bible in the Ancient Synagogue”. In Mulder & Sysling (Eds.), Mikra (pp. 137–159). Robinson, P. M. W. (1994). Collate: Interactive Collation of Large Textual Traditions, Version 2. Oxford: University Centre for Humanities Computing. Shinan, A. (1987). “Sermons, Targums, and the Reading from Scriptures in the Ancient Synagogue”. In L. I. Levine (Ed.), The Synagogue in Late Antiquity (pp. 97–110). Philadelphia: American Schools for Oriental Research. Shinan, A. (1992). The Embroidered Targum: The Aggadah in Targum Pseudo-Jonathan of the Pentateuch. Jerusalem: Magnes. Smelik, W. F. (1995). The Targum of Judges. Leiden: Brill. Smelik, W. F. (2002). “How to Grow a Tree: Computerised Stemmatology and Variant Selection in Targum Studies”. In J. Cook (Ed.), Bible and Computer – Proceedings of the 6th AIBI Congress – Stellenbosch 17–21 July 2000 (pp. 495–518). Leiden: Brill. Sperber, A. (1992). The Bible in Aramaic, 4 vols., IVb (pp. 139–140) (2nd ed.). Leiden: Brill. Staalduine-Sulman, E. van (2002). The Targum of Samuel. Leiden: Brill. Stec, D. M. (1994). The Text of the Targum of Job. Leiden: Brill. Tal, A. (1975). The Language of the Targum of the Former Prophets and its Position within Aramaic Dialects. Tel Aviv: Tel Aviv University. Visser-van Terwisga, M. de (1995–1999). Histoire ancienne jusqu’à César, 2 vols. Orléans: Paragdigme. Wacholder, 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 College. Wattel, E. (1996). “Clustering in Stemmatological Trees: How to Handle a Large Number of Versions”. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 123–134). Amsterdam/Philadelphia: John Benjamins.

Valentin and Namelos discover their parentage Narrative elements in the family tree of an international medieval tale Annelies Roeleveld*, Erika Langbroek*, and Evert Wattel Amsterdam* / Vrije Universiteit Amsterdam

. Introduction During our explorations of Middle Low German texts we came across a tale which is traditonally called ‘Valentin and Namelos’ for short. Apart from two versions in Middle Low German, this tale has come down to us in three other languages, in verse and prose, and the more we looked into these different versions, the more the narrative and linguistic aspects and the family history of the tale started to intrigue us. Stemmatological research into the widely differing and not easily comparable renderings proved to involve feeling our way and testing the ground with each successive step and required the application of unique and innovative methods.

.

The tale

The story of Valentin and Namelos contains several narrative motives often found in medieval verse epics. The main characters are twin brothers who are unaware of each other’s existence, because they were left as foundlings. Valentin is found by a princess and grows up at court, Namelos (‘No-name’) is adopted by a she-wolf and grows up in the wilds. When they meet in a fight, a powerful bond instantly manifests itself. After a great many adventures they discover that they are brothers and find their royal parents. The fictional narra-

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

tive background is the courtly world of king Pippin of France, king Crisosmus of Hungary, and the battles against the (Muslim) heathens.

. The manuscripts The tale has come down to us in four languages/dialects, in verse and prose (we do not take into consideration the tale of “Valentin et Orson”, of which many versions in French and English exist in early prints, but which differ considerably from our tale). We are dealing with: N I and N II – Middle Dutch, 2 verse fragments of 176 lines each, c. 1350 N III – Middle Dutch, verse fragment of 395 lines, of which 16 are barely legible and 22 are completely illegible, 1340–1360 K – Middle Mid German, verse fragment of 52 lines, date unknown S – Middle Low German, verse, 2291 lines, c. 1450 H – Middle Low German, verse, 2613 lines, c. 1476–1481 B – Middle Mid German, prose, 1465 Z – 3 Old Swedish mss, prose, 16th century

This would seem to be a sufficient number of texts for the drawing up of a pedigree, and several scolars have tried to do so since the tale has been the subject of study (Seelmann 1884: xii f.; Beta 1907: 8; Karg 1924: 229; Dieperink 1933: 157; Langbroek & Roeleveld 1987: ii). Matters are much more complicated than they appear, however. a. S, H, B and Z have come down to us more or less complete, but S and H are in verse, B and Z are prose. This makes looking for variant readings difficult and speculative: it is not always possible to distinguish between readings produced by the constraints of rhyme and readings which vary for other reasons. b. Like S and H, manuscripts N I, II and III, and K are in verse, but only fragments of them survive, and unfortunately K does not cover the same ground as the Middle Dutch fragments. The situation is roughly as follows (see Diagram 1). 1–25 are the episodes of the story as defined by Dieperink (1933: 5–11). B and (apparently) N contain only 22 epsiodes; italics indicate verse; ≈ 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 detailed and lenghty manner, using 5 to 7 times as many lines of verse to relate the same episode as S or H, which makes it impossible to do anything like matching line for line. Prose recension B, while less expansive than the Middle Dutch fragments, is still much more elaborate than S, H and Z, and has unique elements and details. d. Not the least complicating factor is that four languages/dialects are involved: Middle Dutch, Middle Mid German, Middle Low German, and Old Swedish.

. The methodological approach In order to obtain a pedigree which reflects the historical connections between the recensions, we are using the stemmatological construction programmes described by Wattel and Van Mulken (1996a, b). These programmes require a database consisting of formulas, each of which describes a version difference between the recensions under consideration. A line of the database contains (1) a textual position, (2) a version difference, and (3) an optional weight factor for the version difference. Where no weight factor is given, a weight of 1.0 is assumed. While the database is being constructed, it is also possible to use category indicators instead of weight factors; weight factors can then be assigned to the categories afterwards and, if necessary, be changed without having to reconstruct the whole data base. When the database is finished, the programmes will supply a pedigree, a set of shock waves, and centrality scores for the manuscripts, as described by Wattel and Van Mulken (1996a, b). The stemma is the most important result for most purposes, but it must be borne in mind that a stemma is an abstrac-

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

tion, an instrument for the understanding the historical process, and not an end in itself. . The unusual text tradition of Valentin and Namelos Applying stemmatological methods to the Valentin and Namelos manuscripts turned out to be very complicated. So far, different texts have always been compared with respect to variant spellings, rhyme and metre, number of verses, choice of words, and grammatical structures. It was impossible to do this for our unusual set of texts, due to the substantial differences in language, length and style. However, in all the recensions the narrative does in fact develop along approximately the same lines, and identical story elements are found in all of them. It occurred to us, therefore, that we should attempt to compare not textual variants, but variants in story elements and narrative development. It turned out that this procedure yielded material just as suitable for mathematical processing as textual variants. . Classifying the material We restricted our variant apparatus to the following categories: category 1 2 3a 3b 4 5a 5b 6

classification important semantic differences minor semantic differences inversions over several lines of text inversions within one line or over adjacent lines important grammatical structures1 important interpolations or omissions minor interpolations or omissions differences in personal or place names

weight factor 3.0 2.0 2.2 1.2 1.0 2.1 1.1 1.5

Of great importance in computer aided stemma construction is the valuation of the version differences. Each of the categories mentioned was given a weight factor, which is listed in the last field of the table. Important differences were given a weight factor 1.0 larger than the corresponding minor differences. These valuation weights are usually a matter of personal experience and expertise, and of the text tradition involved, and small differences in weight will not influence the final stemma construction.

Valentin and Namelos discover their parentage 

If a version difference occurs uniformly through an entire section (as happens with personal names), a version formula is included in the data base two or three times, e.g. at the beginning, at the end, and sometimes in the middle of that section. When the database is finished, the type of version information will not influence the results: the programme does not discover that it is not dealing with the usual information on spelling, rhyme and metre, grammar, and choice of words. The programmes of Wattel and Van Mulken (1996a, b) are especially designed to be able to handle missing information. During the evaluation of a version formula (a line of the database) by the programmes, some recensions and pairs of recensions count positive (because they agree), some negative (because they do not), but the sums of weights and influences is always 0. Where an item is missing in one recension, it will have no influence, and the recension in question will end up in the intermediate range in the evaluation of the version formula for that item. In the end, the position of a recension in the pedigree will not be biased by its number of occurrences in the version formulas. Naturally the results will be more reliable for recensions which occur often in every section of the database for the text tradition. The most important type of information for every pedigree construction are the so-called ‘type 2’ version differences: the ab/cd type. These are version differences in which there are at least two different readings, with at least two manuscripts for each reading. This is also the case for the Valentin and Namelos texts, although many of our version formulas do not contain type 2 information. However, during the computer processing of the different parts of the data base the amount of information was sufficient.

. An illustration For the comparisons, we chose the episodes for which we have Middle Dutch fragments at our disposal, so that we had five renderings to work on: N, S, H, Z and B. Our methods and strategies in finding, formulating and assessing variants is exemplified in the list of variants we eventually drew up for the episode for which we had fragment N I at our disposal. Note that in H and Z the episode is split up: after an introductory report on the situation, it breaks off, to be continued much later on in the tale. In S the folio which should contain a similar introduction is missing, but the episode

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

must have been split up in the same way as in H and Z, for later on in the tale only the rest of the episode occurs. Lines: h: 649-58 + 1458-1535; s: 1202-1276; z: 40,22-27 + 92,6-96,12; b: 89,4090,27; n: 1-176 Number of lines: h 10 + 78; s 75; z 6 + 91; b 34; n 176 lines @ is ‘archetypical’ information (see below under 7). H 651 651 651 655 655 1458 1458 1460 1460 1460 1460 1460 1460 1460 1464 1464 1467 1481 1481 1481 1481 1490 1490 1490 1497 1497 1497 1500 1500 1504 1508 1508 1512 1512

versions hzn/b nb/hz hzb/n hz/bn b/n hsz/bn hsbn/z hsz/bn hs/z hsz/bn hszn/b hszn/b hsz/n hszb/n@ hszn/b hszb/n hsz/b hsz/b hs/zbn hszb/n hsz/bn hszb/n hszb/n hszn/b hszb/n hs/nzb hsbn/z hs/zbn hszb/n hsz/nb hszb/n hszb/n hszn/b hsz/n

type 2 5b 5a 1 2 1 5b 5a 5b 5a 1 5a 3a 5a 2 5a 2 5b 5b 2 2 5b 1 5b 5b 5b 5b 5b 5b 3a 5a 5b 5b 3a

description of difference G in love with Ph /ditto, wants to marry her - - - / G is a traitor - - - / G pulls Ph’s chin, she knocks out 3 of his teeth G feels love, Ph doesn’t notice / Ph doesn’t want him Ph doesn’t respond / hopes to be united with husband episode resumed (long interval) / episode unbroken - - - / G is a fool Ph stays loyal and constant / - - - - - / Ph refuses to become G’s lover - - - / G is angry, swears he will be revenged - - - / “Ph acts as if she were a nun” G says: Ph was not banished home unjustly / - - G says it much later / G says it here - - - / remark by Lica, G predicts she will regret it G brooding on revenge / G thinks of plan to kill Ph - - - / contents of plan; auctorial remark Lica in danger early one morning / one night G goes to Ph / G goes to Ph, finds her asleep he is holding the knife / - - stabs princess to death / accomplishes his design puts knife Ph’s hand cunningly / puts knife Ph’s hand - - - / G goes to bed, awaits daybreak, gets up G says he has been dreaming / king been dreaming - - - / G wants king and himself to go and investigate king is frightened / - - - - - / king gets up - - - / king gets dressed - - - / Ph is holding the knife in her hand - - - / Ph is still asleep G: Ph is guilty, evidence of knife: here / ditto: later G wakes Ph brutally / king cries out, Ph wakes - - - / Ph sees Lica is dead G asks how Ph could do such a thing / - - G asks question here / G asks question later

Valentin and Namelos discover their parentage 

1512 1525 1525 1525 1529 1529 1529 1529 1531 1533 1533 1533 1533

hs/zbn hsz/nb hszb/n@ hsz/bn hszb/n hsz/b hsz/b hszb/n hszb/n hzn/sb hz/n hsz/bn b/n

5b 2 5a 5b 5b 3a 1 5a 5b 5a 2 5b 2

G says Ph should be burned at the stake / - - Ph had the knife / Ph had the knife, so she should die - - - / G points out that wound is bleeding again G binds the knife into Ph’s hands / - - Ph is dragged about / - - Ph dragged about here / Ph dragged about earlier G drags Ph from room / ditto by hair, knights pity - - - / passage of further accusals and denials Ph is very sad / - - Ph is brought before the law / - - G brings Ph before court / G brings Ph into “the hall” - - - / the court sits king has case judged / king explains about murder: what punishment? 649-58 en 1458-1535 (i.e. through the whole episode): hszb/n 6 king of Arabia no name / king of Arabia is Saluber hszb/n 6 his daughter no name / his daughter is Lica hszn/b 6 Phila / Philomena, Philamena hszb/n 6 Gawyn / Glutes

We made similar lists of version differences for the episodes which are covered by 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 written in the same hand, and were once parts of one single manuscript, while N III was discovered elsewhere and was not from the same manuscript. At the time when fragment N III was found, it was thought to be from a different textual tradition than the other two fragments (De Vreese 1892: 145–146). De Vreese comes to his conclusion mainly on the grounds that the order of events is apparently different in this fragment: Namelos goes off to look for his wife before Valentin’s wedding, while in N II (and B, H, S and Z) she comes to look for him right at the end of the tale. As N III is only a fragment, no conclusions can be drawn from this apparent difference: there is nothing to prove that Namelos’ journey is not cancelled or given up prematurely for some reason or another; nor is it impossible that a similar episode also occurred in the text of which we only know the small fragments N I and N II. The facts that king Crisostomus had married again, and that horses and swords are mentioned by name, in N III only, were additional grounds for De Vreese’s conclusion. Again, we do not

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

know that these facts and names did not also occur in the text of which N I and N II are fragments. B, H, S, and Z are shorter and much shorter recensions of the tale, so no conclusions can be drawn from their accounts about whether or not the much more expansive Middle Dutch fragments belong to one single textual tradition. Our stemmatological investigations now seem to show, however, that the three 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 traditional manner, we did have a small passage where we might be able to compare S, H and K (verse), and Z (prose). Z appears to be or derive from a translation of a Middle Low German exemplar very similar to S and H (Dieperink 1933: 31ff., 152ff.; Seelmann 1884: XIIf.), and when we compared the 3 rhyming texts S, H and K (Langbroek & Roeleveld 1998a), we found that K probably also derives from a similar text. However, K is a fragment of only 52 lines and comparison with S and H shows that the translator or copyist eyeskipped 12 lines in the middle of the fragment. In addition, one text is prose while the others are verse, and the texts are in three different languages. We therefore had to do a great deal of puzzling over how to compare the differences fairly. In fact, we had to invent an additional version difference category. There were lines where the prose reading of Z was so different from the verse readings of H, S and K, that they could not actually be compared. With some diffidence, therefore, we introduced category 7: ‘single version information’ and gave it a weight of 2.0, comparable to that of minor semantic differences. Our variant apparatus now fell into to the following categories: category 1 2 3a 3b 4 5a 5b

classification important semanctic differences minor semantic differences inversions over several lines of text inversions within one line or over adjacent lines important grammatical structures2 important interpolations or omissions minor interpolations or omissions

weight factor 3.0 2.0 2.2 1.2 1.0 2.1 1.1

Valentin and Namelos discover their parentage 

6 7

differences in personal or place names single version information

1.5 2.0

The versions for a few lines will illustrate our procedure. * stands for a reading which cannot be compared. H 1186 1186 1186

versions hsk/z hs/k hs/kz

type 7 2 5b

1187 hsk/z

7

1187 1188 1188 1189 1189 1189 1190 1190 1190 1191 1192 1192 1194 1195 1195 1195 1195 1197 1197

1 7 3b 2 5b 1 5b 7 5b 5a 2 5a 5b 7 5b 7 1 2 2

hs/k hks/z h/k/s h/k/sz hsz/k h/k/sz hkz/s hks/z h/ks hks/z hks/z hks/z hk/sz hks/z h/ks hsk/z hs/k hsz/k h/ksz

description of difference al stille, also stille / * al stille / also stille do / - - e scherent, schumes / * e scherent / schumes dat har all blot, daz hor, al blot dat har / * dat har all blot / daz hor / al blot dat har dar / daz / doch was he / was in groter not/in grossir not var/in wyl groter war sach / sach ok sunder wan, al sunder wan / * sunder wan / al sunder wan cruce / røt kors stolte ritter / Falantin - - - / men ... art vil wol / wol wart des, wart / * wart des / wart gemeyt, bereyt / * gemeyt / bereyt do lerde he / he larte yn vp voten / up den voten, 2 føter

This procedure is, again, very unusual, but was also suggested by our unusual textual material, and the results agree perfectly with what could be expected without the help of modern stemmatological methods.

. Archetypical information There were a few places where we found so-called archetypical information, i.e. information which we judged to be close to a hypothetical original recension of the tale. Our criteria for the cases where we considered information to be archetypical were perhaps subjective, but by this time we knew our ma-

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

terial down to the smallest detail and our considerations cannot but have a certain validity. An example of archetypical information is the distribution of the crowns in B, and probably N. In N II the number of illegible lines is not large enough to have contained the news of Pipping’s death and the ensuing distribution of all the crowns, which takes place at this point in H, S and Z, but not in B. B puts the distribution of the crowns at the very end of the tale, by way of happy endings and rightful rewards for all concerned. This is surely the original and logical order of the narrative. Not surprisingly, we found archetypical information in the Middle Dutch fragments, and in one case in B, with their more elaborate recensions of the story, but our choice was never for more ample information only but also for more logical information. The archetypical variants function in the procedure as if they were found in a fictitious additional manuscript; the results position the ‘archetype’ in the pedigree.

. The automatic comparison and the results During the automatic comparison of the variant formulas over the text the programme executes several passes. . The central recension The first pass locates the theoretical graphical centre of the text tradition. This theoretical centre will be the recension which has the largest number of readings in common with the others and is in that respect central to the text tradition. This does not, usually, mean that it is an especially good recension or that it is close to the archetype, but the centre is a popular recension. In this case the centre turned out to be H, the recension in the Hamburg codex known as the ‘Hartebok’. Close to central is S, the recension in the Stockholm manuscript, 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 with all the others. At this stage the programme proceeds from the beginning of a recension and works towards the end. The resulting information is presented as a collection of graphical visualisations for each recension. Clearly, these graphs

Valentin and Namelos discover their parentage 

can only represent points were information is there to be obtained, and so they contain many gaps. The vertical variable in each separate graph ranges between +100% (for total agreement) to –100% (for total difference). For recensions which agree in about half of the version formulas (= lines of the database), the means has a score of 0%. The horizontal variable in these graph gives the numbers of the version formulas. The different sections of version formulas are separated by dotted lines. These sections correspond to the episodes of the tale for which we had the Middle Dutch and Middle Mid German fragments at our disposal (see for the full list of episodes Diagram 1 under 2).

B Z S H

N I 13 13 9 9 9

K N I 14 13 13 14 16 14 16 14 16

N III 19 19 22 22 22

N II 22 22 25 25 25

Recensions H, S, Z, B, + NI: version formulas 1-12, 89-137 Recensions H, S, Z + K: version formulas 13-88 Recensions H, S, Z, B + N III: version formulas 138-164 Recensions H, S, Z, B + N II: version formulas 165-199

In the description of the stemmatological construction programmes in Wattel and Van Mulken (1996), using the line numbers in the text was seen as the most obvious choice, but for the present text tradition with its huge differences between lengthy and concise recensions it makes more sense to use the line numbers in the database, i.e. the version formula numbers. This alternative has no influence on the passes of the programmes. At this stage the behaviour of the graphs makes it possible to locate the corrupted and contaminated parts of the text, and spot positions where possible differences or switches occur in different exemplars as to quires or chapters. The collection of graphs for recension H is presented in Figure 1, and the graphs 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 where an N recension exists. However, in one position H is closer to N than to S: at the end of section N I. From Figure 2 we can conclude that B is the closest neighbour of N, but the similarity between H and S is much stronger than the similarity between N and B: the graph for the comparison of H to S (in Figure 1) dips less often and less low than that for the comparison of N to B (in Figure 2).

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel with ‘NI

variants with ‘K

with ‘NI

with ‘NIII

with ‘NII 100% max –88% mean –100% min

‘N/‘H Comparing ‘H to ‘N

100% max –48% mean –100% min

‘B/‘H Comparing ‘H to ‘B ‘@/‘H

–100% mean –100% min

Comparing ‘H to ‘@ ‘K/‘H

–100% mean –100% min

Comparing ‘H to ‘K

Comparing ‘H to ‘Z

100% max 68% mean –100% min

Comparing ‘H to ‘S

100% max 80% mean –100% min

‘Z/‘H

‘S/‘H 0

10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200

Figure 1. Distance distributions for H with ‘NI

variants with ‘K

with ‘NI

‘B/‘N Comparing ‘N to ‘B ‘@/‘N

with ‘NIII

with ‘NII

100% max 63% mean –100% min 100% max 100% mean

Comparing ‘N to ‘@ ‘K/‘N

‘Z/‘N Comparing ‘N to ‘Z ‘S/‘N Comparing ‘N to ‘H ‘H/‘N Comparing ‘N to ‘S 0

100% max –62% mean –100% min –100% mean –100% min 100% max –88% mean –100% min

10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200

Figure 2. Distance distributions for N I, N II and N III

Valentin and Namelos discover their parentage 

There is no information about the silimarity of K and N, as the investigated sections do not overlap at any point. . The shock waves In the next pass of the programme the ‘shock waves’ are constructed (Wattel & Van Mulken 1996). These shock waves are low at positions where the interrelations between the recensions are stable, i.e. where most or all recensions agree with at least one other recension without switching from one to another. The shock waves peak at positions where corruptions, contaminations and shifts occur: the interrelations between recensions are unstable, as recensions do not agree, or switch from agreeing with one to another of the other recensions. As our material was so unusual, we decided to execute this pass of the programme also and find out what information the shock waves could give us. Figure 3 gives the results of this pass. As 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 waves also disclose that recension B is not very stable throughout, and that between with ‘NI

variants with ‘K

with ‘NI

with ‘NIII

with ‘NII

‘N

‘B

‘@

‘Z

‘K

‘S

‘H

total 0

10

20

30

40

50

Figure 3. The shock waves

60

70

80

90 100 110 120 130 140 150 160 170 180 190 200

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel 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 K is rather unstable. In the version formulas we sometimes added a reading that gave us the impression that it was original (see under 7). We introduced these readings as a fictitious additional recension @, which was processed in exactly the same manner as the existing recensions, so that differences and shock waves were also obtained for @. Since all but one of these archetypical readings correspond to the readings in N, the graph in which @ is compared to N is high whenever it exists, and the shock wave on @ never peaks. . The initial trees In the next pass trees are constructed for the different sections.The simplest tree possible is the quadruple: . The section between version formulas 12-88 does not pose any problem. S and 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 one end, with Z close to this pair, and N and @ at the other end, while B positions itself somewhere in between (Diagram 3).

. The final pedigree The two trees (Diagram 2) and (Diagram 3) in 8.4 have now to be merged into one single pedigree. If we put H at one end and N at the other end of the text tradition, 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).

Valentin and Namelos discover their parentage  N

H S

B @

Diagram 4. H and N at the respective ends S

@

H

B

N

K Z

Diagram 5. Adding K and Z ‘@ 1

4 1 3

‘N

3 2 7

‘B

2 3

8 6

‘K

‘Z

1 2

‘H

2

‘S

Figure 4. A possible pedigree

However, we have to make two extra attachments: a link to K and a link to Z. It was not clear from our first results whether K should be attached closer to H or to Z: the quadruple H-S-K-Z remains unresolved in this combined pedigree. There may be a combined attachment to the line from N to H, which forks into Z and K (Diagram 5). From this we constructed a possible complete pedigree in Figure 4. The small digit in each edge (= connecting line, “branch” in the tree), indicates the distance expressed by that edge; it should be used only as an inidication and has no objective meaning.

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel ‘@ 1

5 1 3

‘N

4 2 6

‘B

3 8 1

‘Z

2 3 6

‘K

1 2

‘H

2

‘S

Figure 5. The final pedigree

Node 2, the one in the middle of the plot with four links, represents the unresolved quadruple mentioned above. When the extra category of version formulas was added which expresses the undisputed differences between prose (Z) and verse (H, S, and K), as described in section 6, the programmes instantly split node 2 and positioned K closer to H and S. When more formulas were added to express the fact that the number of episodes in B, and apparently also in N, differs from the number of episodes in H, S, K and Z, the programmes put B closer to N. We now obtain the final pedigree (Figure 5).

. Validity of narrative elements in stemma construction The results of the computer aided stemma construction show that it is possible to use not only textual but also narrative variants as a basis for an investigation into family relationships between texts. The results of our investigations into narrative elements agree with what we had expected from other indications

Valentin and Namelos discover their parentage 

and, indeed, assumed in our earlier publications (Langbroek & Roeleveld 1997, 1998a, b), see also under 11.

. The other evidence: investigations into rhyme It is generally agreed that the tale of Valentin and Namelos was not originally written in Middle Low German, but as it remained unclear from what language or dialect it was translated into Middle Low German, we earlier investigated the origins of the rhymes in the Stockholm recension (Langbroek & Roeleveld 1998a and 1998b). Two recent editions of the Stockholm recension of Valentin and Namelos were in existence (Geeraedts 1984; Langbroek & Roeleveld 1997), which should guarantee reliable readings. The rhymes were investigated without any preconceived notions about their origin and ‘retranslated’ into Middle High German, Middle Mid German and Middle Dutch to trace perfect rhymes. The 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 1137 rhyming pairs; the rest are orphans. The assessing, and allotting as to possible original languages or dialects, was done on morphological-phonological and on lexical grounds. Most rhyming pairs turned out to be neutral, meaning that ‘retranslating’ them into Middle High German, Middle Mid German and Middle Dutch yielded perfect rhymes of existant words in all three languages/dialects. This was the case for 82.8 per cent of the rhyming pairs, not a surprising percentage for such close linguistic relatives. Of the remainder, 7 per cent yielded perfect rhymes in Middle Dutch only, and 7.7 per cent in both Middle Dutch and Middle Mid German. Only 1.8 per cent argued exclusively for a Middle High German origin. In summary, a percentage of 90.5 of all the rhymes would be consistent with a Middle Mid German (or Middle Franconian) exemplar (or possibly 92.3 per cent if the Middle High German percentage of 1.8 is added). As many as 98.2 per cent of the rhymes, however, is consistent with a Middle Dutch exemplar. Whether the same percentages hold for the Hartebok recension is not yet certain. For comparison of H with S we used our transcript from microfilm. Our very recent diplomatic edition of the Hamburg Hartebok manuscript will have to be consulted for conclusive results.

 Annelies Roeleveld, Erika Langbroek, and Evert Wattel

Notes . Category 4 applies only in cases where the different grammatical structures in question are possible in all the languages involved; it does not, in fact, occur outside the comparison involving fragment K (see under 6). . Category 4 applies only in cases where the different grammatical structures in question are 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) – Middle Dutch, verse, 2 fragments of 176 lines each, c. 1350; editions Kalff 1886, p. 204–220; CD-rom Middelnederlands 1998, Rijmteksten. N III (Bibliotheek der Rijksuniversiteit Gent nr. 2749,9) – Middle Dutch, verse, fragment of 395 lines, of which 16 are barely legible and 22 are indecipharable, 1340–1360; editions De Vreese 1892; Van der Schaaf 1991; CD-rom Middelnederlands 1998, Rijmteksten. K (Kgl. Bibl. Kopenhagen, lost) – Middle Mid German, verse, fragment, 52 lines, date unknown; editions R. Nyerup, Deutsches Museum 1784, vol. II, p. 91–93, Seelman 1884. S (Kgl. Bibl. Stockholm, Cod. Holm. Vu 73, fol. 1r-33r) – Middle Low German, verse, 2291 lines, c. 1450; editions Geeraedts 1984; Langbroek & Roeleveld 1987. H (Hamburg Staats- und Universitätsbibliothek, Cod. 102c in scrinio, known as the ‘Hartebok’, fol. 33r-75v) – Middle Low German, verse, 2613 lines, between 1476–1481; editions Staphorst 1731; Langbroek & Roeleveld 2001. B (Stadtbibliothek Breslau, fol. 304, 13-38b) – Middle Mid German, prose, 1465; edition Seelman 1884. Z (Kgl. Bibl. Stockholm) – 3 Old Swedish mss, prose, 16th century; edition Wolf 1934.

References Beta, E. (1907). Untersuchungen zur Metrik des mittelniederdeutschen Valentin und Namelos. Doct. Thesis Leipzig. Cd-rom Middelnederlands, Woordenboek en teksten (1998). Den Haag, Antwerpen. Dieperink, G. J. (1933). Studien zum Valentin und Namelos. Ein Beitrag zur Geschichte der literarischen Beziehungen zwischen Flandern, Middel- und Niederdeutschland und Schweden zur Zeit der Hanse. Haarlem. Geeraedts, L. (1984). Die Stockholmer Handschrift Cod. Holm. Vu 73. Edition und Untersuchung einer mittelniederdeutschen Sammelhandschrift. Niederdeutsche Studien Vol. 32. Köln, Wien. Kalff, G. (1886). Middelnederlandsche epische fragmenten (Reprint Arnhem 1967). Karg, F. (1924). Die altschwedische Erzählung von Valentin und Namelos. Festschrift für E. Mogk. Halle.

Valentin and Namelos discover their parentage 

Klein, Th. (1997). “Die Rezeption mittelniederländischer Versdichtungen im Rheinland und Augustijns ‘Herzog von Braunschweig”’. Amsterdamer Beiträge zur älteren Germanistik, 47, 79–107. Langbroek, E. & A. Roeleveld (1997). Valentin und Namelos. Mittelniederdeutsch und Neuhoch-deutsch. Herausgegeben, übersetzt und kommentiert von Erika Langbroek und Annelies Roeleveld unter Mitarbeit von Arend Quak. Amsterdamer Publikationen zur Sprache und Literatur 27. Amsterdam: Rodopi. Langbroek, E. & A. Roeleveld (1998a). Valentin bekommt einen Gefährten. Ein Vergleich der Reimpaare in den Handschriften S, H und K. Amsterdamer Beiträge zur älteren Germanistik 50. Langbroek, E. & A. Roeleveld (1998b). Wie reimen sich die Nachbarn? Eine Untersuchung nach den ursprünglichen Reimen in ‘Valentin und Namelos’ in der Stockholmer Handschrift Cod. Holm. Vu 73. Niederdeutsches Jahrbuch 121. Langbroek, E. & A. Roeleveld (Eds.), with I. Biesheuvel & H. Kienhorst (2001). Het Hartebok, Hs. Hamburg, Staats- und Universitätsbibliothek, 102c in scrinio. Middeleeuwse Verzamelhandschriften uit de Nederlanden VIII. Hilversum: Verloren. Schaaf, K. van der (1991). Valentijn ende Nameloos III, een paralleleditie van een Middelnederlands fragment, waarin een handschriftbeschrijving en een onderzoek naar de relatie van de Middelnederlandse Valentijn en Nameloos-versie met de Middelnederduitse. Assen (typescript). Seelmann, W. (1884). Valentin und Namelos, die niederdeutsche Dichtung, die hochdeutsche Prosa, die Bruchstücke der mitelniederländischen Dichtung nebst Einleitung, Bibliographie und Analyse des Romans Valentin & Orson. Niederdeutsche Denkmäler Vol. 4. Norden, Leipzig. Staphorst, N. (1731). Hamburgische Kirchen-Geschichte, Teil I, Bd. 4. Hamburg. Vreese, W. de (1892). “Een nieuw fragment van Valentijn en Nameloos”. TNTL, 11, 140–162. Wattel, E. & M. van Mulken (1996a). “Shock Waves in Text Traditions”. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 105–121). Amsterdam, Philadelphia: John Benjamins. Wattel, E. & M. van Mulken (1996b). Weighted Formal Support of a Pedigree. In P. van Reenen & M. van Mulken (Eds.), Studies in Stemmatology (pp. 135–167). Amsterdam, Philadelphia: John Benjamins. Wolf, W. (1934). Namnlös och Valentin, Kritische Ausgabe mit nebenstehender mittelniederdeutscher Vorlage. Samlingar utgiva av Svenka Fornskrift-Sällskapet. Uppsala.

Index

A abbreviation 170, 172, 186 accidental variation 127–128, 131–132, 134, 142 adaptation 61, 187, 209, 219, 223 addition 137–138, 140, 142 additional contribution 57–58, 82 admission criteria 145–146 apparatus 96, 244, 246, 288, 292 archetype 26–27, 152, 158, 230, 294 archetypical information 290, 293–294 hyparchetype 27, 60 artefact 74, 84, 129, 137, 199 ashkenazi (tradition) 168, 185–187, 189, 191, 196, 206, 273, 275–279, 281 athonite text types 242, 260 attestation 14–16, 19, 21, 29, 33–36, 38–43, 45, 55, 61–62, 64, 67, 70, 77, 79–80 ausgangstext 25 author 14, 25–26, 79, 90, 113, 119, 200 autograph 25–27 auxiliary verbs 181, 197 B bias 151 bible (edition) 100, 104–107, 110, 133, 167, 169–171, 173, 175–176, 185, 193, 196–200, 206, 241, 244, 254, 261, 263, 270, 274, 276–279 bishop of Capua, cf. Victor 129 boniface (St.) 130

byzantine byzantine text 43, 87–91, 94, 96 byzantine tradition 17–18, 49, 87 byzantine witnesses 22, 43–44, 61

C cardiogram 99–100, 104, 106–107, 109–111, 204 categories 169, 171, 176, 182–185, 187, 191, 193, 195, 201, 204 chains of coherencies 34 change in relationships 106, 116 chudov New Testament text type 242, 249, 254, 257, 263 church Fathers 18, 21–22, 78 circle 25, 50, 67–72, 74, 83, 134 circular argument 25 circular development of variants 74 circular edges 67, 72 circular (genealogical) relationship 51, 68, 74 cladistic 5, 152, 198 classes of variants 227–228 classification 169, 171, 193, 243, 249–254, 258–260, 266, 288, 292 cluster 165, 246–249, 254–255, 257–260, 276 cluster analysis 241, 243, 246, 251, 259–260 codex Cassellanus 130–135, 137, 139–140, 142 codex Fuldensis 129–133, 135, 137

 Index

codex Sangallensis 130–131 coherence, coherency 13, 30–33, 35, 37–41, 44–45, 54, 63–65, 70, 72, 77, 79–81, 89, 139 chains of coherencies 34 coherence-based genealogical method 34 coherent chain 43 coherent field 34 directed coherency 33 field coherency 34 genealogical coherence 30–31, 33, 41, 44, 77, 79, 81 imperfect coherency 40, 80 perfect (genealogical) coherency 40–41 pre-genealogical coherency 33 stemmatic coherency 33, 37 undirected coherency 63, 67, 72 coincidental correspondence 16, 30 coincidental emergence of variants 24 commentary 19, 49 computational complexity 147, 154, 156, 252 conflation 193 conflicting data 60 connective variants 28–29, 32 consensus stemma 230–232, 235 contamination 14, 16–18, 22–24, 30, 34, 43, 46, 48–53, 55, 64, 115, 134, 178, 193, 195, 202, 230, 249 contaminated tradition 13, 15, 51, 60, 76–77, 196, 243, 250, 253, 259–260, 279 incidental contamination 8, 109, 191 multi-stage contamination 50, 67 simultaneous contamination 8, 116, 177, 196 successive contamination 99–100, 104, 109, 116–117, 182–183, 203–204

contraction 147–148, 164, 170, 172, 183, 193, 252 convergent evolution 7, 9–10 coptic 18, 79, 89 copulative 170, 181, 183 copyists 128, 134 correction 21, 119, 136, 172, 177–178, 185, 189, 191 corrector 16, 175 critical apparatus 244, 246

D data presentation 246, 252 deep structure 198, 250 design choice 161, 163 deviation 152–155 dialectical development 176, 194 diatessaron 129–132 dichotomy 150, 156, 187, 255 dictation 223 direct copy 130–133 direct (genealogical) relationship 37–38, 70 direction directed coherency 33 directed edge 30 direction of variants 114 undirected coherency 63, 67, 72 undirected edge 64–67, 72 dissimilarity 101 distance distance distribution 100 distance function 251, 258 distance matrix 5 partition distance 230–232, 235 distribution formula 101 DNA 3–5, 7–8, 10 double translation 178

E early version 18, 21–22, 78 eastern tradition 271, 275–276, 279

Index 

edge

14, 49, 55, 60, 77, 81, 120–121, 147, 159, 163–64, 231, 245, 254, 299 circular edges 67, 72 directed edge 30 undirected edge 64–67, 72 Editio Critica Maior (ECM) 14, 17–18, 27, 61, 77, 87, 97 error(s) 4, 26–28, 54, 91, 129, 132, 134–138, 159, 178, 181, 183, 193, 196, 199, 202, 211, 216–217, 221, 236, 272 printing errors 136 scribal error 90, 134, 171–172, 177, 180, 185, 200, 237 evolution 3, 4, 7–10 exemplar 26–27, 49, 78, 89, 92, 99, 110, 116, 123, 133, 169–172, 175, 177–180, 182, 184–186, 188, 191, 195, 199–201, 207–208, 217–219, 222–224, 237, 272, 278, 292, 301 F falsifiability 76 faulty reading 78 field coherency 34 flow, cf. textual flow formula weight 146, 155, 158, 161, 163–164, 251 fragment 52, 63, 74, 286, 289, 291–292, 298, 302 fragmentary witness 71 fragmentation 21, 70, 74, 82 G gender 171, 173, 183, 185, 193, 210–211 genealogical coherence 30–31, 33, 41, 44, 77, 79, 81 genealogical relevance 101, 207, 222, 224 genealogies 132 general textual flow 33, 37 genuine variants 18, 29, 54, 78

global stemma 29–30, 33–34, 37, 43, 46, 49, 51, 53, 58, 63–64, 70–71, 74–76, 81, 83 global textual flow 34, 37–38 graph 30, 64, 72, 77, 81, 83–84, 101–103, 110, 187, 189, 243, 253, 295, 298 H haftarot(h) 185, 270–271, 280 harmony tradition 131, 133 heading line 100–101, 148–150 heart beat 104–105, 108, 111 heliand 130–131 homoioarkton 14 homoioteleuton 14, 136–138, 142 hyparchetype 27, 60 I imperfect coherency 40, 80 incidental contamination 8, 109, 191 inconsistency 89, 114–115, 123, 146, 246 indirect relationship 70, 84 initial text 25–27, 29, 35, 42, 45–46, 52, 61, 63, 81–84 initial tree, initial stemma 147, 158, 164, 186, 188, 190, 195, 237, 250, 252 intermediarity intermediary node 30, 60, 62–67, 71–72 intermediate structure 250 K kings of England 7–9, 228–230, 236, 238 kinship revealing 269, 272, 275 koine 28, 87, 90–92, 96–97 L lacuna(e) 18, 69, 82, 150–151, 222 lanceloet van Denemerken 128

 Index

lapsus 27 latin 18, 89, 129–133, 139–140, 170, 178, 199–200, 280 old Latin Harmony 131 lectio difficilior 41, 79 lectionaries 18, 21–22, 78, 261, 280 level of content 214, 217, 220, 222 level of morphology and syntax 210, 219 level of spelling 209, 219 levels of linguistic variation 207 lexical variation 207, 214, 223–224 lexicon 224 linguistic variation 207 local search 147, 158 local stemma(ta) 13–16, 29, 34–35, 38–40, 61–62, 68–70 local textual flow 34, 42 lydgate 7–9, 228, 238 M majority reading 87–88, 90 majority Text 88 mantel correlation 230 marginal text 78 marginal variants 21 masorah 176 maximal contribution 58 maximum likelihood 5–6 mediate priority 69–70 mediate relationship 70 minimal contribution 57–58, 82 minus 170, 178, 181, 183, 189, 195, 198 mirror copying 223 missing links 23 modern editors 132 morphology 171, 173, 193, 207, 210, 219, 224 morphological variation 211, 213, 223 mouvance 117, 207, 217, 224 multiple stemma 118 multi-stage contamination 50, 67 mutation 4–5

N narrative development 288 neighbour-joining 230, 232–236 new liturgical tetraevangelion 242, 249, 254, 258–260, 263 new philology 224 new Testament 13, 17–18, 24–25, 28, 49, 61, 87, 89–91, 104, 106–109, 129, 140, 241–246, 249, 254, 256–257, 260, 263–266 node 152–154, 158–159, 163–164, 245, 252–254, 257–258, 300 intermediary node 30, 60, 62–67, 71–72 terminal node 75, 250 noise 104, 114, 147, 158, 173, 186–187 non-ancestor 30–31, 60 non-coincidental correspondences 30 nondeterministic 157 non-direct relationship 30–31 nonsense reading 127, 138, 141 number 171, 173, 183–185, 193, 196

O old Latin Harmony 131 old Testament 104–107, 111, 204 old text type 241–242, 244, 249, 254–255, 257–259, 261 omission 14–15, 21, 82, 128, 137–138, 140, 142, 150–151, 171, 198, 212–213, 229, 255 optimal stemma 147, 155 optimal substemma 30–31, 33, 51–52, 55–56, 61–63, 65, 67, 71, 80, 83–84 optimization 157 orientation 119–121, 259 oriented stemma 164, 250 original reading 28, 43, 88, 119–120, 172, 180–181, 204

Index 

original text 14, 24, 29, 41, 88, 90, 119, 179, 193, 277 origin of the reading 114 orthodox corruption 89 orthography 82, 170–171, 183, 193, 195, 272 orthographica 28, 54 orthographicals 128

P parallelism 114, 119–120, 127–128, 136, 138–141, 180–181, 191, 195–196 paratextual element 181, 203 parsimony 5–6 particular textual flow 33, 37 partition distance 230–232, 235 peak 104, 108, 110–111, 182, 204, 297 perfect (genealogical) coherency 40–41 periphery (of a cluster) 249, 258–259 philoxeniana 79 phylogenetic 4–5, 7–8, 10, 228 picardism 122 place of variation 13, 20, 27–32, 34–35, 56, 62–63, 77 plus 170, 178–179, 181, 183, 189, 195 polynomial 157 posterior, cf. also prior 31, 33, 37, 51–52, 55–57, 59–60, 63, 67–71, 76, 84 potential ancestor 31, 33, 37–40, 42–43, 46, 49–51, 55, 57, 59–62, 67–68, 77, 80–83 pre-genealogical coherency 33 predominant textual flow 46–48, 59, 81, 84 preposition 170, 180, 183, 203 preslav text type 242, 249, 254, 257, 260, 262 printing errors 136

prior, cf. also posterior 31, 33, 37, 51–53, 55, 57–60, 62–63, 67–71, 76, 82–84 priority 55–56, 69–70 punctuation variants 227, 238

Q quadruple 122, 155–156, 161, 163–164, 182, 185, 187–189, 204, 298–300 quire separator 110, 117, 122

R random variation 169–170, 177 reading faulty reading 78 majority reading 87–88, 90 nonsense reading 127, 138, 141 original reading 28, 43, 88, 119–120, 172, 180–181, 204 origin of the reading 114 variant reading 138, 150, 168, 187, 201, 261 recombination 7–8, 10 relationship change in relationships 106, 116 circular (genealogical) relationship 51, 68, 74 direct (genealogical) relationship 37–38, 70 indirect relationship 70, 84 mediate relationship 70 non-direct relationship 30–31 resolved tree 159, 164, 250, 252

S sahidic version 79 sample survey 271 scribe(s) 8–10, 14, 23, 25, 27–28, 54, 89, 92, 96–97, 135–136, 138, 169–173, 176, 182, 191–192, 194–196, 198–201, 203,

 Index

208–210, 213, 216–219, 222–223, 227, 242, 245 scribal error 90, 134, 171–172, 177, 180, 185, 200, 237 scribal variations 207 search algorithm 147 sefardi (tradition) 168, 178, 185, 187–189, 199, 204, 206, 273, 275–279 semantic shift 170, 176, 183, 189, 195, 198 separation 170, 172, 183, 185, 193 shock wave 102–104, 108, 110–111, 147, 187, 298 similarity graph 106, 182 similarity score 246–249, 251–252, 257–258 simultaneous contamination 8, 116, 177, 196 single version information 292–293 spelling variants, spelling variation(s) 171, 203, 209–210, 214, 223–224, 227 split decomposition 5, 6, 230 stability 55–57, 67, 70, 72, 80, 84, 191, 245 stemma(ta), cf. also tree 3, 8–16, 24, 59, 72–73, 77, 100–101, 106–107, 109, 114, 123, 127–128, 140, 142, 145–148, 151–158, 168, 171, 181, 185, 187, 197–198, 217, 227–229, 236–238, 243, 245, 250–255, 257–260, 265, 269, 271–273, 275, 279, 287–288, 300 consensus stemma 230–232, 235 global stemma 29–30, 33–34, 37, 43, 46, 49, 51, 53, 58, 63–64, 70–71, 74–76, 81, 83 initial tree, initial stemma 147, 158, 164, 186, 188, 190, 195, 237, 250, 252 local stemma(ta) 13–16, 29, 34–35, 38–40, 61–62, 68–70

multiple stemma 118 optimal stemma 147, 155 stemmatic coherency 33, 37 substemma 30–31, 33–34, 37–38, 46, 49, 51–52, 54–58, 60–63, 65–67, 71, 74–75, 80–81, 83–84, 153–154 stepwise refinement 147–148, 165, 252 story elements 288 substitution 142, 170, 176, 183, 189, 195 subtree, cf. also substemma 253 successive contamination 99–100, 104, 109, 116–117, 182–183, 203–204 syntactic variation 210–211, 214, 223–224 syriac 18, 79, 89, 131–133, 139, 196, 204

T targum

167–170, 176, 179, 181, 191, 194–204, 269–271, 273–275, 277–281 targum Jonathan 198, 269 tosefta Targum 274, 277–280 terminal node 75, 250 test passages 17, 78, 87, 92–93, 95–97 teststellen 17 tetraevangelion 244, 262–264 new liturgical tetraevangelion 242, 249, 254, 258–260, 263 text of the author 25, 119 text type 168, 241–246, 249, 253–263 textual flow 33–34, 36–38, 40–43, 45–49, 51–53, 55–57, 59, 63, 65, 67, 70–72, 80, 83–84 strength of the textual flow 56 textual flow diagram 41, 45–46, 81 textus receptus 90, 246

Index

three level method, three step method 114, 119, 241, 250, 272–273, 279 transcription 116, 223 transposition 7, 181 tree, cf. also stemma 4–6, 8, 10, 13, 46, 48, 71, 81, 114–115, 146–148, 150, 152–159, 161, 163, 167, 168, 170, 181, 185, 191, 193, 199–200, 228, 230, 238, 245, 252–255, 258, 260, 272–273, 275, 285, 298–299 initial tree, initial stemma 147, 158, 164, 186, 188, 190, 195, 237, 250, 252 resolved tree 159, 164, 250, 252 subtree, cf. also substemma 253 unoriented tree 154, 158–159 type 0 (variation) 19 type 1 (variation) 19, 150, 159, 260 type 2 (variant, variation, oppositions) 19, 22, 101, 146, 150–155, 159–160, 207–208, 217, 260, 289 type 3 (variation) 19, 150, 159, 260 type 4 (variation) 19, 150, 160, 249 U undirected coherency 63, 67, 72 undirected edge 64–67, 72 unit of variation 27–28 unoriented tree 154, 158–159 V valuation 146, 155–156, 158–159, 288 variant, variation(s) 14–17, 23, 25–26, 33, 36–41, 43–44, 46, 49–73, 75–76, 79–84, 87–92, 94–96, 100, 104–106, 108–110, 115, 119–120, 123, 141, 145, 149, 157, 172–175, 182–183, 189, 193, 195, 198, 205, 212, 218–220, 229, 232–233, 236–237, 241,

243–246, 250–252, 254–257, 269, 272–273, 275, 279, 292, 294, 300 accidental variation 127–128, 131–132, 134, 142 circular development of variants 74 classes of variants 227–228 coincidental emergence of variants 24 connective variants 28–29, 32 direction of variants 114 genuine variants 18, 29, 54, 78 (levels of) linguistic variation 207 lexical variation 207, 214, 223–224 marginal variants 21 morphological variation 211, 213, 223 place of variation 13, 20, 27–32, 34–35, 56, 62–63, 77 punctuation variants 227, 238 random variation 169–170, 177 scribal variations 207 spelling variants, spelling variation(s) 171, 203, 209–210, 214, 223–224, 227 syntactic variation 210–211, 214, 223–224 type 0 (variation) 19 type 1 (variation) 19, 150, 159, 260 type 2 (variant, variation, oppositions) 19, 22, 101, 146, 150–155, 159–160, 207–208, 217, 260, 289 type 3 (variation) 19, 150, 159, 260 type 4 (variation) 19, 150, 160, 249 unit of variation 27, 28 variant reading 138, 150, 168, 187, 201, 261



 Index

variant selection 167–169, 181, 199 variants in story elements and narrative development 288 variation in word order 211, 213 versional variants 78 version formula 145–146, 149–152, 155, 159–160, 162, 164, 250–253, 260, 289, 295 versional variants 78

victor, bishop of Capua 129 vocalization 170, 172, 193 vulgate 87, 129, 131–133, 139–140, 201 W weight 101, 158–164, 228–229, 237–238, 253, 260, 272, 287–288, 292 formula weight 146, 155, 158, 161, 163–164, 251