Building Stemmas with the Computer in a Cladistic, Neo-Lachmannian, Way

Building Stemmas with the Computer in a Cladistic, Neo-Lachmannian, Way The Case of Fourteen Text Versions of Lanseloet van Denemerken Een wetenschappelijke proeve op het gebied van de Letteren PROEFSCHRIFT ter verkrijging van de graad van doctor aan de Katholieke Universiteit Nijmegen, volgens besluit van het College van Decanen in het openbaar te verdedigen op vrijdag 18 februari 2000 des namiddags om 1.30 uur precies door Benedictus Johannes Paulus Salemans geboren op 10 december 1955 te Maastricht

Promotor: Prof. dr. G.R.W. Dibbets Manuscriptcommissie: Prof. dr. E.M.P. van Gemert (KUN) Prof. dr. Th.F.C. Mertens (UFSIA) Prof. dr. P.Th. van Reenen (VU) 2000 Ben Salemans, Nijmegen ISBN 90-9013480-8 (Ben Salemans) ISBN 90-3730505-9 (Nijmegen University Press)

CONTENTS ACKNOWLEDGEMENTS... 1 1. INTRODUCTION: WHAT IS THIS BOOK ABOUT?... 3 2. SOME CURRENT GENEALOGICAL METHODS... 11 2.1. Introduction... 11 2.2. A Stemma as a Historical Image and as a Tool for Text Reconstruction 12 2.3. Building a Stemma according to the (nineteenth-century) Method of Lachmann... 18 2.4. Building a Chain and a Stemma according to the (twentieth-century) Method of Greg/Dearing... 22 2.4.1. Introduction... 22 2.4.2. Typology of Variations and their Notation... 23 2.4.3. The Importance and the Limitation of Type-2 Variations; the Virtue of Type-1 Variations... 25 2.4.4. Building a Chain with Type-2 Variations: Dearing s Rules for Building Chains with Transformed into a New Algorithm; the Notion End Group... 27 2.4.5. Orienting a Chain into a Stemma... 31 2.5. Current Universal Taxonomical Principles, Biological Systematics and Cladistics... 35 2.5.1. Taxonomical Ordering Strategies... 35 2.5.2. Ordering or Clustering Taxons with Features... 36 2.5.3. Systematics and Text Genealogy... 40 2.5.4. Cladistics and the Principle of Parsimony... 43 2.5.5. Demonstration of Building a Genealogical Tree with PAUP... 43 2.6. Cladistic Implications for the Methods of Lachmann and Dearing... 47 2.6.1. Reconsideration of the Lachmannian Notion Common Error.. 47 2.6.1.1. Lachmann s/maas s Common Error Rule is Only Correct as Long as the Variants are Part of Type-2 Variations... 47 2.6.1.2. The Problem of Building Chains from Complex, Non-Type-2 Variation Formulas like AB:CD:EF... 49 2.6.1.3. The Zwei Zeugen Element Reconsidered... 52 2.6.2. A Cladistic Eye-opener for Lachmannians: Rooted and Unrooted Trees... 54 2.6.3. Criticism of Dearing s Way of Additioning Variations... 55 2.7. The Minimum Number of Three or Four Text Versions... 57 2.8. A Simple Advice for the Study of Contamination... 58 2.9. Conclusion and Summary... 59 3. TOWARDS A NEW TEXT-GENEALOGICAL METHOD... 61 3.1. Introduction... 61 3.2. The Theoretical Framework: Six Basic Text-Genealogical Principles.. 64 3.2.1. The First Basic Principle; the Definition of a Text- Genealogical Variant; Parallelism and Contamination... 64

iv Contents 3.2.2. The Second Basic Principle; the Positivistic Apparatus of Text- Genealogical Variants; a Short Discussion about Objectivity and Quentin s Non-Positivistic Zéro Caractéristique... 71 3.2.3. The Third Basic Principle; the Definition of a Variation Place... 77 3.2.4. The Fourth Basic Principle; the Type-2 Limitation, Partly Dismantled with the Use of End Groups... 78 3.2.5. The Fifth Basic Principle; the Text-Genealogical Use of Differences in Word Order; Additions and Omissions of Words or Verses... 81 3.2.6. The Sixth Basic Principle; the Definition of Text- Genealogical Word Types... 85 3.3. The Formalization: Rewriting the Theory in Eleven Characteristics of Text-Genealogical Variants; the Seventh Text-Genealogical Principle... 90 3.3.1. The Eleven Characteristics of Text-Genealogical Variants... 90 3.3.2. The Seventh Basic Principle of the (Temporary?) Role of the Philologist... 103 3.4. The Implementation: Developing Software from the Formalized Theory... 106 3.5. Conclusion and Summary... 109 4. APPLICATION OF THE METHOD TO THE LANSELOET VAN DENEMERKEN CORPUS... 113 4.1. Introduction... 113 4.2. Short Description of Fourteen Lanseloet van Denemerken Text Versions and their Contents; Bibliographical Remarks... 115 4.3. Demonstration of the Software Treating Verses of Lanseloet van Denemerken... 121 4.4. First Computer Results: All the Detected Text-Genealogical Lanseloet van Denemerken Variants... 133 4.4.1. Introduction: +Comb-formulas and Obs-formulas... 133 4.4.2. The Obs-formulas... 134 4.4.3. The +Comb-formulas concerning Characteristics 5 and 7d and the +Comb-formulas Not Rejected by any Characteristic... 135 4.5. The Removal of Incorrect Lanseloet van Denemerken Variants by the Philologist, Applying Non-Automated Characteristics... 141 4.6. Presentation of the Variation Formulas for Building the Lanseloet van Denemerken Chain... 146 4.7. The Development and Presentation of the Lanseloet van Denemerken Chain with Use of Cladistics... 151 4.7.1. First Attempt to Build the Lanseloet van Denemerken Chain; a Handmade Sketch of this Tree... 151 4.7.2. Judgement of the Handmade Sketch of the Lanseloet van Denemerken Tree: No Contamination... 154

Contents v 4.7.3. Missing Values: Some Extra Variation Formulas are Needed; a Second Handmade Sketch of the Lanseloet Tree... 155 4.7.4. The Lanseloet van Denemerken Tree as Developed by the Cladistic Software Package PAUP... 172 4.7.4.1. Transforming the Variation Formulas into NEXUS format... 172 4.7.4.2. Measuring Distances between Texts... 173 4.7.4.3. The Chain of the Lanseloet van Denemerken Texts as Drawn by PAUP; the Trustworthiness of the Chain.. 174 4.8. The Development and Presentation of the Lanseloet van Denemerken Stemma... 178 4.8.1. Introduction... 178 4.8.2. Determination of the Point of Orientation... 179 4.8.2.1. First Attempt to Find the Point of Orientation... 180 4.8.2.2. Second Attempt to Find the Point of Orientation... 192 4.8.3. Presentation of the Stemma of Lanseloet van Denemerken... 195 4.9. Information Derived from the Lanseloet van Denemerken Stemma and Text Versions... 196 4.10.Conclusion and Summary... 199 5. EVALUATION OF THE LANSELOET VAN DENEMERKEN STEMMA AND THE TEXT-GENEALOGICAL CHARACTERISTICS... 203 5.1. Introduction... 203 5.2. Evaluation of the Lanseloet van Denemerken Stemma by Comparing it with Lanseloet Stemmas in Other Studies... 204 5.2.1. The Part of the Lanseloet Stemma in Leendertz (1907)... 204 5.2.2. The Part of the Lanseloet Stemma in Goossens (1973)... 208 5.2.3. The Part of the Lanseloet Stemma in Goossens (1976)... 216 5.2.4. The Part of the Lanseloet Stemma in Hüsken & Schaars (1984) 218 5.2.5. Results of the Comparison of Our Stemma with the Other Lanseloet Stemmas... 222 5.3. Evaluation of the Text-Genealogical Characteristics... 223 5.3.1. Evaluation of Characteristic 1... 225 5.3.2. Evaluation of Characteristic 2... 229 5.3.3. Evaluation of Characteristic 3... 230 5.3.4. Evaluation of Characteristic 4... 231 5.3.4.1. Evaluation of characteristic 4a... 233 5.3.4.2. Evaluation of characteristic 4b... 235 5.3.4.2.1. Evaluation of the Word Category Adjectives 235 5.3.4.2.2. An Alternative Lanseloet Stemma with Contaminated Texts 04 and 05?... 237 5.3.4.2.3. Evaluation of the Word Category Adverbs. 240 5.3.4.2.4. Diachronical, Parallelistic Changes in Flexion/Casus from 1400 to 1700... 244 5.3.4.2.5. Evaluation of the Word Category Articles.. 247

vi Contents 5.3.4.2.6. The Parallelistic Character of the Gender of Substantives... 251 5.3.4.2.7. Evaluation of the Word Category Auxiliaries 252 5.3.4.2.8. Evaluation of the Word Category Conjunctions 257 5.3.4.2.9. Evaluation of the Word Category Prepositions 259 5.3.4.2.10. Evaluation of the Word Category Pronouns 260 5.3.4.2.11. Implications of the Evaluation of Characteristic 4b for the Emendatio... 264 5.3.5. Evaluation of Characteristic 5... 266 5.3.6. Evaluation of Characteristic 6... 267 5.3.6.1. Evaluation of Characteristic 6a, concerning Small Differences... 267 5.3.6.2. Evaluation of Characteristic 6b, concerning Word Boundaries... 270 5.3.6.3. Evaluation of Characteristic 6c, concerning Nonsense Words, Slips of the Pen or Typographical Mistakes.. 271 5.3.7. Evaluation of Characteristic 7... 272 5.3.7.1. Evaluation of Characteristic 7a, concerning Inflection. 273 5.3.7.2. Evaluation of Characteristic 7b, concerning Differences in Vowels... 274 5.3.7.3. Evaluation of Characteristic 7c, concerning (Personal) Vocabularies... 275 5.3.7.4. Evaluation of Characteristic 7d, concerning Frequently Used Words and Names... 277 5.3.8. Evaluation of Characteristic 8... 278 5.3.9. Evaluation of Characteristic 9... 281 5.3.9.1. Evaluation of Characteristic 9a, concerning Rhyming Conventions... 281 5.3.9.2. Evaluation of Characteristic 9b, concerning Duplicate Rhyming Words... 284 5.3.10. Evaluation of Characteristic 10... 286 5.3.11. Evaluation of Characteristic 11... 287 5.3.11.1. Evaluation of Characteristic 11a, concerning Added or Missing Words... 287 5.3.11.2. Evaluation of Characteristic 11b, concerning Added or Missing Verses... 294 5.3.12. Summary of the Evaluation of the Characteristics... 296 5.4. Conclusion and Summary... 297 6. FINAL REMARKS... 299 REFERENCES... 303 SAMENVATTING IN HET NEDERLANDS... 313 INDEX... 323

Contents vii LISTS OF FIGURES AND PICTURED LANSELOET VERSES... 339 List of Figures, Including their Captions... 339 List of Pictured Lanseloet Verses... 348 CURRICULUM VITAE... 351 APPENDICES ON CD-ROM (in file apps.pdf) (to be read with Adobe Acrobat Reader, available for free on the Internet; see: http://www.adobe.com/products/acrobat/readstep.html or http://www.adobe.com/) Appendix A: Description of the Eight Steps the Software Performs: from the Fourteen Single Text Versions to the Variation Formulas Appendix B: Guide to the Interpretation of the Computer Results and Output Appendix C: The Computer Results: the Synoptic Lanseloet van Denemerken Text Versions and the Variation Formulas Appendix D: Presentation of the Variation Formulas, Ordered by Characteristics Appendix E: Appendix F: English Translation of Parts of Salemans (1987) on Cladistics The Fourteen Synoptic Lanseloet van Denemerken Text Versions (or: Appendix C without the Variation Formulas)

viii Contents

ACKNOWLEDGEMENTS First of all, I would like to thank Prof. Dr. Jan van Bakel (University of Nijmegen) for his impressive and inspiring lectures and scientific lessons. My promotor Prof. Dr. Geert Dibbets (University of Nijmegen) taught me the first principles of editing old texts. I admire him for his wisdom and his patience with me. I will always be grateful to Prof. Dr. Piet Buijnsters (University of Nijmegen) for sharing his immense knowledge and appreciation of old books. Dr. Paul Wackers (University of Nijmegen) opened up my heart to the beauty of medieval literature. Dr. Willem Ellis (University of Amsterdam), a biologist, was always willing to answer my stupid questions about biological ordering methods. Prof. Dr. Thom Mertens (Ruusbroec-genootschap in Antwerp) and Dr. Ton Duinhoven (University of Amsterdam) were always prepared to criticize preliminary versions of my papers about text-genealogical matters, including this book. I discussed many stemmatology matters under pleasant conditions with Prof. Dr. Pieter van Reenen (Free University of Amsterdam) and his group of Dutch textgenealogists, amongst which the kind and wise mathematician Dr. Evert Wattel (Free University of Amsterdam) and the Dutch text-genealogical pioneer Prof. Dr. Ton Dees. I thank Prof. Dr. Vittore Branca (University of Padova) for sharing his wise insights with me during my stay in Rome in May 1998, when I was invited to give a lecture on automated textual criticism at the Accademia Nazionale Dei Lincei. I am also greatful for the heartening support of Prof. Dr. R. Hauer (University of Utrecht), Dr. D. Hertzberger (Nijmegen) and Dr. L. Bellersen (University of Nijmegen). Three friends followed my research and life closely with much patience. Dr. Peter-Arno Coppen (University of Nijmegen) helped me to keep a general methodological overview of my research, which started in 1985. Dr. Margot van Mulken (University of Nijmegen) and I discussed during many days, evenings and even nights, the fundamentals of text-genealogical research and life, often in our Nijmegen pub t Haantje. Thank you, Margot, for these very pleasant, inspiring and encouraging hours. Dr. Willem Kuiper (University of Amsterdam) was always willing to lend me an ear concerning stemmatological or private affairs, no matter what time it was; a true friend. I am also greatful for the friendship of the editors of Neder-L, the free electronic Internet-magazine about Dutch language and literature, which I started in 1992 (URL: http://baserv.uci.kun.nl/~salemans/). Besides the already mentioned Willem Kuiper and Peter-Arno Coppen, the editors are Dr. Piet Verkruijsse (University of Amsterdam) and Prof. Dr. Marc van Oostendorp (Meertens Institute and University of Amsterdam). I thank Maastricht for just being there. The same goes for its football club MVV. Thank you, Willy Brokamp, Jo Bonfrère and Erik Meijer. I thank Bob Dylan, Green Day, Paul McCartney and Wolfgang A. Mozart for their music.

2 Acknowledgments It is good to know that in hard times some friends stay around. Roland de Bonth, Patrick Leijzer, Frans Schaars, Hans and Conny Schoonbrood, Peep Stalmeijer and Remy Wolfs, thank you. The most important people for me are my children Bart, Milou and Joost. I

1. INTRODUCTION: WHAT IS THIS BOOK ABOUT? In the Middle Ages copyists transcribed texts by hand. During the laborious transcription process they introduced - intentionally or unintentionally - errors and other new, unoriginal elements in the copy texts they were producing. If these new texts were copied again, new errors could be introduced, etc. The invention of the printing press around 1450 did not change this process much. After all, book printers (type-setters, bookbinders) introduced unoriginal elements in their works as well. The general result of copying texts was that after a time several different versions of one original text existed, while the original text was lost. Philologists of all ages, even before Christ, saw themselves confronted with the problem how to remove these unoriginal elements from the existing texts. Before 1700, textual criticism, the art of reconstructing old lost texts, was not very systematic. This can be demonstrated by the first printed version of the Greek New Testament. The first complete edition of the Greek New Testament was produced by Erasmus of Rotterdam. It was printed in 1516 on Froben s presses in Basel. For centuries it has been accepted as the ultimate Greek New Testament (NT) text: the Textus receptus. Erasmus developed the Greek NT from a few (fragmented) Greek NT versions which were present in Basel and its neighbourhood. He selected the best parts of them rather arbitrarily - he simply chose the parts which he judged, on the basis of his taste, to be original - and glued them together into one book. To fill the missing parts, he even used Latin sources, which he translated into Greek. In the eighteenth century, philologists like Bengel, Bentley, Griesbach, Mill and Wettstein 1 started to fight the authority of the textus receptus. However, they did not succeed in developing methods with which original and unoriginal elements in text versions could be detected. Professor Karl Lachmann demonstrated in 1830 that Erasmus s Greek NT was composed incorrectly. He proposed a new text-critical method, known as the method of Lachmann. This nineteenth-century Method will be discussed in detail in 2.3. At this point, I will give a concise introduction. The heart of the method of Lachmann is that we must know the relationships of text versions, before we start to correct (or emend) unoriginal parts in them. In short, it is a method in two steps: first, in the recensio phase, a pedigree or stemma of the text versions must be developed; second, in the emendatio phase, (un)original elements in these texts can be detected with the use of the stemma. How can a stemma be developed during the recensio phase in a Lachmannian way? The basic Lachmannian thought is sound and simple. If a serious error is introduced into a text version - e.g. a couple of verses are missing -, it is likely that the descendants of that text version will show the same common error. It is 1 More thorough information about the history of textual criticism and text-critical philologists like Bengel, Bentley, Erasmus, Griesbach, Lachmann, Mill and Wettstein can be found in Aland & Aland (1971), Kenney (1974) and Timpanaro (1971).

4 Chapter 1. Introduction assumed that all the texts with the missing couple of verses go back to the same common ancestor in which this error occurred for the first time. Once we have detected more common errors, we will be able to draw the stemma or pedigree. One can compare this with a kind of unique disease or DNA sequence which is passed on from the father or mother to the children and their children, etc. The occurrence of the disease, or DNA sequence, may serve as a guide to find genealogical family relationships. Likewise, if one can find common errors in text versions, the family pattern or relationships between the text versions will become clear and the stemma can be produced. Imagine that we have six text versions: text A, B, C, D, E and F. Suppose that in certain text passages one or more common errors in text A and B have been detected, while the other text versions C to F have other words (readings) in common, different from the common errors. Then we may conclude that both texts A and B have a common forefather, in which the errors first showed up. We label this (lost) common ancestor of both texts as a. Text a cannot have been the forefather of, for instance, text C, because we assume that one or more common errors present in A and B do not occur in C. In the same line of thought, suppose that unique common errors are present in texts C and D (leading to common ancestor b ), and others pop up in texts E and F (leading to c ). Furthermore, C, D, E and F have common errors, which gives rise to the thought that these four texts must share a common forefather d. We assume that the six texts go back to one common original text: text O. All these common errors lead to the next stemma, as an end result of the recensio: O a b c A B C D E F (α) (β) (γ) (γ) (δ) (α) Figure 1. Example of a stemma. How can we use this stemma, the genealogical pattern, as a tool for text reconstruction? The stemma is used for that purpose during the second phase: the emendatio. Suppose that the first line of text A starts with α ; at the same place B has β, C and D read γ, E starts with δ, and F has α. In other words, we have four different variants (i.e. different words or readings): α appears in two texts, β in one text, γ in two texts, and δ in one text. These variants are presented at the bottom of fig. 1. When we look at variant γ, we see that it is present in texts C and D. Therefore, it is likely that the common ancestor of C and D, text b, must have had variant γ as well. Now we look at variant α, which occurs in A and F. The (first) common ancestor of both texts is O, the lost d

original text. According to the method of Lachmann we can assume that the lost original text had the reading α. In this way we have been able to reconstruct a part of the lost original text with the use of the stemma. (Later on, in 5.3.4.2.11, this sketch of the emendatio phase will be criticized.) As we will see in 2.4, the method of Lachmann has been criticized because it is hard to detect common errors in a scientific (= verifiable / repeatable), way. It is unclear how it can be determined that a variant is an (unoriginal) error. The Lachmannian way of determining errors was based upon subjective judgements of variants beyond scientific control. Furthermore, its (vague) principles were not performed consequentially or consistently. Sometimes, for instance, differences in word order are used as common errors; sometimes they are not; therefore, Lachmannian judgements about the originality of variants often have an ad hoc character. As will be discussed in the same section ( 2.4), twentieth-century philologists like Greg (1927), Quentin (1926), Dearing (1974) and Dees (1975) proposed an alternative, better approach. They demonstrate that a stemma can be built in two steps. First, these modern text-genealogists develop a deep-structure of the stemma, the so-called chain, using variants that do not have to be judged as to their originality. Second, they produce a stemma from the chain. The advantage of this modern two-step method will be clear: the judgement about the originality of the variants is less important. Often, modern philologists use type-2 variations to build their chains. We speak of a type-2 variation, if at a certain place in the text versions precisely two variants occur and if each variant is present in at least two text versions. Generally, I am convinced that complex variations, with three or more competitive variants, are almost useless for the development of chains and stemmas. In other words, I think that philologists should in general only work with the mentioned special type of variance. This severe limitation is called the type-2 limitation. What is this dissertation about and which are its merits? In the first place, this dissertation offers a global introduction to stemmatology, the art of building text-genealogical trees. It gives an introduction to and a critical overview of several existing methods of building text-genealogical trees, the socalled chains and stemmas. In discussing these methods, we will see that it will be sometimes necessary to adapt basic notions of these methods. Of course I do not intend to give away all the results of my research in this introduction, but in 2.6.1 it will be demonstrated, for instance, that the Lachmannian common errors may be used only under very special conditions (namely in a so-called type-2 environment ). In this book, current text-genealogical methods will be compared and a new method of building trees will be established. 5

6 Chapter 1. Introduction Secondly, this study is interdisciplinary. In this dissertation, we will have a look at the way biologists build their genealogies for animals and plants. From this outsiders view we hope to get a clearer picture of strenghts and weaknesses of several textual-stemmatological methods. In 2.5 we will introduce cladistics, currently one of the leading biological ordering methods. The core of cladistics is the permanent question which elements or characteristics in a species can be used to develop genealogies. The simple lesson taught by cladistics is that we must be very careful in using characteristics for genealogical research. For example, the fact that both swallows and flies have wings, does not imply that these birds and insects belong to the same family. The characteristic having wings is not a trustworthy genealogical informant. Again, text-genealogists can learn from cladistics that they must be very careful in choosing variants to be the building stones of chains and stemmas. Thirdly, two hot items in current textual stemmatology will be discussed: in 2.8 the problem of contamination and in 2.6.3 the type-2 limitation. Normally a text version is a copy of one other text version. It is, however, possible that a text version X is produced from two or more texts Y and Z, for instance, because text Y was incomplete or because text Y was considered to contain incorrect passages. X is then a bastard text or a contaminated text. It may show, almost unpredictably, contradicting variants pointing to a descendance from different families. Contamination causes bias and hinders the development of chains and stemmas. In 2.8 a simple advice will be given to text-genealogists confronted with contamination. The type-2 limitation is a strict limitation, since it prescribes the use of only special variants. We can derive text-genealogical information from these variants only if the text versions show precisely two variants, each present in two or more texts. This implies that more complex situations, for instance with three or four different variants, are useless for the development of text-genealogical trees. Dearing (1974) claimed that a solution to break this severe type-2 limitation was present. He developed a set of mathematical rules which enable us to derive (type-2) information from these complex situations. I still do not intend to give away all the results of my research in this introduction, but in this dissertation, Dearing s approach will be falsified. Dearing s stemmatological ideas have been quite influential. They have been used, for instance, in the development of parts the stemma of the Bible. By using Dearing s approach, parts of the original Bible text have been reconstructed in a possibly incorrect way. Fourthly, this book does not concentrate on the emendatio, but on the recensio, the art of building chains and stemmas. As an example, the stemma of the fourteen versions of the medieval drama text Lanseloet van Denemerken will be produced. I stress that no emending attempts will be undertaken to restore or reconstruct (fragments of) the lost Lanseloet original. My reason for not

performing emendatio activities is simple. If we consider the classical Latin and Greek languages as dead, non-altering languages, it may be possible to restore an original text from around the year 100 on the basis of younger copies dating from around 1200. We assume that a classical Greek word in a text from the year 1200 looked the same in the year 100. A universally accepted standard Dutch language did not exist in the Middle Ages or the Renaissance. In the Low Countries, several dialects were spoken and written, which evolved in the course of time; these dialects were not dead. Therefore, it is quite dangerous to restore the lost Lanseloet text on the basis of the existing younger copies. We simply cannot be sure what a seventeenth-century word looked like in the original text from, say, the twelfth century. We do not even know the dialect of the original. This does not make the Lanseloet stemma worthless for text-critical purposes. It may still allow us, for instance, to reconstruct more abstract themes, subjects, etc., in the original Lanseloet van Denemerken text. Fifthly, this book will discuss differences between inductive and deductive textgenealogical research. Roughly speaking, we can say that science is either inductive or deductive. As we know, in inductionism the observation of (objective) facts is the basis for the development of theories. Often, mathematicalstatistical (inductive) techniques and computer programs are helpful in deriving knowledge from facts. Deductionism is based on (subjective) hypotheses, which are compared to observed facts. The facts may show that the hypotheses are incorrect, in which case the hypotheses are falsified or need to be adptated. If the facts are totally in agreement with the hypotheses, we say that the hypotheses are confirmed - not proved. The theoretical ideas in this book are hypotheses. They cannot be proved; they can only be falsified or confirmed. Logically, the correctness of the Lanseloet pedigree in this book cannot be proved neither; the tree is the result of the (automated) application of the hypotheses. In 3.2.2 we will discuss the opposition of inductive and deductive science. In the last decades a still growing number of scientists within the humanities seem to adhere to the law of objectivism. Their credo seems to be: science must be objective; knowledge must be derived in an objective way from objective facts; subjective science is a contradictio in terminis. This dominant philosophy, this paradigm of objectivism, has lead (too) often to a rejection of subjective thoughts and hypotheses, because they supposedly have an unscientific aura. Subjectivity, however, is not unscientific by definition. One fundamental of scientific research is that a scientist must show precisely what he or she is doing. Subjective ideas must be expressed in such way that they can be understood and criticized by other scientists; additionally, other scientists must be able to verify (and repeat) the procedures described by these subjective ideas. Once these ideas have met these criteria, they cease to belong to one person; they become intersubjective and scientifically acceptable. Anyway, science is, in my view, not 7

8 Chapter 1. Introduction about objective measurement of facts; it lies in the interpretation of the outcome thereof. Therefore, science is necessarily intersubjective. So far, we saw that this study covers five themes: 1. it offers a global introduction to stemmatology; 2. it is interdisciplinary and pays attention to biological-cladistic genealogical concepts; 3. it discusses hot text-genealogical items like contamination and the type-2 limitation; 4. it is concentrated on the recensio, not on emending text passages; 5. it pays attention to differences between deductive and inductive text-genealogical research. These five themes are the environment in which the main subject of this book is discussed: my method to build textgenealogical trees, which I call automated deductive stemmatology. Automated deductive stemmatology is a new text-genealogical approach, in which the computer is used to perform and test a set of (subjective) deductive hypotheses to recognize textual variants, with which trustworthy text-genealogical trees can be built. This subject covers the largest part of the book: in chapter 3 the basic textgenealogical principles and characteristics are explained and developed, in chapter 4 they are applied and in chapter 5 they are evaluated. Before sketching what automated deductive stemmatology is about, I want to make a sidestep to describe my motivation for developing a deductive method. In retrospect, I realize that this motivation grew out of three overlapping periods in my research. First, when I started my text-genealogical research, I was fascinated by one fundamental question: why do stemmatologists claim that text-historical trees can be built with all kinds of textual variants? I understood how Lachmannians used their specially selected common errors, most often quite eye-catching variants, as hereditary scars passed on to the descendants. Once we have found these scars we can determine the text-historical relationships between the text versions. Unfortunately, the Lachmannian selection of common errors was not clear and repeatable, which made it unscientific. The second phase started when I was studying current, modern alternatives for the method of Lachmann. I was, and still am, surprised by the easy, nonchalant way in which modern inductive stemmatologists use variants. Often the status of variants, the textual differences, seems to have become unimportant to them. They simply consider each textual difference as an objective, easily observable (objective) fact, although they sometimes exclude small or unimportant variants for unexplained reasons. They gather these objective facts and introduce them in statistical-mathematical software which builds, in an objective way, a tree out of them. But is such a tree a chain or a stemma, a text-historical tree? In inductive research, the objective facts must be related to the goal of the research. If I want, for example, to predict the weather, I can gather all kinds of objective facts in and around my house: paperclips, stones, papers, etc. It is obvious that I will not be able to predict the weather with these facts, even though

they are objective. In other words, a goal-oriented justification is necessary for the selection of objective facts in inductive research. I dare to say that, until now, a scientifically necessary justification for the use of all the variants as building stones for historical trees has not been presented in inductive stemmatology. Some inductive, statistical stemmatologists admit that their trees are not historical trees, but trees which show the spread of the variants in the text versions. In that case, I simply do not see the virtue of such trees. The third phase began when I became acquainted with biological cladistics. Cladists warn us that we must always ask ourselves whether a characteristic (in our case a variant) has the power to reveal the historical relationship of animals or plants. Only a very few characteristics have this relationship-revealing power. Cladists are convinced that a single convincing characteristic can provide more trustworthy information about the historical relationship than a thousand vague characteristics. They warn their inductive-statistical colleagues that statistical techniques like the Law of great numbers can filter out important characteristics. The three phases taught me that my first question, about how text-genealogical trees can and should be built, was still unanswered. As we saw, the answer to this question is not only important for deductive stemmatologists, such as myself, but also for statistical-mathematical, inductive stemmatologists. They are obligated to justify their choice of variants as well. In other words, the answer to my first question is relevant to both deductive and inductive stemmatology. We can even say that it connects both camps; both deductionists and inductionists must explain why they use (certain) variants as building blocks for their text-genealogical trees. In chapter 3, I will sketch in global terms some fundamental principles or hypotheses about stemmatology. These hypotheses should not to be seen as a well-balanced model, but rather as a theoretical framework with which a proper model for the detection of text-genealogical building blocks should be developed. Text-genealogical variants are differences in text versions that can be used to develop chains and stemmas. Later in chapter 3, I will derive concrete (sub)characteristics of text-genealogical variants from the hypotheses. This derivation is called the formalization process. While the hypotheses of the theoretical framework have an open or vague character, the characteristics derived from them are concrete, and can be applied to detect text-genealogical variants in the Lanseloet corpus. The set of text-genealogical characteristics can be seen as my method or theory to find text-genealogical variants to develop the Lanseloet tree with. Also in chapter 3, I will explain how these characteristics can be transformed into computer software. This transformation from the theory into software is called the implementation process. As we saw, the nineteenth-century method of Lachmann has been criticized, because its detection of common errors was not verifiable or repeatable. Furthermore, the method of Lachmann has an ad hoc character, because certain detection principles are not applied consistently and persistently to all the variants. Logically, I wanted the application of my 9

10 Chapter 1. Introduction characteristics to the Lanseloet texts to be scientific: verifiable, repeatable and consistent or persistent. In order to meet these criteria, I chose to let the computer to perform the characteristics, because it is an excellent apparatus to execute complex instructions rigourously, precisely and quickly. In chapter 4, the fourteen Lanseloet van Denemerken texts will be entered into the computer software that I wrote in the computer language SNOBOL/Spitbol. As far as I know, this was the first time that a computer was able to analyse texts according to a deductive text-genealogical theory. The scientific advantages of the computer as analysis instrument is evident. Once the theory (the characteristics of variants) has been programmed into the computer it is repeatable, it will be performed consistently and it can be checked afterwards. The main advantage of letting the computer perform a theory is that we can be sure that, in our Lanseloet case, it will apply all the programmed characteristics to all the ten thousands of variants consequently and quickly. It is almost impossible to treat such a large amount of variants by hand without making mistakes. The automation of the procedure to detect the variants to build a text-genealogical tree with, is also important for inductive text-genealogists, because they usually work with variants which are classified by hand. The computer output, specifically the Lanseloet text versions plus the computer-generated variation formulas built with the automated characteristics, is too large to be printed as an Appendix on paper. Therefore, I put it on the cd-rom. However, parts of the output of Appendix C are offered in in this book (see the List of Pictured Lanseloet Verses on p. 348). All variation formulas (dealing with precisely two competitive variants) will be ordered by the computer. They are presented in Appendix D on the cd-rom. Many two-variant formulas will be rejected by the computer because the variants concerned are not in accordance with one or more automated characteristics. We will only use the two-variant formulas that are in total agreement with all the characteristics. These formulas are presented in 4.4 to 4.7. Using the formulas, the chain of Lanseloet van Denemerken will be produced in two separate ways: by hand, using a simple algorithm explained in 2.4.4, and with the cladistic software package PAUP. Eventually, the Lanseloet stemma will be derived from the chain. In chapter 5, the stemma and the characteristics will be evaluated. We will compare the tree with the Lanseloet stemmas developed by other researchers. We will also investigate whether the chain and stemma are trustworthy and whether the characteristics can be confirmed, rejected or adapted. In the future, we can investigate if and how the text-genealogical characteristics can be applied to other texts as well. Notice, that the characteristics in this book are developed for the Lanseloet van Denemerken text versions (and other Middle Dutch texts). Undoubtedly, some characteristics will have to be reconsidered or reformulated before they can be applied to texts in other languages.

2. SOME CURRENT GENEALOGICAL METHODS 2.1. INTRODUCTION Generally, textual criticism aims to restore a lost original text based on younger (handwritten or printed) copies of the text. As a tool for text reconstruction, a genealogy or stemma of the exististing text copies is indispensable; it displays the mutual relationship between the texts and can be used to trace original text elements. In the last centuries several methods of building stemmas have been developed. One important theoretical study on text genealogy is W. Greg s Calculus of Variants (1927). In this standard work, Greg formulated some fundamental mathematical rules by which stemmas can be built. V. Dearing elaborated Greg s Calculus of variants in his famous Principles and Practice of Textual Analysis (1974). Both Greg and Dearing stated that there are several types of textual variation. Variation is the phenomenon that different versions of one text show different readings at certain places in the text. Greg claimed that only so-called type-2 variations give direct information about the shape (or chain) of a stemma: (..) it will be apparent that it is only such variation as we see in type 2 that is fundamentally significant (Greg 1927:23). This type of variation occurs when text versions show, at the same place in a text, precisely two competitive variant readings and when each variant reading (or variant) is represented in at least two text versions. Working with type-2 variations implies that four or more text versions are needed. Both Dearing and Greg prefered to work with type-2 variations. Additionally, Dearing developed an algorithm which adds up other types of variation formulas. The result is sometimes that new, extra type-2 variation formulas are created. This makes Dearing s method of dealing with variants more powerful than Greg s, since it accepts more types of variation as source of information to construct a chain. However, at the end of this chapter we will see that Dearing s method of adding up type-3 and type-4 variation formulas, which results in new type-2 formulas, is not adequate. In 3.2.2, we will also study an alternative method for building text-genealogical trees with, Quentin s zéro caractéristique method. In chapter 2, some current text-genealogical methods will be discussed and explained for the reader who is unfamiliar with text-genealogical methods. As such, it offers a rough introduction to the world of Greg and Dearing. Necessarily we will examine the method of Lachmann (also known as: the common error method), one of the oldest ways of generating stemmas. It works with so-called common errors: text versions which show the same common error go back to the same ancestral text. We will see that this is correct only under special circumstances (namely in the case of type-2 variations). Therefore, we will formulate a more accurate definition of the common error. A point of criticism by Bédier, one of the firmest opponents of text-genealogical methods, we will discuss as well.

12 Chapter 2. Current Genealogical Methods Biologists have ordered plants and animals into genealogies for centuries. It is possible that their ordering methods contain eye-openers for text-genealogists. That is why we will also examine current biological tree-building methods. We will see that the (biological) cladistic method does offer several interesting insights, which will be incorporated into this study. We will see that the cladistic software package PAUP can be used successfully in our text-genealogical fields. 2.2. A STEMMA AS A HISTORICAL IMAGE AND AS A TOOL FOR TEXT RECONSTRUCTION In this study we consider an original text to be the text which the author meant to put on paper (or vellum, etc.). 2 Everyone, who has ever copied a text, knows how hard it is to copy it without making mistakes; textual differences, variants, are introduced into the copy. This results in an inadequate text copy, which we call the text version. The more often a text is copied the greater the number of nonoriginal, deviant text versions. When we want to study a medieval text, we are very often confronted with the problem that the original text has been lost, while younger, varying, versions still exist. The absence of the original version challenges us to reconstruct it based on the younger versions. Like archaeologists and palaeontologists, we try to rebuild or give an impression of an original form by using its relics which are present in the existing text versions. The first step in text reconstruction is the determination of the genealogical interdependence of the versions, which is usually presented in a genealogical tree or stemma. In 2.3 and 2.4, we will see how stemmas can be constructed. Now, in the current section, we will concentrate on the way a stemma is used as a tool to reconstruct a lost original. To illustrate this, we use the stemma in fig. 2. That stemma is a fake stemma, only drawn for explanatory purposes. In this tree, which must be read top-to-bottom, 3 lost O-900 at the top represents the lost original text, which we are trying to reconstruct. Lost O was produced in the year 900 A.D. The lines in the stemma are the lines of descent. The (arbitrarily chosen) characters A, B, C, D, E & F at the end of these lines of 2 See Salemans & de Bonth 1990-91:212-215. The definition of the original text says that it is the text which the author meant to put on paper. It does not say that it was the autograph, the text which was actually put down on the paper by the author. In other words, according to this definition the original text never existed on paper. According to this view, the author is considered to be a copyist of the book in his head. While copying his imaginary book, he can make errors, just like any other copyist. Logically, the autograph may contain errors. It is impossible to reconstruct a perfect original text if it contains errors or imperfections. Ergo, if we assume that the original text did not contain errors, it must refer to the perfect text in the head of the author. 3 Biological stemmas are presented in the opposite way: the original species stands at the bottom, while the derived species are presented above the original; the root of the tree is presented at the bottom.