«The stemma of the story of Sinuhe or: How to use an unrooted phylogenetic tree in textual criticism Carsten Peust, Konstanz Abstract When a stemma ...»

LingAeg 20 (2012), 209-220

The stemma of the story of Sinuhe

or: How to use an unrooted phylogenetic tree in textual criticism

Carsten Peust, Konstanz

Abstract

When a stemma is constructed according to the traditional practice of textual criticism, one continually

needs to make originality statements, i.e. decisions about which of two different readings is original

and which is innovative. This kind of decision is hard to make and can be regarded as the major

challenge in stemma building. This also means that numerous instances of textual deviations, namely those which do not allow for originality statements, must be left aside.

I support here the use of an alternative method, so far unused in Egyptology, which does not require originality statements during the first step of stemma construction. The result of the first step is an unrooted rather than a rooted stemma. Only in a second step, the unrooted tree is assigned an orientation. This procedure makes textual criticism easier, more objective, and more reliable at the same time. I exemplify this method by reconstructing a stemma from eight manuscripts of the story of Sinuhe.

Traditional stemma construction Textual criticism is a method of dealing with texts transmitted in several manuscripts that display variant readings because of either copying errors or intentional text changes.1 This method was established in the early 19th century in particular by Karl Lachmann, which is why I will refer to it as “Lachmann’s method” from now on.

Lachmann’s method has been applied most often in classical and medieval philology as well as theology, but there have also been a number of applications by egyptologists, for the most part on the religious text corpora of the Coffin Texts and the Book of the Dead.2 The first step of Lachmann’s method consists of exploring the genealogical relations of the manuscripts and graphically representing them as a rooted phylogenetic tree or family tree (stemma). To use the terminology of graph theory here, the stemma consists of nodes, each of which represents a manuscript (either an attested one or a reconstructed ancestor), as well as edges (lines that connect the nodes), each of which represents the process of copying one manuscript from another, which is where text changes were introduced. The tree is rooted: There is one node at the top (the reconstructed archetype),3 and each node adds one or more changes to the whole subtree  I wish to thank Jean Winand who provided valuable comments on an earlier version of this paper.

1 Contrary to what is sometimes claimed, the principles of stemma building are completely independent from the specific factors that cause textual changes.

2 See the recent comprehensive overview by Backes (2011), to which Werning (2011, specifically vol. I: 51-82) is now to be added.

3 The archetype is the common ancestor of the known manuscripts. Since many manuscripts have usually been lost, the archetype may be considerably younger than the author’s original. The only 210 Carsten Peust that depends from it, so that the changes (or “errors”) accumulate towards the bottom nodes.

The fundamental assumption in stemma building is that all manuscripts that share a common error derive from a common ancestor which introduced the error. Put in terms of tree representation, all nodes sharing a common error, and only these, are to be located below one node which is assumed to have introduced that error. Each error thus helps to establish one subsection of the tree. By examining a number of different errors, a tree is successively constructed which is, hopefully, free of any contradictions.4 After that, the textual history of the manuscripts can be read off the stemma, and conclusions can be drawn about the original text (archetype) located in the top node.5 To be somewhat more precise, the philologist needs to solve three tasks while

constructing the stemma according to Lachmann’s method:

(1) Finding text changes that are unlikely to be made by different copyists independently (the change must be, as I would put it, unreproducible).

(2) At the same time, the text changes should be so grave that succeeding copyists cannot easily have corrected them back into the original text (the change must be uncorrectable).

(3) Last but not least, the philologist needs to be sure about the direction of the changes. One can use only those textual differences for which it can be judged which of the variant readings is closer to the original and which one added an error. To use the terminology introduced by Jürgens (1995: 10) here, one has to rely on digressions (or errors, variants with known direction) rather than on differences (variants with unknown direction).

While all these three tasks involve some amount of subjective judgement on the side of the philologist, requirement (3), the originality judgement, is the most difficult one to fulfill. I would even say that it is almost unsolvable prior to the reconstruction of the stemma, at least in the field of Egyptology.6 What philologists try to do here is to

look for changes:

● which are obvious misunderstandings or deteriorations of a text, way to go further back than the archetype is by internal reconstruction, e.g. by the emendation of implausible passages, but this is beyond the task of stemma building.

4 If contradictions are found, they may indicate several things: Either, certain manuscripts were copied from more than one ancestor (contamination, for which see below). Or, the philologist did not select his textual changes well, so that he included changes which could be introduced independently by more than one copyist, or which copyists who encountered that error in their exemplar succeeded to correct back into the original text. As such complications are met, the task may be reformulated in a more modest way so as to construct not a tree void of contradictions, but a tree in which the number of contradictions is minimized. Conversely, it may happen that more than one tree can be reconstructed because not enough errors were found to decide on a particular tree.

5 The methodology of how to draw conclusions on the archetype by means of a stemma is not discussed here since I am not attempting that in the present paper. The principles are rather selfevident and are discussed at length in the literature on textual criticism (e.g. West 1973).

6 A large part of West’s (1973) book deals with how to identify corruptions in Greek and Latin classical texts. I am not in a position to assess how reliable such judgements really are in classical philology, but our knowledge of the Egyptian language is clearly not sufficient to make decisions of a similar kind.

The stemma of the story of Sinuhe 211 ● which are obvious omissions or repetitions (e.g. by aberratio oculi), ● which seem to be motivated by the models of younger linguistic strata of the language, ● or finally, one applies the rule of thumb to consider the lectio difficilior as primary, assuming that copyists tend to introduce readings that are easier than what they found in the original (so-called banalisation or trivialisation).

I believe that all such judgements must be suspected as highly subjective and unreliable, at least much more so than judgements on the first two requirements.

Stemma-like phylogenetic trees are used in other sciences as well, for example to represent the evolutionary connections between organisms in biology or to represent the relationships of genetically related languages. The originality judgement is the hardest requirement to fulfill in historical linguistics as well. It is necessary, in view of a lexical or grammatical difference, to decide which of the variant forms is innovative and which is inherited. Historical linguists use the phrasing that genetic groupings must be based on “common innovations” rather than on “common retentions”. In practice, this decision is extremely difficult to make in the absence of historical records, which is the major reason why no phylogenetic tree of even such a wellknown family as the Indo-European languages has so far been agreed upon.

Constructing an unrooted tree In order to overcome this obstacle, I would like to advocate the use of another method which is not new but has never before been discussed in Egyptology. This method, which I call “Greg’s method” after its inventor, was applied e.g. by Greg (1927), Dearing (1974), Dees (1976)7, Salemans (2000) and Wattel (2004), among whom Salemans provides the most accessible presentation and is the best reading to start with. I will only use the core idea of the method as already established in Greg’s original work. Some refinements and elaborations introduced by the subsequent authors are certainly helpful in more complex cases but need not be taken into account for my present purpose.

Greg’s method of tree reconstruction simply omits the third requirement, namely the originality judgement of variant readings, and uses only the first two requirements in selecting textual differences. As we drop the third requirement, the result will be an unrooted tree rather than a rooted tree. The unrooted tree still shows relationships between manuscripts but makes no assumption about the directionality of edges, nor does it indicate where the archetype is located. The representation in form of an unrooted tree can be rotated or mirrored without any change in its meaning. After an unrooted tree has been constructed, the root of the tree may be identified in a second step, as will be described below.8 7 With no reference to Greg; this seems to be an independent discovery.

8 Dees (1976: 485) describes the procedure as follows: “on ne considérera, dans une première phase, que les structures non-orientées, en nombre beaucoup plus réduit; la deuxième phase consistera à choisir, dans l’ensemble exactement déterminé des orientations possibles, celle qui convient. Il est vrai que cette dernière opération peut être très délicate, mais on sait au moins quelles sont les alternatives à considérer.” 212 Carsten Peust Consequent to this, many more variants can be exploited in Greg’s method than in Lachmann’s method because no originality judgement is required. One specific requirement of Greg’s method should be noted, however. The Lachmann-like stemma reconstruction can be based on text passages attested in three or more preserved manuscripts, provided that a directionality judgement is made.9 In contrast, Greg’s method must be based on text passages with at least two variant readings each of which is attested in at least two manuscripts.10 That is, only text passages with four witnesses, at minimum, can be exploited.11 This also becomes evident by considering the graphical representation of nodes in an unrooted tree (see figure 1): While three nodes have one single representation in an unrooted tree, distinct groupings of nodes in an unrooted tree only become possible with four nodes.12

Figure 1: All possible unrooted trees of sizes 3 (left) and 4 (right)

I impose two restrictions on the trees here and throughout this paper with the aim of

limiting the combinatorial number of possibilities to be considered:

(1) I assume that no preserved manuscript is the exemplar from which another manuscript represented in the tree was copied. In terms of tree representation, this means that preserved manuscripts are always represented as terminal nodes rather than internal nodes.

(2) I assume that no more than two manuscripts represented in the tree were copied from the same exemplar. In terms of tree representation, this means that the tree is bifurcating, i.e. no more than three edges connect to a node.

These are reasonable restrictions that greatly simplify the construction and handling of the trees without the danger of introducing any major damage to the reconstruction.13 9 At minimum, one needs to find two manuscripts A and B sharing a common error against a third manuscript C which preserves the genuine text. Provided that the variant reading of A and B is judged secondary, there is sufficient argument for grouping A and B together against C.

10 Singular readings can never reveal genealogical relationships, neither in Greg’s nor in Lachmann’s method of textual criticism. Greg (1927: 19) states: “Since every manuscript contains variations from its immediate source, any reading supported by one manuscript alone may have originated in that manuscript, and such a reading therefore cannot, without further analysis, throw any light on the relation of the manuscripts of the collateral group”.

11 Greg (1927: 21) calls this fact the “ambiguity of three texts”.

12 In a tree with four nodes, the possible groupings are [AB][CD], [AC][BD] and [AD][BC]. A grouping such as [AB][CD] implies that either in [AB] or in [CD], but we do not know which, a

text change was introduced. ––– With larger trees, the number of possible groupings rises quickly:

For n terminal nodes, it is (2n–5)! / ((n–3)! · 2n–3).

13 If one of these restrictions should be mistaken for a given instance of manuscript transmission, this would only introduce a local fault into the reconstructed tree: (1) If it should indeed have occurred that a preserved manuscript served as the exemplar of another manuscript, our reconstruction The stemma of the story of Sinuhe 213 The procedure of constructing an unrooted tree by Greg’s method can now be described as follows: We collect text passages ● attested in at least four manuscripts, ● with exactly two14 variant readings each of which is attested at least twice, ● where neither of the variant readings could easily have been created from the other more than once independently (the difference is unreproducible), ● and where the distance between the variant readings is so substantial that none of them could easily have been corrected back into the other by succeeding copyists (the difference is uncorrectable).

The last two points may be summarized by saying that only significant variants must be used.

An unrooted tree is then drawn and each textual variant is assigned to an edge so that all manuscripts on either side of that edge agree with one another regarding their readings of the variant.15 It should be ensured that there are no contradictions,16 and also that the tree is the simplest possible tree to fulfill these principles.