Readings in Computational Stemmatology

Introduction

What follows is a to-do list in the form of a thin, unsatisfying, much too English-language-oriented, somewhat (Greek and Latin) Classics biased, bibliographical essay. This page may, circumstances permitting, grow over time as I collect references and record brief remarks on a few of them. I have a lot to add and it will take time to dump my private list of resources into something worthwhile, still longer to find more resources, and even longer still to say anything sensible about them. Eventually---again, circumstances permitting---this page may be of use to someone other than me.

If you're impatient for an excellent, recent, and far more comprehensive (and more richly multilingual) bibliography, I highly recommend Trovato (2017). Unfortunately it doesn't provide a full bibliography, but it makes up for it with a select bibliography supplemented by short bibliographical essays throughout the book paired with specific topics. Most's bibliographical supplements to his translation of Timpanaro (2005) also contains much helpful recent material.

You can find all the references on this page collected at the bottom of this page. You can find my complete bibtex file here.

A full history of edits to this page can be found here.

If you see an especially glaring error or omission, please feel free to let me know at chris@chrisyoung.net.

Traditional Textual Criticism

Introduction to Textual Criticism

An accessible and engaging introduction to the whole world of ancient, medieval, and renaissance scholarship and textual criticism can be found in Reynolds and Wilson (1974) (now in it's scrupulously updated fourth edition, Reynolds and Wilson (2013)).

A superb introduction to most of the topics discussed on this page, including both traditional Lachmannian stemmatology and computational stemmatology, is Trovato (2017). Trovato gives a spirited, lucid defence of neo-Lachmannian stemmatology and offers some skeptical remarks about (biologically oriented) computational approaches to textual criticism. If you have any interest in the field you should read this book.

Karl Lachmann, working in the second half of the 19th Century, is associated with a leap forward in rigour in the establishment of classical texts. Timpanaro (2005), whose relatively recent translation from the Italian we owe to Glenn Most¹, traces the development of the method now associated with Lachmann's name. One irony: Lachmann is neither the first to develop the method now associated with his name nor does his work provide an especially good example of the method in practice. Although the main text of Timpanaro (2005) is devoted to tracing the historical development of the method, the appendices to the work wrestle with various technical challenges, and are still worth reading. (At a slight remove from our main topic, Timpanaro (1976) has also been translated into English.)

Earlier introductions to the practice of textual criticism include Maas (1958) and West (1973). Both books are starting to show their age, the former in particular. But both are worth reading, including Maas (1958), which is a little gem of a book. West (1973) has a deeper discussion of the phenomenon of contamination, if I recall, and is a great read too. An important role in 20th century debates about stemmatological method was also played by (the Italian language, and still untranslated) Pasquali (1952). I have not been able to review Reeve (2011) yet, but it looks like a valuable collection of papers.

A very different review of the history of textual criticism (focusing on New Testament textual criticism) and biology is offered by Lin (2016) (see this review for a bit more). I hope to comment on this strange book at some point.

Finally, I do not discuss here the "Coherence Based Genealogical Method", which is all the rage in New Testament studies.

Lachmann's Method

Editing a text is traditionally divided into recensio, selectio, and emendatio. Recensio refers to the construction of a stemma --- a family tree relating witness to the text --- selectio to the selection from manuscript variants in a way constrained by the stemma, and emendatio to the use of conjecture to repair cruxes. This page is entirely concerned with techniques, old and new, for the first stage only.

And now for Lachmann's method, as it is often called:

1. Collate your manuscripts against some reference text that you hope is not too far from what the author wrote. Departures from the reference text are provisionally considered "errors." As work continues and readings are reassessed our best guess as to which readings are errors may change. What stays constant is that shared readings are only considered evidence of provenance when they are errors.

2. Discard polygenetic errors, that is, those that may easily occur spontaneously across a tradition --- or may be easily corrected by a copyist --- and so are unlikely to carry information about provenance: "only monogenetic errors should be used as indicative errors" (Trovato (2017), p. 56).

3. Take the pattern of indicative errors and use it to build a tree called a stemma. If B contains all the errors of A and adds some further errors ("separative errors") of its own, we conclude that B is a child of A. If A and B each contain separative errors not shared in common with any other manuscript but also conjunctive errors together against the rest of the manuscripts, posit a parent for both that is the source of the conjunctive errors. Add more manuscripts to the tree in the same way.

The resulting stemma, if it can be reconstructed with confidence, provides the basis for an edition of the text. A stemma can cut through a great deal of noise and focus attention on the essential choices between variants.

This first step of editorial practice (redactio) along with subsequent refinements is also sometimes called neo-Lachmannian stemmatology. An alternative to this is the so-called "best manuscript" method championed by Bédier, which selects a "best manuscript" as the basis for a text and departs from it only when absolutely necessary.

We call the root of the stemma the archetype, and usually designate it with a ω. Reconstructed nodes are often called hyparchetypes and represented by Greek letters. Consider the following stemma, which follows these conventions:

And suppose that we observe a number of variants at a particular location in the text, where for convenience each variant is assigned a number:

Manuscript	Variant
A	1
B	2
C	2
D	3
E	4
F	4

Without a stemma, an editor has to consider 4 possible variants at this location. But the stemma suggests that D, E, and F can be discarded as witnesses. For they are copies relying entirely on C, whose readings we already possess. We may confine our attention, then, to a simplified stemma:

Now our attention is restricted to a subset of variants:

Manuscript	Variant
A	1
B	2
C	2

Variant 2 is represented in twice the number of manuscripts as variant 1. No matter. Both variants have an equal probability (as far as the stemma is concerned) of being in the archetype of the stemma. The editor must now choose. The stemma can say no more.

Now consider a trifurcating tradition:

and the following table of variants for a particular location:

Manuscript	Variant
A	1
B	2
α	2

We must suppose that our archetype, ω, contained variant 2 and that it is A which is in error. Otherwise we must suppose that a monogenetic variants occurs twice by coincidence.

A stemma may in many cases therefore reduce the number of variants that need to be carefully considered by an editor. Later, it eases the burden on readers, since far fewer variants need to be reported in the apparatus of the text.

It is important to remember that even the most impeccably grounded stemma is a representation of the logical relationships between the surviving manuscripts, not a claim about the specific details of actual historical transmission. Single lines may connect two manuscripts to a hypothesized "parent" manuscript, but this does not imply any confidence about the number of steps between those manuscripts and the parents.

To review:

A stemma allows us to reduce the number of witnesses to the text that need to be considered.: Wherever a manuscript is found to depend on another it may be discarded.

The number of manuscripts that contain a given reading is irrelevant: What matters is the logical relationship between the witnesses, more specifically, which ones provide independent testimony about the archetype.

The value of a witness is determined solely by the stemma: There is the value a manuscript has at auction and the value it has for an editor. The former has to do with age and provenance and beauty and chains of ownership and so on.² The latter is solely a function of what other witnesses have survived and how they are related. If I copy the original of a tradition and then lose the original, my copy will fetch nothing at auction (my handwriting is a disgrace, my paper cheap) but it is a witness to the original that is independent of every other, and this makes is extremely valuable. This is true even if my copy is sloppy, since the errors I make will be different from those made in other lines of descent and I will likely preserve readings lost in other parts of the tradition. A copy of my copy is of no value --- we already have the first copy. If that first copy is lost, my copy of that copy is suddenly promoted in worth as a witness.

The number of errors in a manuscript is irrelevant to its value as a witness: Of course a large omission in a manuscript will prevent it from weighing in on affected passages but in general an error-ridden manuscript may, in virtue of its place in the stemma, offer readings independent of other witnesses and so have great value.

Recentiores non deteriores: That is, more recent manuscripts are not worse. Another way of putting all this is that the age of a manuscript is irrelevant to its value as a witness. Now, it may well turn out in practice that older manuscripts offer a more direct route to the archetype than later ones, or that later manuscripts will turn out to be copied from some older and still surviving manuscript, and so can be eliminated from a stemma. Nevertheless, the veneration of old manuscripts at the outset of an investigation in textual criticism is the confusion of auction value with editorial value.³

If this rather abstract argument grounded in stemmatatological reasoning fails to convince, see Browning (1960) for helpful discussion of evidence from paleography and indirect testimony about Byzantine and medieval copying practices. He collects a number of examples in which later manuscripts preserve now lost readings from sources which were clearly independent of all other surviving witnesses, and concludes that "[w]hat emerges from the above examples is that it was not an exception for a thirteenth or fourteenth century scholar to have access to early tradition, it was the rule" (p. 18). Note finally that it is not just manuscripts that fall under this principle: Reeve (1983) argues that early editions should be considered witnesses to the textual tradition as well. And see Reynolds and Wilson (2013), p. 218 for further examples and discussion.

Now consider several common complaints about Lachmann's method:

Circularity: We assume we know what the errors are and then use them to build a tool which will aid us in distinguishing between good and bad readings (that is, errors!).

Scalability: A complicated tradition for even a medium sized text contains an enormous quantity of information in the form of variants and metadata about those variants. Scholars have groaned and sometimes collapsed under the weight.

Insufficiency: It is not always clear in practice which variants are monogenetic and which polygenetic. A conservative approach may leave us with "a much reduced data set, where the correct stemmatic theory is bound to be underdetermined by data," as a friend recently complained to me.

Bédier's Paradox of Trifurcation: The Bédier mentioned above noticed a century ago that very few published stemmas contain trifurcations let alone higher order splits. They consist almost exclusively of bifurcations. Why? There seems to be no rational explanation for this, since manuscripts were surely copied multiple times from the same exemplar.

For Bédier the phenomenon suggests that stemmas are figments of the imagination, rather than rational reconstructions. A bifurcation allows an editor an irresistable degree of freedom in the construction of a text, since the stemma itself can't determine the correct reading. A higher order split by contrast ties an editor's hands because the reading of the archetype can usually be inferred.

Besides wishful thinking, Bédier also thought that the technique itself has a methodological bias in favour of bifurcation. (I will say more about this in a later revision.) His influence loomed large over the development of 20th Century textual criticism and is still felt, especially in French scholarship.

Contamination: The method as stated above assumes that the manuscripts are copied from single exemplars. But the reality is more complicated, indeed far more complicated. As a friend remarked to me recently some medieval manuscripts are more like critical editions themselves and they represent a lot of work not just in copying but in collating other witnesses to the text. These collations are often noted in the margin or above the line and when these manuscripts were themselves copied scribes were at liberty to choose between readings. Past a certain point it's not clear that Lachmann's method can determine a workable stemma.⁴

Let us take these in order.

One response to the concern about circularity is to affirm that there are clearly monogenetic errors and that in sufficient quantity they can be used to reconstruct a stemma that can then be useful in determining between harder cases. The long and often implausibly successful story of philology provides some support for this.

I will have more to say about the issue of scalability below.

Bédier's Paradox of Trifurcation was apparently a spur to Pasquali (1952) as well as to Timpanaro (2005) (see, especially his Appendix C, where he wrestles at length with this problem), and Trovato (2017) credits Bédier with stimulating much of the development of 20th Century stemmatology in reaction to his criticisms.

Trovato (2017) features a long and interesting discussion of Bédier's Paradox of Trifurcation. There is the point stressed by Timpanaro that it can be difficult in practice to distinguish between bifurcation with contamination and trifurcation. But I think Timpanaro sensed that this was only a partial explanation for the phenomenon. The best response to the problem is rooted in work that realistically simulates manuscript reproduction and survival. For this, see especially Weitzman (1987). Assuming realistic rates of manuscript loss (discussed carefully and with examples by Trovato (2017)), most stemmas will consist of bifurcations only. Contra Bédier, the preponderance of stemmas featuring bifurcation can be explained by the way that radical loss shapes a tradition, and not by the wishful thinking of scholars.

Contamination presents the most serious challenge to any stemmatological method, traditional or computational. Maas's famous pronouncement "Gegen die Kontamination ist kein Kraut gewachsen" ("No remedy has been developed against contamination") is often quoted in discussions of the subject.⁵

The Mathematics of Stemmas

Buneman (1971), an early paper on the mathematics of manuscript descent, was also important in the development of modern phylogenetics. A trio of papers by Colin Flight: Flight (1990), Flight (1992), and Flight (1994), works out theoretical problems with stemmas and contamination. See also Hoenen et al. (2017).

Computational Stemmatology

The field of computational/mathematical stemmatology has flourished in recent years, motivated especially by the concerns about circularity and scalability noted above.

As of this writing, Roelli (07 Sep. 2020) is not yet published and the table of contents is not available, but the summary on the publisher's page is promising.

Biologically-based and Traditional Computational Stemmatology

The parallels between problems presented by manuscripts and their descent and by biological sequences (nucleotides and proteins) and their evolution are striking. Not only is there error introduced by copying in both cases, but the phenomenon of contamination has an analogue in gene transfer. See Howe et al. (2004) for a useful discussion of parallels between phylogenetics and stemmatology.

Trovato (2017) usefully distinguishes between computational approaches that aim to automate the more labourious aspects of Lachmann's method and those that aim to apply biological software to the problem. It is very clear that the dominant approach to computational stemmatology is biologically-inspired. He cites Salemans (2000) (which I have not yet read) as an excellent example of the former approach. He is highly critical of the latter, dominant, approach.

An important early spur to the development of the field was a challenge issued in 1991, which is reviewed in Robinson and O’Hara (1992).

In a future revision of this essay I will have much more to say about specific tools and approaches.

The Use of Artificial Traditions

A sensible first step in the development of algorithms and techniques for handling manuscript traditions is probably to generate artificial traditions to various specifications so that we can observe how well different techniques fare under different conditions (e.g., levels of contamination, mss loss, etc.).

This is not an original idea. A friend informs me that Pasquali mentions that Kantorowicz had the idea in 1912: "He asked a batch of his students to make copies of a Latin document, and his task was to edit it properly; his conclusion was to stress how fallible this process is." The idea has been carried forward in more recent work, including Spencer, Davidson, et al. (2004), Baret et al. (2006), Roos and Heikkilä (2009).

Other Work

This essay is a draft-in-progress and what follows is a mostly unstructured dump of readings, many of which I haven't gotten to you.

A trove of interesting papers can be found in Van Reenen et al. (2004). These include, but are not limited to Howe et al. (2004), Mink (2004), Schmid (2004), Wattel (2004), Spencer, L Mooney, et al. (2004).

I list the remaining papers in this category in chronological order:

Silva and Love (1969)

O’Hara and Robinson (1993)

Gjessing and Pierce (1994)

Schmid (2004)

Eagleton and Spencer (2006)

Andrews (2009)

A number of resources for textual criticism are online:

[http://www.textualscholarship.org/](http://www.textualscholarship.org/] (the source of some of the references on this page).

Interedition

Centre for Manuscript Genetics

European Society for Textual Scholarship

I list the remaining items in this category in chronological order:

Silva and Love (1969)

O’Hara and Robinson (1993)

Gjessing and Pierce (1994)

Robinson and O’Hara (1996)

Salemans (2000)

Mooney et al. (2001)

Spencer et al. (2002)

Spencer et al. (2003)

Howe et al. (2004)

Mink (2004)

Robinson (2004)

Schmid (2004)

Spencer, L Mooney, et al. (2004)

Spencer and Howe (2004)

Spencer, Wachtel, et al. (2004)

Robins (2007)

Windram et al. (2008)

Andrews (2009)

Schmidt and Colomb (2009)

Schmidt (2009)

Dekker and Middell (2011)

Roos and Zou (2011)

Andrews et al. (2012)

Howe et al. (2012)

Andrews and Macé (2013)

Nassourou (2013)

Hoenen (2015a)

Hoenen (2015b)

Tools

Manuscripts often contain information that is both genealogically relevant and tricky to capture. The Text Encoding Initiative attempts to standardize the many issues that arise in trying to manage multiple copies of a text.

The Interedition Github page contains a number of repositories for projects dealing with textual criticism, including especially the collatex project.

The upshot of all this: I think it's absolutely worth making low-effort experiments with biological software, as a check on our results. But what I'd prefer to focus on computationally is developing a tool that allows us to automate the tedious parts of traditional Lachmanian stemmatology. It should support queries and outputs relevant to this task. E.g.,

Build a stemma using separative and conjunctive monogetic variants Given a proposed stemma, print out all counterevidence to that stemma (that is, list all conjunctive monogenetic variants that don't fit the stemma and so suggest a bad stemma or contamination). Allow repeating the above while easily altering (perhaps in another column of the spreadsheet?) the base text, and so assumptions about which readings are in error. Identify all cases when counterevidence to the stemma would in fact support the stemma if we altered a reading in the base text. Allow all the above while excluding manuscripts so that we can quickly experiment with subsets of the tradition. Allow us to build a stemma with a subset of manuscripts and then add manuscripts one by one in a specified order. Give us complete visibility into the process and data supporting any particular conclusion Allow error checking of variants and reporting List all the editorial choices left open by a given stemma

A command line tool (that is, one you run from a terminal and not a graphical user interface) to do this would be manageable I think. A richer interface on a website seems to me like a lot more work, but then it's been a long time since I built a website. Perhaps Nick will disagree about how much more work a richer interface would be.

Requirements for the suggested command line tool: A column in the spreadsheet for departures from our base text (Bekker's edition) A way (another column?) of distinguishing in our spreadsheet between monogenetic and polygenetic variants

Phylogenetics

A rather mathematically demanding introduction to Phylogenetics is:

Semple and Steel (2003)

An introduction to phylogenetics which succeeds in being less intimidating without apparent sacrifice of mathematical rigour is:

Huson et al. (2010)

As the book's title suggests, its emphasis is on phylogenetic networks as opposed to simply trees, which should make it of special interest of stemmatologists.

And speaking of phylogenetic networks, don't miss The Genealogical World of Phylogenetic Networks, a blog that is chock full of references and interesting posts.

Complete List of References

Unless otherwise specified, original content on Christopher Young's website is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License.
Contact me at chris@chrisyoung.net.