Thursday, October 24, 2013

The Genealogy of Manuscripts and the CBGM

Whilst working on my paper for the upcoming SBL in Baltimore (abstract here), I wrote a section which I cannot use in the paper, but would love to get some comments on. It is not about the Coherence Based Genealogical Method (CBGM) as such, but about a potential misconception of its most fancy result, the Global Stemma.

"Once we have produced a global stemma there may be an unintended side-effect for the unaware reader, which, in an ironic twist, repeats the error we have tried to avoid by doing away with the notion of text-types.

A problem in the use of text-types is that the voice of a whole group of manuscripts is reduced to a single voice, even when it is clear that, in the current state of play, there is no evidence that the whole text-type goes back to a single ancestor. To reduce manuscripts to simply being members of, say, the early Alexandrian type is simplifying a complex reality to the level of becoming only a near-truth.

However, with a global stemma a similar abuse of the constructed genealogy is possible. The error is this, that manuscripts that are not on the top layer but on the second, third or lower level, are seen as being dependent on the manuscript(s) above them and therefore can be discounted (as per Westcott-Hort). But of course, in the CBGM this is not the case, since the rationale behind the global stemma is that it only gives the closest 'potential ancestor' but not all 'potential ancestors'. I assume that many of the manuscripts on the higher levels still need the Ausgangstext (initial text) as one of their potential ancestors to explain all their variants and are therefore still independent witnesses to that initial text. Yet the danger of reduction is real: instead of falsely reducing individual witnesses to being mere members of a text-type, witnesses could be reduced to having little or no independent value because they do not have the Ausgangstext as their closest ancestor.

The main reason for this apparent misconstrual of the meaning of the global stemma is that the term 'genealogy' is employed in its own singular way — not as indicating the ancestry of the text in the strict meaning of the word, but rather as pointing to a text with a higher proportion of earlier readings. And the difference between these two is huge! The term 'genealogical relationship' as used in, say, the Introduction to the Editio Critica Maior (ECM) suggests much more than it is actually supposed to mean."


  1. If one follows the genealogical principle of the classical textual critics like Paul Maas, then I'd say an important purpose of the stemma is to enable the critic to identify and perform the step of eliminatio codicum descriptorum. Maas hold that his is applicable both to an abschrift and to the situation where its hyparchetype can be reconstructed without its help. Of course, contamination creates big problems with this.

    Despite the name "stemma", I don't view the CBGM Global Stemma as a stemma codicum in the traditional sense but more as a text flow diagram where the chart indicates the relationship, not of manuscripts, but of the states of their text.

    Also, it is my understanding that each state of the text can have a number of potential ancestors (i.e., ancestral states of text, not manuscripts), not all of which can be shown on the diagram. In fact, I think the image only shows the first potential ancestor.

    So, yeah, though the CBGM uses words likes "genealogical," people should not assume that it is meant in the traditional stemmatic sense where manuscripts can be eliminated or "discounted." But my understanding of the CBGM is still that of an outsider looking in.

  2. If my understanding of the CBGM is correct then the diagram you've attached above isn't the global stemma, but is rather a local stemma for 1 Peter 1:1/8.

    An example of the global stemma (or at least an excerpt) can be found on slides 566ff (where Mink is comparing global and local stemmata) here

    The global stemma does show multiple lines from all neceesary ancestors (as derived from the optimal substemmata).

    If that's incorrect then please let me know!

    (reposted to correct link)

  3. Correction to my comment - it's not a local stemma, but a textual flow diagram for 1 Peter 1:1/8.

  4. Andrew, you are right, and the terms should be distinguished. Still, the textual flow diagram is the best prediction of the shape the global stemma will take (when following the red lines in Mink's presentation). Why I chose this particual flow diagram is that there is only one tiny variant in a witness somewhere at the bottom of the flow, so that the overall structure approximates the main relationship between any two witnesses within the global stemma.

    On a related note, has anyone a nice image of the global stemma of the text as given in ECM2? Or hasn't it been published yet?

  5. Dirk, I don't think it's been published yet because of computing power required. In the closing paragraph of Mink's "Contamination, Coherence, and Coincidence" (p. 205), he says, "The current challenge, however, is the construction of optimal substemmata for witnesses with large numbers of potential ancestors. In many cases overwhelming numbers of possible substemmata emerge."

    Since the connections in a global stemma are "absolutely determined by the connections in the substemmata" (Stemmatology II, p. 75), it seems to follow that they haven't been able to complete a full global stemma yet. I say "full" because certain sections have been constructed and these shouldn't change because their substemmata are secure (see page 76 in Stemmatology II).

    But I should probably add my own "if I understand CBGM correctly" caveat.

  6. Textual flow diagrams and global stemmata must definitely not be confused. The ways of producing textual flow diagrams and of constructing a global stemma are completely different and, accordingly the claims of them are qintessentially different.
    Therefore, it is impossible to base the assessment of the method of constructing global stemmata on textual flow diagrams, which only show the most closely related (= the most agreeing) potential ancestors of each witness in an attestation. Yet, regard, please, potential ancestors are not stemmatic ancestors (as found as sources of a given witness in a global stemma resp. in an optimal substemma). Genealogical coherence as documented in textual flow diagrams meets other conditions than stemmatic coherence in a global stemma.
    Textual flow diagrams may be an often more or sometimes less good prediction of which witness is one of the sources of which witness in a global stemma. Yet, an absolutely reliable prediction of stemmatic ancestors is impossible. The reasons are (i) that an optimal substemma (= every element of a global stemma consisting of a witness and its stemmatic ancestors) must show a combination of ancestors which is as small as possible and, therefore, often does not contain the most closely related potential ancestor which we see in textual flow diagrams, and (ii) that an optimal substemma must be philologically plausible at every place of variation. Therefore philological reasoning is crucial in the procedure leading to optimal substemmata.
    In this context, potential ancestors play an important heuristic role. Yet, in CBGM, prior variants of variants in a given witness are not necessarily found in potential ancestors only. Please, regard the paragraph "Prior variants found only in non-ancestors" (Studies in Stemmatology 59-63).

    Dirk, as regards the abstract of your SBL paper: Is the different positioning of 33 and 2344 caused by genuine CBGM procedures or different basic assessment of variants (which, of course, would result in different values for prior/posterior variants in CBGM)?
    You wrote "in many cases this leads to statistically insignificant outcomes". Which are the statistical approaches your statement rests upon? What is the null hypothesis for testing statistical significance? Which test? Which significance level? I ask because CBGM makes no statistical statements on samples and populations.
    I cannot attend the SBL meeting, but I am very interested in your paper.

  7. Thanks Gerd for your encouragement, and I will do my best to make every effort to discuss the paper in detail with you at some stage (first have to finish it, for starters). Where statistical relevance comes in, is when we take into account the confidence we have in a particular local stemma. In many cases, for small variants, this may be 100%, in other cases this will be considerably less. More detail will be in the actual paper.
    So far, the CBGM has always stayed clear of including confidence levels, and I think it will be hard to incorporate them throughout the method.
    As for 33 and 2344, the point is not so much that there is anything wrong with the principles or procedures of the CBGM, but rather that there are some issues with the hardness of the data set. The, at times, rather 'soft' data that manuscripts provide (in the sense of judgements that already need to be made at the data gathering stage), are turned into 'hard' data for use within the CBGM.

  8. Thank you very much, Dirk.