The CBGM, invented by Gerd Mink, is not the easiest method to understand. I think we would all agree on that. Various attempts have been made to explain it including mine and Tommy's. Given the learning curve it takes to understand it, misunderstandings are inevitable. I addressed some of these in my PhD thesis. But what is the most common one? And what does the inventor think is the most common misunderstanding of his own method? Here is Mink's answer from the recent Festschrift for Holger Strutwolf:
The most common misconception when using the CBGM is that the role of potential ancestors in constructing stemmata is not understood, and the connections in textual flow diagrams are read like connections in a stemma. However, one must resist the suggestiveness of these graphs. The textual flow diagram is not a stemma. (p. 579)
I would agree with Mink on this. I found this to be the case in my dissertation. Here is what I say there in my chapter on the Harklean text:
textual flow diagrams should not be used for the purpose of studying the text’s overall development. Their simplicity can have a mesmerizing effect. But their clarity can become a hindrance to their proper use when it tempts one to make more of the distinct relationships than is appropriate. Most importantly, they should not be treated as stemmata. (p. 88)
In our intro to the CBGM, Tommy and I have a subsection in ch. 4 devoted just to this point. There we say this:
The fact that there is always far more genealogical data than is shown in the textual flow diagrams brings us to our second caution: a textual flow diagram is not a stemma. Textual flow diagrams reduce and simplify the total genealogical picture, somewhat like a map of the London Underground. They are very good for studying coherence at a point of variation, but they are not good for studying the history of the text on a larger scale. Because a textual flow diagram usually connects each witness with one potential ancestor and does so by agreement whenever possible, we need to resist the temptation to interpret it as a traditional stemma, giving us a map of the text’s historical development. (p. 92)
So, heed the warning: Do not use textual flow diagrams as if they were stemmas. They are neither designed nor intended for use in making simple historical judgments about manuscript relationships. Along with that, do not use them to try to critique the CBGM as being non-historical. In short, do not use them for historical judgments in a box with a fox or in a house with a mouse, do not use them Sam I am!
I do not like CB&M, Sam I Am.....
ReplyDeleteYou might not want to read to the end of the book where it turns out that "textual flow diagrams as stemmas" turn out to be really delicious.
ReplyDelete"In short, do not use them", well said :P
ReplyDeleteseriously though, my major criticism is that CBGM is essentially a black box. Is there any documentation on how the algorithm is analysing the decision made by the textual critic?
If that’s your major criticism, you’re in luck. There is no algorithm and the way the tools use the editors’ decisions has been explained many times now in books and articles (for instance). The CBGM software uses basic SQL databases (I once downloaded them as large but simple Excel spreadsheets) and uses the most basic math like addition and division on them. Perhaps by “algorithm” you just mean computer code? As to it being a black box, the whole thing has been explained multiple times and can be run on your own computer using either Joey McCollum’s open-cbgm or INTF’s. It’s only a black box to those who don’t have the time or desire to learn how it works.
DeletePeter, even SQL queries can act as algorithms, especially if running joins and aggregates. I suspect what you've downloaded is data that has been transformed in some way before putting it in a simple excel spreadsheet. I've seen some of the aggregate data from INTFs ECM online, its clear that functions have acted upon that data before being presented to me.
DeleteEither way, I've ordered your book on the kindle and I'll be doing a thorough dive. If you like, I'd love to have you on the channel to help clarify CBGM!
I'd be happy to.
DeleteThe best way to entirely dispel the "black box" belief is to actually use the CBGM from start to finish. You could do this with my virtual research environment, https://apatosaurus.io/. You start by making textual decisions for every reading by every witness. Then that data is passed to Joey McCollum's open-cbgm. All of it is open-source and on GitHub. You can also see INTF's implementation of the CBGM on GitHub. Though mine and Joey's will be easier for you to follow. You should use the tools and follow the source code as you do.
DeleteFrom a separate source: "Mink repeatedly insists that CBGM is transparent because all decisions are documented. Yet documentation alone does not guarantee accessibility. The datasets involved comprise hundreds of thousands of relationships, processed by proprietary or semi-proprietary tools unavailable for independent replication.
ReplyDelete"Transparency without practical auditability is functionally indistinguishable from opacity."
To paraphrase Jerry Maguire, part of the solution is "Show me the code!"
That separate source is simply incorrect.
DeleteSo why not simply release the CBGM code and dispel the "black box" concerns?
DeleteThat was released years ago.
DeleteWhere?
Deletehttps://github.com/SCDH/intf-cbgm
Deletehttps://scdh.github.io/intf-cbgm/
https://ntg.uni-muenster.de/
You know what? I'm gonna start using textual flow diagrams as stemmas even harder.
ReplyDeleteI think that you are quite right that this is a common misunderstanding. They are not intended to show the progression from one *manuscript* to another. However, if they do not intend to show the development of one *textual state* to another, it is difficult to grasp what might be meant by “textual flow,” or how they might contribute to our understanding of textual history. Mink is very clear that “The CBGM presupposes that genealogical relationships between witnesses are evidenced by genealogical relationships between variants.” Based on Mink's view of contamination, the CBGM attempts to determine the overall flow between states of text based on the total number of prior and posterior readings found in each witness. Mink’s view of contamination is that “normally contamination seems to have had only minor influence on each individual copying process. It is a consequence of the loss of so many links between surviving witnesses that they appear to be heavily contaminated.” My basic critique of the CBGM is that I do not think that this view of contamination is true. If it is not, then it is difficult to see what textual flow diagrams can reveal about textual history or textual development, no matter how careful one is not to take them as a stemma of manuscript relationships. Of course I do not dispute the former part of the sentence—most instances of copying occurred without substantial contamination. My argument would rather be that, when substantial contamination did occur, it was often between dissimilar witnesses. This means that we cannot take the *overall* agreement or lack of agreement between witnesses as an indication of the “coherence” of readings.
ReplyDeleteAnd I think that contamination happened between similar as well as dissimilar witneses. My first project in textual criticism was suggested to me by no other than Maurice Robinson (who comments in this thread) and it resulted in my first journal article published here (in the journal I now edit): https://jbtc.org/v07/Wasserman2002/Wasserman2002.html (soon this link will not work since articles are migrating to a new website). I examined a very special reading in the pericope of the adulteresss in 34 manuscript, a special reading which seems to demand some sort of genetical relationship. Interestingly, those manuscripts which had the reading were indeed closely related but found in various groups, so the reading seems to have been both copied from exemplars and also found its way into other archetypes from which it was copied. I concluded: "This study was based on the recognition that certain kinds of variation are highly significant when constructing a stemma of textual relationships. We have already mentioned Gordon D. Fee's postulate that variants should be 'weighed' as well as counted. This is because some agreements, as Fee writes, 'seem to demand textual relatedness, either by immediate ancestry or by ultimate relationship with a text type parent or the original text itself.' However, this study implies a fourth possibility that Fee did not mention, namely that significant variants such as the peculiar Patmos-addition in John 8:8-9 have the potential to gain influence on their own and to find their way into otherwise unrelated MSS. Such processess would be facilitated in places like the monasteries of Mount Athos, where there was plenty of contact between scribes and MSS. This adds another dimension to the problem of what should be given priority, whether 'weighing' or 'counting'. The more genetically significant a reading, the more potential it has to affect unrelated MSS. Therefore weighing and counting remain complementary methods when determining genetic relationships among MSS." This reasoning which was before I knew anything about the CBGM, but that method actually both counts and weighs (two sides of "connectivity").
DeleteI completely agree that contamination also does happen between similar witnesses! I even agree that it is likely that it happens more frequently between similar than dissimilar ones. What I disagree with is Mink's assertion that "It is a consequence of the loss of so many links between surviving witnesses that they appear to be heavily contaminated." My primary contention is that there are enough examples of such "heavy contamination" to distort the results of textual flow (a particularly clear example of this being 2737 having the ATEXT as its closest ancestor).
DeleteThat is an edited text which I hear will be removed from the ECM of Matthew altogether.
DeleteYes, it has thankfully now been removed—but that was after Stephen Carlson made the same point I made above. It still illustrates the point that textual flow diagrams do not handle well contamination from distant manuscripts since a relatively thin layer of corrections from earlier manuscripts caused a copy of the TR to have the ATEXT as its closest potential ancestor. The problem is that this sort of "editorial" contamination didn't only happen with copies of the TR. It also happened with a good number of Byzantine manuscripts (e.g., for Jude, 424, 2401, and the marginal readings in 605, all of which have been corrected at least as thoroughly as 2737/372). This means that early readings that are also corrections in largely Byzantine manuscripts will show as less coherent than they should be, causing, in at least on case (1 Peter 4:16), a reading with no early attestation to be preferred as the initial text on the grounds of coherency.
DeleteJust today I received Christina Kreinecker's reminder for the conference on "Scribes, Transmission, and Bilingual Traditions: A Fresh Examination of GA 629" in Louvain next year, June 2026.
DeleteThe use of Greek/Latin GA 629 in ECM is discussed by Andrew Patton in a nice article in "TC", 30(2-25), pp. 37-55.
The Greek text of GA 629 is changed according the Vulgate in the 14th century.
Patton, p. 49: "The inclusion of GA 629 in the ECM apparatus therefore offers useful data about the rransmission of the text of the New Testament for future research on the text and history of the biblical text even though it will not carry much weight for decisions about the initial text."
Sirleto is a linking pin between 629 and 2737/372. The cardinal possesed 629 and he is also certainly probable the editor of the text of 2737/372.
When I personally evaluate readings I do not trust the diagrams blindly; to me it is important to take into account the philological nature of the variant. If, for example, a haplography is very possible due to homeoteleuton, it is no surprise that a shorter reading will show many incoherencies... so the coherence is one tool in the toolbox. Interestingly, very few decisions are actually made on the basis of genealogical coherence and this in a final phase. I was a bit surprised when I realized how few editorial decisions were involved.
Delete