Sunday, June 17, 2018

An Interview with Stephen Carlson: Cladistics and the Text of Galatians

Dr. Stephen C. Carlson

Joining us today is Dr. Stephen C. Carlson, a professor at the Australian Catholic University. A former lawyer, in 2012 Dr. Carlson received a Ph. D. at Duke University; his dissertation consisted of an attempt to reconstruct the transmission-history of the text of Galatians, using principles from the biological field of cladistics.  In the process, Carlson produced new collations of the of the text of Galatians in Papyrus 46, ℵ, B, A, C, D, F, G, and several minuscules, including 1780 (Codex Branscombius, from c. 1200) which is housed at Duke University.
His investigation involved the testimony of 94 witnesses across 1,624 textual contests.  He also made a new compilation of the entire book, disagreeing with NA27 thirteen times.  Let’s ask him some questions about all that.

(1) Dr. Carlson, thanks for taking the time for this interview. What was the most surprising thing you learned in the course of researching the text of Galatians?

Carlson: Thanks for taking the trouble to read my work carefully and propose such well thought-out and intelligent questions. There were a lot of little surprises but nothing really big. I suppose the most surprising thing is how few times I ended up disagreeing with the critical text. (Also how little evidence of theologically motivated changes there was, particularly in the road up to the Byzantine text.) The biggest differences were Gal 2:12 (“when he came” vs. “when they came”) and Gal 2:20 (“faith of/in God and Christ” and “faith of/on the Son of God”).  Almost everything else involved word-order and word-choice variants.

(2) What is cladistics?      

Carlson:  Cladistics is a method for inferring a family tree for a group of entities that were generated by a genealogical process.  If we have a way of judging which of two different proposed family trees fit the data better (i.e., a family tree of manuscripts that puts very similar manuscripts together is better than one does not), then the method of cladistics says that we should try as many of the possible genealogies that we can until we find the “best” one. Moreover, cladistics proposes a metric for identifying good family trees:  they’re the ones that minimize the number of scribal errors that must have happened throughout the transmission history.  This is called “maximum parsimony.”

(3) A lot of attention has been given lately to the Coherence-Based Genealogical Method.  Your approach seems to involve less data-manipulation, and fewer initial steps that steer  results.  What can you tell us briefly about (a) what an unoriented stemma is, (b) how you made the unoriented stemma for Galatians, and (c) how cladistics guides the process of starting with an unoriented stemma and ending up with an oriented stemma?

Carlson:  Good question. We ordinarily think of a stemma or family tree as having a chronological direction. It starts with the original, proceeds to its copies, and then to copies of the copies. In cladistics, we favor the stemma that minimizes the number of scribal errors, but this begs the question as to what is a scribal error and what is the “authorial” reading (also known as the “true” or “original” reading).
The CBGM, as I understand it, is a development of Kurt Aland’s local genealogical principle that there are places in the text where we can definitely tell which readings are errors and build from there.  The CBGM is a more rigorous application of this principle, identifying “potential ancestors” based on how many locally genealogically prior readings a manuscript has in relation to the other.  Of course, the value of this approach depends on making good judgments about the priority of one reading to another – although there is some provision for feedback and updating one’s judgments about genealogical priority.
My approach defers the question of priority to the back end of the process.  The fundamental insight for this is in fact 50 years old; it is this:  you don’t need to know which variant is more likely to be a scribal error in order to determine the basic shape of the stemma. Regardless of which reading you think is older than another, the stemma will have the same shape, that is, it will identify the same manuscripts as being close (or distant) relatives as before.
The only difference is that the stemmas will have different orientations.  A textual critic that likes the Western text and thinks its readings are prior to those of non-Western manuscripts will orient the stemma in such a way that the Western manuscripts will at the top, while a textual critical who likes 1739’s readings a lot can orient the stemma in such a way that 1739 and its relatives will be at the top.  However you orient the stemma, there is only one “unoriented stemma” that most efficiently arranges the manuscripts with respect to one another.
It turns out that finding this unoriented stemma is something that computers are very good at, and it does not need any judgments about which reading is better or more prior than another.  Once the best unoriented stemma has been found (of those looked at), the textual critic can move in and identify the best way to the orient the stemma based on the critic’s understanding of which readings are more likely authorial or scribal.
            So to answer your specific questions: (a) the unoriented stemma is the stemma that most efficiently accounts for the patterns of readings (but not their priority) among all the manuscripts that you’re looking at; (b) I made the unoriented stemma by creating a matrix of manuscripts and variation units that describe what the variants readings are and fed it to a program I wrote for finding the stemma that most efficiently accounts for the distribution of the variant readings; (c) the cladistics procedure ends with discovery of the best-found unoriented stemma: the rest is old fashioned text critical work dependent on the critical judgment of the textual critic.

(4) For a long time, textual critics have been comfortable working on the premise that community of error implies community of origin. Does a stemmatic approach confirm that principle, or should that principle be modified? Or to put it another way, did you find a significant amount of accidental or coincidental agreements among essentially unrelated witnesses?

Carlson:  I would say that a stemmatic approach implements this principle. Cladistics does recognize that there can be a large amount of accidental coincidences among relatively unrelated witnesses, but as long as these coincidences are not patterned more strongly than genetic agreements, the maximum parsimony principle still works. I would say that within the medieval transmission of the Byzantine text, most of the coincidences in non-Byzantine readings are in fact accidental and the result we get in the stemma is not so much in falsely identifying close relatives when they are not but rather an inability to identify close relatives at all.

(5) (Long question.) A normal principle in textual criticism is, “Prefer the reading that best explains its rivals.”  But sometimes plausible explanations can be given for two rival variants. In those cases, some textual critics have resorted to making decisions based on the character of groups, preferring the reading supported by witnesses that tend to have readings that best explain their rivals. Metzger’s Textual Commentary seems to frequently describe that resort being taken; Alexandrian readings are often preferred because they are Alexandrian, which is a little frustrating, because at some point the assumption of the relative degrees of reliability of witnesses ends up confirming itself. But what alternative is there?  Can cladistics contribute something to the resolution of finely-balanced contests?

Carlson:  Well, normally in textual criticism there are two kinds of evidence, internal and external. Where the internal evidence is not decisive we need to look at the external evidence (and vice versa).  A thorough-going eclectic might eschew external evidence, but I don’t know that the alternative is besides giving up.  If we knew the history of the text, we should use it. Metzger and other “reasoned eclectics” would claim that the reason that Alexandrian readings are preferred in these cases is that in other cases Alexandrian readings seem to be good.  What I think cladistics/stemmatics can contribute is being consistent in the use of external evidence, which doesn’t always happen.  In addition, there can be variants where contests are finely balanced both internally and externally.  Those I think we should try the best we can.

(6) When measuring the relative degrees to which one witness’ contents are related to other witnesses’ contents, how did you handle the problem of gaps in the materials, such as missing pages in manuscripts, incomplete quotations in patristic writings, or the inability of some versional evidence to express distinctions between rival readings that appear in Greek?

Carlson:  The method does allow for gaps in the witness, which can be coded as “missing.”  It doesn’t really affect the algorithm or the maximum parsimony principle, but with less data the result will be less precise.

(7) In the most recent N-A/UBS compilation, the testimony of minuscules 1739 and 1175 seems to have been given more weight, and thus had more impact, than one might have expected in the General Epistles. Would that be justified in Galatians, too?

Carlson:  In my research for Galatians, they’re of secondary weight.  1739 turns out to be as close to the archetype as A but not as close as 01 (Sinaiticus) or 33.  It is an important manuscript with an early text, but not as good as Zuntz claimed for Paul (results could differ in other letters).  1175 turned out to be a mixed manuscript:  one source is related to 33 and the other source is Byzantine.

(8) The text produced by Westcott and Hort in 1881 has been criticized for being based on a genealogical method that lacked a real-life genealogy. Do we now have a real-life genealogy for the text of Galatians?

Carlson:  Having a genealogy for Galatians is indeed my claim. Of course, it is not “real-life” but a proposal based on the evidence of 92 witnesses. The difference over Hort is that it is more precise and less hand-wavy than the vague groups that Hort predicted.

(9) Tell me about parsimony.  Did you find the “expense” of a particular family-tree of witnesses consistently the same across different segments of the text? Also, inasmuch as, as you acknowledged in the dissertation, “history can be messy,” can parsimony ever give us anything more than a vague that-a-way wave of the hand?

Carlson:  My impression is that the text of Galatians is less stable (and hence would have a higher “expense”) in the first two chapters than in the last four.  As for parsimony, it is like Occam’s Razor: all other things equal, we go for the simplest, not because it is more likely to be true but because we don’t really know or have a justified basis in claiming how the true reality is specifically more complicated.

(10) What did you find out about minuscule 1854 that was interesting?

Carlson:  Out of the 92 witnesses of Galatians I studied, a full two-thirds of them were Byzantine, and most of these had scribal errors that differed from the Byzantine prototype in about a dozen or so places. 1854 is interesting in that it only deviated in four (and in some of those four, the Byzantine Text itself is divided).  Since 1854 is an eleventh-century text much younger than some uncial Byzantines of the 800s, it suggests that some important and early exemplars of the Byzantine were still available to be copied centuries after they were made.

(11) Sifting through the textual commentary on select variant-units embedded in the dissertation, I noticed that you occasionally favored a Byzantine reading over the reading of Vaticanus. Do you think this trend would continue if a cladistics-based approach were applied to the entire Pauline corpus?

Carlson:  Those cases are mainly based on how I evaluated the internal evidence, while cladistics helps me the external. I do think that the critical text over-values Vaticanus when it omits short words.

(12) What are our top three Greek manuscripts of Galatians?

Carlson:  Sinaiticus (01), Vaticanus (B), and Claromontanus (D), but I think in terms of good combinations of manuscripts instead of good manuscripts.  Choose any one manuscript from {D, F, G}, one from {B, P46}, and one from {01 33 A 1739 1611 1854 (Byz)}, and adopt the reading supported by at least two of the three.  That will give you a very good approximation of the archetype just from those three.  The three closest are B 01 33, but since 01 and 33 are too closely related, you will need a Western (or even a Byzantine because of its readings shared among Westerns) instead of 01 or 33 to get a better text than what the combination of B, 01, 33 alone can give you.
(13) In Galatians 2:16 you rejected the Byzantine non-inclusion of δε even though the Byzantine reading is supported by 1739, the Harklean group, and Papyrus 46.  Do you still have the same view, and if so, could you briefly walk us through your reasoning for that?

Carlson:  This is a very close case in my opinion.  In favor of the inclusion of δέ we have D*FG d b vg; B; 01; C 1241S, and Chrysostom.  For non-inclusion of δέ we have P46, 33 1175, AP, 1739 Ψ hark Byz.  In terms of external evidence, aside from the Westerns (DFG), every other group is split, so there’s slight weight on the inclusion.  Transcriptionally, the omission of the connective particle δέ looks harder than its inclusion, but as Royse has shown, the omission of such little words was fairly common in the earliest period.  Intrinsically, i.e., in terms of what Paul meant, the best interpretation of the non-inclusion is the same as one with the inclusion of δέ, so that’s not much help.  I ended up favoring the slight external evidence for the inclusion, thinking that its omission would have been more common in the early period.  But it’s not a judgment I would bet the farm on.

(14) Regarding a reading in Galatians 5:21, you claimed that a deliberate insertion (of φóνοι, “murders”) was more likely than its accidental omission via simple parablepsis. Could you elaborate on the basis for your reasoning there?

Carlson:  I think both possibilities did happen in the transmission of the text. On my stemma, it was inserted twice and omitted twice.  Ultimately, I went for the intrinsic evidence, what I think fits Paul’s argument the best, and the sin of “murders” seemed somewhat out of place compared with the other sins of the Galatians.  But, yeah, another variant I’m not betting the farm on.

(15) Robert Waltz, the custodian of the Encyclopedia of New Testament Textual Criticism website, has finished a compilation of the text of Galatians, and although his approach employed different principles from yours, I think he would concur that the text of 4:25a is “all over the map.”  You gave this passage some attention back in 2014.  Should the phrase “Sinai is a mountain in Arabia” be considered an interpolation?  Or is that conclusion a case of cutting the knot rather than untying it?

Carlson:  I’ve argued that it came from a marginal note in an early (second-century) authoritative edition of Paul’s letters.  Textual critics are by and large skeptical of such solutions.  I suppose it is cutting the knot (if any reading is original, I would favor that one starts with τὸ γὰρ Σινᾶ) rather than untying it, but maybe the existence of the knot in the first place is because of such a marginal note.

Thanks, Dr. Carlson, for sharing your research with our readers.