Re: M. A. Robinson's recent article
- Dear Jonathan,
Thank you for your interesting post. I am inclined to agree with Maurice Robinson's point that the UBS Greek New Testament (= Nestle-Aland text) is artificial in the sense that, being eclectic, nothing quite like it exists among surviving witnesses of the NT.
As you know, I believe that Streeter was right to believe that there were local texts. I've recently finished a paper which sets out an analysis based on the 44 variation units found in the UBS Greek New Testament (4th ed):
Everything in this paper is provisional, mainly because the number of variation units is so small. (The analysis is restricted to Hebrews alone. Hopefully, it will not be too long before the same kind of analysis is applied to much larger data sets and to other parts of the NT.) That said, it is interesting to note that MDS maps seem to support a geographical reading of the data.
Other things being equal, each textual cluster has an equal claim to representing the most ancient recoverable text. (However, other things are not equal, I think.) I discern a number of textual clusters in the MDS maps. Saying how many there is represents a challenge -- the data does not seem to support any particular number of clusters. One possibility follows:
1. P46, B
2. Sinaiticus, A, C, M33, M81, M436, M2464, cop-sa (Sahidic Coptic), arm (Armenian)
3. it-ar, it-b, it-comp, vg-cl, vg-st, vg-ww, cop-bo (Bohairic Coptic), geo-1 (First Georgian)
4. D, it-d (two sides of the same Greco-Latin codex)
5. K, L, and the rest
The kind of analysis upon which this partition is based takes no account of genealogy. It simply shows where texts lie relative to each other in what I call "textual space." So the question of which text gave rise to the others remains unanswered. This is the question that we all need to take a fresh look at.
I suggest two different approaches for getting at the earliest recoverable text:
1. Choose one local text as prior then use it as the basis of the reconstructed "initial" text (to use the INTF's terminology). (I would but my money on cluster 1, above, for admittedly speculative reasons given in the paper.)
2. Distill the hypearchetypes which stand behind the local texts then use these to get back to the initial text.
The INTF's CBGM represents a third approach, and I believe that it has much merit. I share WW's concern about circularity when the results of one iteration of the method are used to inform another iteration. The INTF is fully aware of this concern but still thinks it worth doing. Looking at things from a statistical perspective (I'm an amateur), I am reminded of Bayesian analysis where prior knowledge is factored into the decision-making process. In my opinion, some useful information is extracted by the first iteration which can validly be used for another iteration. However, I suspect that there is a law of diminishing returns. It is like the physical law which prevents one from building a perpetual motion machine -- you can't get something from nothing (unless you are God).
If all of these approaches come up with the same text then our work will be complete and we can look for something else to do. However, I suspect that a number of the variation units will remain hard cases -- we do not always have enough information to make a confident decision.
On the topic of confident decisions, I believe that the UBS approach of assigning A, B, C, D categories to decisions is useful. My (perverse?) approach is to go further and attach probabilistic meanings to the categories:
A = beyond reasonable doubt (conf > 95%)
B = probable (50% < conf < 95%)
C = a few competing possibilities (5% < conf < 50%)
D = many competing possibilities (conf < 5%)
I think that the ideal of truthfulness calls for us to indicate the confidence held in each textual decision about what reading should be preferred at each variation unit, and the A, B, C, D scheme seems to me simple yet powerful enough to suit.
The question then arises as to how the rating for a reading is obtained. I think that a Bayesian approach, where each piece of evidence contributes something to the result, is worth considering. All of the received wisdom concerning criteria can be applied provided we can find an appropriate reliability rating for each criterion. For example, the "reading which appears in the most diverse places" might be given a reliability rating of 0.75, while the "shorter reading if not likely to have been caused by parablepsis" might get 0.55. (A rating of less than 0.5 means that the criterion is more often wrong than right.) If we can come up with a set of independent criteria and agree on reliability ratings then we could use them as pieces of information to help us choose the best reading for each variation unit. (Independence is important. The "reading which best explains the others" is conditioned by what you believe concerning the development of readings and is therefore not independent of criteria such as "prefer the shortest.") Using a suitable (Bayesian) formula to combine the results would give us a probability that the decision is right (and an indication of which reading best explains the others, all things appropriately considered). The A, B, C, D scheme noted above could then be used to classify each preferred reading.
--- In email@example.com, "Jonathan C. Borland" <nihao@...> wrote:
> Dear Wieland,
> Perhaps I did a poor job summarizing the article. I'm not sure if Dr.
> Robinson is a member of this list. If he is, perhaps now would be a
> good time to chime in. I do not wish to put words in his mouth, but as
> one of his former graduate students, I will briefly make a couple
> On Dec 1, 2009, at 4:35 PM, Wieland Willker wrote:
> > I have no problem with the fact that, at times, a verse as a whole
> > in the NA
> > is not extant in any surviving manuscript. This can happen. And
> > about 100
> > out of about 8.000 verses is not much.
> The critical issue of the article is whether or not NA27 reflects a
> "'test tube text' which never existed at any time or place" (Aland/
> Aland rule number nine). The article essentially demonstrates that in
> more than 100 whole verses NA27 reflects a "test tube text," i.e., a
> sequence of readings that "never existed at any time or place," or at
> least there is no evidence that it did. In this sense it is a new text
> with no descent, created 1900 or so years after the originals.
> Perhaps 100+ whole verses with zero manuscript support does not sound
> too bad. Yet the presence of this very situation indicates that if one
> were to consider whole verses with only singular, dual, or triple
> manuscript support, the likelihood is that he would find hundreds more
> whole verses with only diminutive support. Opening up to NA27, page
> 220 (Luke 18:17-29), e.g., one finds several readings of the text, not
> to mention whole verses, so minutely supported.
> Is current textual theory saying that textual transmission was so
> mixed that sequence of readings in the actual hard evidence
> (manuscript data) is no longer relevant?
> Sequence of readings important to Dr. Robinson (and others) in part
> because the primary argument against the Byzantine Textform is _not_
> that its readings are not old. P45, the oldest witness in Mark,
> appears to agree with the Byzantine Textform at least as often as it
> does NA27 (if not more so). The main complaint of scholars is that the
> Byzantine Textform's sequence of readings is not found in any "early"
> Alexandrian manuscripts. Do they mean the sequence of one or a few
> verses? Of course not, since over half of the verses of the NT in both
> the NA27 (assuming its "initial text" status) and Byzantine Textform
> are identical (excluding orthographic differences). They mean the
> sequence of variants of the entire NT!
> So Dr. Robinson asks: Is sequence of readings important, or isn't it?
> If it is, then only the Byzantine Textform represents a continuous
> sequence of readings based on actual manuscript evidence, while NA27
> does not. If sequence of readings is no longer important, then the
> Byzantine Textform may be assumed to have arisen from a now-lost
> majority of 2nd and 3rd century manuscripts from all over the Empire,
> and claims that its entire sequence of readings is not found only in
> early Egyptian papyri are totally irrelevant.
> Jonathan C. Borland