Monday, January 23, 2017

Textual Criticism as Rhetoric

From Richard Tarrant’s new book Texts, Editors, and Readers: Methods and Problems in Latin Textual Criticism (reviewed in BMCR here):
To classify textual criticism as a form of rhetoric is a way of highlighting the fact that its arguments depend on persuasion rather than demonstration. Textual critics cannot prove that their choices are correct; the most they can hope to do is lead their readers to believe that those choices are the best available ones.

Facts do, of course, play an important part of textual criticism. But in the end the facts cannot yield a definitive answer, only a relative probability, which is where the critic needs to employ rhetorical argument.
This point is not, of course, unique to textual criticism. All study of the past, certainly the ancient past, trades in probabilities and textual criticism is nothing if not a historical endeavor. So long as we are okay with probabilities, there is little here to really fuss about. Where the debate can be had is about just how probable any of our particular text critical judgments are. That, however, takes us beyond mere rhetoric since some judgments are better than others. Still, I take Tarrant’s quote as a welcome reminder that judgment is always involved.

For more along the same lines, see Gary Taylor, “The Rhetoric of Textual Criticism,” Text 4 (1988): 39–57.


  1. I agree with you that the need to frame most conclusions in terms of probability is not unique to textual criticism. To a greater or lesser degree, I think that's true of most fields.

    The real problem, however, is the general inability of people to correctly perceive probability. Study after study shows that we are (most likely) terrible at it.

    There was a commercial on TV a few years ago that I loved. Two accountants in suits and briefcases walking down a city street. Then something really bizarre passes them by - it might have been a hopping kangaroo in a frilly dress or something like that - and the first one turns and says "wow, what were the odds of that!" to which, with baffled amusement, the second says, "why, 100% of course!"

    More seriously though, the first year philosophy class example I keep going back to is the accountants and the lumberjacks. You stop by a large hotel in a big city, it goes. At the hotel there are two conventions going on: the National Association of Accountants, and the Brotherhood of Lumberjacks. The former group has 900 members in attendance. The latter group has a mere 100 members in attendance. You see a man walk through the lobby, he's large, burly beard, wearing blue jeans, boots and a plaid shirt. What's his most probable identity? Is he a lumberjack or an accountant? Almost every student will answer "he's a lumberjack of course" whereas the real odds are 9-1 that he's an accountant. Our perception of probability can be so easily misled by this or that attractive but ultimately irrelevant feature.

    I had to deal with this in my dissertation. There's a group of textual critics who conclude that it's most probable if not certain that the entirety of the original text has survived in one ms or another.

    Yet when we look at the statistics of the transmission history, the number of mss we have lost - that failed to survive - suggests otherwise. Conservatively we have lost at least half of the manuscript copies ever produced, and more liberal estimates put the size of the loss at upwards of two-thirds or even higher. By the most conservative estimates then we are still dealing with the loss of the majority of the evidence. As I asked then, is there any other field of study where you could lose a majority of the evidence and yet still conclude that you had lost nothing substantial? Doesn't such a great loss make it automatically more likely that at least one or two original readings were lost? And yet this group of critics continues to be by this or that attractive feature misled to the conclusion that somehow, against all odds, the original has made it through at every point. The problem there is not a disagreement of facts, its a difference in the perception of probability.

    1. Ryan: "is there any other field of study where you could lose a majority of the evidence and yet still conclude that you had lost nothing substantial?"

      Since I know you're not suggesting that we've lost the majority of the words of our NT, let me ask you: what's the smallest number manuscripts that it could theoretically take to preserve all the original words of our NT? In theory, of course, the answer is one.

      Now, I actually agree with you that it's possible the original reading has occasionally been lost. Case in point is Mark's original ending. But the difficulty with your argument for conjecture from lost evidence is that you're trying to base probability on precisely what we don't have. So, while I actually would agree with you that it is not hard to imagine the original text being lost in some cases despite our vast number of MSS, you have a hard sell. The reason is that those text critics who disagree with you can always point to the evidence we do have whereas you are stuck trying to get them to see what can't directly be seen. In other words, it's an argument from silence and arguments from silence are always harder to make.

  2. Thanks Peter. I think I disagree with you though on the "argument from silence" designation.

    Not that I think you're doing this, but in my experience that designation is more often simply a rhetorical trick. It's a way to frame the situation so that the argument goes in your favour.

    You could compare it to the line that "you can't prove a negative". That's a great tool for an argument: as soon as you can show that your opponent has the negative premise, then they automatically have no way to beat you!

    The fallacy of that, of course, is that almost any premise can be framed as either a positive or a negative. The positive claim "it is sunny out" is effectively the same as the negative claim "it is not overcast", and vice versa. So if you have two canadians arguing about the weather (good luck finding only two!), does the first one to phrase his claim positively automatically win?

    So you see the problem with that kind of thing.

    I think the "argument from silence" often works in the same way. For what it's worth, I don't think I'm giving an argument from silence, I think I'm making a positive claim about the state of the existing ms base.

    But more on that in a minute...

    1. Ryan,
      Your argument is faulty in numerous ways. I will highlight a couple. First, while most, if not all, would accept we have lost a significant number of manuscripts, it is not manuscripts that we are trying to restore, but the text which the manuscripts are but a vehicle for, hence Peter's answer, one is hypotheticallly correct. Second, you seem to discount the amount and tenacity of the evidence, read text, we do have.
      The concept of the text itself being what we are attempting to restore seems to be the most difficult idea to grasp with textual criticism. Oh, we all know it is the text, we just act and proceed like it is the manuscripts themselves which are inspired.


    2. Ryan, I don't think your weather example is actually an argument from silence; it's just a negative vs. positive which is not the same.

      Your own historical argument for conjecture is something like: we have lost x% of manuscripts some of which were the only tradents of the original text in some cases. Therefore we have lost the original text in some cases. Now, that may be true (I certainly think it is possible), but it is still an argument from silence.

      The "silence" is in your premise where it is assumed that manuscripts to which we don't have access contained original readings. But if we don't have access to them, how do we know they had original readings? From the argument as stated, we don't. We need other evidence. So the *historical* part of your case rests on an argument from silence. Again, I don't think your conclusion is necessarily wrong. But I can see what others aren't convinced by it. This does touch on Tarrant's point about rhetoric in some ways. We have to convince others that our historical reconstruction is more than just possible. We have to convince them that it's better than theirs.

      But let me know if I've misstated your case here. You left me hanging with your "in a minute..."

  3. Sorry Peter, the phone rang just that minute!

    But what I was thinking was about the character of the existing ms base.

    Let me give a silly illustration (because the internet needs more of those).

    A couple weeks ago I was getting ready to teach a class. I did a final review of my syllabus and then, satisfied, printed it out and put it in my folio. When I arrived on campus, I stopped at the photocopier, stuck my syllabus in the autofeed, and made a stack of stapled copies. Later that day in class we were doing the obligatory syllabus tour, and to my embarrassment when we turned over to page 6, I discovered that my printer had, apparently, malfunctioned and only printed the first line. Page 7 was fine, but p 6 was basically missing entirely. Random printer malfunction. That part is true, sadly, that actually happened. The silly illustration part is this:

    Lets say now I'm preparing my new book to go off to the publisher. It's 100 pgs long, and I print it off on the same printer, stuff it in a big manila envelope and drop it in the mail. A couple days later I get a call from the publisher: 23 of the 100 pages are basically blank!

    Now, if raising children has taught me anything, its that doing the same thing over and over again and expecting different results is not, contrary to popular wisdom, the definition of insanity, but rather the definition of normalcy.

    So I print off another copy on the same printer and mail it off. Then I start to worry that the second package may also be incomplete, so I quickly print off a third copy and drop it in the mail. But then I read in the paper that Canada Post is experiencing labour troubles and having wild cat strikes and there's rumours of mail bags being dumped in the canal, so thinking it's better to be safe than sorry, I print off a few more copies and stuff them into manila envelopes and mail them off. Maybe I mail them from separate post offices just to give them a better chance.

    So now what do we have? We have a great number of manila envelopes all traveling through the mail, all containing incomplete copies of my book, but each one of them is incomplete in a different way: we don't know for certain for each copy how many pages or which pages are missing.

    Over at my publisher's office he's still sitting there with his 77 page incomplete book. He's really hoping that the next manila envelope that comes contains the missing 23 pages. Sure enough another envelope arrives, and inside is a new copy. Sure enough that copy also has its own blank and missing pages, 21 of them in fact. But the 21 pages missing from the second copy are not exactly the same as the 23 missing from the first one. Thankfully though there is some overlap, and so the publisher is in fact able to retrieve 5 of the 23 originally missing pages from the second envelope. So now he's only missing 18 pages. So now he's really hoping that another manila envelope comes that contains the missing 18 pages. The third envelope I sent was lost in the mail though, thrown in the canal. Fortunately I sent a forth, and it makes it through to the publisher. That copy, it turns out, is missing a phenomenal 36 pages, but by sheer luck, among the 64 pages it does have are 2 of the missing 18! So now my eclectic publisher is missing only 16 pages.

    So you see how this would progress, yes? My publisher is holding out hope that subsequent manila envelopes will, eventually, contain all of the missing pages. So, to bring this back home, what are the odds that this will happen?

  4. It's hard to say the exact odds, of course. You might say that it depends on how many copies I have mailed him. Every time he's received another manila envelope so far, he's managed to recover a few more of the missing pages and get a little closer to his goal of a complete book. So it would make sense that he simply needs to receive more manila envelopes then, and eventually he'll get all the missing pages, right?

    But no, that's not right. Remember that the distribution of the missing pages was random - i.e. we could never tell when my printer was going to malfunction and which or how many pages it would leave blank. So even if he received a hundred, or a thousand more envelopes from me, there's no guarantee that all 1000 of them wouldn't be missing some of the same pages.

    If you've ever played Bingo, then you know for certain what I mean. You can get 4 out of 5 numbers early in the game, and then sit there in vain waiting for that last number. You keep thinking "I'm only one number away! Surely as he calls more and more numbers, eventually my number *has* to be called!" And yet as the numbered balls are randomly read out of the basket, ball after ball your number just doesn't seem to come up. And you'll sit there waiting on that last number through ten times as many called balls as it took to get the first 4 numbers, and you'll keep on waiting right up until someone else calls bingo and wins the prize! It doesn't seem to make sense to you, but that's just how random chance works. Much like the random distribution of ms corruption.

    Now, you might argue with me on that point - you might say "come on Ryan, surely if we just send enough manila envelopes, eventually all the pages have to arrive, just by sheer weight of numbers, right? Doesn't increasing the number of manila envelopes increase the chance of all the pages arriving?"

    I would still say that it doesn't, but we can argue that later, because my point is really the other side of that: even if increasing the number of envelopes did increase the odds of getting all the pages - even if I conceded that, which I don't - but even if it did, surely you can agree though that the opposite would also be true? That as we decrease the number of manila envelopes in the mail, that the odds of getting all the pages would also decrease? Of course you agree, because you simply cannot argue that. The fewer envelopes my publisher receives, the fewer chances he has to get all the missing pages, and having fewer chances at something is the very definition of a lower probability.

    And that's the essence of ms loss. As we decrease the number of mss that survive the transmission process, we decrease the odds that all the original words make it through.

    And now that transmission process is done. All the mss that were ever going to be written have been, and all the ones that were going to be lost have been, and we can now step back and do the post mortem, and the results of that is that we lost the majority of the envelopes. So I ask again, how, outside of a flawed perception of probability, could we still think that all the original words likely arrived safe and sound nevertheless?

  5. As for me, I'm still mired in the lumberjack issue, with visions of Monty Python filling my head (in which case I would definitely label the bearded fellow as described an accountant)....

    However, in regard to the main point, there is a reason to postulate a greater likelihood of the original reading being present in our existing MSS, and that indeed is the quantity of extant data that has been preserved -- particularly in contrast to the extremely limited number of works of Latin antiquity (or even Greek classical antiquity) that virtually force conjecture by their paucity.

    To put it statistically: even assuming, say, that 60% of all Greek NT MSS that ever existed have been lost, the odds remain extremely favorable that the remaining MSS would preserve in relative proportion the same readings that might have appeared in those now-lost documents; if so, then there is no good reason to postulate much if any supposed "loss" of original readings as matters now stand.

  6. Thanks Maurice, now I'm going to be singing Monty Python all afternoon!

    By the way, thanks for sending that article over, I'm halfway through it and so far enjoying it immensely.

    As for this argument, I think I would reply that it's not quantity of the mss that would prove your case, but the quality. We must weigh, not count, no? (though that may get us into an entirely different discussion on which I suspect you and I will forever agree to disagree...)

    But by quality I primarily have in mind their internal redundancy.

    That is, let's think of it this way: the historical events of manuscript transmission have effectively taken the original text and distributed it randomly but redundantly among a large number of vessels. All of those vessels contain some of the original text, but none of them contain all of it. They all contain some, however, and in fact we could say that all of them contain most of the original text. So that majority-of-the-text (note those hyphens!) that is redundantly carried in all of them, we would surely have a high probability of being able to restore that even if only a few of the vessels survived, right? Right, because even just a few of them would still have that majority-of-the-text that they all shared in common. So their high redundancy in that regard would ensure preservation despite a high degree of loss.

    But what about the minority-of-the-text that they did not all share? That minority quantity of the original text which, through random distribution, was conveyed by a much smaller number of vessels? That is the text about which I think we should be concerned.

    The preservation of a great number more vessels is not helpful if none of them carried the text in question. And if the text in question is carried in only a small number of vessels, then it stands to reason that the more vessels we lose, the greater the risk is that we loose all of the vessels that had that text. Thus total loss, or "primitive corruption"

    So that brings us back to the character of the existing mss: what is their quality? How redundant are they? Remember, if they are highly redundant, then the high degree of loss I'm alleging would not matter.

    To answer that question, as I've argued before, I agree that the majority is highly redundant. But you and I have discussed the minority of readings that are not distributed across the surviving ms base with high redundancy. Rather, some readings that survive today survive in just one, two, or a small handful of mss. What that tells me is that there were readings which were not distributed with high redundancy at all, and it is these low-redundancy readings which are given the highest likelihood of loss by our great loss of manuscripts.

    Now, of course, the virtue of your majority text position is that you get to easily discount those readings and say they were not part of the original, and so far as you see it no part of the original text suffered low redundancy distribution. That makes it easy for you! But for those who follow the eclectic method and have in their printed texts readings which did barely make it through with very low redundancy, those critics don't get off so easily!

  7. I would just point out that a great deal about the number of manuscripts and the preservation of the original text in them depends on how they are related. That is what Dan Wallace was getting at in his review of Ryan's book. Dan's conclusion didn't convince me because he too relied too much on silence, but he is on the right track in working through the issue genealogically.

  8. Ryan: "But what about the minority-of-the-text that they did not all share?...That is the text about which I think we should be concerned.

    As you note, this might become more of a problem for eclectic praxis where some readings are based on only a small handful of MSS (or one or two or even none), whereas it would not be or be less of a problem for those who might argue reconstruction of a particular and generally well-attested type of text in a putative archetype (not only Byzantine, but even Alexandrian or -- as per Heimerdinger and Rius-Camps -- the Western).

    Yet for whatever reason, even most eclectics continue to function on an assumption that the original text in almost all instances has been preserved among the extant MS base, and whatever their presuppositions on this matter, these would be important for understanding their praxis and general reluctance to embrace conjecture.

  9. Another though not rhetorical-criticism-geared remark from that book that may be of interest: 'Editors of classical texts have no difficulty in defining their aim as that of reconstructing the author's original version, while at the same time recognizing that, given the evidence available, that aim can never be fully achieved.' It seems that not a few people from the NT guild are very quick to affirm the latter (which I can and do as well) while eschewing the former.

    1. '... Classical editors' fixation on the author's original does not necessarily betoken ignorance of or hostility to the questioning of that concept in other areas of textual criticism, but is rather the result of the circumstances in which classical literature has been preserved.'

    2. Now I'm inclined to think that the NT texts seem to have been preserved in conditions much more akin to those of the classics rather than 'semi-classical' or 'multi-versional' texts, let alone modern literary works where in his Tarrant views the concept of a single original text may indeed seem problematical (e.g. What's more 'original'—a book before or after copy-editing?)

  10. Ryan, one moment please.

    900 accountants, about 1% of them dress lumberjack style, at most. So at most there are about 10 men at that meeting who are accountants with

    "large, burly beard, wearing blue jeans, boots and a plaid shirt".

    If you do not accept my 1% statistic, simply look at a picture of the AICPA convention.

    In fact, 1% is very, very generous. The real number is .1%. That is, you might average one person at a 1,000 man convention dressed that way.

    100 lumberjacks, 2/10 of whom dress lumberjack style and have beards and are big and burly. (the other 8/10 wear jacket, or sport shirt, or regular shoes or sneakers or don't have a beard). Maybe 20 people at the convention fit your description.

    So there are about 10 accountants who fit that description (we are padding the accountant side) and about 20 lumberjacks in the building.

    Please explain why you say there is a 9/10 chance that you are seeing one of the 10 accountants rather than the 20 lumberjacks who are in the building and fit the description.

    And I could take your example to reductio ad absurdum by having a convention of Native American Indian chiefs with headdress and feathers. You really think it is 9/10 you are seeing an accountant and not a Native American?


    I've always claimed that textual critics are very weak on probability and statistics. And I accuse Daniel Wallace of statistical illiteracy in his handling of textual arguments.

    Ryan, I consider myself reasonably adept at probability concepts and I agree with you 100% that the standard probability sense can be deceiving. This has come up in the Jesus tomb studies (prosecutor fallacy) and many other areas.

    And I also enjoy your kangaroo post-facto 100% probability example. The inability to understand post-facto probability concepts is endemic.

    Your little example blew right by.

    If you demonstrate your 90% accountant, I will tip my hat to you. If not, my point is proven once again, more emphatically, since we have many top textcrits here.



  11. Hi Steven! Thanks for your question. I hate to say it, but I think you're actually proving my case for me.

    Why did you instinctively assume that only a low percentage of accountants would be burly fellows dressed in plaid?

    Is it ok for me to say that you assumed that because that is in line with your experience (and the pictures you have seen of AICPA)? I think so.

    But all that shows is that you subconsciously expect reality to conform to - or at least interpret reality in light of - your own past experiences.

    But your experiences are not going to be the same as others'.

    Take myself, for example. I live in the north country. Out where I live, 9 out of 10 accountants are burly fellows dressed in plaid. (The only accountant I know who does not fit that image is my father, who works down in the city).

    Bottom line then, your personal - subjective - experiences would lead you to have an expectation that is counter to the statistical probability. I'd say, then, that's a perfect example of the point I was trying to make: our own thoughts and preferences distract us and mislead us from seeing the actual odds of a conclusion.

    As the line I love to quote from krister stendahl goes, we are more often misled by what we think we know, than by our lack of vision.

  12. Bonus: the famous monty hall problem is another great example of this. The statistical reality is that you should switch, every time. It is a mathematical fact that the odds favour switching. The fact that most people intuitively think that the odds are still only 50:50 at best and that they're just as well off either switching or not-switching is a perfect example of how the typical human brain crashes and burns when faced with the task of assessing probability. Try it and see. Try it with someone else. I've done it with students. You can almost see the cognitive dissonance rippling across their face as they try to wrap their head around the facts: "but..but..but.." they stammer..."there's only two doors left, how can the odds not be 50:50?!?!" They just can't accept it, it makes no sense to them. Even after you explain it, it will still "feel wrong" to them. Our preferences, our experiences, our "common sense", it's all just so strongly embedded in our thinking, it really is next to impossible to recognise contrary facts, even if they come up and smack us.

  13. > Ryan
    > Why did you instinctively assume that only a low percentage of accountants would be burly fellows dressed in plaid?

    Simply because nationwide and worldwide that is 100% true. It is not instinctive, it is analytical observance.

    Maybe in 1% of accounting conventions the number is significantly higher, but that does not affect the overall probability, because you did not frame the question as a conference in the northwoods of Minnesota. (As an example of a location with a higher quotient of lumberjacks.)

    It would be amazing that you pull this over the textcrits, if I did not know from experience that they tend to be probability-challenged.

    Your bogus claim:

    > whereas the real odds are 9-1 that he’s an accountant.

    Is absurd and ridiculous. Even in the northern Minnesota convention site, the odds would likely be much lower.

    And it definitely is not true in the large majority (remember this is probability) of conventions, New York, Pittsburgh, Chicago, Indianopolis, etc.

    I am just amazed that yhou could pull over such an absurd example on this team of textcrits here. And not one called you on it.

    You have confirmed once again that textcrits tend to be abysmally weak in probability and statistics.

    Your assertion is so off-base I do not even have to check it with my brother (who is one of the top probability guys in the world, Erdos number of 1. We discuss stuff like post-facto probability at times.)


  14. Steven Avery

    > whereas the real odds are 9-1 that he’s an accountant.

    I'll lessen my critique a tiny bit. There is the question of the distribution of lumberjack conventions, which could be in Minneapolis, Chicago, Orlando, the Adirondack Mountains and other places. The distribution of places that have both lumberjack and accountant conventions would still tend to the big cities, since you have 900 accountants as the place.

    In some places, it is possible that a good number of the accountants dress with full beards, flannel, etc. They would still be a smaller pct than actual lumberjacks, making the 9 to 1 figure false. In other cases (a convention in Florida, a convention Chicago) the accountants would tend to be your normal accountant look.

    So I will acknowledge that the very fact of the distribution of locations of lumberjack conventions changes the probability a bit, but your bald assertion:

    > whereas the real odds are 9-1 that he’s an accountant.

    Remains false and absurd.

    Unless you are talking about one specific spot where the accountants are known to dress and look like lumberjacks. (You claim there is such a spot, you are probably exaggerating, but the question is about the general distribution, not one unspecified spot.)


  15. Ryan did point us in one fascinating direction.

    The Monty Hall problem is a brain teaser, in the form of a probability puzzle (Gruber, Krauss and others), loosely based on the American television game show Let's Make a Deal and named after its original host, Monty Hall. The problem was originally posed in a letter by Steve Selvin to the American Statistician in 1975 (Selvin 1975a), (Selvin 1975b). It became famous as a question from a reader's letter quoted in Marilyn vos Savant's "Ask Marilyn" column in Parade magazine in 1990 (vos Savant 1990a):

    Suppose you're on a game show, and you're given the choice of three doors: Behind one door is a car; behind the others, goats. You pick a door, say No. 1, and the host, who knows what's behind the doors, opens another door, say No. 3, which has a goat. He then says to you, "Do you want to pick door No. 2?" Is it to your advantage to switch your choice?

    Feel free to think this through on your own account first. Wikipedia left out an important piece of information. Monty is always opening up a door with a goat, he is not randomly choosing.


    There is no real relationship to a bumbled artificial lumberjack attempt with the fascinating Monty Hall problem, which has a status as a "cognitive illusion". And even caused Paul Erdős to stumble.

    A good vid on it is at:

    National Museum of Mathematics
    Math Encounters - What's Behind Door Number Two: The Monty Hall problem, reconsidered
    Jason Rosenhouse

    Start around 21:55 and go to at least 36:00 or so.


  16. Peter,
    Somebody -- Epp or Kilpatrick or one of those folks -- once gave an interesting example of how a side-effect of some text-critical methods is to imply that some of the NT text has been lost. His idea, iirc, ran along these lines: in Acts, a reading from Codex D was selected by eclectic editors X number of times per Y pages of the MS. Epp (I think) argued that one could surmise that it is probable that if Codex D were in pristine condition, the editors would continue to select its readings at the same rate -- meaning that they would have adopted Z readings from pages that are no longer extant. Which would mean, in turn, inasmuch as we don't have those readings, that they have been lost.

  17. There are two fields where we have lost most of the evidence and yet still can come to conclusions: paleontology and history. 99.9+ % of the bones of dead animals have been eaten or destroyed, and yet we can still come to conclusions based on the what is preserved. 99.9+ of people in past centuries died without writing anything that has been preserved about what they saw in their lifetime. 99+% people who experienced the American Civil War and died did not write anything about it that we have preserved. Yet, can we saw that we don’t know for sure about anything about the Civil War because we lost 99% of the evidence. So you can’t measure the tentativeness of your conclusions solely by the evidence you don’t have; you also have to take into account what you have. One way to measure reliability of a prediction or conclusion is by test and training sets. Applied to New Testament manuscripts, you use a subset of manuscripts, the training set, to predict what the originals said. Then you use that prediction on the other manuscripts, the test set, and see how well it predicts what those manuscripts say. You could do this for all changes, only changes that are not obviously wrong (like misspellings), or only changes that have meaning. That would give an estimate of how much you can trust the prediction.