Sunday, June 15, 2014

Method and the logically possible; a reply to Alex C

Alex C says the following in a reply of mine (here). The interested should look at the whole thread. I lift the comment here for discussion as I think it highlights an important point: that if you ignore what GG has wrought over the last 60 years then your work does not address the question: what is the structure of FL?

Maybe yes and maybe no is all that I am claiming, and all that the dialectic needs. You need to have a definite no to sustain your argument.

(I think ECP effects are a good example of something that might be a reflex of some structural property of the class of grammars-- so I think this is the easiest one for you to make your argument. If you can't do it for this, then you can't do it for any).

You mistake the dialectical lay of the land. You are defending a position that both you and I consider unassailable: viz. that it is logically possible that you are barking up the right tree. Of course it is. It is also logically possible that the Lochness Monster exists, that there is no human induced global warming, that the earth is flat, and many other wonderful possibilities.  So my argument is not and never has been that it may not be possible that your methods will yield results relevant to FL. I had a much more modest claim, in fact two: (i) that as of yet it has NOT yielded any that address what I take to be the core phenomena pertinent to describing the basic architecture of FL and (ii) that there is little reason to think that they ever will lead in such directions. Let me comment on each.

Re (i), I think that here we are in agreement. I have asked you to pick out any of the effects discovered over the last 60 years and show how any of them could be explained in ways fundamentally different from those that GGers have pursued. You cite the ECP as possibly a tough nut for you to crack and as my best case.  I don’t see it this way. I think all of the examples of effects that I have cited are interesting cases. And as I never like to take advantage of others (well, sometimes, but not always), that’s why I offered you a discussion of any of the others.  But alas, till now, you have remained mum. I take this to mean that you have nothing on offer. You have claimed in past comments to have different fish to fry. I agree. And they are clearly not my fish or anything like my fish. Thus, the point stands: you have nothing on offer for the GG discoveries.  From what I can tell, however, you do offer votive candles: you hold out the promissory HOPE that some day, one day, you will get around to these questions. You counsel patience and forbearance. I grant you both, but I don’t grant you my attention or interest. Call me when you get somewhere.

Re (ii), I have no reason to think that you will get anywhere. Why do I think this? Not because it is logically impossible that you will. But because you haven’t come close yet! You see, in empirical domains being logically coherent is a very low bar, and a position that even our illustrious flat earthers and climate deniers can hurdle with ease. What one wants is not only this, but some reason for thinking that this position should be taken seriously. A position that we should allot time to developing.  This is a much higher hurdle. Here’s what it requires: a result or two bearing on the questions of interest. I think that I have been quite straightforward about what these questions of interest are. I’ve listed concrete cases galore. I’ve asked you to discuss just a handful to make your case that doing things your way gets us anywhere with these. To repeat, to date you’ve refused. Fine. So your position is logically coherent, but to date it is sterile (at least as regards what I (and GG) take to be the questions of interest). Note, this leaves little room for interesting “dialectic” as you put it. None actually. Or, the dialectic will look a lot like this. Entertaining? Yes, very. Productive? Not really.

There is a third reason that I can add to the two above: I believe that there are better ways of developing concrete models of real time learning than the ones you put on offer. I’ve cited examples of these also. People like Berwick, Wexler, Fodor, Yang, Lidz, Dresher, etc. are doing just that. Their work is responsive to the data that linguists have uncovered over the last 60 years. Their work is not merely logically coherent, it is even relevant! In short, there are better models on offer, ones that offer concrete reasons for optimism in that they address the questions of interest.

One last point: There is a kind of argument that seems to appeal to philosophers and mathematicians. It goes something like this: if you cannot show that a position is logically incoherent then it’s worth taking seriously. I never liked this argument, at least when advanced by philosophers (see here for a good example of this move). In the non-mathematical sciences, you need a lot more than this. To defend a line of research you need to demonstrate that it will get you somewhere with respect to the questions you are interested in.  In lecture 2 (which I am in the middle of) Chomsky noted that programs are not true or false. But they can be premature or sterile.  Part of the game is to show how the program you are recommending will lead towards answers to questions you are interested in. I believe that I have been very clear about what sorts of questions interest me and that emerge from the history of GG. These are questions that reflect the history of GG and its many discoveries. They suggest a research program that builds on (rather than ignores) these very non-trivial discoveries. I believe that any approach to the study of FL that does otherwise will get us nowhere. Indeed, failure to do this means that you don’t think much of these results regardless however much you deny this. Like in much else, it’s the walk not the talk that counts. So, as I keep saying, obviously raising the hackles of many, there can is no compromise on this. You need to choose!!. Either you take these results of the last 60 years to be directly relevant to current work, indeed the empirical foundations on which further work will build, or you don’t. If you don’t, then, from where I sit, there is little else for us to discuss. We are just doing different things. We may seem to be talking about the same things, but we aren’t. It’s logically possible that I am wrong about this. But I wouldn’t bet on it, and bets are what research is all about.


  1. I think we are talking at cross purposes -- the discussion from which you lifted my quote was about what Pullum and Scholz called the Indispensability part of the Poverty of the Stimulus of the Argument. This is a premise of the argument. I don't think that "maybe yes, maybe no" is enough justification for this premise.

    This was in the context of Ewan's arguments against the "input-driven" idea. To put my cards on the table, I agree almost entirely with Ewan's arguments in this case, but think he goes slightly too far; I think one can rescue a technical notion of surfacey/objective/empiricist.

  2. @Alex C:
    As your reply used my locution ("yes and no") to your comment before mine, it seemed to me (and still does) that you were addressing MY comment. There is no reference to Pullum and Sholz (PS) in the give and take. Moreover, I was responding to your claim that I need a definite "no" to sustain my argument. As I argue above, this simply misunderstands the problem and the argument. If the misunderstanding is yours or PS's is irrelevant to the point at issue. It's a mistake regardless of who holds it.

    Two more points: I do not think you completely agree with Ewan but for your demurrals re input/Empiricism. I am pretty confident that Ewan would find tackling the effects as central to the project. You clearly do not. How do I know? Because you never do and seem to find it premature if anyone does. The sophisticated version of your position is that the effects GG has discovered MIGHT one day become important to address. But they are a distraction right now. This means that you do not think that we ought to build on this empirical foundation right now and try to see what features of FL are responsible for them at this point in time. Maybe when messiah comes it will be time to turn to these more difficult issues. But for the time being, fougetaboutit! That's a view I understand. I think it wrong and scientifically counterproductive, but I get it. I do not get a view that says the following pair of things: these results are interesting and central but we ought not to address them and anything that does is certainly wrong because something sometime in the future will demonstrate this. This is pure obfuscation.

    So, let me repeat: those that ignore the results of the last 60 years are engaged in a very different project. I believe that this project, if it takes itself to be addressing the structure of FL will fail for the same reason that flat-earthism failed, and that climate science denial fails, viz. it ignores the central relevant facts of the matter. A methodology that is happy with putting these data aside is unlikely IMO to get anywhere. I have said why. You disagree. But I knew you would. This is for the benefit of others, not you and me.

    1. It is refreshing to see that Norbert recognizes what is scientifically counterproductive in [his construal of] the position of others [note for Alex C. - I do not think you hold the position ascribed to you so please don't pounce at me]. Maybe this is the first step towards recognizing that his own position is of the same kind? He talks of the results of the last 60 years, yet each time when pressed for presenting results that are uncontroversial he resorts to either insults or silence. In his reply to Alex he revealed that he relies on faith. Faith [into the Chomskyan position being right] so great that [as revealed in an earlier reply] no empirical discovery could shake it. There is of course nothing wrong with faith when one talks religion but how about science?

      It is somewhat ironic that not Alex C. disputes [all of] the results of the last 60 years [otherwise he hardly would waste any of his time here] but Chomsky does. His reply to McGilvray in SoL was a pretty clear indication. But an even stronger indication is a recent talk he gave to the Vatican Foundation. [report is in "The Tablet" February 1st 2014 - maybe someone can provide a link?] where he encourages us to embrace mysticism [one is also taken aback by the remark attributed to Andrea Moro]. So maybe instead of attacking Alex C. for asking very reasonable questions Norbert could dedicate a blog to explaining how mysticism and science go along. Maybe it is not logically impossible that they do but one would like to see a pretty convincing argument...

  3. The dialectic is something like this.

    The POS goes something like this:
    A) you can only learn the ECP from ECP effect sentences
    B) adults have the ECP
    C) ECP effect sentences are vanishingly rare to nonexistent in the PLD
    D) ECP is innate.

    A) is the indispensability part. It is a premise of the argument so you probably want it to be true, if you intend to use the POS to convince anybody of anything. I think A is already problematic since you can't say what you need to have to learna nything without a learning algorithm. But you don't claim D) you claim that something else, deeper and more general is innate (ECPX).
    So the argument looks like:

    A2) you can only learn the ECPX from ECP effect sentences
    B2) adults have the ECPX
    C) ECP effect sentences are vanishingly rare to nonexistent in the PLD
    D2) ECPX is innate.

    A2 now looks really weak. To be concrete suppose that ECPX is something like
    "forward crossed composition is not available". Then A2 is just false.

    Now as it happens, I think ECPX is the sort of thing that probably is innate -- something is innate for sure, and every model I am interested in has some built in class of grammars, and some built in biases. So this is a better example to consider the logic of the argument, since we broadly agree on the context. If we swap ECP for aux-fronting in polar interrogatives in English, we get the same structure of the argument but we might get distracted by the empirical premises etc.

    1. @Alex C.: Thanks for this explanation. Just to make sure I understand before i comment further: you say "ECPX is the sort of thing that probably is innate -- something is innate for sure, and every model I am interested in has some built in class of grammars, and some built in biases." Does innate here refer to domain specific or to domain general?

    2. It is neutral as to whether it is specific to language or not.

  4. I don't think these "my theories are better than your theories" discussions are very helpful but I think this does need a comment.

    "I have no reason to think that you will get anywhere. Why do I think this? Not because it is logically impossible that you will. But because you haven’t come close yet!"

    This is an argument to never even start to develop alternatives. Chomsky and others argued very convincingly for 50 years that distributional learning techniques were inadequate in principle and as a result nobody looked at them for about 50 years. I don't think that you can then rely on their poor state of development as an argument for not looking at them. We have only had principled CFG inference algorithms since 2006, and MCFG algorithms since 2009 and strong learning algorithms since 2013. I am very happy with the progress. Indeed looking at the Berwick and Chomsky critiques that you have linked to, they criticize the first papers for limitations that have now already been fixed in papers that they do not cite*. For me that's a good rate of progress.

    *To be scrupulously fair the strong paper did not come out until after the Berwick et al papers were published.