Saturday, October 20, 2007

Looking For Alternative Gurus

In the paper, C&M mention that a context free grammer can in no way represent the interior of a speaker/hearer, that we have to instead posit something akin to a transformational model. I'm not sure I entirely agree with this. Really, context free-grammars have done quite a bit of work in the NLP domain, and (S -> NP VP) is quite a standard thing to see, and if given enough of the rules with sufficient generatilty to allow for embedding and such, I don't see why this can't cover the range of acceptable sentences in a given language. I mean, do we really need transformations? We've gotten rid of D- and S- structure, so why not dispense with the entire paradigm? There are theories out there that do with out such frivoloties. Shouldn't we as informed consumers expand our intake and shop around? Quite frankly, if it's the goal of this class to critique the foundations of the Generative enterprise, shouldn't we be reading actual critiques? Lakoff, Pollard and Sag...."Women, Fire, and Dangerous Things" is actually a great read. I just don't think rehashing what Chomsky has to say will give us any insight into the potential flaws of the discipline that maybe others with alternative formulations may have already noticed.


Asad Sayeed said...

As to your question about context-free grammars, experience has shown that to cover anything close to the range of phenomena in natural language, you need a very large number of rules, especially once you take things like scope, binding, etc, into account.

This is OK for engineers, who are largely interested in "good enough" approximations to Actual True Reality. Even then, some engineers do pine for a better way to do things.

My understanding about why we don't read people like Lakoff in courses like this is that it's not exactly clear that his object of study is the same as the object of study that we're interested in. ie, some people use the word "language" to mean something other that what we mean.

Philip said...

Sarah, the goal of this seminar is not to critique the foundations of the generative enterprise. The goal is to evolve a reasonably clear idea of what "computation" means in the context of how we study language, or at least to define the space of possibilities for what it can mean, and to increase the ability of students to take a computational perspective on linguistics research.

Phrases like "how we study language" and "linguistics research" are very broad, but no seminar can be all things to all people. The intent is to focus on issues that are of primary interest to this set of participants.

Now, what's emerged over the course of the semester so far is that it's hard to discuss the role of computation without opening up discussion of the basic foundations of the Chomskian enterprise. This is natural, both because of the discipline's mathematical and computational roots (of which Miller and Chomsky 1963 is a great example), and because the conceptual vocabulary of current linguistic theory overlaps heavily with the conceptual vocabulary of computation per se. For example, it is pretty much impossible to have a substantive conversation about (or within) current linguistic theory without talking about observable data, input and output, representations, operations on representations, the simplicity or complexity of operations, etc., all of which are core issues when talking about computation.

Because of this, and admittedly because of the proclivities of some participants -- particularly faculty members, myself included -- the conversation has tended toward big issues and wrestling with the foundations. But the overall aim here, in class and on the blog, is a carefully considered discussion that improves our understanding of where and how computational thinking fits in. Overall criticism of the generative enterprise does not meet this criterion, and, for what it's worth, I'm not aware of a computational or formal description of Lakoff's ideas that would make them amenable to the kind of discussion we want to have. I could see discussing transformations versus other mechanisms for characterizing dependencies, but for what it's worth this is well trodden ground.

sarah a. goodman said...

My only point is why we're limiting to discussing things from a Generative standpoint. There are other theories out there that may address so-called mental computations and language. Why not look at some of those? I just assumed from the syllabus that in this examination, there'd be some critiquing obtaining relative to the pervasive borrowing of certain buzzwords like, well, 'computation'. As for myself, I'm personally not prepared to say we compute things mentally, as I don't think biological science is even close to telling us how we think. But then again, I don't consider myself a cognitive scientist. Given that, maybe I'm simply hoping the class will take a turn. That's all.

Asad Sayeed said...

I do recall that we have mentioned Bresnan, etc, from time to time, and they belong to linguistic perspectives that have diverged from what is often called the "Chomskyan" consensus, such as it is. Is that what you're talking about?

The premise of the course as I understand it is that we want to see whether we can describe human language as a computational procedure in a way that is linguistically realistic. You can argue whether "computation" is the right word for it, but we have ample arguments that we went through near the beginning of the course to show that it is a promising path to take.

Some might argue that until some of these unspecified other, non-computational perspectives provide an alternate analysis of the same observations that is stronger than what we have now, computation is the only game in town. I myself would actually be willing to argue that it is a trivial point that "computation" is the way to talk about it.

Certainly, eg, physicists, biologists, and even psychologists have no trouble using representational systems (that they sometimes don't directly call computational, but are) to reason about the idealizations they necessarily use. This is, at minimum, no different.

Your persistence in this critique is intriguing, because it's not clear to me what exactly you are aiming for. It's not obvious to me what nontrivial form an "anticomputational" argument would take. Certainly, it's extremely difficult for me to consider an "anticomputational" argument (or a relevant analogue of one) wielded in critique of any other science. Could you perhaps be more specific? Then maybe we could "take a turn."

Asad Sayeed said...

Furthermore, and in summary, if you're going to "look for an alternative guru", then you'll have to present a reason why the current consensus---whatever that is---is inadequate in some fundamental way. Otherwise, why would anyone bother?

For instance, I suspect most people attending the seminar would agree that the current state of understanding is inadequate and in need of improvement (like most sciences), but that we have no choice but to build on what has been done before until someone has an adequate critique of the foundations.

I am suspecting that you are dissatisfied with the foundations. Consequently, bringing forth the underlying critique that underlies your dissatisfaction may assist us all.