Thursday, July 06, 2006

Pinker and Jackendoff vs Chomsky, Hauser, and Fitch: Background   posted by agnostic @ 7/06/2006 09:02:00 AM

By popular request [1] and on the heels of Darth's interview with Steven Pinker, I'm going to post on the Chomsky, Hauser, & Fitch (CHF) vs Pinker & Jackendoff (PJ) debate on language evolution. For the relevant background, see here, here, and here (& the links therein). There are two papers we haven't discussed yet: the latest reply from PJ (available here as Jackendoff & Pinker 2005, "The nature of..."), as well as this paper (pdf) by Scott Atran, which in part deals with the CHF vs PJ debate. In a few days, I'll address these evolutionary issues in detail, but in this preliminary post, I'll review some basic linguistics. I'll gossip & provide sociological details along the way, since this isn't a sober journal, and these things are must-knows for non-linguists. First, here is an old post of mine outlining what the different components of language are. It's important to keep these components straight, as much of the CHF vs. PJ debate centers around which ones are "uniquely linguistic." Here is a simple diagram of how the parts work in the standard "T model" of Chomsky's particular view:

First, there are words in the lexicon (our mental dictionary), which are then combined into phrases and sentences via syntactic operations (e.g., "stick a direct object to the right of its verb"). This full sentence is then fed into two separate systems: one for speech production (this form is sometimes called PF for "Phonetic Form"), and the other for semantic interpretation (sometimes called LF for "Logical Form"). For those that have been following CHF and PJ's back-and-forth over the Faculty of Language in the Narrow sense (FLN) vs the Faculty of Language in the Broad sense (FLB), CHF's hypothesis is that only the stuff that's going on in the vertical line in the diagram ("narrow syntax") is what constitutes FLN -- the horizontal branches at the top represent interfaces with systems of the brain used for not uniquely linguistic purposes, and homologues or analogues of these not uniquely linguistic systems may be found in other species. PJ's hypothesis is that, while these non-syntactic components (i.e., the horizontal branches) may interface with other systems of the brain, and while they may have homologues or analogues in other species, that doesn't exclude the possibility that they've been crafted by natural selection such that their present form is uniquely human and uniquely linguistic.

Just what lies behind this "narrow syntax," on CHF's view? It is recursion, the operation that allows multiple embeddings:

FLN is just recursion.
Chomsky believes that [FLN is just recursion].
Pinker laments that [Chomsky believes that [FLN is just recursion]].
I'm not surpised that [Pinker laments that [Chomsky believes that [FLN is just recursion]]].
ad infinitum

Over the past 10 years or so, Chomsky's approach has been the Minimalist Program, which seeks to identify the sine qua non of syntax, and he believes it is the operation "Merge" -- basically, stick two things together to get a single, hierarchically structured clump of stuff. Note how that allows recursion: merge the verb "sleep" with the adverb "furiously," and you get back yet another verb, "sleep furiously." For those of you who've read some syntax before, maybe just from The Language Instinct or what have you, this is a big change for Chomsky & his fellow travelers, as before the "syntax" part of the T-model included the separate levels of "deep structure" and "surface structure," plus "transformations" that mapped one syntactic level to the next -- for example, simplifying, the deeper, unpronounced sentence You saw who? was transformed by a movement transformation that took the pronoun who and placed it at the front, before the sentence was sent off for pronunciation & interpretation: Who did you see? Within the Minimalist Program approach, by hypothesis, there are no special levels of D-structure and S-structure, and the number of transformations (like "Merge") is greatly reduced.

As a sociological aside, let me note that this "insight" was proposed decades before by practitioners of the camp within Generative Grammar known as Categorial Grammar (related approaches include Lexical-Functional Grammar and Head-driven Phrase Structure Grammar). In their view, syntactic operations are simply applying a function to an argument. For example, a transitive verb is a function that wants a direct object as one argument, as well as a subject argument to result in a complete sentence. Simplifying, consider the following derivation of the sentence Chomsky likes syntax, where S = sentence, NP = noun phrase, the part to the right of a slash represents the argument of the function, the part to the left of a slash represents the result of function application, and the subscript just below the slash shows in which direction the function looks for its argument.

Thus, likes (and any transitive verb) is a function that first looks for a direct object noun to its right, and once applied to the argument syntax, has closed off that argument slot, becoming something like an intransitive verb (i.e., likes-syntax). Then this likes-syntax function looks for a subject noun to its left, and once applied to the argument Chomsky closes of its remaining argument slot. With no open argument slots, we are left with a complete sentence. Note that functions in this approach are curried, meaning they take their arguments one at a time, starting with the outermost layer.

Now, "applying a function to its argument" is bascially the same thing as the Minimalist Program's "Merge," and it too allows recursion: the adverb "furiously" is a function that takes the verb "sleep" as argument and returns yet another verb, "sleep furiously." If this approach was proposed decades ago, why didn't anyone realize this? It's because, within the broad Generative Grammar paradigm which all descendents of Chomsky adhere to, there are rival sub-theories. The particular one proposed by Chomsky himself was at first called Government & Binding, and he has revised this with his work in the Minimalist Program. To be frank, with only the slightest of exceptions, close to zero syntacticians working in Chomsky's particular framework read, let alone cite & debate, the work of those working outside this particular framework yet still within the Generative paradigm. It would not be an exaggeration to say that Chomsky's status is like that of Marx or Freud at the height of their popularity, such that his followers demonstrate an almost religious cult mentality. If Chomsky proposes that the field do X, the next day, without skipping a beat, everyone drops what they were doing and picks up X. A linguistics professor once told me that, when they went to see Chomsky lecture at MIT, the room was so packed and the audience's attention so reverential, that it was as if God were holding court. To be fair, Chomsky is only minimally responsible for this -- it's true that he also doesn't cite or debate the rival Generative groups, but it's really his closest followers who have magnified this fault and made their enterprise almost like a cult of the charismatic leader.

This may seem like pointless gossip, but it's crucial to understand Chomsky's status in linguistics to get why so many either become quasi-disciples or sworn enemies, leaving marginalized those who agree with his broad Generative and Nativist approach but not the particular details. Because of the history of uncritical acceptance of, indeed devotion to whatever Chomsky proposes, their arguments (especially on evolution) need to be taken with a large grain of salt. Moreover, one advantage that CHF and likeminded thinkers claim on their side of the debate is that strong adaptationist approaches like those of PJ haven't illuminated theoretical linguistics, while weaker adaptationist approaches like their own have -- namely, the idea of purging FLN of as much as possible resulted in boiling transformations down to the Merge operation and shifting away from D-structure and S-structure. But as already noted, this barebones approach is nothing new to any of the rival syntacticians who had always looked askew at the runaway process of baroque architecture in the Government & Binding theories, so claiming that there was something special about the CHF view of language evolution that has illuminated theoretical linguistics is not particularly accurate -- after all the rival views were not informed by evolutionary considerations, just a skeptical attitude toward Rube Goldberg machines and less zealous devotion to God. Though let me note that a semanticist told me that the same cult mentality once prevailed in semantics until the death of the counterpart of Chomsky (namely, Montague) -- afterwards, the field opened up a lot.

So, mull over this background material, and by all means ask for clarification in the comments. The next installment will specifically address language evolution and the continuing CHF vs PJ debate.

[1] I know we don't discuss language that much here at this blog, but if there's anything linguistic that the readers are interested in knowing about, just shoot me an email and I'll try to post about it.