Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!newshost.marcam.com!zip.eecs.umich.edu!caen!sdd.hp.com!hplabs!hplntx!curry
From: curry@hpl.hp.com (Bo Curry)
Subject: Re: What's innate? (Was Re: Artificial Neural Networks and Cognition
Sender: news@hpl.hp.com (HPLabs Usenet Login)
Message-ID: <D3yHxq.5nG@hpl.hp.com>
Date: Mon, 13 Feb 1995 20:51:26 GMT
References: <3h66fb$c85@mp.cs.niu.edu> <D3n9IJ.GBz@hpl.hp.com> <D3so0B.A4L@cogsci.ed.ac.uk> <1995Feb11.013632.5037@news.media.mit.edu>
Nntp-Posting-Host: saiph.hpl.hp.com
Organization: Hewlett-Packard Laboratories, Palo Alto, CA
X-Newsreader: TIN [version 1.2 PL2]
Lines: 115

: In article <D3so0B.A4L@cogsci.ed.ac.uk> jeff@aiai.ed.ac.uk (Jeff
: Dalton) writes
: >But of course the UG is uiversal: it's constructed to be universal.
: >(Surely Chomsky et al change their account of the UG when they encounter
: >something the current version gets wrong or doesn't explain.)  So
: >it matters what Chomsky's theories look like: it matters that they're
: >not incredibly _ad hoc_.
: >
: >-- jd

Marvin Minsky (minsky@media.mit.edu) wrote:
: Yes, but there's something more serious even than that.  It seems to
: me that the rhetorical act of labeling the observed regularities--such
: as they are--as resulting from a "language instinct" has served to
: produce some strange and drastic results.  Let's begin with the
: problem of the unacceptability of an incompleted conjunction: what's
: wrong with

: "Whom did you see Mary and?"

: A first problem is that an "and" conjunction can apply, as Terry
: Winograd observed in his thesis, to any place in the referred-to
: antecedent structure--namely, to something like "I saw Mary and
: *Jack*" or to "I saw Mary and *so I ran away before she could see me*.
: I don't think we're nearly so likely to have the same trouble with the
: more restricted preposition ""with".  Consequently, it is not
: surprising that all languages evolve, commonsensically or memetically,
: rather than genetically, to discuorage that usage.

: Why?  Because to resolve that particular ambiguity, you'd have to
: compare the plausibilities of two different tree-searches over 
: your mental representations of the situation you're envisioning.  In
: the course of doing so, you'd need extra recursion or
: short-term-memory capacities that young children might very likely not
: have available during the language-learning stage of development.
: Thus the illusion of an innate syntactic constraint would emerge from
: limitations of other, quite non-linguistic limitations.  

: Note that these need not be "built in" at all.  On the contrary, the
: "constraint simply comes from something esle *not* being available.  
: I'm sure teen agers could quickly learn to use "Who did you see Mary
: and?" if the adults simply agreed not to use many utterances of
: extended supra-verb conjunction that would naturally lead to
: conjunctive questions like "Who did you meet and where did you go?

But not all structures which suffer from "processor overload"
are ungrammatical. People quickly get lost with recursive
center-embedded clauses such as "This is the cheese that the
rat that the cat that the dog bit chased ate", though they
are perfectly grammatical.

I agree that possible grammars are limited by the limitations
of the parsing machinery, but I don't see why that implies either
(a) that the grammar parsing machinery is not (at least partially)
specific to grammar, or (b) that the limitations of the machinery
are prior to the limitations of the syntax (rather than co-evolving).

: Now I'll try to be more constructive.

: *************FLAME ON***************

: The most impressive feature of the UG movement is its extraordinarily
: mind-numbing sterility with regard to making plausible theories.  Even
: in the brilliant exposition of Pinker's book, my cursory scan reveals no
: concern whatever with the sorts of questions that I think would occur
: to every other sort of scientist.  Suppose we take UG at face value.
: The rules are not learned, but they are already built into the brain,
: except for the selection of certain "parameters".  Then there are an
: obvious set of questions that you'd feel impelled to ask and in return
: get at least some hypotheses about the nature of and the workings of
: that machinery. Here are a few questions you'd think they'd ask:

: 1. What machinery  selects what clues from the ambient sounds and
: "knows" precisely which parameters to set to what?  (In other words,
: how can it parse before it can parse?)

: 2. What machinery connects those rules with a speaker's semantic
: machinery?  

: 3. How might knowledge be represented so that those rules could be
: applied to it?  What knowledge-structure manipulating procedures are
: available for applying them.  

: 4. What limitations -- on processing complexity, cache memory
: capacity, addressing space constraints, etc., constrain those other
: processes?

: 5. When you hear an utterance, what sort of machinery must be engaged
: to convert the expression into a knowledge representation?  

: 6. What sorts of searches are used to resolve the ambiguities.

: <General complaints about the scientific sins of "Chmoskyists" deleted>

: In particular I conjecture that the observed regularities of
: UG are not, in general linguistic at all, but mostly are side effects
: of the constraints of type 4. 

This conjecture remains, however, to be demonstrated.

I'm not defending "Chomskyists", or the research agenda of linguists.
You will surely agree that the truth of the UG hypothesis
does not depend upon the personal merits of its adherents.

I'm interested in this subject precisely because I'm interested
in the connections among (a) the empirical UG hypothesis, (b)
the results of psychophysical and neuroligical brain studies,
and (c) the results of AI. In other words, perhaps *someone*
could investigate the relationship between internal knowledge
representations in the brain, and grammatical representations
of knowledge in language, even if the "Chomskyists" will not.
In such an investigation, the regularities of syntax observed
by linguists supply important empirical constraints.

Bo
