Biological and cognitive realism

[George Lakoff]

Dear Tom,

I'll start with something I forgot to mention.

Dan Jurafsky's parser (see his Berkeley thesis)
is set up to accord with the psycholinguistic data on language processing.
It uses constructions that include semantic and pragmatic information.
Those not familiar with Dan's work should be. In addition, there's going to
be a conference at ICSI at Berkeley after Cogsci next August on
psycholinguistically-based language processing. Dan JUrafsy and Terry
Regier are running it.  The point is that there is a community of
computational linguists who have higher standards than just getting tree
parses.

Now back to your posting.

Tom, I said grammar does exist. As has Langacker. What we say is that it
exists as constructional pairings of cognitive semantics and phonological
representations (word order included). The exact nature of grammar is an
empirical question. We argue our position empirically. See, for instance,
my CLS 86 paper on the frame semantic control of the coordinate structure
constraint, where I argue that there is no autonomous syntactic coordinate
structure constraint.

If you want some details, take a look at Ron's Foundations of Cognitive
Grammar (about 1000 pages of details), as well as the rest of the
literature in the field. If you want a real short intro, start with Ron's
Concept, Image and Symbol.

Also, I can't imagine how you could possibly have read the last 120 pages
of Women, Fire and Dangerous Things -- the most extensive study of
there-constructions ever done by far -- and still think I don't believe in
grammatical constructions. Try reading it for evidence for constructions
and against autonomous sytax.

What I said was that AUTONOMOUS syntax didn't exist. That's a very very
different claim than that grammar does not exist, which is ridiculous.

It seems to be obviously true that syntax is not purely autonomous. If
syntax were autonomous, it could take into account no input from anything
else, such as meaning, general cognition,perception, processing
considerations, etc. In short, it would be as Chomsky has always
represented it, a grammar box with no input, only output. This is necessary
if Chomsky's basic metaphor is to be accepted, namely, that a sentence is a
sequence of formal symbols, a language is a set of such sequemces, and
grammars are devices for generating such sets. The theory
of formal grammars requires that rules be stated only in terms of symbols
in the grammar. In short, no external input that is not in the formal
language can be looked at by the rules of a formal grammar.
        Such a view, if it is to be instantiated in the brain would require
a brain module
(or some complex widely distributed neural subnetwork) WITH NO INPUT!  But
there is nothing in the brain with no input. Such a view is biologically
impossible.
        What we have proposed instead is that grammar is based on other
nonlinguistic
cognitive abilities and that neural connections and bindings bring about
grammar. We try to show exactly how.
        I suggest you read the discussion by Gerald Edelman in Bright Air,
Brilliant Fire
called "Language: Why The Formal Approach Fails", starting on p. 241 for a
neuroscientist's view of the issue. Edelman is director of the Neuroscience
Institute
at Scripps in La Jolla.

        Next, I was not attacking functionalists at all. I happen to like
and teach much functionalist research. I think functionalism has
contributed a great deal.

        Indeed, I was not attacking anybody. As I said, I think there are
few enough linguists doing serious research in any school, and despite
theoretical differences, lots of folks have lots of things to teach us.


>
>1. GRAMMATICALIZATION AND SYNTACTIC CHANGE:

Cognitive linguists have been working on grammaticalization and syntactic
change for quite a while. But it requires a nonautonomous view of grammar.
Traugott has written for decades on the need for pragmatics and metonymy is
accounting for grammaticalization. Kemmer has argued overwhelmingly that
semantics (of a cognitive rather than formal nature) is required for
grammaticalization. Heine has been arguing that conceptual metaphor is
necessary to understand grammaticalization.
That has been confirmed in great detail by Sarah Taub, in one of the most
extensive studies of grammaticalization to date -- on Uighur. Any serious
look at the grammaticalization literature in recent years will see the role
of cognitive semantics -- especially metonymy and metaphor -- in
grammaticalization.



>2. ACQUISITION: As Dan Slobin has shown, the semantics precedes the
>acquistion of much of grammar.

As for the acquisition of semantics, there I suggest you read Terry
Regier's THE HUMAN SEMANTIC POTENTIAL, MIT Press, 1996. Regier's acquistion
model shows that spatial relations concepts and terms (not counting 3D and
force-dynamics) in asignificant range of the world's languages can be
accomplished on the basis of structured neural models --
with the structure given by models of neural structures known to exist in
the brain:
topographic maps of the visual field, orientation-sensitive cells,
center-surround receptive fields, and so on.  The point is that
NONLINGUISTIC aspects of brain structure can be used to learn conceptual
and linguistic elements. Regier also shows how this can be done with NO
NEGATIVE INPUT. Regier did his work in our (Feldman's and my)  group
several years ago.
        For recent work on related topics from the ICSI Lzero group, check
out the Lzero website at icsi.berkeley.edu/lzero. Other topics have
included the learning of verbs of hand motion, aspect, and metaphor. The
point is to do this in a way that is biologically and cognitively
realistic.



>
>3. PIDGINS AND CREOLES: We have been observing for a long time the
>peculiar consequences of having no morpho-syntax

Ron and I of course recognize morphosyntax. We just give a nonautonomous
theory of it.
I happen to like most of the work you cited, much of which is not set in
terms of autonomous syntax.


>4. CROSS-LANGUAGE TYPOLOGICAL VARIATION:

As Bill Croft, Suzanne Kemmer, Ron Langacker and others have been writing
for years,
a linguistics based in cognitive semantics does a better job at getting at
cross-linguistic
typological variation. Functionalist studies of classifier languages
(Collete Craig)
and speech act indicators (Nichols and Chafe) show that such typological
cannot be done with an autonomous syntax that does not admit semantic
factors.






>
>5. DISCOURSE: Again, Ron and I support a nonautonmous theory of grammar and of
constructions, Our work fits very will with the work you cite by Wally,
Sandy, Jack, and other of our friends and colleagues. I can't imaginine any
of them supporting an autonomous syntax.
        Indeed, the whole idea of the CSDL (Conceptual Structure, Discourse
and Language) conferences was to bring together the cognitive and
functionalist approaches into a unified group.
        The first two conferences were enormously successful. I know you
didn't attend either, Tom, but why come to this year's conference at
Boulder. It should be a good as the the others.
        For those who have bought it yet, I recommend the proceedings of
the first conference, CONCEPTUAL STRUCTURE, DISCOURSE, AND LANGUAGE (edited
by
Adele Goldberg and published by CSLI publications, distributed by Cambridge
University Press).


>
>6. COGNITIVE PSYCHOLINGUISTICS: Same point again. The Nijmegen group has
>been studying universals of spatial relations, which requires a form of
>cognitive semantics (not formal semantics or formal syntax). Brian and Liz
>have been arguing for years against autonomous syntax and the language
>box.
        For an extensive study from cognitive psycholinguistics supporting
the results of cognitive semantics, read Ray Gibbs' THE POETICS OF MIND,
published by Cambridge University Press. If those experiments don't
convince you, I don't know what will.




>6. NEUROLOGY: We all know localization is a complicated, that grammar
>in adults is distributed across many "modules". Sure, the modules bear
>little resemblance to their Jerry Fodor name-sakes. They interact, they
>talk to each other, they are NOT encapsulated, they collaborate with
>"cognitive" modules (attention, activation, memory, intention, pragmatic
>zooms, etc.).

Exactly the point. No syntax module without input from the reset of the
brain, hence no autonomous syntax.


But however widely distributed, portions of this complex
>mechanism can be knocked out selectively by lesions. What is it, George,
>that aphasics have lost, exactly?

The Damasios have answered that: Connections. Not localized modules.

You study their transcribed discourse
>(Menn and Obler, eds 1990, e.g.), and you notice that
>  (i) the lexicon is there. nouns, verbs, adjectives.
>  (ii) the coherence of discourse is still there (referential coherence,
>       temporal coherence, all the measurables).
>So what is it that is NOT there?

Again, connections. Drop a note to Liz Bates at UCSD for all her many
surveys of arguments against the syntax module. Agrammatic patients have
been shown to be able to make grammaticality judgments (Linebarger, et al).
Liz cites her own work with an agrammatic patient in Italy ( a well-educted
architect) who could not repeat a grammatical sentence, but could only say
one word: the Greek grammatical term for the syntactic phenomenon! The
point: It is now well established that agrammatism is not the wiping out of
a supposed "syntax module".

About connectionism, there is a big big difference between PDP
connectionism and structured connectionism. I was explicitly talking about
the latter. The former cannot account for most linguistic phenomena.

>
>7. EVOLUTION:

Your remarks about iconicity attest to the inadequacy of autonomous syntax.
Iconic constructions require the pairing of form and meaning. That cannot
be done in autonomous syntax. For a discussion in my work, see Chapter 20
of Metaphors We Live By. A magnificent study of the role of cognition,
especially cognitive semantics, in iconicty
is now in progress -- Sarah Taub's dissertation on ASL. If you have never
heard Sarah
talk on the subject, you should. Invite her up to Oregon as soon as possible!

In the course of evolution, layers have been added to the brain. Higher
cognition is done in the neocortex, which is furthest from direct bodily
input and which takes input from layers closer to bodily input. The study
of conceptual metaphor show that that huge system is grounded in the
perceptual and motor system and that abstract concepts tend to be
conceptualized in bodily terms. This is just what one would predict given
the evolutionary structure of the brain.

        Tom, your posting was useful because all the topics you mentioned
are important, and indeed support the cognitive position (and other work on
nonautonomous grammar, like most functionalist work). I'm sorry you didn't
understand what I was saying. I've written a lot on the subject, so I
thought it would be clear, but maybe it wasn't for those not into the
cognitive literature. For that, I apologize. I hope this clarifies the
position.

        I'm also glad that Joan and Dan are really into your work Tom. I
hope that they been reading other functionalists, and maybe they'll get to
the cognitive literature too that way.

        I agree with Tony Wright. Serious discussion is needed. It should
be based on serious reading, of course.

        Gotta get back to book writing. Take care, Tom, and Happy New Year.
You're invited for a beer next time you're in Berkeley or I'm in Eugene or
if you decide to go to Boulder for CSDL. Let's try to talk things out
calmly and in detail.

Best wishes,

George