LINGUIST List 2.85

Thursday, 21 Mar 1991

Disc: Cognitive Linguistics

Editor for this issue: <>

Directory

1. Wayne Cowart, Cognitive Linguistics
2. Bill Poser, cognitive linguistics
3. "Daniel L Everett", falsification and usefulness
4. Vicki Fromkin, Re: Cognitive Linguistics
5. Scott Delancey, Re: Cognitive Linguistics

To Fred the Beagle:

 Autonomy claims feel a bit contentless and
invulnerable (at least to a workaday Popperian) because
they are metatheoretical. Thus the only 'results' that
can endorse or falsify them are those that emerge slowly
as theories crystalize and gain credibility. Note:
"theories". Autonomy claims can't really be tested
against any one theory because they are claims about
relations among theories. So you have to wait for at
least two theories to mature before questions about their
relations can be settled. This will never bring the rush
you get when a theory crumbles under the weight of a
single observation. And that hardly ever happens
anywhere anyway.
 As compelling accounts of specific phenomena become
available, the answer to the autonomy question will
become increasingly clear. Progress will be swiftest, I
suspect, if the focus can be on those phenomena. A
sufficient variety of hunches about the metatheoretical
issues is already on the table.
 The terminology doesn't matter. Good theories can
survive assaults by tendentiously named alternatives.
 So I'm with you; let's stay out of this debate.

Wayne Cowart

 I am frankly shocked at both the content and the virulence
of some of the spate of responses to my comments on "cognitive linguistics"
and the modularity debate, mainly because they are based largely
on gross misrepresentation of what I said. I am accused of claiming
that non-linguists know nothing about language or linguistics and
can make no contribution. A concise example is Alexis Manaster-Ramer's
reference to:

 my brethren who cannot seem to imagine that you people,
 however intelligent you may be, have any right to say anything
 about language. 

Brother Alexis here summarizes the view that some of the other writers
appear to hold.

In point of fact, I said nothing of the kind. I merely said that
linguists typically know more about language than other people do,
and that the gap in overall level of knowledge about language between
linguists and others is large. Nowhere in my message did I make the
sort of extreme universally-quantified statement that I am accused
of making.

 I reproduce here in its entirety the main part of my message
relevant to this issue so as to put this ridiculous distortion to rest:

 Now, this is not by any means to say that all non-linguists are
 ignorant, but it is to say that that it is very important, in
 any psychological investigation of a behaviour, to have a good
 idea of what that behaviour is like, and that linguists play a
 crucial role in investigations of linguistic behaviour by
 providing most of the facts and generalizations about the
 structure of language that need to be accounted for.

Please note that I explicitly denied that "all non-linguists are
ignorant", and that I claimed merely that linguists provide "MOST
of the facts and generalizations about the STRUCTURE of language..."
[emphasis added]. I did not claim that non-linguists do not provide
any of these facts and generalizations, and I explicitly limited
my claim to structural matters, as I am fully aware that much of
what is known about the mental processing of language, for example,
is due to psychologists, not linguists.

In this connection I also call attention to my statement that:

 It is true that linguists are not, in general, particularly
 well versed in psychology and up on the literature on
 modularity.

Moreover, the claim that I deny the right of any non-linguist
to say anything about language is directly refuted by my statement
that:

 the input of anyone with relevant knowledge is welcome

Similarly, there is no basis in my message for the claim that
I am opposed to a variety of approaches to the modularity issue.
Indeed, the brunt of my message about modularity was that the
issue is NOT settled.

In sum, I do not hold the views imputed to me, and my message cannot
reasonably be interpreted as it has been. I believe that those
who have misrepresented me in this way owe me an apology.

Turning to substantive matters, let me first respond to Jon Aske's
suggestion that "formal" linguists are being silly by ignoring
all the neat phenomena he is interested in. I agree that many
of these phenomena are interesting and deserve more attention.
Indeed, lest this seem mere rhetoric, let me direct his attention
to my paper "MA" (a Chinese character that I can't reproduce here)
in the volume _Interdisciplinary Approaches to Language: Essays in
Honor of S.-Y. Kuroda_, edited by Carol Georgopoulos and Roberta
Ishihara, Kluwer, 1991, pp. 449-458, in which I discuss the
semantics of a Japanese morpheme from a perspective drawn from
the work of Elinor Rosch and George Lakoff. Specifically,
I propose that this morpheme restricts the denotation to the
cognitive reference point, which explains, among other things,
why it cannot be added to stems that lack a prototype (a reference
point that lies within the extension). So you see, I have some
interest in these things myself and have read Lakoff's book and
some of Rosch's papers. I'm (sort of) one of you, Jon.
[If you are actually moved to read the paper and can't find a copy of
this nice but incredibly expensive Festschrift, Paul Kay has an ms. and
I have put a copy of the published version in the mail to George Lakoff.]
But my point was that if you want to take a strong anti-modular stand,
you can't just find that SOME aspects of language aren't modular
and leave it at that, especially when the claims in favor of
modularity are based on other aspects. If it turns out that
semantics isn't modular, that doesn't mean that syntax isn't,
and if the claims you're trying to refute are about syntax,
you haven't met the challenge by talking about semantics.

With regard to Fred (=youngheevm.epas.toronto.edu)'s statement
that there have been non-autonomist attempts to account for
locality principles, let me cite the relevant sentence:

 For most of the generalizations posited by linguists,
 especially formal ones (e.g. locality principles), I
 have seen no attempt at functional explanation.

Note that I was talking about functionalism there, not modularity.
I'd certainly be interested in references to functionalist
approaches to locality principles (especially in phonology and
morphology).

Second, consider Robert Goldman's point about Roger Schank and the
related point by George Berg. I don't think I'm being unfair to
Schank. Schank waffles on what he considers syntax and what semantics,
and backs down on precisely this point, introducing considerations of
word-order, while still maintaining that what he is doing is
syntax-free. Now he may be right that you don't need to do the
same sort of parse that syntax-based parsers do, but certainly
word order is standardly considered to be syntactic. And I did
say that it was only the pure form of his theory that fails on these
sentences.

I think that the parsing procedure that Goldman describes
implicitly makes use of syntactic information. At least for the
sake of argument, I will accept the claim that some sort of
representation that says that the verb "see" requires an agent
and a patient is a semantic representation. (The point is
arguable - once you start considering differences between
obligatory and optional arguments and so forth you may decide
that this kind of argument structure is really syntactic.) The
crucial thing that Goldman's parser does in addition to trying to
satisfy this argument structure is to do it in a certain order.
It looks for the agent first, then for the patient. In doing so,
it makes use of the knowledge that in active sentences with no
dislocation (topicalization etc.) the agent comes first, i.e. a
word order fact. Hence, I claim that this parser makes use of
syntactic information, albeit of a limited sort. To see that this
isn't a universal property of conceptual organization, we have
only to consider such a parser for a language with OVS word order
(e.g. Hixkaryana). To make it work, the parser will have to be
modified to look for the patient first, then the agent. Yet I
hesitate to say that the conceptual organization of speakers of
Hixkaryana is any different from that of speakers of English (at
least not on this basis). Do we really want to say that their
conceptual abilities are different, or that they have a different
notion of "see"? It seems much more plausible to say that their
language has a different syntax.

Schank may have been trying to show that semantics plays a larger
role than it had been granted, but he did so by making a much
stronger and much more dubious claim about the lack of need for
syntax.

Whether or not Schank has a linguistics degree isn't the point.
The point is the seriousness with which such notions are taken by
others in the field. I certainly don't claim that anyone with a
linguistics backgroud is right or reasonable.

On the general point of sophistication in this area, I'm well
aware of the work of many AI people, and that some Natural
Language Processing people (perhaps a better term as some don't
like to be associated with "AI") are quite sophisticated
linguistically. But the gap overall is still quite large. And I
would say it is even larger when one looks at morphology,
phonology and phonetics, closer to my own area. For example,
most work on morphological parsing until a few years ago was
ungeneralizable because the people writing it had little idea of
the range of morphological systems. (Richard Sproat of Bell Labs
has a nice discussion of this in a tutorial paper on
morphological processing that is supposed to appear as an ACL
publication, or maybe already has.) An example from phonetics is
the statement in Douglas O'Shaughnessy's book _Speech
Communication: Human and Machine_ (p. 62) that "Except for trills
and ingressive sounds ... English provides good examples of
sounds used in various languages...", which gives the misleading
impression that the phonetic inventory of English covers most
known speech sounds. In addition to trills and ingressives,
English lacks glottal ejectives, rounded front vowels,
retroflexes, uvulars, bilabial fricatives, pharyngeal fricatives,
nasal fricatives, and pharyngealization, among others. If we
consider distinctive oppositions we may add to this list still
other categories, such as aspiration, nasalization, and voiceless
sonorants. This is, overall, a very good book on speech
technology, but this statement and others like it show a very
limited knowledge of the range of speech sounds and inventories
in the world's languages. (This bit is lifted from my review,
in press, in the Journal of the International Phonetic Association.)
Although I certainly chose extreme examples, I have a great deal of
contact with AI people, psychologists, and others in linguistics
related fields, and although there are some who are very knowledgable,
I am constantly appalled at how little so many people know.

In response to Margaret Fleck, while I am not a student of
computer vision, I do know some people in the field and have read
Ballard & Brown's _Computer Vision_, so my image is not based on
pure hearsay. Some people in the field do care about human
processing, but the people working on humans frequently call
themselves psychologists or psychophysicists or
neurophysiologists, restricting computer vision to people who
are, literally, interested in vision by computers, and that is
how I interpreted her message. I certainly agree that there are
parallels between different cognitive subsystems and that
comparison is worthwhile. Analogies are interesting and often
productive, but generally not probative. My query was why
Margaret thought that computer vision was essential, or whether
perhaps it was just another field that shed light on cognition in
general. It seems to me that she has answered that the study of
human vision sheds light on the general issues, with which I
fully agree.

I also agree that visual/auditory interaction is interesting.
Indeed, another example is the McGurk effect, whereby subjects
presented with a voice saying [ba] and a video image of a speaker
articulating [ga], hear [da]. (McGurk, H. & MacDonald, J. (1976)
"Hearing lips and seeing voices," Nature 264.746-748.)

Let me close by agreeing with Margaret's observation that people
in different areas know too little about other areas. My previous
message was about a particular case of this, namely that
non-linguists frequently do not bother to learn enough about
linguistics. But the point certainly holds true in other directions
and between other subfields.

For Susan Newman:

Popper and Feyerabend certainly do not exhaust the alternatives for
understanding or analysing scientific research. There are many other
ideas, some quite persuasive, and "naive falsificationism" is too easy a
trap to fall into for one to protest too loudly against theories which
are not readily falsifiable by one's own, perhaps idiosyncratic,
criteria.

Much of my research is spent in the Amazon, trying to learn something
about the languages spoken there. In field work of this type, where one
is face to face with a nearly overwhelming array of new cultural and
linguistic facts, it would be counter-productive for anyone to limit
themselves to one body of literature, such as formal linguistics, even
when it is their intention to publish their research results
primarily in formal linguistics journals. Formal linguistics of the
chomskyan type does not claim to study "language" anyway (but *linguistic
competence*), so the wider the reading the better in field work.

Lakoff, Fillmore, Hopper, Givon, Chafe, and the many other so-called
"functionalists" or "cognitivists" have produced a significant body of
literature. While I cannot honestly claim to have read it all, I have
read much of it. It is often illuminating and I know that I have come
to appreciate and maybe even understand things in Amazonian languages
through these perspectives that I would not have learned through a
purely formal perspective. Nevertheless, it is true that these accounts
very often leave me less than satisfied because, when all is said and
done, they can explain just about anything. Functional accounts can
often serve to explain equally well two, mutually contradictory
hypotheses. This is not the case in formal accounts. In this sense,
Functional Linguistics and Cognitive Linguistics are often less than
helpful and most frequently suggestive rather than conclusive, precisely
because they can't be falsified (at least I haven't seen how to do it
and have lost interest in trying). Similar problems for functionalist
approaches in evolutionary theory, not unlike the claims of so-called
Cognitive Linguistics, are discussed in various works by S. Gould.

I seem to have started something. I think that's good. Better to
be talking to each other. I think. Maybe not. One day I will come back
into the arena -- Vicki

Though I'm sure everybody's (getting?) tired of the argument, I
think it's worthwhile to point out that the term "cognitive", in
the specific sense of describing approaches to linguistics that
seek explanations for general linguistic prinicples in terms of
general ( = nonspecific) prinicples of cognition, has been in use
for quite a while--the first published use that I know of being
Lakoff and Thompson's paper in BLS 1 (1975 !!). (Which is about
the same time that I first noticed the term "theoretical linguistics"
being used in the restricted sense of "mainstream generative
linguistics"). Why is everybody getting so excited about it now?

Scott DeLancey
University of Oregon