LINGUIST List 4.961

Fri 19 Nov 1993

Disc: Psycholinguistics

Editor for this issue: <>

Directory

1. Daniel S. Jurafsky, Re: 4.952 Psycholinguistic evidence in linguistics
2. , 4.952 Psycholinguistic evidence in linguistics
3. , psycholinguistics
4. Joseph P Stemberger-1, Re: 4.952 Psycholinguistic evidence in linguistics

Message 1: Re: 4.952 Psycholinguistic evidence in linguistics

I thought I'd mention that many of us working on models of parsing or
interpretation for linguistic theory do consider psycholinguistic evidence as
essential in building a processing model, and in addition consider our models
to directly instantiate the structures or principles countenanced by our
respective linguistic theories.

A number of researchers, (Ford, Bresnan, and Kaplan (1982), Gibson (1991),
Jurafsky(1992) and others), following Chomsky's Competence Hypothesis and
Bresnan and Kaplan's Strong Competence Hypothesis, have been building models
based on the assumption that there is some tight relationship between some
internal psychological model of the knowledge of linguistic
structure (competence) and some model of linguistic processing (performance).
While this is certainly not necessarily true (as researchers
like Berwick and Abney have argued), I believe it may very well be true,
and that it is a good and useful working hypothesis.

In particular, it seems to me that assuming that the structural relations
we derive from our linguistic theories play the role of the structural
ingredients in our models of linguistic processing (and hence available to
psycholinguistic experiments) is a stronger hypothesis than the contrary (that
there is no necessary relation between the two).

The vast array of recent psycholinguistic work on sentence processing
is a testimony to the health of the field, and the importance with
which this data is being taken.

Just a thought re: David Pesetsky's defense of what I, like
many others I think, consider to be indefensible: the way in
which psycholinguistic evidence (or what passes for such) is
frequently used in ling. but only when it fits, and the way it
is dismissed as irrelevant the moment it does not fit. Or actually
two thoughts:

First, I think that purely "internal" linguistic data is often
treated this way too. Whole areas of fact, which were supposedly
crucial one day, suddenly become irrelevant.

Two, while he is right that there is not an agreed-upon methodology
for using such data in linguistics, that is precisely what makes its
casual use so irresponsible. Those who use such data should, I think,
commit themselves to at least some degree to some kind of methodology,
and, moreover, should be willing to engage in serious discussion of
said methodology. But claiming crucial support for your theory
so long as the data seem to work and then refuse to acknowledge
that they seem to support somebody else's theory when they turn
out the other way is not any kind of methodology.

Steven Schaufele puts the psycholinguistics question into a historical
perspective in an interesting way, but I think it is not sufficient to say
that linguistics in Europe developed out of humanistic disciplines like
classics and philology, while in America it developed from anthropology and
psychology came in only with Chomsky.
 Linguistics has been cognitively oriented long before psychology itself
was an academically established discipline. The somewhat speculative
grammaire raisonnee tradition was cognitively oriented, and so was
Wilhelm von Humboldt's linguistics (which focused much more on empirically
observed cross-linguistic data). The psycholinguistics of the psychologist
Wilhelm Wundt was widely noticed among linguists. The Neogrammarians had a
clear cognitive orientation, and so did the European structuralism of Trubetzkoy
and Jakobson (note that Jakobson 1944 was one of the first linguists to bring
together data from aphasia and linguistic theory).
 The only influential anti-cognitivist, anti-psychological school was
Bloomfield and the post-Bloomfieldians.
 This does not mean that those earlier linguists necessarily paid a lot of
attention to research in psychology (though some did), but I think that many
of them would have accepted the view that what they were doing was part of
a larger attempt to understand the human mind (at least among other things;
why should there be a contradiction in saying that linguistics is both
a branch of psychology and a branch of sociology?).

Martin Haspelmath, Free University of Berlin

Message 4: Re: 4.952 Psycholinguistic evidence in linguistics

Just a reminder about comments that various people have been making about
"falsification" of linguistic concepts based on psycholinguistic evidence.

First of all, you get "falsification" of linguistic concepts using only
internal evidence as well. And sometimes it turns out that the concepts
weren't really falsified at all. Pesetsky points out the trace stuff in
syntax, where early psycholinguistic work suggested that the idea was
wrong, but some recent work has shown that it might be a useful concept,
using both linguistic & psycholinguistic data. I can think of cases in
linguistics where linguistic data was used to reject a hypothesis (such as
that most phonological rule ordering can be derived through simulateous rule
applications), which then went unexplored for years, until it surfaced
again in a new guise that people get excited about (this time with
simultaneous constraint satisfaction). "Falsification" is always a tricky
business.

Second, to falsify something, you must be very clear on what the
predictions of the concept are, and on how easy it will be to spot.
Suppose I have a syntactic theory that predicts a difference in the amount
of time that it will take to comprehend construction A vs. construction B.
And suppose that I do the experiments and find no differences. Well, how
BIG a difference was predicted? The theory probably will not be very
specific; the difference could be 500 msec or only 2 msec. And the smaller
the actual differences, the harder it will be to detect. Failing to find a
predicted difference is an instance not of negative results, but of null
results, and it's hard to interpret. Sameness is in principle impossible
to demonstrate.

On the other hand, if the theory says that two things will be the same,
and psycholinguistic evidence says that they're different, then the theory
has some explaining to do. Differences can in principle be demonstrated.

But this is the way it is with internal evidence as well. We can always
say that there's no need to distinguish between A & B, that they have
identical representations. But the next issue of LI may prove that A & B
must have quite different representations, because of properties that
had not been previously known.

I think that the value of psycholinguistic evidence is that it gives us an
additional source of evidence, often of a fairly different nature, to test
the same hypotheses that we've been looking at with internal evidence. the
more different sources of evidence we have that argues for a conclusion,
the more confidence we have that the conclusion is right. And the more
different sources of evidence conflict, the more nervous we should be.

 ---joe stemberger