LINGUIST List 9.457

Wed Mar 25 1998

Disc: NLP and Syntax

Editor for this issue: Martin Jacobsen <martylinguistlist.org>

Directory

1. Pius ten Hacken, Re: 9.383, Disc: NLP and Syntax
2. Peter Menzel, NLP and Syntax

Message 1: Re: 9.383, Disc: NLP and Syntax

In reaction to Phil Bralich's posting in Vol. 9-383 I would like to
restate my point briefly as follows:

The goal of linguistics is an explanatory account of the data. A
descriptive account of the data is not the goal of linguistics. It is
not even an intermediate goal. It is only a side effect of the search
for an explanatory account of the data.

As a result, a full descriptive account of the data has by itself very
little value scientifically. It does not indicate any degree of
maturity of a scientific theory.

Therefore a partial explanatory account is better for a linguistic
theory than a full descriptive account.

If you do not accept this view of linguistic theory, you cannot make
sense of much of scientific practice. Some examples: Why do so many
articles and conference presentations START with a presentation of the
data rather than ending there ? Why do linguists never use a
parser-generator to get the most efficient CFG for their data set (or
suggest a more efficient one) ? Why do so many articles and
presentations take data from different languages into account or apply
a theory developed for one language to data from another language ?
The answers to these and many other questions lies in the above
description of the relationship between explanatory accounts,
descriptive accounts and the goal of linguistic theory.

The implementation of a syntactic theory as a parser can only test its
amount of descriptive adequacy. Therefore, there is no reason for
linguists to accept it as a valid criterion for the evaluation of a
linguistic theory.

Of course a parser aims at a full descriptive account of the
data. This indicates where the aims of linguistic theory and parsers
diverge.

I apologize if readers find this a repetition of points I made in my
earlier postings. Phil Bralich's last reaction suggests, however, that
for at least one reader I did not make my position clear enough so
far.

Pius ten Hacken

==================================================================
==================================================================

Dr. Pius ten Hacken
Institut fuer Informatik/Allgemeine Sprachwissenschaft
Universitaet Basel
Petersgraben 51 || Tel. +41-61-267'33'38
CH-4051 Basel || Fax +41-61-267'32'51
Switzerland || email: tenhackenubaclu.unibas.ch

web page: http://www.unibas.ch/LIlab/staff/tenhacken

==================================================================
==================================================================

Well, I certainly seem to have provoked a minor storm with my
comments. Let me try to answer the most egregious misunderstandings
or misinterpretations.

Bralich takes me to task on three issues: the distinction I drew
between theoretical and computational linguists (linguistics); my
comments on native speaker intuition and linguistic theory; and my
doubts concerning the applicability of algorithms and math in general
to descriptions of linguistic and other psychological processes (as
does Sparkman). Arnold doubts my understanding of modularity, and
seems to think that, therefore, my comments about mathematical and
algorithmic descriptions of linguistic and psychological processes are
invalid.

At bottom, the whole disagreement rests on different intepretations of
the terms "native speaker intuition" and "psychological reality",
which, as I indicated, are used in more or less the same sense by most
linguists. (Remember, by the way, that I already said that linguists
are not always too clear what they (we) mean by this concept.) I
don't think there's any point in trying to argue (or trying to figure
out) what Chomsky "actually meant" by this concept; what interests me
is how we, as linguists (practicing scientists) have understood the
term, and have applied it in our work.

Now, certainly, Bralich is correct in his observations concerning
native speaker intitions about actives and passives not being
paraphrases of each other, and the other remarks he makes about
speaker perception of relationships between sentences. Nor do I doubt
that he (and Bickerton, whose work in sociolinguistics I know well)
are theoretical linguists. (I guess I did not formulate my
distinction too clearly: It should have been the one I made later;
i.e., between computational and speaker linguists and linguistics.)
Thus, my question remains: What is, for them, psychological reality?
>From Bralich's remarks on my discussion, I conclude that his notions
of this concept do not correspond with mine. In retrospect, it seems
to me that many arguments among linguists turned on this point of
psychological reality, with people making different assumptions about
the "reality" part of it without clarifying them -- perhaps even to
themselves.

In my own work, I've found this to be a key concept; one that, over
the years, has become, first more and more psychological, and then
more and more neurophysiolocal. For, in the long run, psychology has
to be founded on neurophysiology. And there are, in a sense, my
answers to Bralich and Arnold: Psychological reality, in the long run,
comes down to "neurological" reality. To be sure, under certain
assumptions of what constitutes psychological reality, formal syntax
of the type Bralich (or GB, or minimalist syntax) propose are
"psychologically real". For me and others they are not, because these
proposals don't take into considerations the way minds (and brains)
work. Bralich accuses me that, with remarks like these, I want to
throw out math and logic, and other algorithmic achievements of the
mind/brain. Far from it ! But the fact that some of us are capable
of performing these feats does NOT mean that our mind/brain habitually
and normally works like this.

In this connection, a -- not so minor -- aside: The story of how
easily children learn language(s) and complex social systems, and how
difficult it is for (most of) them (and adults) to learn (the by far
not as complex) algebra, logic, etc, is well known. Some people
conclude from this that the proposed modularity of mind treats
language differently, and thus enables us to learn this most complex
of human achievements, language, easily. Now, aside from the fact
that this assumption does not explain why we, as human beings, learn
complex social systems quite as easily as we learn language, and learn
to survive in a very complex physical world almost equally easily,
there is no evolutionary reason for assuming a seprate "linguistic
component" of the mind/brain. Again, on the contrary: We do know that
nature and evolution seem to be "extremely good" at adapting already
existing mechanisms for new, useful-to-survival purposes. It stands
to reason to assume that this also happened with our language ability,
our social ability, etc. If evolution had created some all new and
powerful processors for language, then there would be reflexes of this
found in the brain. But they are not. And in this sense, I say that
modularity of the mind (in Fodor's terms) is an interesting game, but
it has no "neurological reality". (Remember: All psychological
processes must ultimately have a neurological basis.)

Arnold states that (Fodor-type) modulyrity of mind is "an interesting
theory", and does not need to have close neurological correspondence.
This raises the question of what, to him, is an "interesting theory"?
One that makes claims concerning the structure of the mind in the face
of neurological evidence to the contrary?!?

There remain only two final remarks. The first is to Bralich and
Spackman: Of course we should not stop trying to capture aspects of
reality using math, in fact, much work has been done describing
certain aspects of brain processes using mathematical models. To
reiterate my point: That does not mean that we use algorithms in
dealing with reality, including language. For the third time, on the
contrary: Recent (and even not-so-recent) psychological evidence
indicates quite strongly that in order to understand anything, we
construct model(s) of it. Neurological evidence indicates that we do
this using vector coding and vector processing in a multi-dimensionsal
space, with (oversimplifying) neuron serving coding points, and neural
networks serving as processors. One major advantage of this line of
research is that it finally allows us to get away from notions like
"language of thought", which explain all human thought in terms of
language, and words in terms more words.

The second point takes up on one of ten Hacken's remarks: The notion
that data are independent of theory, and that there can be a
theory-independent description of them went out with logical
positivism. Today, psychologists, philosophers, and neurologists all
agree that even "perception is heavily theory laden". This does not
invalidate Bralich's description of (English) syntax, but it does put
it inside a certain theoretical framework, one which is different from
that of some linguists who hold different theories. As to whose
theory is better, the future may tell; as to whose theory is more
stringently formulated, chances are that the one that can be
implemented on a computer is. But, to echo (paraphrase) ten Hacken:
Given what we know about how mind/brain operate(s), a theory that can
be formalized to run on a computer is not likely to correspond to how
speakers deal with (learn, use, store, process, etc) language.

My response seems to have turned out, once again, longer than I
anticipated.

Peter