LINGUIST List 11.321

Wed Feb 16 2000

Disc: Newmeyer: Language Form & Language Function

Editor for this issue: Anthony Rodrigues Aristar <aristarlinguistlist.org>

Directory

1. Kevin R. Gregg, Re: 11.191, Disc: Newmeyer: Language Form & Language Function
2. Lotfi, Disc: 'Generativia': a democratic 'nation'?
3. John A. Goldsmith, RE: LINGUIST List Daily Summary for Wed Feb 9 2000

Message 1: Re: 11.191, Disc: Newmeyer: Language Form & Language Function

>Kevin R. Gregg wrote (LINGUIST 11.169):
>
>>Grammaticality is a
>>technical term within a theory of grammar, hence it's the theory that tells
>>us what's grammatical or not. What a native speaker, including the
>>theoretical linguist, can tell us, with absolute authority, is whether the
>>sentence is acceptable, to that speaker; but the whole community of
>>English-speakers could with one voice reject 'The rat that etc.' as
>>unacceptable, without--simply in virtue of that unanimity-- impugning in
>>the least the veridicality of the theory that marks it grammatical.
>
>Essentially you have just said that a theory of grammar:
>
>a) cannot be tested by empirical means
>b) is completely arbitrary to the whims of the grammar theorist
>c) or else, only exists as a Platonic Ideal
>
>Though this may be an interesting philosophical position, it has serious
>problems as a basis for a science. Though I'm sure many of us would enjoy
>constructing logical Ideals of what we think language should be, (and some
>professional linguists do), this doesn't hold much interest to those of us
>interested in language as an empirical phenomenon in the real world.
>Empirical grammatical theories MUST account for the data actual usage of
>language by real speakers; there is no other empirical basis upon which to
>construct or evaluate them. (If you can think of another, I'd be glad to
>hear about it.)
>
>Marc Hamann
- -----------
 I think there may be a confusion here between 'empirical' and
'empiricist'; there certainly are other sources of empirical evidence for
grammatical theories besides utterances of speakers. Some people use their
intuitions, some people use ERP data, some do reaction-time studies. As
Fodor says, 'The data for a theory are *just whatever confirms its
predictions*, and can thus be *practically anything at all* (including, by
the way, bits and pieces of other theories)' (his emphasis). 
 There are also, in linguistics as in other sciences, non-empirical
reasons for preferring one explanation over another: the fact that there
is a plausible processing explanation for the unacceptability (not
ungrammaticality) of rat-type sentences, coupled with the lack of a
non-arbitrary syntactic criterion for ruling them out, is itself evidence
for their grammaticality: ceteris paribus, one chooses the simpler
explanation.
 Acidity is a technical term within a chemical theory, hence it's
the theory that tells us what's acid or not; schizophrenia is a technical
term within a theory of mental disorders; white dwarf is a technical term
within astronomical theory, etc. There is, of course, no arguing with a
claim as to what a theory MUST do; but why should linguists feel any more
obliged to explain what laymen like me say about grammar than astronomers
should about white dwarfs? If you want to try to explain the actual usage
of language, far be it from me to stop you. But it does seem to me that
one is going to have one's work cut out to produce a *grammatical* theory
that will account, say, for the absence of the following utterances in most
corpora:
 a) the rat the cat the dog chased bit died
 b) shut up, officer, or I'll knock your teeth down your throat
 c) he may have been being followed
 d) colorless green ideas sleep furiously
But do what you must.

ref: Fodor, J.A. 'The dogma that didn't bark (a fragment of a naturalized
epistemology)' Mind 100:201-20 (1991)


Kevin R. Gregg
Momoyama Gakuin University
(St. Andrew's University)
1-1 Manabino, Izumi
Osaka 594-1198 Japan
tel.no. 0725-54-3131 (ext. 3622)
fax. 0725-54-3202

Dear Linguists,
As Newmeyer (1996, Generative linguistics: A historical perspective)
points out there is little evidence that 'organizationally, American
linguistics (let alone linguistics worldwide, Lotfi) is controlled by
generative grammarians; infact, their influence is disproportionally
small' (p. 34). He mentions some evidence (like the fact that non-
generativists dominate linguistic programmes at many major American
universities, that generativists receive only a small minority of the
grants, that the Lingustic Society of America is far from being a
generativist-dominated organization) in support of this observation.
|
All right! Generativists are moderate both in power and finance. But
this does not necessarily mean that they are always liberal when/
wherever in power. I intend this posting to Linguist to be one on the
subject of how democratic generativist-dominated institutes are and to
what extent they observe (or should observe) democratic traditions of
the society at large in their controlling of the whole field of
generative linguistics.
|
Academic societies are not too dissimilar from other human societies we
are all members of. They need leaders, financial resources, organs to
control, ... and who knows, perhaps even an army to defend against
potential intruders and a police force to supress rebels! Interestingly
enough, academic societies are not typically on a better record than
many third-world societies in terms of the democratic distribution of
power among their members and the observation of human rights! We
consider it a right for any member of a human society to have a share
of power, and to enjoy the freedom of speech. Even far-right
politicians do not dare any more to say that people don't understand
what's right and what wrong. But we usually forget that as linguists
practicing a particular brand of linguistics, we are members of one and
the same academic society, and then as members we have our own rights
to defend.
|
Speaking of rights, I'm reminded of some potential dangers to the
rights of the citizens of Generativia -- the land of generativists!
I mention three of these (possible) rights below.
(a) All human languages are equal. Then speaking or studying a
 European language, esp. English, does not make you different from
 the rest. Other languages cannot be reduced to 'marginal cases' or
 mere supportive evidence for what you find about English.
(b) There is to be no cultural bias against non-European scientific
 cultures. We're still far from having a global culture of science
 in terms of styles of argumentation, analysis, and presentation.
 Doing science the way they do in (Western) European and
 American universities does not necessarily make you a better
 scientist.
(c) Dominant journals of the field represent the whole 'nation'.
 Publication criteria are not to be biased against minorities.
|
As a linguist with a Persian background, an Iranian academic culture,
and plenty of unorthodox generativist ideas, I have happened to be
exposed to what I can't help labelling cultural and academic pre-
judices. I invite Linguists to join a discussion about the political
status of the field, democratization of the science of language, and
academic rights within the field of (generative) linguistics.
I think it is specially important to draw a line (however imprecise and
blurred it may prove to be) between science, which perhaps cannot be
sacrificed at any price, and non-science (culture, and 'politics'
included) in this respect.
|
Just as a specific case to begin with, I think the field can afford an
electronic journal to accept a range of submissions, store them some-
where on the web, and put their abstracts on the list to seek peer
reviews and subscriber comments. This will definitely ease up
publication for minorities, though (and significantly enough) making
it more difficult for 'people in high places' to keep a firm grip on
academia in terms of what linguists think and invite others to think
about.
|
Best regards,
|
Ahmad R. Lotfi, Ph. D
Chair of English Dept.
Azad University, IRAN.

I'd like to weigh in with a different view about the (ir)relevance of
native speaker grammaticality judgments to generative grammar. There
is another way of evaluating a linguistic model (i.e., a grammar of a
language, not a theory of grammars) than to seek predictions of
grammaticality judgments.

This alternative is based on using the grammar to associate a
probability with any sentence it generates; there are many ways to do
this, some of them natural and straightforward to syntacticians,
others less so. The simplest use for that is to allow one to then
choose between grammars on the basis of an already-established data
set of some hefty size. Select the grammar which assigns a higher
probability to the corpus.

That's not really the smart way to use that probabilistic information,
though, in judging between two grammars. Rather, you really want to
establish some apriori relative probability of the two grammars,
independent of data; and then for each grammar, take the product of
the grammar's probability times the probability it assigns to the
corpus; and see which of the two scores higher on that composite
score.

What is an apriori probability for a grammar? Within generative
grammar, that's a well-known concept -- corresponding roughly that of
Chomsky's evaluation metric [EM]; to make this more explicit, we might
assign probability = exp ( -1 * EM (grammar)).

This notion was first formulated by Ray Solomonoff around 1957, after
reading Chomsky's 1956 Three Models.., and has become standard in the
computational field, in recent decades.(See e.g. Charniak's 1993 book
on Statistical Language Learning.), though the connection to chomskian
evaluation metric is not. See e.g., The discovery of algorithmic
probability, R.J. Solomonoff, in Computational Learning Theory,
ed. Paul Vitanyi, Springer, 1995, esp. pages 9-10.

Ideas such as these have been codified as well in what's known as
Minimum Description Length (J. Rissanen's term).

Using grammars to predict speaker's grammaticality judgments is a poor
use for linguistic theory. There are better ways to do it.

John Goldsmith