[Corpora-List] Looking for linguistic principles

[John Goldsmith]

Hi Rob, I was responding to the question about the history of the subject -
I think that was what the original query concerned: what had been said (by
Halle, by Chomsky), and whether those early arguments had been answered or
simply ignored.

So the first thing to be clear on is that nobody argued that structuralist
distributional techniques led to inconsistent results. The Hallean argument
against the phoneme took the form: if you apply American structuralist
principles, then you are faced with the conclusion that the same
phonological rule will sometimes be labeled a "rule of allophony"
(regulating the allophones by which a phoneme is realized) and sometimes a
rule at a higher level, a rule of morphophonology. As Anderson is at pains
to point out, the fact that data exists showing this was not new; Bloch had
famously argued the same point a number of years earlier. What was new was
the *conclusion* that Halle drew from the facts--a conclusion that was part
of a program that he was beginning to enunciate with Chomsky.

In reading the literature written in the last thirty years concerning the
disputes (especially between the structuralists and the generativists), one
must bear in mind that how the dispute is described is very different
depending on whether the writer identifies with the winner (the
generativists) or the losers (the structuralists). The generativists were
especially interested in defeating the structuralists, and so they invested
intellectual effort in succinctly summarizing what they took to be the
*essential* points in the structuralist program, and then arguing that at
least some of those points were (simply) false. The structuralists, from
their point of view, generally took the points not to be essential, but to
be working points, hypotheses which, like any others, would be modified
minimally--but as necessary--to maintain a theory that was adequate. In
addition, structuralists disagreed over a wide range of issues; the two
leading theorists (Zellig Harris and Charles Hockett) took very different
views with regard to the question (for example) as to whether linguists'
grammars should be understood a scientific models of what goes on in a
speaker's nervous system. Harris said No; Hockett said Yes. Chomsky said the
structuralists in general said No, and in that he was, in my opinion, simply
wrong but wrong on purpose; and that purpose was what I just mentioned: he
was invested in a program of defeating structuralism as he saw it, not
improving it.

We live in a very different world now, phonologically speaking; the problems
that generative phonology placed at the center of the world of phonology
were largely problems that the structuralists called "morphophonology", and
they did not investigate those problems with the same energy, rigor, or
depth that they did the problems of (their) "phonology". 

If the discussion is going to turn from one of history to one debating where
we should go now, that is indeed a different question; and in my opinion,
machine learning has advanced enough that it can play a serious role in
understanding phonology better. However--and this is a serious however--
just as phonologists need to learn machine learning seriously so that they
can use it and evaluate it, so too do computational phonologists need to
understand what successes phonological theory has had, where those successes
are genuine and where they are less substantial and more window-dressing. 

Yes, I do think an evaluation metric is critical for all theories of
phonology, and I think that something along the lines of Minimum Description
Length analysis is the right direction to go: the two terms of an MDL
objective function correspond to grammar complexity and goodness of
grammar-to-data fit, two notions that should be critical to any linguistic
scientist --

Best, John   

-----Original Message-----
From: Rob Freeman [mailto:lists at chaoticlanguage.com] 
Sent: Saturday, October 15, 2005 12:25 AM
To: John Goldsmith
Cc: 'Stefan Bordag'; CORPORA at UIB.NO
Subject: Re: [Corpora-List] Looking for linguistic principles

Hi John,

Thanks for the refs. but are any of these a solution to the problem (...that

the application of distributional methods to language was shown a long time 
ago to give us inconsistent results)?

Anderson says:

"The persuasiveness of Halle's original argument really rests crucially on 
one's willingness to take seriously the need to get rules right." (p.g. 15)

It may do, but given a desire to get rules right, distributional methods
seem 
to come unstuck. Can we use them or not?

Perhaps you are presenting generative phonology as the pattern for a
solution. 
But generative phonology doesn't deny the inconsistent results problem. It 
accepts there are multiple inconsistent results and seeks to find a 
"evaluation metric" which can be used to select between them.

Do you think this "evaluation metric" is the solution, and something all 
distributional methods for finding grammar should use?

-Rob

On Friday 14 October 2005 22:34, John Goldsmith wrote:
> ...
>
> The best discussion of the content, background, and impact of Halle's
> argument is to be found in Stephen Anderson's paper (
> http://bloch.ling.yale.edu/Public/Royaumont.pdf).
>
> I have a detailed webpage -- from a course I did last year -- on the
> development of early generative phonology from its structuralist
> antecedents:
>
http://humfs1.uchicago.edu:16080/~jagoldsm/Webpage/Courses/HistoryOfPhonolo
>gy/index.htm
>
> There is a discussion of Harris's views in my paper in the current issue
of
> Language (available also at
>
http://humfs1.uchicago.edu:16080/~jagoldsm/Webpage/Courses/HistoryOfPhonolo
>gy/index.htm )
>
> And a brief overview of the history of this area in a paper by Bernard
Laks
> and myself, at
>
http://humfs1.uchicago.edu:16080/%7Ejagoldsm/Papers/GenerativePhonology.pdf
>
>
>
> The controversy you refer to did not speak to the question of
> distributional methods in phonology or elsewhere; that was a separate
> issue, and the perspective that Chomsky criticizes was his interpretation
> of Harris (inaccurate, in my view), and Harris took what other linguists
of
> the period (like Charles Hockett) thought was a wildly extreme position,
> though they recognized that he did it in part in order to see the
> consequences of adopting a strong methodological position.
>
> John Goldsmith