[Corpora-List] Looking for linguistic principles

[Stefan Bordag]

Dear Rob,

On Sat, 15 Oct 2005, Rob Freeman wrote:

> On Saturday 15 October 2005 16:59, Stefan Bordag wrote:
>> ... I unfortunately have not read enough about
>> phonology to make literate comments on that part of the debate that went
>> on there.
>
> OK. I think you should take a look at it, though. I see no reason to believe

As I said, that's what I am doing now. I am also checking with outher 
sources and it takes some time.

> the problem is limited to phonological representations. It strikes me as the
> single most important issue in distributional methods then and now.
>
>> ...I guess that the potential of clustering and
>> this contrastive method of comparisons (which are really independent of the
>> language level used) is what Chomsky didn't understand, although this
>> sounds almost unlikely.
>
> I agree. I think Chomsky understood it good.
>
> He understood it, but found it led to rules which lacked generality.

It more and more occurs to me that what's missing is a precise notion of 
'generality'. Currently I don't see any missing generality in the 
possibility to find any kinds of rules automatically. But perhaps I 
misunderstand something. I am also not sure, how clustering and the 
potential for abstract rule learning it bears might have looked like back 
in the 60s.

> Chomsky's conclusion was this lack of generality was fatal to the method. We
> don't have to draw that conclusion. Like Syd Lamb we can explain it as
> evidence language rules are not "linear", or other. But we gain nothing by
> ignoring the observation.

And here it's the problem of what was meant with linearity. As Roy Harris 
also discusses at length, Chomsky had a problem with Saussures statement 
that language obeys the principle of linearity. But Saussure did not mean 
'linear rules' by that! Even more, Sassures linearity was restricted to 
syntagmatic relations only, while the examples discussed in the phonology 
debate include mostly paradigmatic relations (which are based on 
syntagmatic relations again).
What generativists call the surface structure and what all their rules 
also aim to construct is a *linear representation* of any utterance. So 
in truth, the principle of linearity does not interfere with the 
generality of rules. In fact, since restricting search space, it 
even facilitates the learning of general rules, but that's a different 
topic.

> John Goldsmith seems to be focusing on the historical impact of the
> observation, pointing out that it was not so much that the problem forced us
> to accept rules, but rather that the problem only presented itself once you
> started to think about rules. But he is not disputing the fact that
> distributional methods were shown to result in rules which were not general.
> Beyond that his position seems to be the classical generativist one that this
> is evidence everything comes down some innate language facility which selects
> between possible representations. As far as I can tell. John can correct me
> if I am wrong.
>
> What is in "The language instinct debate"? Once again does it dispute the
> observation of a lack of generality in distributionally derived
> representations, or only the innatist conclusions Chomsky drew from it?

I don't know, it has been suggested to me and I ordered it yesterday. But 
according to the description on amazon.com it seems to invalidate all the 
points made against structuralism with lots of evidence.

Best regards,
Stefan Bordag
p.s.: I am not entirely sure whether a discussion like this should be 
posted in this list (or each message thereof). Perhaps we should continue 
it privately and at the end post a summary to keep traffic low for the 
participants.

-- 
---------------------------------------------------------------------
- Bordag Stefan, sbordag at informatik.uni-leipzig.de                  -
- Institut fuer Informatik, Abt. Automatische Sprachverarbeitung    -
- Universitaet Leipzig                                              -
---------------------------------------------------------------------