from a review...

[Stephen M. Wechsler]

At 2:13 AM +0100 4/11/02, G Tsoulas wrote:
>It should perhaps be pointed out that Chomsky actually said that it
>would be pointless to identify the set of sentences with convergent
>derivations with the set of `grammatical' or `well formed' sentences,
>simply because the latter are not well defined classes (see for example
>the discussion around the notion "degrees of grammaticalness" in Aspects)
>. Now, does anyone really disagree with this?  Personnally at least, I
>really
>don't see what's that got to do with the `rules' of scientific inquiry,
>evaluation of theoretical proposals (well, ok, but see below), or what
>Paul Postal accepts as counterexamples to his theory.  After all, what
>exactly is the formal characterisation of the notion 'grammatical
>sentence' beyond the one that says that  it is a sentence generated by the
>grammar (has a convergent derivation) without being circular ?  It is
>undeniable that we may talk of sentences as being  readily
>accepted by native speakers and so on, but there is a host of
>ancillary assumptions/idealisations attached to this (informal) idiom
>(ideal speaker/hearer, homogeneous speech communities etc.) with which,
>again, I have no problem though I am not so  sure that they are
>accepted by many of the readers of this list, witness Ginzburg and
>Sag(2000)

Right.  Another point Chomsky has made is that there are other forms
of evidence besides native speaker intuitions that could and should
be brought to bear in judging the adequacy of a grammatical model
(psycholinguistic evidence; cross-linguistic comparison;
neurolinguistics, etc.).  That is presumably what he has in mind by
the phrase 'known empirical justification' when he writes, in the
"(in)famous fn 7":

`The concepts "well-formed" and "grammatical" remain without
characterization or known empirical justification...'  (Chomsky 1995,
p. 213 fn 7)

For example, in Chomsky 1986 (Knowledge of Language) he argues that
there could be good grounds for choosing between two extensionally
equivalent grammars, e.g. if one involves mechanisms that are better
supported by other forms of evidence besides native speaker
intuitions.

Makes sense to me... and, apparently, to others on this list who have
argued the pros and cons of various formal mechanisms (e.g. defaults)
on the basis of computational properties, i.a.

Montalbetti's citing fn 7 to justify ignoring speaker judgments is
mysterious, to say the least.  Maybe he was kidding. (?)

Steve