And:
>> >7. One of the points in Culicover's book _Syntactic nuts_, which,
>> >most unusually, is a book I can recommend very highly to both
>> >formalisty and cognitivisty members of this list, is that many
>> >representational issues in syntax are not relevant to acquisition,
>> >because whatever the representation it can only be learnt on the
>> >basis of the observable facts.
>> ## I had a little chat with him in which he said something like this. I
>> find this a very odd view. The nature of what's represented in the
>> learner's mind surely is relevant to the question of how it's learned?

>I don't understand what you find odd. If representation R accounts for
>data facts A, B and C, but R can be learnt only on the basis of A, B
>and C having been observed (and recognized as data facts), then
>independently of arriving at R the learner already knows A, B and C.
## If A, B and C are observable, they must be outside the learner's mind;
so the question then is HOW the learner 'knows' them. I assume that in some
sense they must be represented (by representation R), so R, not A, B and C,
is the representation in the learner's mind. Suppose A-C are three
utterances of the word DOG. On the basis of these three experiences the
learner builds a representation of this word. The representations that we
construct in our descriptions are intended to be models of this
representation; and if you can show that it has characteristics that aren't
included in our models, these need to be revised until they fit exactly. Is
this different from your view?

>
>> >8. If your goal is to account for acquistion and to model what
>> >a speaker must know in order to be able to speak the language,
>> >then models that are entirely content with construction-specific
>> >rules, without seeking deeper patterns, are probably entirely
>> >adequate to the task: the WG/CG/HPSG-type model is like a
>> >particularly thorough-going version of a descriptive reference
>> >grammar.

>At any rate, my contention is that constructional approaches are less
>impelled to seek deeper patterns than nonconstructional approaches, because
>the constructional approach has the ability to describe specifics without
>seeking out generalities.
## I think you're talking about the social psychology of linguists here -
the PR of the schools, if you like. One of the attractions of construction
grammar is the possibility of handling details, but it's up to individual
linguists to choose whether to work on details or on broad pictures.
Likewise in Minimalism or HPSG - some people aim at the details, others aim
at global architecture. It's a matter of personality and what's socially
approved, not theory. Surely?

>
>But more particularly what I think that if your goal was to write a grammar
>sufficient for a computer (or a person) to use a language like a native
>speaker, then it's not necessary to seek out deeper patterns. Yet the
>deeper patterns are there, and they show up in our intuitions about
>data the like of which we have not encountered before (such as pied-piping).
## You mean because native speakers look for generalisations which may not
make any difference to their use? Yes, I agree. But this is a question of
how, in fact, the learner represents the facts. Just as I said above, it's
crucial for us linguists to know what actually goes on in people's heads,
as it's the only way to choose between alternative grammars.

>
>> My ideal is to find deep patterns which unite linguistic patterns with
>> non-linguistic ones.
>
>I don't think this has been one of the areas of WG's greatest success. I
>can only immediately think of the "raising in syntax, semantics and
>cognition" paper, whose arguments I was not very impressed by.
## My impression is that I've spent a lot of time thinking (and I thought
writing) about the similarities between linguistic and non-linguistic
thought. There's a whole chapter about it in English Word Grammar (Chapter
4), Jasper's and my paper about cycling is about it, and every time I talk
about default inheritance I point out that it's the same process that we
use in general thought. I may not have made a good case, but I have tried!

>
>I'm suggesting that a fluent command of a language can be gained
>without necessarily arriving at all possible generalizations,
## Yes, I'm sure that's true.

or at all
>generalizations that are necessary for accounting for our grammaticality
>intuitions.
## How can that be? It sounds like a contradiction: you can have intuitions
without the generalisations that are necessary for explaining them?

>I wonder what Joe will have to say about this. If storage is cheap,
>why should the learner-speaker who has achieved fluency still
>press on to seek a more optimal grammar? I can't see any functional
>pressure at work here.
## I can't answer for them, but for me the most important fact that drives
generalisation is a very basic drive to spot generalisations, which seems
to be built in (like curiosity - which may be the same thing). E.g. as I've
no doubt mentioned before, there's sociolinguistic evidence that we go on
spotting morphological similarities among words through our lives (e.g.
spotting that "kept" contains a suffix which sounds like the one in
"walked"), and someone must have spotted the similarity between "hamburger"
and "ham sandwich". The general drive to generalise probably has an
enormously high survival value - it allows us to generalise from past
experiences into the future - but the individual achievements of
generalisation don't necessarily do us any good at all. (E.g. who cares if
"adultery" contains "adult"? But it's a generalisation waiting to be made:
every word that starts "adult" is to do with grown-ups.)


Richard (= Dick) Hudson

Phonetics and Linguistics, University College London,
Gower Street, London WC1E  6BT.
+44(0)20 7679 3152; fax +44(0)20 7383 4108;
http://www.phon.ucl.ac.uk/home/dick/home.htm