Joe:
> And Rosta wrote:
> >
> > Joe:
> > > Very cool example and construction. What you're saying is actually
> > > another argument for accepting constructions.
> >
> > I don't see why it is either cool or, in Goldberg's very good
> > definition, a construction.
>
> One of the things that Goldberg neglects to point out (but something
> I mention in my thesis) is that conventionalization also makes for
> constructions.

How come?

> (1) What did you do it like that for?
>
> Even though I didn't turn up any "what... like that for" strings
> in the BNC, intuitively it does seem conventionalized. This is not
> to say that it does not have variants; e.g.:
>
> (2) What did you go and do it like that for?
>
> There is perhaps a more compelling reason for saying that (1) is
> a construction (or better yet, a prototypical variant of a somewhat
> more abstract construction). The reason has to do with the fact that
> WHAT... LIKE THAT FOR? structures have their own specific pragmatics.
> That is, when someone says (1) to you, it's not a straightforward
> question, but a reproach.

I'm not convinced of this particular example, but I'll grant that they
exist in principle. But all sentences carry with them faint memories
of their previous utterances, which influences the way they get
interpreted when re-uttered. I'm a bit unhappy about trying to build
this into the grammar.

> Thus, even on Goldberg's definition, we're
> dealing with a construction (or, probably more accurately, a specific
> variant of a construction): the meaning of the whole is not the mere
> sum of the parts.
>
> In any event, with regard to your reply to Jasper, I think you might
> be right that what is needed is special rule for WHAT... FOR? Or maybe
> we could posit a string that looked something like this:
>
> (3) What DO(aux) NP V (like that) for?
>
> This would more or less account for (1) as well as:
>
> (4) What did you hit him for?

I think you just need a special rule for FOR, stating that it describes
a purpose, that its complement is WHAT and must be extracted. It doesn't
merit being classed as a construction.

> > BTW, it is interesting to ask what, given Goldberg's definition, is
> > the difference between a construction and an idiom. I'd say that a
> > construction, if strictly compositional in form but not in meaning,
> > must involve a restriction on the compositional meaning, while an
> > idiom must be strictly compositional in form but can have an
> > utterly noncompositional meaning.
>
> It depends on what you mean by "utterly noncompositional meaning."
> Ray Gibbs has done some experiments (both off-line and on-line)
> that suggest that people do analyze the imagery conveyed by
> idiomatic expressions. (Of course, semantic compositionality is
> a matter of degree, so that would be another factor to take into
> account.)

What I mean is that in a construction the compositional meaning can be
added to, and hence restricted, by the construction, but not overridden.
There is no such constraint with idioms (on the relation between
idiomatic and compositional meaning).

--And.