feature structures vs. feature structure descriptions

[Lutz Gunkel]

At 03:14 30.01.99 -0500, Carl Pollard wrote:

>Why do you say the feature structures themselves form a language?
>This seems to me like saying first-order models form a language,
>for which first order logic is a metalanguage.

I do not say that feature structures in general form a language. But we
agree that certain feature structures serve as models or idealizations of
linguistic expressions of a given natural language L. Suppose now that
there is an (infinite) set L' of feature structures, each of which
modelling some linguistic expression of L. Then we can regard L' itself as
an (idealized) language, because it is just an idealization of some natural
language L. (In this - and only in this - sense I meant that feature
structures might form a language.).
A grammar of L' (G(L')), as I see it, is a theory of L'. The grammar G(L')
may be formalized by using some feature logic. The statements of the
grammar G(L') - formulas of the feature logic - form a metalanguage
relative to L'. L' itself in turn is the object language of G(L').

Lutz