status of words; HPSG and CG

[Carl Pollard]

Dear Dean Mellow,

I've always taken the strict lexicalism of HPSG to mean that

A. there are two disjoint kinds of signs, words and phrases

B. what the words are is determined independently of what the phrases
   are

C. the schemata that license/deduce/generate phrasal signs from other
   (lexical or phrasal) signs "know" whether the signs they operate on
   are lexical or phrasal but not how they were built up from other
   signs.

(I'm not necessarily endorsing these assertions, just telling what I
think the term "strict lexicalism" has generally been taken to mean in
the context of HPSG.)

Thus, A entails that if the same string is analyzed as both a word
(V say) and a phrase (VP say), then one can tell the two apart without
looking at their phonology or daughters (say, by whether the sort
is word or phrase, or by the value of a feature such as LEX).

B means, for example, that one can think (if one is so inclined) of the
lexicon as generated from a set of basic words by lexical rules that
are distinct from the phrase-generating schemata.

C means that a phrasal schema (e.g. the English head-complement
schema) has access to the fact that its head daughter is indeed a
word, but not to any sub-word morphemes from which it might have been
constructed, unless the mechanism that constructed it makes this
information indirectly available by stamping some feature value on
it. This qualification, of course, might reasonably be seen as vitiating
the empirical bite of strict lexicalism. Similarly, a description-level
(i.e. Meurers-style) lexical rule can tell that its "input" (= its
sole daughter) is indeed a word.

Note that C is rejected by categorial grammarians, in the sense that
once the lexicon is determined, the syntactic rules cannot distinguish
which of the signs it operates on are lexical and which are phrasal,
so that CG could not have a rule that expressly combines a lexical
head with phrasal complements.

Regards,

Carl