discussed in Chapter 12) raise similar problems of constituent incompatibility on the syntax–semantics side
too.) Again, the difficulty comes from mistakenly insisting on the unity of the sign. However, it seems
altogether feasible to adapt HPSG's basic formalism to the parallel architecture: phonological, syntactic, and
semantic combination could be run side by side, and kept connected not by being united in a single sign but
by somewhat looser interface components.- Section 6.2 mentioned that HPSG assumes a strict distinction between lexical and phrasal combination. The
reasons for this are the usual ones: the semiproductivepatternsmustbe listed in the lexicon, and therefore, in
theinterestsofa homogeneousmorphology,allmorphologyisdonethere.Againthereis noparticular reason
for maintaining thisassumption,given the presentanalysis. The proper division, I havesuggested, is between
semiproductive and irregular forms, which are listed, and productive forms, which are the result of free
combination but may be listed as well.An HPSG/CG versionof this approach has in fact been proposed by
Koenig and Jurafsky (1994).^95 - Finally, HPSG means“Head-Driven Phrase Structure Grammar.”“Head-driven” means that the basic
combinatorial principles in a phrase are determined by its head. However,theconstructions in section 6.6 are
precisely cases in which the argument structure of a verb phrase is driven not by the verb, but rather by the
lexical verb phrase that constitutes the constructional idiom. That is, these constructions violate the Head
Constraint of section 5.9. In order to adapt HPSG to these constructions, it is necessary therefore to slightly
weakenthe“head-driven”aspect ofthetheory, but ina waythatseems on thewholenatural. Such weakening
appears in Goldberg's (1995) approach to constructions, and to some extentin Sag's (1997) HPSG treatment
of relative clauses.
On the whole, although these proposed modifications to HPSG/CG are important, I think they are more cosmetic
than substantial. It should be an interesting challenge for the practitioners of these theories to work out the
consequences.
LEXICAL STORAGE VS. ONLINE CONSTRUCTION 195
(^95) A placewherethe“lexicalism”ofHPSG leads to results thatIfind counterintuitiveis in thetreatmentoftheclusters ofobjectcliticsin Romancelanguages, for examplethe
italicized items in (i).(i) French:
Mariele luidonne.
Marie it to hi mgives (‘Marie gives it to him.’)
Miller and Sag (1997) show how HPSG can treat these clitics as attached to the verb“in the lexicon,”so thatle lui donneis a lexical unit inserted into phrasal syntax. However,
some Slavic languages such as Serbian have similar clitic clusters which are not attached to the verb, but rather occupy an independent syntactic position in the sentence;
hence they cannot be treated as lexical addenda to the verb (Spencer 1991). More generally, Romance clitics are clear instances of free combination: we don't want to say
every verbis listed inthelexiconwitha full paradig mofcliticco mbinations. Inthepresentapproachtheyare thereforetreatedlike allother productivemorphology, perhaps
in the morphosyntactic tier.