The Syntactic Structure of French Auxiliaries Anne Abeillé and Danièle Godard While a consensus has been reached about the monoclausality of the Romance construction with an auxiliary verb and its verbal complement, questions remain about its syntactic structure. We focus here on French auxiliaries—the past tense auxiliaries (avoir and être), and the passive auxiliary (être)—which are unique in French in contributing only tense and aspect and triggering obligatory clitic climbing. Three syntactic structures have been proposed for such auxiliaries: a VP complement analysis, a verbal complex analysis, and a ‘flat’ VP analysis. We show here, working within a head-driven phrase structure grammar framework and basing our arguments on classical constituency tests, bounded dependencies, and lesser-known properties of a subset of manner adverbs, that the flat-structure analysis is to be preferred for tense auxiliaries, which take as their complements the bare participle as well as the complements subcategorized by this participle and ‘inherited’ from it. In contrast, the passive auxiliary, which we identify with the copula, has a predicative complement with different realizations: either an ordinary phrase, ‘saturated’ for its complements, or a ‘partial’ complement, where the predicative head lets some or all of its complements be inherited by the auxiliary. Our analysis allows for a solution to the well-known problem of auxiliary selection, which, we argue, should not be taken as an indicator of syntactic structure but is best handled via lexical constraints.* 1. Romance auxiliaries Auxiliary verbs have been extensively studied in Romance languages. A consensus has been reached about the monoclausality of the construction with an auxiliary verb and its verbal complement (Aissen & Perlmutter 1976/83, Davies & Rosen 1988, Fauconnier 1983, La Fauci 1994) but not about its syntactic structure. This is the case more generally for the set of complex predicates, to which auxiliary constructions belong. In the Romance languages, heads of complex predicates include restructuring verbs (absent from modern French), causative and perception verbs, and verbs with a predicative complement, in addition to auxiliaries. We focus here on French auxiliaries, which are unique in French in contributing only tense and aspect and triggering obligatory clitic climbing: they are the past tense auxiliaries (avoir and être), and the passive auxiliary (être).1 Three syntactic structures have been proposed for auxiliaries in the Romance languages, and French in particular: they have been analyzed as taking a VP complement (Guéron & Hoekstra 1988, Moore 1991 for complex predicates in general, Pollock 1989 for French auxiliaries), or forming a verbal complex (a subconstituent) with the participle (Emonds 1978 for French), or heading a ‘flat’ VP (Legaré & Rollin 1976 [End Page 404] for French). We show that the flat structure analysis should be preferred for tense auxiliaries, which take as complements the bare participle, as well as the complements subcategorized by this participle and ‘inherited’ from it. Our arguments are based on classical constituency tests, on bounded dependencies, and on lesser-known properties of a subset of manner adverbs. The passive auxiliary, though, which we identify with the copula (Couquaux 1979, Milner 1986), has a predicative complement with different realizations: either an ordinary phrase, ‘saturated’ for its complements, or a ‘partial’ complement, where the predicative head lets some or all of its complements be inherited by the auxiliary. This partial complement analysis makes sense of some generally overlooked cliticization and extraction facts. Our proposals are formalized in the framework of head-driven phrase structure grammar (HPSG, Pollard & Sag 1987, 1994), but they are largely independent from the chosen framework: It would be possible to implement other analyses of the French auxiliaries in HPSG, and conversely, our analysis could be implemented in other frameworks. Crucial to our treatment are the conception of lexical items as specifying precise syntactic and semantic information about their arguments, and the use of typed feature structures, where information can be shared between partial structures. This allows a straightforward representation of the sharing of subcategorization (argument inheritance), which has proved successful in the analysis of Italian restructuring verbs (Monachesi 1993, 1999), and for modal and auxiliary verbs in German (Hinrichs & Nakazawa 1994, Kiss 1992, Pollard 1996), Dutch (Bouma & van Noord 1998) and Korean (Chung 1998). But argument sharing per se...