Abstract for Alain Rouveret's talk

The content of resumptive pronouns
Alain Rouveret (Paris)
February 24, 2005 (Thursday)
4:15 PM - ; Room 6417, the CUNY Graduate Center

Basing their analysis on English, Lebanese Arabic (L.A.) and Chinese, Aoun and Li (2003) argue that three strategies are necessary and sufficient to derive relative clause constructions:
(1) a. Promotion analysis, also labelled Head-raising analysis,
b. Matching analysis, i.e. wh-operator movement analysis,
c. Direct base-generation (no-movement) analysis.
The diagnostic properties which support the divide in (1) are reconstruction effects and locality effects. The constructions derived via (1a) display the full range of reconstruction effects.. Neither the relatives involving operator movement nor the base-generated ones exhibit any reconstruction effects. As for locality effects, they are observable in the relatives whose derivation involves movement, i.e. in those which illustrate (1a) or (1b), but not in the base-generated ones. In their approach, the gap / resumptive divide is orthogonal to the typology in (1). In L.A., reconstruction effects can be detected in definite resumptive relatives (provided that they do not occur in an island context), but not in definite relatives in island contexts nor in indefinite relatives. They argue that a distinction should be made between "true" resumption - the only strategy available in island contexts - and "apparent" resumption - which involves the presence of a movement chain linking the resumptive and the Head. The primary aim of this talk is to motivate a different conception of the articulation between resumption, reconstruction and locality, taking Celtic relatives as a paradigmatic case.
(2) The derivation of resumptive relatives constructions (RRCs) does not involve any movement of the resumptive pronoun or of a null operator in overt or in covert syntax (cf. Awbery 1977, Harlow 1981, McCloskey 1979, 1990, 2001, 2002, Rouveret 1994, 2002).
(3) The referential dependency between the resumptive element and the Head is established through a syntactic process, taking place in the narrow syntax, and does not result from an LF interpretive procedure.
Both claims have been challenged in the literature (cf. Boeckx 2003, McCloskey 1990). Part of the uncertainty surrounding the proper analysis of RRCs in part stems from their apparently contradictory behavior with respect to the standard diagnostic properties (which underlie (1)).
(4) a. Subjacency effects are observed in Welsh RRCs. No such effects can be detected in Irish RRCs.
b. Successive cylicity effects manifest themselves in Irish and Welsh RRCs.
c. Welsh RRCs display some reconstruction effects, but not the full range of reconstruction effects.
I intend to argue that this complex behavior can be made sense of once certain recent minimalist theoretical concepts are adopted: the operation Agree, which allows a syntactic connection to be established between two objects (more precisely between an uninterpretable feature that needs to be valued and an interpretable one that can value it) and which is a precondition on Move; the notion of Phase, whose primary motivation is to define the domains appropriate for the successive cyclic displacement of expressions and which is involved in the Phase Impenetrability Condition, which imposes a strict cyclicity condition on syntactic operations,.
Once these notions are adopted, certain characteristics, previously taken to be reasonable diagnostics that movement is involved in a derivation, can now be interpreted as reflecting constraints on the operation Agree itself. I will make the claim in (5) and show that, combined with (6), it opens the way to an elegant account of the gap / resumptive divide in Celtic relatives and to an explanation of the morphosyntactic differentiation of complementizers in long-distance relativization structures in Irish and in Welsh.
(5) Agrees applies phase by phase in a cyclic fashion.
(6) The link between the resumptive pronoun and the relative periphery results from the operation Agree, triggered by the presence of "active", uninterpretable features on the relative C and on the pronoun.
If the establishment of resumptive and movement dependencies both involve Agreement with relative C, it is natural to assume that the availability of each strategy is directly linked to the availability of the appropriate complementizer in the numeration, i.e. of the bundle of features triggering Agreement, and that the microvariation between Welsh and Irish partly reflects differences in the feature composition of the relevant complementizers. This conclusion is trivially supported by the fact that Irish has a specialized C to introduce RRCs, while Welsh exclusively resorts to the declarative complementizer in the same contexts. It will be shown that in Irish, apparently unbounded resumptive dependencies are mediated by the establishment of a tense dependency, involving the V-v-T complex and the tensed marked complementizer at each cycle, while in Welsh, the mediating dependency specifically involves the verbal inflection, which is pronominal in nature, and the complementizers, which are endowed with phi-features. In conclusion, successive cyclicity effects should not be viewed as an exclusive property of the structures derived by movement.
In the second part of the talk, reconstruction phenomena in RRCs will be considered and will be shown to be perfectly compatible with a non-movement analysis of these structures. The key observation is the one in (4c). Some reconstruction effects can be detected in RRCs, but not the full range of reconstruction effects. More precisely:
(7) a. Principle C effects are obviated in RRCs.
b. Reconstruction effects occur in pronominal binding constructions (where a quantifier internal to the relative takes scope over a pronoun contained in the nominal antecedent) and in anaphoric binding constructions.
As far as movement structures are concerned, Chomsky (1993) observes that the notion of reconstruction is redundant in a theory in which the trace left by a displacement operation is a copy of the moved element, but that, at the same time, it is necessary to introduce a Preference Principle to account for the asymmetric reconstruction behavior of anaphoric relations and of the structures falling under Principle C. In the analysis of reconstruction effects in resumptive constructions much depends on what we take resumptive pronouns to be. Following many semanticists (cf. Heim & Kratzer 1998, Elbourne 2001), as well as Frieidin & Vergnaud (2001), I will assume that when we see a definite pronoun on the surface, we are actually dealing with a definite description. More precisely (Freidin & Vergnaud 2001):
(8) Underlying representation of definite pronouns [DP [+ def] phi NP ] with the agreement features of the nominal expression and NP the silent NP part component.
To account for the asymmetry in (7), I will propose that a Preference Principle is also at work in resumptive constructions:
(9) Preference Principle (for resumptive structures) Given Pron = [DP [+ def] phi NP ] a definite pronoun functioning as a resumptive, only the [+ def]- and phi-components of Pron (i.e. D) and the N head of the NP-component are syntactically active at LF for interpretive purposes. But (9) only holds up to convergence. It holds when Principle C is concerned. But it is overriden when anaphoric binding or pronominal binding impose the representation of the antecedent in the internal position. In the latter cases, the NP-component of the definite pronominal expression is represented at LF. This leaves us with the two options in (10): (10) a. pron = [DP D [NP N DP PP ]] b. pron = [DP D N ] (10b) is relevant for (8a) and (10a) for (8b). It is important to determine the exact status of (9). An interesting possibility is that (9) reflects a preference for the late insertion of nominal complements (their late adjunction to the N head), along the lines of what Lebeaux (1988) and Chomsky (1993) have proposed to account for the argument/adjunct clause asymmetry with respect to reconstruction. This of course supposes that all nominal complements have an adjunct status. Time permitting, we will examine some of the difficulties raised by more complex reconstruction data (the interaction between principle C and pronominal binding discovered by Aoun, Choueiri & Hornstein (2001)).
If the Phasal Agreement analysis of RRCs is on the right track, successive cyclicity effects should not be viewed as an exclusive property of the structures derived by movement. As for the reconstruction effects detected in these constructions, they can be dealt with within a non-movement analysis, once the content of pronouns is made explicit. More generally, it appears that the correlation between movement and reconstruction should be loosened, contrary to what (1) presupposes