Modelling possessor constructions in LFG: Hungarian and English
John Payne, Erika Chisarik
University of Manchester, UK
ABSTRACT Hungarian noun phrases have two structurally distinct possessor positions, as illustrated in [1a,b]: [ 1] a.a kiraly lany-a ART king.NOM daughter-POSS[3S] b. The possessor construction in [1a], standardly called the nominative possessor construction, has the possessor in a form unmarked by any postnominal case suffix and directly preceding the possessum. By contrast, the possessor construction in [1b], standardly called the dative possessor construction, has the possessor marked with the dative case suffix and followed immediately by an article. In both constructions, the possessum agrees with the possessor in person and number. The two constructions are incompatible: it is impossible to have both a nominative and a dative possessor in the same noun phrase. As shown by Laczko (e.g. 1995), action nominalizations require the theme argument to be expressed either as the nominative or dative possessor, and the agent as a participial modifier. On the basis of their incompatibility and similarity in function, existing analyses (Szabolcsi 1994, E.Kiss 1999 etc.) treat the two constructions as related by movement. The LFG analysis of Laczko naturally eschews movement, but treats the two possessor positions as functionally equivalent: it is then natural to account for the incompatibility of the two constructions using the principle of coherence (e.g. Bresnan 2000). One problem not addressed by such approaches, however, is the fact that the two possessor positions are not semantically equivalent. In particular, we demonstrate that the range of semantic relations expressible by the dative possessor constuction is a subset of the semantic relations expressed by the nominative possessor construction. For example, excluded from the dative possessor construction are semantic relations such as quality and apposition: [2] a. b. [3] a. b. The situation is somewhat analogous to that in English, where similar relationships are excluded from the s-genitive construction: *happiness’s minutes/*Budapest’s city. The major difference is that the s-genitive and of constructions are not mutually exclusive. In order to model the mappings involved, we propose a principle that noun phrases permit two [–r] argument functions: SUBJ and NCOMP. The SUBJ function parallels the SUBJ function in clauses, but NCOMP is special to noun phrases only and distinct from the sentential function OBJ, not least in the range of semantic roles it encompasses. In Hungarian, the dative possessor maps to SUBJ and the nominative possessor to NCOMP. Similarly, in English, the s-genitive maps to SUBJ and most of-phrases to NCOMP. Given a thematic hierarchy for noun phrase arguments analogous to, but distinct from that required for clauses, the NCOMP function is a suitable mapping for the widest set of semantic roles, while the SUBJ function is specialised to higher roles. In English, as in Hungarian, a single [–r] argument of an appropriately high role can map either to NCOMP or to SUBJ. When two [–r] arguments are present in English, there is an asymmetry in that the higher role must map to SUBJ and the lower to NCOMP. This asymmetry does not arise in Hungarian, where we argue that two [-r] arguments are blocked by a principle akin to the asymmetrical object constraint of Bresnan and Moshi (1990). References Bresnan, Joan 2000. Lexical-Functional Syntax. Oxford:Oxford University Press. |