Steven Frisson,1 Martin J. Pickering2 & Brian McElree3
1 University of Antwerp, 2 University of Edinburgh, 3 New York University
Many adjective-noun modifications involve simple predicate conjunctions in which the denotation of the adjective is intersected with the denotation of the noun. For example, a "strong fireman" is someone who is a fireman and who is strong (cf. Frazier, 1999; Kamp & Partee, 1995). This interpretation is called "intersective". However, this type of rule cannot describe the likely interpretation of many adjective-noun combinations. For example, a "strong applicant" is usually not interpreted as someone who is an applicant and who is strong, but rather as someone whose application looks strong. The preferred interpretation of these latter adjectives involves picking out a subset of a set denoted by the noun and is therefore called "subsective" (see also Jackendoff, 1997; Pustejovsky, 1991, 1995, for an explanation in terms of type shifting).
Here we examine whether adjective-noun interpretation is determined by a strategy of adopting the default (local) interpretation first or whether global contextual information determines which interpretation (be it intersective or subsective) is selected first. In an eye-tracking experiment, we contrasted adjective-noun modifications (e.g., "strong applicant") with subsective default interpretations following a global context that either supported the intersective (1a) or subsective (1b) interpretation. In (1b), the local and global information is congruent; whereas in (1a) they are incongruent. A third condition (1c) examined modifications with an intersective default interpretation with a congruent global context.
(1a) Even though he had lifted weights for years, the strong applicant was not impressive.
(1b) Even though he had the right qualifications, the strong applicant was not impressive.
(1c) Even though he had lifted weights for years, the strong fireman was not impressive.
The results showed difficulty during initial processing of (1a) as compared to (1b) and (1c), which did not differ. This indicates that the global context in (1a) did not completely determine the initial interpretation, and that it did not suppress the local interpretation of (1a). The data are consistent with a serial view in which the local interpretation is computed first with global information coming in later or with an alternative parallel view in which both analyses compete for the final interpretation.
References
Frazier, L. (1999). On Sentence Interpretation. Dordrecht: Kluwer Academic Publishers.
Jackendoff, R. (1997). The Architecture of the Language Faculty. Cambridge, MA: MIT Press.
Kamp, H., & Partee, B. H. (1995). Prototype theory and compositionality. Cognition, 129-191.
Pustejovsky, J. (1991). The generative lexicon. Computational Linguistics, 17, 409-411.
Pustejovsky, J. (1995). The Generative Lexicon. Cambridge, MA: MIT Press.