Distinguishing Generation and Selection of Modifier Attachments: Implications for Lexicalized Parsing

We investigated the lexicalization of syntactic knowledge in two reading experiments, one using a moving window paradigm and one measuring eye movements. Our study was motivated by Spivey-Knowlton and Sedivy’s (1995) finding that verb class, and by implication frequency information, determined how easily a prepositional phrase (PP) was attached to a verb phrase (VP) or to a direct object noun phrase (NP). Because they found longer reading times for NP-attached PPs than for VP-attached PPs following action verbs, and the reverse pattern for psych/perception verbs, their data seem to demonstrate that adjunct slots and their frequencies are encoded in the lexical entry of the verb. Such detail is consistent with a highly lexicalized parser, in which both argument and adjunct slots are represented in the lexical entry of each phrasal head. However, we came to different conclusions after using stimuli that were very similar to those used in Spivey-Knowlton and Sedivy, but that compared temporarily ambiguous sentences with unambiguous versions. We essentially replicated the Spivey-Knowlton and Sedivy results using temporarily ambiguous sentences, though the results were not robust in the item analyses. In contrast, with unambiguous versions, we found lexical frequency effects for argument PPs, but not for adjunct PPs. These findings indicate that the lexicalization of syntactic knowledge is limited to argument structure. More generally, this work offers additional support for the distinction between structure generation and selection, and illustrates how factors differentially affecting those processes may be explored empirically.