The Late Assignment of Syntax Theory (LAST) proposes that a preliminary analysis of statistically valid information elicits an initial meaning/form hypothesis (Townsend & Bever, 2001). The initial meaning/form hypothesis consists of probable thematic roles that follow from low-level cues-based segmentation of words into phrases. The grammar uses this hypothesis to synthesize a detailed syntactic structure that comprehension system compares with a memory representation of speech. We report on two lines of research that support LAST.
LAST proposes that semantic properties of verbs can influence the initial meaning/form hypothesis. Telic events (e.g., drink a lemonade) require an endpoint in time. A verb internal delimiting argument such as a patient can signal this endpoint (Dowty, 1979; Jackendoff, 1997; Pustejovsky, 1991). In passive sentences, the delimiting phrase appears before the verb. Atelic events refer to events that are on-going, without any specific endpoint in time (e.g., drink lemonade). Thus, atelic events do not require a delimiting event. We presented four pictures on a screen either simultaneously with the verb or two seconds after the sentence, and asked subjects to indicate the picture that goes with the sentence. With presentation of the test pictures simultaneously with the verb, accuracy was greater for telic passives than for telic actives, but for atelic events the reverse was true. With post-sentence presentation of the test pictures, telic passives were harder than telic actives, but there was no difference for atelic events. Thus, the passive voice facilitates the early stage of comprehension for telic passives. But just after the synthetic point telic passives have a more complex representation, and hence are harder than telic actives. We interpret this evidence to support the hypothesis that the semantic properties of verbs are a part of the initial representation that is formed during sentence comprehension.
LAST proposes that a detailed syntactic structure appears late during comprehension. Since speech contains explicit markers for wh-fillers, wh-gaps are filled immediately, but since there are no explicit markers for NP-fillers, assignment of antecedents to NP-traces occurs late. On-line monitoring for a change of speaker (Townsend & Bever, 1991) showed that response times for post-gap target words were faster for wh-movement than for NP-movement. Post-sentence reports, however, showed no difference in accuracy of identifying target words in sentences with wh- versus NP-movement. We interpret these results to show that a detailed syntax, consisting of assignment of antecedents to NP-traces, appears relatively late during comprehension.
References
Dowty, D. (1979). Word meaning and Montague grammar. Dordrecht: D. Reidel Pub. Co.
Jackendoff, R. (1997). The architecture of the language faculty. Cambridge, MA: The MIT Press.
Pustejovsky, J. (1991). The syntax of event structure. Cognition, 41, 47-81.
Townsend, D. J., & Bever, T. G. (1991). The use of higher-level constraints in monitoring for a change in speaker demonstrates functionally distinct levels of representation in discourse comprehension. Language and Cognitive Processes, 6, 49-77.
Townsend, D. J., & Bever, T. G. (2001). Sentence comprehension: The integration of habits and rules. Cambridge, MA: MIT Press.