Language comprehension consists of several operations ranging from the identification of individual words to the construction of a suitable interpretation for a complete text or utterance. Although much progress has been made in understanding different facets of this system, little is known about semantic composition, that is, how a contextually suitable interpretation for a sentential expression is derived from the products of lexical and syntactic analyses. Compositional processes provide the crucial interface between lexical and syntactic processing on one hand and discourse and text comprehension on the other. The goal of this project is to advance our understanding of this critical interface by identifying the essential processes and knowledge structures used in composition.
A traditional view of composition holds that the lexical representations of sentential constituents are simply combined in a manner that is informed by syntactic structure. However, recent formal analyses of common and seemingly simple expressions suggest that a substantially more complex mechanism is needed to compute contextually appropriate interpretations. Compositional processes appear to enrich the meaning of expressions by modifying default interpretations of individual constituents (sense extension). Often, this is accomplished by generating semantic structure not explicitly represented in the sentence or discourse. Recent studies of on-line sentence processing have provided behavioral evidence in support of this position, demonstrating that expressions argued to require an enriched form of composition are costly to process (McElree, Traxler, Pickering, Seely, & Jackendoff, 2001; McElree, Pylkkänen, Pickering, & Traxler, submiited; Pickering, McElree, & Traxler, 2004; Traxler, McElree, Williams, & Pickering, submitted).
To better understand compositional operations, experiments are designed to behaviorally dissociate simple and enriched forms of composition, and to identify the properties that distinguish different forms of composition. On-line sentence processing measures (eye-movement patterns, self-paced reading times, and time-course functions derived from the speed-accuracy tradeoff procedure) are used to contrast the processing of expressions hypothesized to depend on different types of information and to engage different kinds of compositional operations. These experiments seek to identify the knowledge structures used in composition, and to determine whether some compositional processes are more modular than others, because they draw upon local lexical properties rather than world knowledge. Experiments examine how local (phrase-internal) and global constraints affect composition, and whether there are aspects of the compositional process that operate in a context-insensitive manner.
Broader Impact. This research is expected to provide insight crucial for the development of an articulated model of composition, one capable of elucidating how novel, creative interpretations are computed from relatively static lexical representations and how a contextually relevant interpretation is formed from words with multiple yet related senses.