Context effects in coercion: Evidence from eye movements

Four eye-movement monitoring studies examined the processing of expressions argued to require enriched semantic composition (Pustejovsky, 1995). Previous research found that noun phrases denoting entities (e.g., the book) were diYcult to process following verbs that require event complements (e.g., begin). Expressions like began the book may be diYcult to process because they require complex operations to construct an event sense (e.g., began writing the book), they engender competition between alternative interpretations (cf. began reading the book), or they require a costly retrieval operation to recover a suitable activity (e.g., reading). Introducing the activity before a target expression did not eliminate the processing cost (Experiments 1 and 2), but introducing the entire event sense did (Experiments 3 and 4). These are incompatible with accounts that would attribute the observed cost to the retrieval or selection of an implicit activity in the event sense of the expression. They suggest that interpretation is costly when composition requires the online construction of a sense not lexically stored or available in the immediate discourse