The On-Line Effects of Semantic Context on Syntactic Processing

This experiment tested the claim that on-line syntactic processing is autonomous and not affected by semantic context. Subjects heard sentence fragments containing syntactically ambiguous phrases, and the first clause of each fragment biased the listener toward one of its readings. Naming latencies to a visually presented probe word which was either an appropriate or inappropriate continuation of the sentence fragment were measured. Latencies were longer for inappropriate probes, suggesting that well before the clause boundary is reached, syntactic decisions can be influenced by prior semantic context. This result is predicted by an on-line interactive model of sentence processing. A basic issue in the study of sentence perception and comprehension concerns what Forster (1974) has termed the "autonomy of syntax"; that is, whether the syntactic structure of a sentence is computed without reference to its meaning. This concept of an independent syntactic processor is a crucial component of a major current model of sentence processing (Bever, Garrett, & Hurtig, 1973; Fodor, Bever, & Garrett, 1974; Forster, Note 1). According to this serial model, lowerlevel analyses (phonetic and lexical) are performed as the clause is heard and are used to construct preliminary hypotheses as to the syntactic structure of each clausal unit. These syntactic hypotheses cannot be affected by the meaning of prior clauses until the end of the clause, where a final deep structure representation is assigned and integrated with existing semantic hypotheses. When the syntactic deep structure of a clause is ambiguous, the model predicts that all syntactic structures are