Predictive analysis vs. segmentational analysis in sentence perception

Three experiments involving ongoing sentence perception are described. A nonword detection (NWD) latency procedure was employed without a concurrent task (Experiment I), with a concurrent comprehension task (Experiment II), and with a concurrent recall task (Experiment III). Predictions were made based on two plausible models of sentence perception--predictive analysis (PA) and segmentational analysis (SA). Both characterize the hearer as actively imposing a grammatical structure on the input. PA constructs a surface structure representation sequentially (i.e., on a word-by-word basis). SA partitions off clauselike units before establishing the structure of smaller within-clause constitutents. The NWD latency data generally support a PA system (particularly when a concurrent task is used). The results of a nongrammatical condition in Experiment III confirmed that the general findings were not artifactual.