Tests of a holistic chunking model of sentence memory through analyses of noun intrusions

Abstract

A holistic chunking model of sentence acquisition and retrieval is described and tested by a prompted sentence recall procedure. In this procedure, subjects first study a list of unrelated sentences and later receive single-word prompts to cue sentence recall. The model assumes that (1) words in sentences are grouped into propositions during acquisition, (2) the propositions are encoded holistically and later retrieved as units, and (3) the retrieval of one proposition does not automatically lead to recovery of other propositions in a sentence. The model was tested by patterns of intrusion errors. Noun intrusions for elements within a recovered proposition were always related conceptually to the presented nouns, even when a noun violated the co-occurrence restrictions of the verb (e.g., the tray loved the house). In contrast, noun intrusions for elements outside of the scope of a recovered proposition were often unrelated to presented nouns. It was argued that patterns of intrusion errors provide more appropriate tests for sentence structure than do quantitative patterns of correct recall, at least from the framework of the holistic chunking model.