How is the memory search of a two-clause complex sentence in immediate memory carried out?

Abstract

An item recognition task was performed with 32 native English speaking, adult, right-handed subjects who listened to eight two-clause complex sentences presented to the left ear, each immediately followed by a probe word presented to the right ear. The subjects indicated whether or not the probe word occurred in the sentence and their recognition latency was measured. An analysis of variance was performed on recognition latency as a function of the three independent variables: (a) the serial position of the target word, early or late, within (b) a main or subordinate clause, in (c) initial or final clause position. The findings of this experiment were: (a) a word in the final clause is recognized significantly faster than a word in the initial clause; (b) for subordinate clauses, subjects take longer to respond to a target word occurring late in the clause than to a target word occurring early in the clause; for main clauses, subjects take longer to respond to a target word occurring early in the clause than to a target word occurring late in the clause. Present storage models of sentence processing and memory search models are inadequate to account for all the data. A combined storage-search account was proposed. A serial self-terminating model of clause accessing, with final clause search occurring prior to initial clause search, fit the data better than a simultaneous search of both clauses. Clauses are searched either in a primary or a secondary buffer, depending on clause type (main or subordinate) and clause position (initial or final) in the sentence. To explain the difference in mode of search between main and subordinate clauses, it was suggested that main clauses exhibit a property of primacy over subordinate clauses.