Semantic and phonological competition in the language production system

Abstract

At its most basic level, a model of language production must describe the processes involved in a real time mapping from a conceptual notion of what is to be said onto a well formed string of sounds that can then be communicated to others. In the tradition of looking at the distributions of speech errors as a window into the architecture of the language production system, investigations of experimentally elicited number agreement errors have provided a useful paradigm for gathering production data. Any time two words within an utterance vie for control of the form of a third word in that utterance, these two words can be said to be in competition for control of the third word. By investigating the factors governing this competition, we can infer constraints on the cognitive architecture of the language production system as a whole. The present dissertation presents five studies on experimentally elicited speech errors, specifically, errors of number agreement between subjects and verbs. Experiments one, two and three investigate the role of semantic factors in the agreement process by manipulating the animacy of the nouns within the complex subject, their degree of semantic overlap, and the plausibility of their relationship with the sentence predicate, respectively. Experiments four and five investigate the role of phonological factors by manipulating the phonological overlap and surface frequency of the nouns within the complex noun phrase. Results indicate that while semantic factors can readily influence the computation of agreement, phonological factors do not. However, only lexical level semantic information (animacy, semantic overlap) appears to play a role, sentence level semantics (plausibility) show no effects. Overall, results converge with previous work suggesting that the flow of information through the production system is incremental, and that there is minimal feedback between phonological processes and semantic or syntactic levels of processing. Results did diverge from previous work in that we propose that feedback may in fact be necessary between conceptual and lexical-semantic levels of processing. In addition, our results argue for an activation based model of the agreement process itself.