Invited talk 5

Accounting for good enough conversational speech

Antje S. Meyer and Florian Hintz

Experimental psycholinguistics has mainly concerned the production of grammatically correct fluent utterances. Yet, as we illustrate, conversational speech is riddled with disfluencies, repairs, and grammatical and lexical errors, and so comprehensive models of speech planning should accept them as typical results of speech planning and account for them. Based on experimental work and corpus analyses, we argue that the typical disfluency of conversational speech arises from a combination of two factors, (1) the complexity of the cognitive task, in particular the requirement to combine speech planning and listening to the interlocutor, and (2) the fact that disfluent speech is usually good enough to attain the speaker’s communicative goals. Though this holds for all speakers, there may, in addition, be stable differences between speakers in their ability to produce conversational speech and hence in their conversational fluency. To assess this hypothesis, we analyzed brief monologues of a large sample of speakers, provided in two sessions, a month apart. We found good between-sessions correlations for most measures of fluency, suggesting a trait of conversational fluency. We then related indicators of fluency to performance indicators in a battery of tests of linguistic and domain-general skills. In spite of good test-retest reliability of the test battery, we only found weak relationships of domain-general skills (in particular working memory) with fluency indicators. Notably, measures of linguistic knowledge were not related to the fluency indicators. We discuss the theoretical implications of this work and its limitations. We propose that links between indicators of linguistic and domain-skills on the one hand and indicators of fluency of a person’s spontaneous speech are weak because, consistent with the requirements of everyday conversation, speakers focus on planning utterance content rather than aiming for fluency.