Effects of Conversational Pressures on Speech Planning

Abstract

In ordinary conversation, speakers experience pressures both to produce utterances suited to particular addressees and to do so with minimal delay. To document the impact of these conversational pressures, our experiment asked participants to produce brief utterances to describe visual displays. We complicated utterance planning by including tangram figures that prohibited easy lexicalization. Participants completed the task in either the presence or absence of an addressee and also under circumstances of natural or explicit time pressure. Results suggested that speakers produce richer utterances with addressees present but that they do so efficiently, without sacrificing planning time. We propose a good-enough view of the language production system: We suggest that, much like the comprehension system, speech planning processes flexibly adapt to external task goals.

Acknowledgments

This material is based on work supported by National Science Foundation Grant 0325188. We thank Bill Wenzel for his significant contributions to the collection and analysis of data. We also thank Susan Brennan, Arthur Samuel, and two anonymous reviewers for their helpful comments during the development of this project.

Notes

Matthew E. Jacovina is now a postdoctoral fellow at the School of Education and Social Policy and the Department of Psychology at Northwestern University.

¹ As pointed out by an anonymous reviewer, the claim that addressees elicit longer descriptions without sacrificing planning time (or, as we put it here, the claim that effects of addressee presence on initiation time effect do not arise in proportion to the effects of addressee on description length) can be evaluated statistically. We did so by including a covariate term in the linear mixed-effects models for both the description length measure (for which we included a mean-centered initiation time covariate term) and for the initiation time measure (for which we included a mean-centered description length covariate term). Neither model produced significant effects of the covariate terms, and including these terms did not result in significant chi-square differences in the fit of each model. To put it another way, initiation time did not predict description length and its presence as a covariate did not help explain the other effects. It was the same for initiation times: Description length did not predict initiation times, and its presence as a covariate did not explain the other effects.