Conceptual Effects of Audience Design in Human–Computer and Human–Human Dialogue

ABSTRACT

Speakers in conversation often tailor referring expressions in ways that reflect beliefs about the knowledge and perspectives of specific addressees. In this study we examined whether such audience design can influence speakers’ conceptualizations of referents beyond the conversation. In a Wizard-of-Oz paradigm, participants identified novel images either for an ostensible computer dialogue application or for a human partner as part of a referential communication task. Subsequently, they independently sorted the task images into discrete groups. Consistent with audience design, when interacting with the computer a greater proportion of participants’ descriptions in the dialogue task focused on literal, geometric features compared with descriptions for the human partner. Importantly, participants’ postdialogue sorts were also more shape-based after human–computer interaction. Different ways of attending to objects for communicative purposes can affect how speakers construe objects for themselves.

Acknowledgements

We thank Darren Gergle and Gregory Ward for their valuable input on this project and three anonymous reviewers for their helpful feedback on an earlier version of this work. We are also grateful for the critical contributions of Shirley Roitberg and Lola Less, who served as experimental confederate in our Wizard-of-Oz paradigm.

Notes

¹ A similar logistic mixed model using purely analogical descriptions as the outcome measure revealed an effect of Partner (b = −6.42, p < .03), with more analogical descriptions with the human partner (M = 0.18, SD = 0.39) than the computer partner (M = 0.08, SD = 0.26) and a marginal effect of Feedback (b = −5.66, p = .051), with more analogical descriptions after minimal feedback (M = 0.18, SD = 0.39) than conceptual feedback (M = 0.10, SD = 0.30). However, these two factors did not interact, and there were no effects involving Trial. Thus, the main effects for this analysis mostly mirror those for purely geometric descriptions.

² Note that the converse is not necessarily true. That is, we cannot definitively say that less similarity to the baseline sorts means that participants were spontaneously considering analogical information. In such cases we can only say that participants were less likely to be organizing their image groupings around geometric features.