When to Elicit Feedback in Dialogue: Towards a Model Based on the Information Needs of Speakers

Abstract

Abstract. Communicative feedback in dialogue is an important mechanism that helps interlocutors coordinate their interaction. Listeners pro-actively provide feedback when they think that it is important for the speaker to know their mental state, and speakers pro-actively seek listener feedback when they need information on whether a listener perceived, understood or accepted their message. This paper presents first steps towards a model for enabling attentive speaker agents to determine when to elicit feedback based on continuous assessment of their information needs about a user's listening state.

Design and evaluation of postural interactions between users and a listening virtual agent during a simulated job interview

“…Wang et al 41 proposed utilizing varied goals and roles to go "deeper" than just generic feedback. Buschmeier et al 42 attempted to determine the timing of feedback by considering "information need" of the speaker, which necessitates keeping track of the narrative. In the SEMAINE project 43 , recognition modules of emotional state and the user's words are added to guide the generation of specific responses.…”

Confidence in uncertainty: Error cost and commitment in early speech hypotheses

“…The early semantic information can be used for producing grounding signals [ 38 , 40 ]. Vice versa, processing the user’s actions during the artificial agent’s utterance enables the agent to estimate the user’s understanding [ 41 , 42 ], to adapt its utterance and to explicitly elicit user feedback if needed [ 43 – 45 ]. These accounts aim for integrating incremental hypotheses into a larger, multimodal model of intention recognition.…”