Reprompts as error handling strategy in human-agent-dialog? User responses to a system's display of non-understanding

摘要

In speech based technical systems, a `reprompt' can be deployed as a verbally non-explicit and semantically unspeciflc practice of making a failure-to-understand transparent. Users' repeats or rephrasings of their previous answers might lead to further non-understandings, resulting in further reprompts by the system. On the basis of a Wizard-of-Oz video corpus in a schedule management setting with an embodied conversational agent and the special user groups of elderly and mildly cognitively impaired persons, we investigate in a conversation analytical approach the interactional impact of three-fold reprompts on subsequent user actions to an appointment suggestion. We focus especially on the type of user actions during the course of multiple reprompts in a confirmation/disconfirmation context. Analysis reveals more fine-grained user response types, testifying that all users ratify the first reprompt. After the second and third one, users tend to either add problem manifestations or initiations of the relevant next move. Or they substitute their previous answer by these types of actions. While additional or substituting problem manifestations call for more specific and linguistically restricting error handling practices, the user-initiated next moves are technically exploitable as implicit cues for confirmation in the presented special yes/no-context.