Conversational Re-Prompting in Natural Language Dialog

Abstract

Natural language understanding technology has reached a level of sophistication where it can be profitably employed in interactive voice response systems in telephony. This paper describes a call routing application, where callers state a request in unconstrained, natural speech. The system then routes the call to the correct destination system or attendant. If there is a problem in understanding, then the caller must be re-prompted. This paper looks at two cases of re-prompting in the second turn of dialog based on the caller's response in the first turn: (a) when a caller's initial request is to speak to a real person instead of stating the reason for their call, and (b) when callers are too vague in their initial response. A strategy of conversational re-prompting is introduced which fits into the greater naturalness of the dialog, and, we show, increases the performance of the system in terms of successful fulfillment of user requests.