Robust Understanding in a Dialogue System

摘要: This paper describes a system that leads us to believe in the feasibility of constructing natural spoken dialogue systems task-oriented domains. It specifically addresses issue robust interpretation speech presence recognition errors. Robustness is achieved by combination statistical error post-correction, syntactically- and semantically-driven parsing, extensive use context. We present an evaluation using time-to-completion quality final solution suggests most native speakers English can successfully with virtually no training.