Concept Type Prediction and Responsive Adaptation in a Dialogue System

摘要: Responsive adaptation in spoken dialog systems involves a change system behavior response to user or situation. In this paper we address responsive the automatic speech recognition (ASR) module of system. We hypothesize that information about content utterance may help improve for utterance. use two-step process test hypothesis: first, automatically predict task-relevant concept types likely be present using features from context and output first-pass ASR utterance; then, adapt ASR's language model predicted user's run second pass ASR. show that: (1) it is possible achieve high accuracy in determining presence absence particular post-confirmation (2) 2-pass with type classification can lead to improved performance utterances.