A statistical prediction model of speakers’ intentions using multi-level features in a goal-oriented dialog system

A dialog system is an intelligent program that helps users easily access information stored in a knowledge base by formulating requests in their natural language. A dialog system needs an intention prediction module for use as a preprocessor to reduce the search space of an automatic speech recognizer. To satisfy these needs, we propose a statistical model to predict speakers’ intentions. The proposed model represents a dialog history, with various levels of linguistic features. The proposed model predicts the user’s next intention by giving the linguistic features as inputs to a statistical machine learning model. In experiments conducted in a schedule management domain, the proposed model showed a higher average precision than the previous model.

► A statistical prediction model of speakers’ intentions using various levels of linguistic features. ► Multi-level features; morpheme-level, discourse-level, and domain-level features. ► Good usefulness; both reducing the search spaces and increasing the precisions of ASR systems.