Machine Learning for Shallow Interpretation of User Utterances in Spoken Dialogue Systems

We investigate to what extent automatic learning techniques can be used for shallow interpretation of user utterances in spoken dialogue systems. This task involves dialogue act classification, shallow understanding and problem detection simultaneously. For this purpose we train both a rule-induction and a memory-based learning algorithm on a large set of surface features obtained by affordable means from an annotated corpus of human-machine dialogues. Using a pseudo-exhaustive search, the parameters of both algorithms are optimized. The shallow interpretation task turns out to be a difficult one, partly since there are 94 types of user answers. The best overall accuracy (exact match) obtained was 73.5%, which is a significant improvement over the baseline. The best average precision and recall for dialogue act classification was 91.2%, for classifying slot types 86.8% and for detecting communication problems 91.0%.