Speech intention understanding based on decision tree learning

This paper proposes a method of speech intention understanding based on a spoken dialogue corpus to which the intention tags are given. The intention tag expresses the task-dependent intention of the speaker, and therefore, the proper understanding enables a spoken dialogue system to take appropriate actions. We have tagged about 35000 utterances in the CIAIR incar speech database. In our method, several decision trees for intention understanding are constructed. By constructing decision trees and using them at the same time, the strong amount of characteristic features related to intentions can be retrieved, and it can also be robustly coped with the diversity of the utterances. An experiment on inference of utterance intentions has shown 73.1% accuracy.