Modelling of Non-Linguistic Utterances for Machine to Human Communication in Dialogue

抄録

Non-Linguistic Utterances (NLUs) present a potentially useful alternative communication channel between humans and machines. NLUs are potentially cheaper, and easier to implement, and not limited to the constraints of natural language and therefore may be appropriate in situations such as assisting tourists with various language backgrounds and needs. An experiment was done to establish ranges for NLU parameters such as pitch, duration, amplitude, and timbre. Subjects listened to randomly produced NLUs and selected applicable dialogue descriptors from: Positive, Negative, Greeting, Apology, Thanking, Hesitation, Question, Approval, Disapproval, Hushing, None of the Above. Factor analysis yielded 3 major factors, which were labeled as follows: 1) Affirmative vs Negative, 2) Questioning, and 3) Meaningful vs Indeterminate. NLUs with lower pitches, downward pitch patterns, and simpler timbres were found to be more Negative. Those with upward pitch patterns were more likely to be identified as a Question. In future work, experiments will be used to develop a model of NLU inference and interpretation within a Dialogue in terms of the dominant descriptors and to test the model in applications for tourist support systems.