Tom Williams and Rafael C.Nunez and Gordon Briggs and Matthias Scheutz and Kamal Premaratne and Manohar N. Murthi
Ibero-American Conference in Artificial Intelligence
Understanding Indirect Speech Acts (ISAs) is an integral function of human understanding of natural language. Recent attempts at understanding ISAs have used rule-based approaches to map utterances to deep semantics. While these approaches have been successful in handling a wide range of ISAs, they do not take into account the uncertainty associated with the utterance's context, or the utterance itself. We present a new approach for understanding ISAs using the Dempster-Shafer theory of evidence and show how this approach increases the robustness of ISA inference by (1) accounting for uncertain implication rules and context, (2) fluidly adapting rules given new information, and (3) enabling better modeling of the beliefs of other agents.