Constructing a lexicon of action

Tokunaga Takenobu (Tokyo Institute of Technology)

Okumura Manabu (Tokyo Institute of Technology)

Saito Suguru (Tokyo Institute of Technology)

Tanaka Hozumi (Tokyo Institute of Technology)

MMO2: Multimodal Lexicons & Corpora

This paper describes a Japanese speech dialogue system that enables a user to interact with agents in a virtual world and proposes a design framework for building a lexicon of action. This lexicon is used to realize the behavior of the agents in response to the user's commands. The lexicon has two levels - a macro and micro level. The system uses the macro-level lexicon, which is similar to a conventional plan library, to translate the user's goal to a sequence of basic movements. This process is the same as conventional planning with symbol manipulation. The micro-level lexicon is used to translate the basic movements into animation, which is represented by a sequence of avatar postures. We discuss how to define a set of basic movements and how to make these basic movements reusable.

Dialogue system, Virtual agent, Agent action, A lexicon of action

44.pdf