Multi-agent-based adaptive av interface

Abstract EN

In order to build adaptive interfaces, we need adaptive interaction and dialogue handling methods.

We have presented an advanced model for interaction and dialogue management to support adaptive natural language Audio Visual (AV) interface.

Our multi-agent-based Natural Language (NL) interface is a software application environment that breaks up NL interpretation into a community of collaborating, learning agents.

It allows users to control AV appliances in NL, rather than using remote control devices.

It learns and remembers the way a user does things, customizes its performance to match the user’s behavior.

This paper shows at first the basic feature of AV agent system, and then reports the implementation and experimentation for Japanese version, which connect multi-agent-based NL interface with actual appliances and Sound Recognition Engine (SRE).

By the experiment, our system works well; it provides an impressive degree of accuracy, measured as the percentage of requests that translate into the operation intended by the user.

But we consider that the miss recognition of SRE should be absorbed more by the multi-agent system to make this system easier and, comfortable to the users.

Therefore we propose an absorption theory by learning the habits of the SRE and the users, and then absorb the recognition errors of SRE after a time of training.