Proposed benefit-harm value based utterance learning model for human-machine communication control system

Abstract

To make control actions efficient and simple, kinds of human-machine communication control systems have been developed to replace manual control. However, traditional systems have two open problems on text processing part: first, output returned by traditional systems cannot cover enough key information restored in database; second, they cannot learn utterance of manipulator automatically, which may cause misunderstanding of manipulator's order. Inspired from biology, this paper proposes a conception of benefit-harm value. And with it we design a novel system whose output covers more possible key information and that is able to learn utterance of manipulator through training. In experiments we test how many necessary keywords the outputs of traditional system and our system can cover respectively. Finally we ask volunteers to give scores to both systems for the sake of demonstrating satisfactions to their utterances.