An approach to mental image based understanding of natural language: Focused on static and dynamic spatial relations

Abstract

It must be rather difficult for ordinary people to communicate with robots using special technical languages. Therefore, it must be more desirable for them to use natural language (NL) for such a purpose because it is the most conventional among them. This work proposes a methodology for natural language understanding through an AI system named Conversation Management System (CMS) based on Mental Image Directed Semantic Theory proposed by M. Yokota. CMS is intended to enable a robot to understand NL in the same way as people do, and actually can reach the most plausible semantic interpretation of an input text and return desirable outcomes by employing word concepts, postulates, and inference rules. Recently, the authors have applied several spatial terms in English language, for example verbs, prepositions (e.g. between, along, left, right, and so on). We found that the methodology is outstanding from conventional approaches with the attempt to provide robots understand NL based on mental image model. This paper focuses on how CMS understands static spatial (3D) relations expressed in NL.