Building an Autopoietic Knowledge Structure for Natural Language Conversational Agents

Abstract

This paper proposes a graph structure called an augmented semantic network (ASN) which is an extension of the ordinary semantic network (SN). We have developed an experimental conversational agent system and a knowledge structure based on the ASN that can hold a larger number of rules for replying to user utterances and modifying some parts of the rules when necessary. The system operation has shown that the knowledge structure is capable of implementing well-studied conversational models and becoming an autopoietic system. Autopoiesis means the systemic nature of life activity as discussed in the field of life science. An autopoietic system will be able to reproduce its elements as a result of their activity. Although the system will have to be capable of other functional natures to become an autopoietic system, the additional flexibility and extensibility enabled by the ASN-based knowledge structure might be necessary for realizing autopoietic and intelligent conversational agents.

The SN graph structure consists of a vertex set and an edge set whose elements each connect two elements in the vertex set. ASN edges are also able to connect elements in the edge set. The knowledge structure permits concept synthesis by utilizing the ASN’s edge modification ability. It removes the restriction on giving meanings from the outside to all concepts in the knowledge structure. Each element of rules represented by the ASN graph structure has a concrete meaning that can be synthesized by other elements of the rules. This capability might further the development of an autopoietic knowledge structure system.