Extending a lexicalist functional grammar through speech acts, constructions and conversational software agents

Abstract

This paper advances a model of conversational agents in a computational framework that builds on the notion of speech act performatives from discourse within a functional model of grammar. We describe the language specific elements of the intelligent conversational agents paradigm and how it can be usefully employed in the modelling of human language in software through use of agent-embedded speech acts. The linguistic model employed is Role and Reference Grammar (RRG). We propose a view in which a conversational agent has ‘internal’ and ‘external’ models to support the speech acts. The internal model of the agent is concerned with the internal state of the agent, based upon the intersection at any given time on the agent’s internal beliefs, desires, and intentions, known as BDI states. The external model of the agent is composed of an interaction model with its world (human and other agent). Importantly, the conversational agent also has a language model in software that is related to its interaction model to support bi-directional communication in human language through speech acts. For this, we employ the RRG model to motivate the design of the language model and use the RRG bi-directional linking system. We present a framework that connects the intelligent conversational agents paradigm to the RRG model of language. This has significance in that it has potential for use with linguistically oriented ontological semantics modelling and as a framework for testing of hypotheses on languages to support claims of adequacy within a functional approach. It also extends the breadth of computational work within RRG.