Towards Artificial Communication Partners with a Multiagent Mind Model Based on Mental Image Directed Semantic Theory

In recent years, there have been developed various types of real or virtual robots as artificial communication partners. However, they are to play their roles according to programmed actions to stimuli and have not yet come to understand or imitate delicate mental functions of their human partners such as Kansei, one of the topics in this chapter. Kansei evaluates non-scientific matters such as art, music, natural scenery, etc. by Kansei words (e.g., ‘heartcalming’, ‘fantastic’, ‘grotesque’) (Fukuda et al, 1998; Sugita et al, 2003; Butz et al, 2005) and ‘Artificial Kansei’ namely ‘Kansei for robots’ is expected to play a part in ‘artificial or robotic individuality’ (Tauchi et al, 2006). The author has proposed a human mind model consisting of Stimulus, Knowledge, Emotion and Response Processing Agents, intending the intelligent system IMAGES-M (Yokota, 2005a; Shiraishi et al, 2005) to understand and imitate miscellaneous human mental functions involving emotion processing as well as knowledge processing originally aimed at. For example, Kansei is defined and realized as tight collaboration of Knowledge and Emotion Processing Agents. This multiagent mind model is much simpler than Minsky’s (Minsky, 1986) and its most remarkable feature is that the agents involved communicate with one another by exchanging and computing mental images represented in the formal language Lmd developed for integrated representation and computation of multimedia information (Yokota, 2005, 2006) while other multiagent systems were designed to employ special programming languages for inter-agent communication (e.g., Labrou et al, 1999; Vieira et al, 2007). IMAGES-M is originally intended for integrated multimedia understanding as knowledge processing for intuitive human-robot interaction such that may happen between ordinary (or non-expert) people and home robots (Yokota et al, 2008). Here, ‘integrated multimedia understanding’ means especially ‘multimedia understanding based on such a knowledge representation common to multiple information media (i.e., natural language, picture, music, gesture, etc.) that can facilitate cross-media operations (Yokota et al, 1984; Eakins & Graham, 1999; Kherfi et al, 2004; Yokota & Capi, 2005a)’. For ordinary people, however, natural language is the most important because it can convey the exact intention of the sender to the receiver due to its syntax and semantics common to its users, which is not necessarily the