What charaterizes intelligent systems: Issues of language

Abstract

Summary form only given. Human intelligence is based on the functions of language. From this fact it is very natural to characterize an intelligent system by language: capability of handling language or capability of working with language. Discussing some issues of language related to intelligence, we focus on language functions of the brain in this talk. There are three approaches to elucidate the language functions of the brain: top-down, intermediate, and bottom-up. In this talk, we shall first see some examples of top-down and intermediate approaches, and then discuss a bottom-up approach. As a top-down approach we show computational models of language in which four principles are realized, and as an intermediate approach, we introduce the stratification in the brain shown by aphasia studies. As a bottom-up approach, we have conducted brain experiments, aiming to elucidate the difference of brain activities during processing ideational meaning and interpersonal meaning. In the experiments, two kinds of sentences with and without honorific expressions were sequentially shown to subjects, where texts with honorific expressions contain interpersonal meaning and texts without honorific expressions mainly hold ideational meaning. Electroencephalograms were measured to acquire event-related potentials. Then using equivalent current dipole source localization method, we analyzed the neuronal activation of the brain. It was found that the brain activities for understanding sentences with honorific expressions are different from those produced for understanding sentences without honorific expressions. Finally we shall discuss a brain mechanism possibly used for learning how to speak.