An Attempts to generate internal semantic concepts by autonomous segmentation of sequential

Abstract

Elucidation and implementation of consciousness on machines have not yet been realized, while understanding mind by attempt to synthesize the mind is important. In this study, we attempted to generate a semantic concept by considering the mutual-segmentation-hypothesis and the characteristics of information processing in a living neuronal network, by referring to the model of the language acquisition process of a human infant. The purpose is to build a model of a biological neuronal network that generates primitive consciousness from a series of input stimuli that reflect the situation of the outside world. As a preliminary step, we developed a simple model of autonomous generating of semantic concept in silico. Stocked inputted words as nodes in a co-occurrence network were applied clustering based on two criteria, the meaning of the pleasant-unpleasant state of the system, and these generated clusters were defined as memory of the system. Then the pleasant-unpleasant state of the system fluctuates the memory of the system depending on the newly input words, and the words co-occurred on the semantic network were output as a response according to this pleasantness. There were cases where words related to input words was outputted, however, it did not always naturally associated words necessarily output. The results of learning depends on the learning source, it suggested to be required learning with sentences that have much simpler meanings for meaningful output.