Abstract
In this paper, we propose a technique of vector conversion of word and linguistic information processing neural network (LIPNN) with abilities of memory, recall, and inference of sentences expressed by semantic network form as one of the methods to treat natural language. The LIPNN receives Japanese texts, and outputs semantic network form of triple representation. First, the proposed network extracts a relation between words by CaboCha (Japanese dependency structure analyzer) and converts a natural language into a semantic network form of triple representation. Then thesaurus is used to convert a word into vector expression and input it into the LIPNN. Because of a word converted into vector expression, the proposed network is able to handle the word that does not have learned. The LIPNN is a three-layered structure to mem-ory and recalls triple representations and sentences. And also it can infer without control from the outside by newly proposed inhibit activation method. In the computer simula-tion, we carried out two kinds of simulations to confirm the network is able to handle the unlearned words and abilities of memory, recall, and inference of sentences in the network.
