Natural Language Processing Neural Network for Analogical Inference

Abstract

In this paper, we propose a neural network which learns knowledge from natural language documents and analogize. In proposed network, firing patterns of neurons are memorized in Form Inference Memory and Meaning Inference Memory. When a similar firing pattern is appeared, a memorized pattern is retrieved from Memories. Analogical inference is allowed by this processing. In learning process, natural language documents are analyzed by Japanese dependency structure analyzer, CaboCha. The results are used in network connection learning. By these processing, the proposed network can learn and analogize. We confirmed the effectiveness of the proposed network.