Natural language processing neural network for recall and inference

Abstract

In this paper, we propose a novel neural network which can learn knowledge from natural language documents and can perform recall and inference. The proposed network has a sentence layer, a knowledge layer, ten kinds of deep case layers and a dictionary layer. In the network learning step, connections are updated based on Hebb's learning rule. The proposed network can handle a complicated sentence by incorporating the deep case layers and get unlearned knowledge from the dictionary layer. In the dictionary layer, Goi-Taikei, containing 400,000 words dictionary, is employed. Two kinds of experiments were carried out by using goo encyclopedia and Wikipedia as knowledge sources. Superior performance of the proposed neural network has been confirmed. © 2010 Springer-Verlag Berlin Heidelberg.