Automatic segmentation of Chinese chunks using a neural network

Abstract

Chunk analysis method is a new idea proposed in recent years in developing new machine translation methods. Chinese chunk analysis plays a very important role in machine translation from Chinese to foreign languages. The automatic segmentation of Chinese chunks is the basis of Chinese chunk analysis. It is important for processing real text and corpus construction. This paper presents a method for automatic segmentation of Chinese chunks based on 3-layers neural networks. The corpus we used has been processed with Chinese word segmentation and phrase identification and tagging. In the neural networks model, the input data is the segmentation situation of every character and its combinations with neighbor characters in a Chinese sentence. The output is the segmentation results of every character in a Chinese sentence. The preliminary results show that the method is feasible and effective.