A Stochastic Technique to Obtain Training Data for Word Segmentation

Unlike western languages, there exists no word boundary in Japanese. This is why we face to hard problems to analyze documents in Japanese very often. More difficulty arises in expertised domains such as medical, mechanical, computer science documents. In this work, we discuss how to obtain pseudo test corpus based on Markov process Monte Carlo Method (MCMC), given small amount of test data. In this environment we show nice results using our approach.