Improved method to detect erroneous characters of Japanese sentence using Markov chain model

摘要

In optical character recognition and continuous speech recognition of a natural language, there are erroneous characters that have been wrongly substituted, inserted and deleted. We have previously proposed a method to detect and correct these en-ors using Markov chain model. In this paper we propose a improved method to detect erroneous characters wrongly substituted or inserted characters using two threshold values of Markov chain probability. and to correct these erroneous characters using 2nd-order and 3rd -order Markov chain models. From the results of the experiments. It is concluded that this method is useful for detecting as well as correcting these erroneous characters