Semantic Error Detection and Correction in Bangla Sentence

Detection and correction of errors in Bengali text is essential. In general, Bengali text error can be classified into non-word error and semantic error (also known as context sensitive error). Till date, auto-correction for semantic error in Bengali sentence is challenging since there is no significant research works on this very topic. In this paper, we bring out the concept of Semantic Error detection and correction. We have developed a method that can detect and correct this kind of errors. Semantic error includes typographical error, grammatical errors, homophone errors, homonym error etc. Our goal to this study is to develop an approach to handle multiple semantic errors in a sentence. We have used our own built confused word list by edit distance and apply Naïve Bayes Classifier to detect and correct typographical and homophone error. For a candidate word from a sentence, we pick out a set of words which is a collection of confused words. We use all other neighbor words as features for each word from confusion set. Then we apply naïve theorem to calculate the probability and decide whether a target word is error or not. We have used 28,057 sentences to evaluate our model and we have achieved more than 90% accuracy. All data corpora used to evaluate the model are built by us. We strongly believe that the problem we have solved may shed light on the advancement of Bengali language processing significantly.