A Bangla Spell Checking Technique to Facilitate Error Correction in Text Entry Environment

Spell checker is a common tool used in different text entry techniques for error free writing. Recent advances in spell checking techniques in different languages significantly decreases word and sentence level typos from writing text. But developing an optimized spell checker in Bangla language is still a great research challenge due to the complex rules of Bangla spelling. To overcome the challenge we propose a new technique of spell checking considering both word and sentence level errors to facilitate the text correction techniques of Bangla language. We use a hybrid approach by combining edit distance algorithm and probability based N-gram language model for the error detection and correction tasks. A large corpus is built containing almost 0.5 million words and 50,000 n-gram sentences to detect and correct the errors. We evaluate the performance of our proposed method with different test data collected from online sources. Again, the efficiency of proposed technique is estimated by conducting a pilot study involving users of different ages. The accuracy rate of our proposed spell checker is approximately 97%. The validity of our claims have been proved through comparison with other Bangla spell checkers.