Myanmar Spelling Correction Based On N-grams Model

Abstract

This paper describes Myanmar spelling correction intended for real-word errors and non-word errors. There are three main modules in this paper. They are error detection,candidates generation, error correction. Dictionary look up method is used for detecting errors, Levenshtein Distance Algorithm is used for generating candidates and N-grams model is used for correcting errors. There can be human-generated misspellings which can be distinguished into three groups (i) Typographic Errors(Non-word error) (ii)Phonetic Errors(Cognitive error) (iii) Context Errors(Real word errors). This spelling correction can solve all of these three misspellings problem and the main contribution of this paper is to solve the context errors using n-grams model in sentence level. Moreover, this spelling correction can solve the pali misspelling errors. Experimental results show that each of error types can be solved by this spelling correction. The general accuracy of all error types is greater than 85%.