A String Similarity Measure Based on Orthographic and Phonetic Similarity for Spelling Correction

The most commonly used string similarity measure for spelling correction is minimum edit distance (MED), which is based solely on the orthographic similarity between two strings. In order to overcome this shortcoming, this paper presents a more sophisticated similarity measure that considers both the orthographic and phonetic similarity between two strings. To demonstrate the effectiveness of the proposed measure, we implement and test a spelling correction system and apply it to two languages, namely English and Korean. We investigate the useful features for each language through various experiments and achieve 10-best accuracies of 95.2 for English and 97.4 for Korean.