The adaptation of an English spellchecker for Japanese writers

It has been pointed out that the spelling errors made by second-language writers writing in English have features that are to some extent characteristic of their first language, and the suggestion has been made that a spellchecker could be adapted to take account of these features. In the work reported here, a corpus of spelling errors made by Japanese writers writing in English was compared with a corpus of errors made by native speakers. While the great majority of errors were common to the two corpora, some distinctively Japanese error patterns were evident against this common background, notably a difficulty in deciding between the letters b and v, and the letters l and r, and a tendency to add syllables. A spellchecker that had been developed for native speakers of English was adapted to cope with these errors. A brief account is given of the spellchecker’s mode of operation to indicate how it lent itself to modifications of this kind. The native-speaker spellchecker and the Japanese-adapted version were run over the error corpora and the results show that these adaptations produced a modest but worthwhile improvement to the spellchecker’s performance in correcting Japanese-made errors.