An Interactive Proofreading System for Inappropriately Selected Words on Using Predictive Text Entry

Hideya Iwasaki, Kumiko Tanaka-Ishii

Abstract: Predictive text entry systems on computers like kana-to-kanji conversion provide a mechanism that enables users to select among possible words for a given input. Mistakes in selection are relatively common, and they introduce real-word errors. A proofreading system is thus needed to detect and correct real-word errors on a computer without imposing troublesome operations on users. To this end, a practical proofreading system for Japanese text is proposed. The system automatically detects possible real-word homonym errors, and for each detected word, suggests substitution candidates of the same pronunciation. The user can either choose the most appropriate one or leave the original untouched. The system uses an algorithm based on the Naïve Bayesian method. Although the proofreading system was implemented for homonym errors in Japanese text, its design concept and algorithm are also applicable to other languages. The client program of the proofreading system is implemented on the Emacs text editor and works in real time.