Multitask Learning for Chinese Grammatical Error Detection

Abstract

Chinese as a Foreign Language (CFL) learners often make grammatical errors such as missing words, selecting wrong words and wrong word order due to language negative migration. In this paper, we propose a neural sequence labeling model with a supplementary objective for Chinese grammatical error detection. We use the manually labeled dataset written by CFL learners to train the models. This multitask learning model has better performance than other sequence labeling model because it can learn the bias in the label distribution and learn richer features for semantic composition.