Automatic Error Correction on Japanese Functional Expressions Using Character-based Neural Machine Translation

Automatic Error Correction on Japanese Functional Expressions Using Character-based Neural Machine Translation Jun Liu1, Fei Cheng2, Yiran Wang1, Hiroyuki Shindo1,3, Yuji Matsumoto1,3 1 Nara Institute of Science and Technology, Japan 2 National Institute of Informatics, Japan 3 RIKEN Center for Advanced Intelligence Project (AIP), Japan {liu.jun.lc3, wang.yiran.ws5, shindo, matsu}@is.naist.jp fei-cheng@nii.ac.jp Abstract Correcting spelling and grammatical errors of Japanese functional expressions shows practical usefulness for Japanese Second Language (JSL) learners. However, the collection of these types of error data is difficult because it relies on detecting Jap- anese functional expressions first. In this paper, we propose a framework to correct the spelling and grammatical errors of Jap- anese functional expressions as well as the error data collection problem. Firstly, we apply a bidirectional Long Short-Term Memory with a Conditional Random Field (BiLSTM-CRF) model to detect Japanese functional expressions. Secondly, we ex- tract phrases which include Japanese func- tional expressions as well as their neigh- boring words from native Japanese and learners’ corpora. Then we generate a large scale of artificial error data via substitu- tion, injection and deletion operations. Fi- nally, we utilize the generated artificial er- ror data to train a sequence-to-sequence neural machine translation model for cor- recting Japanese functional expression er- rors. We also compare the character-based method with the word-based method. The experimental results indicate that the char- acter-based method outperforms the word- based method both on artificial error data and real error data. 1 Introduction The Japanese Language has various types of functional expressions which consist of more than one word including both content words and functional words, such as “を踏まえて (based on), に違いない (no doubt), てはいけない (must not)”. Due to the various meanings and us- ages, spelling and grammatical errors are often made by JSL learners when they use Japanese functional expressions in their writings. We ob- served some example sentences in Lang-8 Learner Corpora1 and summarized some typical types of spelling and grammatical errors of Japa- nese functional expressions, including word se- lection error (S), missing word error (M), redun- dant error (R), and word spelling error (W). Some example sentences of grammatical errors are shown in Table 1. Much previous research has paid special attention to the automatic detec- tion of Japanese functional expressions (Tsuchi- ya et al., 2006; Shime et al., 2007; Suzuki et al., 2012) while relatively few grammatical error correction applications have been developed to support JSL learners. Given this situation, auto- matic grammatical error correction of sentences written by JSL learners is essential in Japanese language learning. In this paper, we define a new task of correct- ing spelling and grammatical errors on Japanese functional expressions as follows. Given a phrase of a Japanese functional expressions and its neighboring words, our system aims to correct errors inside this phrase. For instance, a phrase “ 行くましょう。 (Let’s go.)” where the Japa- nese functional expression is in bold will be ex- pected to be corrected as “行きましょう。”, because the correct usage of Japanese verb con- jugation rules in this phrase depends …

Content truncated to 500 words