Neural network approach to grammatical error correction without annotated data

Kazuya Uekado, Susumu Mori, Taku Harada, Hayato Ohwada

研究成果: Paper査読

抄録

Grammatical Error Correction (GEC) is an important role for English Learners. In recent years, the competitions of this task were held. Therefore, researches about automatic GEC are very active. Many research about this task uses a machine translation method. Although machine translations output a high performance, training and parameter tuning need a large amount of annotated data. Generally, collecting annotated data is very expensive. We should avoid using annotated datasets. This paper proposes a method without annotated data. Our method uses only native data because obtaining native data is easier from websites. Language model and classification correct errors. The language model can correct multiple errors and a classification method improves the performance of corrections. For training model, we use neural networks because these methods achieved success in various natural language processing tasks. As a result, our approach got the decent performance on CoNLL-2014 test datasets. We show an effectivity of automated GEC without annotated data.

本文言語English
ページ58-62
ページ数5
出版ステータスPublished - 2019
イベント10th International Multi-Conference on Complexity, Informatics and Cybernetics, IMCIC 2019 - Orlando, United States
継続期間: 12 3月 201915 3月 2019

Conference

Conference10th International Multi-Conference on Complexity, Informatics and Cybernetics, IMCIC 2019
国/地域United States
CityOrlando
Period12/03/1915/03/19

フィンガープリント

「Neural network approach to grammatical error correction without annotated data」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル