Grammatical Error Correction using Robust Word Representation Learning and Deep Neural Networks

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K16117
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: Tokyo Metropolitan University
• Principal Investigator: Komachi Mamoru 首都大学東京, システムデザイン研究科, 准教授 (60581329)
• Research Collaborator: Kaneko Masahiro ; Zhang Longtu
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 深層学習 / 単語分散表現 / 文法誤り訂正

Outline of Final Research Achievements

This study proposed an approach to apply deep learning to grammatical error detection and correction. The sentences written by language learners differ from those written by native speakers in that they may make mistakes in the words themselves or in the context of which they are used. Taking these differences in context into account, we built a mathematical model for representing words and used deep learning to detect and correct grammatical errors. We also proposed a method for error detection using a contextualized language representation model learned from a large amount of text data written by native speakers, and achieved the state-of-the-art accuracy in English grammatical error detection.
On the other hand, for Japanese and Chinese, we proposed a model that reconstructs word sequence by decomposing kanji into radicals, and demonstrated its effectiveness in Japanese-Chinese neural machine translation.

Free Research Field

自然言語処理

Academic Significance and Societal Importance of the Research Achievements

本研究では英語学習者の文法誤り検出について、学習者の文章の誤り方を考慮して単語をモデル化することと、ネイティブが書いた大規模な文章データから獲得した文脈つきの言語表現モデルを用いることが、それぞれ有効であることを世界で初めて示し、いずれの研究においても当時の世界最高精度の精度を達成することができました。本研究は世界を代表する英語学習者の文法誤り検出の研究の一つです。