Exploiting Unlabeled Data for Neural Grammatical Error Detection

摘要

Identifying and correcting grammatical errors in the text written by non-native writers have received increasingattention in recent years. Although a number of annotated corpora have been established to facilitate data-driven gram-matical error detection and correction approaches, they are still limited in terms of quantity and coverage because lmmanannotation is labor-intensive, time-consuming, and expensive. In this work, we propose to utilize unlabeled data to trainneural network based graminatical error detection models. The basic idea is to cast error detection ms a binary classificationproblem and derive positive and negative training examples from unlabeled data. We introduce an attention-based neuralnetwork to capture long-distance dependencies that influence the word being detected. Experiments show that the proposedapproach significantly outperforms SVM and convolutional networks with fixed-size context window.