Source code for paper: Improving Grammatical Error Correction via Pre-Training a Copy-Augmented Architecture with Unlabeled Data - View it on GitHub