SAU’S Submission for CCMT 2021 Quality Estimation Task

This paper describes our submissions to CCMT 2021 quality estimation sentence-level task for both Chinese-to-English (ZH-EN) and English-to-Chinese (EN-ZH). In this task. We follow TransQuest framework which is based on cross-lingual transformers (XLM-R). In order to make the model pay more attention to key words, we use the attention mechanism and gate module to fuse the last hidden state and pooler output of XLM-R model to generate more accurate prediction. In addition, we use the Predictor-Estimator architecture model to integrate with our model to improve the results. Experiments show that this is a simple and effective ensemble method.