作者: Chris Hokamp , Iacer Calixto , Joachim Wagner , Jian Zhang
DOI: 10.3115/V1/W14-3341
关键词:
摘要: We describe the DCU-MIXED and DCUSVR submissions to WMT-14 Quality Estimation task 1.1, predicting sentencelevel perceived post-editing effort. Feature design focuses on target-side features as we hypothesise that source side has little effect quality of human translations, which are included in 1.1 this year’s WMT shared task. experiment with QuEst framework, our past work, three novel feature sets. Despite these efforts, two systems perform poorly competition. Follow up experiments indicate poor performance is due improperly optimised parameters.