Paper: A Re-examination on Features in Regression Based Approach to Automatic MT Evaluation

ACL ID P08-3005
Title A Re-examination on Features in Regression Based Approach to Automatic MT Evaluation
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2008
Authors

Machine learning methods have been exten- sively employed in developing MT evaluation metrics and several studies show that it can help to achieve a better correlation with hu- man assessments. Adopting the regression SVM framework, this paper discusses the lin- guistic motivated feature formulation strategy. We argue that “blind” combination of avail- able features does not yield a general metrics with high correlation rate with human assess- ments. Instead, certain simple intuitive fea- tures serve better in establishing the regression SVM evaluation model. With six features selected, we show evidences to sup- port our view through a few experiments in this paper.