Paper: Stochastic Iterative Alignment For Machine Translation Evaluation

ACL ID P06-2070
Title Stochastic Iterative Alignment For Machine Translation Evaluation
Venue Annual Meeting of the Association of Computational Linguistics
Session Poster Session
Year 2006
Authors

A number of metrics for automatic eval- uation of machine translation have been proposed in recent years, with some met- rics focusing on measuring the adequacy of MT output, and other metrics focus- ing on uency. Adequacy-oriented met- rics such as BLEU measure n-gram over- lap of MT outputs and their references, but do not represent sentence-level informa- tion. In contrast, uency-oriented metrics such as ROUGE-W compute longest com- mon subsequences, but ignore words not aligned by the LCS. We propose a metric based on stochastic iterative string align- ment (SIA), which aims to combine the strengths of both approaches. We com- pare SIA with existing metrics, and nd that it outperforms them in overall evalu- ation, and works specially well in uency evaluation.