Centroid-Based Efficient Minimum Bayes Risk Decoding
Nara Institute of Science and Technology, Japan and National Institute of Information and Communications Technology (NICT), National Institute of Advanced Industrial Science and Technology · Nara Institute of Science and Technology, Japan · Division of Information Science, Nara Institute of Science and Technology · National Institute of Information and Communications Technology (NICT), National Institute of Advanced Industrial Science and Technology
PDF 由论文原始站点提供，PaperCompass 不保存论文文件。DOI 10.18653/v1/2024.findings-acl.654 ↗
摘要
Minimum Bayes risk (MBR) decoding achieved state-of-the-art translation performance by using COMET, a neural metric that has a high correlation with human evaluation.However, MBR decoding requires quadratic time since it computes the expected score between a translation hypothesis and all reference translations.We propose centroid-based MBR (CBMBR) decoding to improve the speed of MBR decoding.Our method clusters the reference translations in the feature space, and then calculates the score using the centroids of each cluster.The experimental results show that our CBMBR not only improved the decoding speed of the expected score calculation 5.7 times, but also outperformed vanilla MBR decoding in translation quality by up to 0.5 COMET in the WMT’22 En\leftrightarrowJa, En\leftrightarrowDe, En\leftrightarrowZh, and WMT’23 En\leftrightarrowJa translation tasks.