Tuning as Linear Regression

Marzieh Bazrafshan,  Tagyoung Chung,  Daniel Gildea
University of Rochester


Abstract

We propose a tuning method for statistical machine translation, based on the pairwise ranking approach. Hopkins and May (2011) presented a method that uses a binary classifier. In this work, we use linear regression and show that our approach is as effective as using binary classifier and converges faster.