Minimum Error Training of Log-Linear Translation Models

Recent work on training of log-linear interpolation mod-els for statistical machine translation reported perfor-mance improvements by optimizing parameters with re-spect to translation quality, rather than to likelihood ori-ented criteria. This work presents an alternative and more direct training procedure for log-linear interpola-tion models. In addition, we point out the subtle inter-action between log-linear models and the beam search algorithm. Experimental results are reported on two Chinese-English evaluation sets, C-Star 2003 and Nist 2003, by using a statistical phrase-based model derived from Model 4. By optimizing parameters with respect to the BLUE score, performance relative improvements by 9.6 % and 2.8 % were achieved, respectively. 1