Hypothesis ranking and two-pass approaches for machine translation system combination
Given a number of machine translations of a source segment, the goal of system combination is to produce a new translation that has better quality than all of them. This paper describes a number of improvements that were recently added to the JHU system combination scheme: (i) A hypothesis ranking tech- nique which orders the system outputs, on a per-segment ba- sis, according to predicted translation quality, thus improving a subsequent incremental combination step. (ii) A two-pass combination procedure, which first produces several combi- nation outputs with the given translations, and then performs one more combination step with these new outputs. Results from the NIST MT09 informal system combination evalu- ation on Arabic-to-English and Urdu-to-English 1 show that both approaches offer significant BLEU and TER gains over a baseline JHU combination scheme.