... alignments we extractedwere better according to all metrics we used, itshould not be too surprising that they yield better translation performance, but perhaps a better trade-off can be achieved ... per-formance. We tested this hypothesis on six differ-ent language pairs from three different domains, andfound that the new alignment scheme not only per-forms better than the baseline, but ... 986–993,Columbus, Ohio, USA, June 2008.c2008 Association for Computational Linguistics Better Alignments = Better Translations?Kuzman GanchevComputer & Information ScienceUniversity of Pennsylvaniakuzman@cis.upenn.eduJo˜ao...