From Better Translation Units to Better Reordering: Soft Linguistic Constraints in Statistical Machine Translation

old_uid15825
titleFrom Better Translation Units to Better Reordering: Soft Linguistic Constraints in Statistical Machine Translation
start_date2015/06/16
schedule11h30
onlineno
location_infosalle de conférences
summaryThis talk focuses on some ways to gainfully apply source-side syntactic and semantic constraints in statistical machine translation. First, I will show how soft syntactic constraints on phrase selection during decoding can yield gains, going back to some of my own doctoral dissertation work. Keeping soft semantic constraints on translation units to a separate talk, I will next show how incorporating a different kind of soft syntactic constraints into a hierarchical phrase-based translation system improves reordering, and yields further significant gains (ACL 2014). Last, I will present soft semantic reordering constraints. Although useful, in our experiments their gain was limited in the presence of soft syntactic reordering constraints. Therefore we will examine in detail the behavior differences between the two linguistic reordering constraints.
responsiblesGrau