The HDU Discriminative SMT System for Constrained Data PatentMT at NTCIR10
Patrick Simianer, Gesa Stupperich, Laura Jehl, Katharina Wäschle, Artem Sokolov, Stefan Riezler · 2013
We describe the statistical machine translation (SMT) systems developed at Heidelberg University for the Chinese-to-English and Japanese-to-English PatentMT subtasks at the NTCIR10 workshop. The core system used in both subtasks is a combination of hierarchical phrase-based translation and discriminative training using either large feature sets and ℓ1/ℓ2 regularization (for Japanese-to-English) or variants of soft syntactic constraints (for Chinese-to-English). Our goal is to address the twofold nature of patents by exploiting the repetitive nature of patents through feature sharing in a multi-task learning setup (used in the Japaneseto-English translation subtask), and by countersteering complex word order differences with syntactic features (used in the Chinese-to-English translation subtask).