ZZX_MT: the BeiHang MT System for NTCIR-9 PatentMT Task

Wenhan Chao,Zhoujun Li

2011-01-01

Abstract:paper, we describe ZZX_MT machine translation system for the NTCIR-9 Patent Machine Translation Task(PatentMT). We participated in the Chinese-English translation subtask and submit three results, which correspond to three different models or decoding algorithms respectively. Both of the first two are phrase-based SMT approaches integrating the BTG constraint into reordering models, and the last one is a hybrid system, which is an SMT system while using an example-based decoder. As a PBSMT, ZZX_MT uses a Tree-Tree model to combine the global and local reordering together. In this model, we will use the BTG for the global reordering, and assume the structure of the BTG tree is independent of the local reordering of the two adjacent phrase pairs, so we can design the global and local reordering models respectively. Also, we provide a tree isomorphism model to use the source-side parser tree to constrain the BTG tree, and provide a local reordering model. Through the tree isomorphism model, we can incorporate the linguistic syntactic knowledge into the formally syntax-based translation model. In order to consider the global reordering further, ZZX_MT provides an example-based decoder, using the translation example to constrain the translation sentence's structure.