A Unified Model for Soft Linguistic Reordering Constraints in Statistical Machine Translation

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1123–1133, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics A Unified Model for Soft Linguistic Reordering Constraints in Statistical Machine Translation Junhui Li† Yuval Marton‡ Philip Resnik† Hal Daum´e III† †UMIACS, University of Maryland, College Park, MD {lijunhui, resnik, hal}@umiacs.umd.edu ‡Microsoft Corp., City Center Plaza, Bellevue, WA yumarton@microsoft.com Abstract This paper explores a simple and effec- tive unified framework for incorporating soft linguistic reordering constraints into a hierarchical phrase-based translation sys- tem: 1) a syntactic reordering model that explores reorderings for context free grammar rules; and 2) a semantic re- ordering model that focuses on the re- ordering of predicate-argument structures. We develop novel features based on both models and use them as soft constraints to guide the translation process. Ex- periments on Chinese-English translation show that the reordering approach can sig- nificantly improve a state-of-the-art hier- archical phrase-based translation system. However, the gain achieved by the seman- tic reordering model is limited in the pres- ence of the syntactic reordering model, and we therefore provide a detailed analy- sis of the behavior differences between the two. 1 Introduction Reordering models in statistical machine transla- tion (SMT) model the word order difference when translating from one language to another. The popular distortion or lexicalized reordering mod- els in phrase-based SMT make good local pre- dictions by focusing on reordering on word level, while the synchronous context free grammars in hierarchical phrase-based (HPB) translation mod- els are capable of handling non-local reordering on the translation phrase level. However, reorder- ing, especially without any help of external knowl- edge, remains a great challenge because an ac- curate reordering is usually beyond these word level or translation phrase level reordering mod- els’ ability. In addition, often these translation models fail to respect linguistically-motivated syn- tax and semantics. As a result, they tend to pro- duce translations containing both syntactic and se- mantic reordering confusions. In this paper our goal is to take advantage of syntactic and seman- tic parsing to improve translation quality. Rather than introducing reordering models on either the word level or the translation phrase level, we pro- pose a unified approach to modeling reordering on the linguistic unit level, e.g., syntactic constituents and semantic roles. The reordering unit falls into multiple granularities, from single words to more complex constituents and semantic roles, and of- ten crosses translation phrases. To show the ef- fectiveness of our reordering models, we integrate both syntactic constituent reordering models and semantic role reordering models into a state-of- the-art HPB system (Chiang, 2007; Dyer et al., 2010). We further contrast it with a stronger base- line, already including fine-grained soft syntac- tic constraint features (Marton and Resnik, 2008; Chiang et al., 2008). The general ideas, however, are applicable to other translation models, e.g., phrase-based model, as well. Our syntactic constituent reordering model con- siders context free grammar (CFG) rules in the source language and predicts the reordering of their elements …

Content truncated to 500 words