A linguistically annotated reordering model for BTG-based statistical machine translation

Abstract

In this paper, we propose a linguistically annotated reordering model for BTG-based statistical machine translation. The model incorporates linguistic knowledge to predict orders for both syntactic and non-syntactic phrases. The linguistic knowledge is automatically learned from source-side parse trees through an annotation algorithm. We empirically demonstrate that the proposed model leads to a significant improvement of 1.55% in the BLEU score over the baseline reordering model on the NIST MT-05 Chinese-to-English translation task.

Backward and trigger-based language models for statistical machine translation

“…This article significantly extends our previous work (Xiong, Zhang and Li 2011) in four major aspects. Firstly, we thoroughly compare the single backward language model with (1) the single forward language model (FLM) and (2) the combination of the forward and backward language models by integrating them into the decoder at the same time.Secondly, we investigate different trigger pair selection criteria in order to maximize the performance of the MI trigger model.Thirdly, we explore distance-dependent and syntactically informed triggers in order to investigate whether such triggers are able to obtain further improvements.Finally, we validate the robustness of the proposed models on several language pairs, including Chinese-to-English, Spanish-to-English and Vietnamese-to-English.…”

Tree kernel-based SVM with structured syntactic knowledge for BTG-based phrase reordering

“…In this paper, phrase reordering is recast as a classification issue as done in previous work (Xiong et al, 2006(Xiong et al, & 2008Zhang et al, 2007a). In training, we use a machine learning algorithm training on the annotated phrase reordering instances that are automatically extracted from word-aligned, source sentence parsed training corpus, to learn a classifier.…”

“…For significance test, we use Zhang et al's implementation (Zhang et al, 2004). Baseline Systems: we set three baseline systems: B1) Moses (Koehn et al, 2007) that uses lexicalized unigram reordering model to predict three orientations: monotone, swap and discontinuous; B2) MaxEnt-based reordering model with lexical boundary word features only (Xiong et al, 2006); B3) Linguistically annotated reordering model for BTG-based (LABTG) SMT (Xiong et al, 2008). For Moses, we used the default settings.…”

“…We generate the boundary word features from the extracted reordering instances in the same way as discussed in Xiong et al (2006) and use Zhang's MaxEnt Tools 2 to train a reordering model for the 2 nd baseline system. Similarly, we use the algorithm 1 in Xiong et al (2008) to extract features and use the same MaxEnt Tools to train a reordering model for the 3 rd baseline system. Based on the extracted reordering instances, we generate the four structured features and the linear features, and then use the Tree Kernel Tools (Moschitti, 2004) to train our kernel-based reordering model (linear, tree and composite).…”

“…In phrasebased method, local word reordering 1 can be effectively captured by phrase pairs directly while local phrase reordering is explicitly modeled by phrase reordering model and distortion model. Recently, many phrase reordering methods have been proposed, ranging from simple distancebased distortion model (Koehn et al, 2003;Och and Ney, 2004), flat reordering model (Wu, 1997;Zens et al, 2004), lexicalized reordering model (Tillmann, 2004;Kumar and Byrne, 2005), to hierarchical phrase-based model (Chiang, 2005;Setiawan et al, 2007) and classifier-based reordering model with linear features (Zens and Ney, 2006;Xiong et al, 2006;Zhang et al, 2007a;Xiong et al, 2008). However, one of the major limitations of these advances is the structured syntactic knowledge, which is important to global reordering Elming, 2008), has not been well exploited.…”