TranslationModelBasedCross-LingualLanguageModel.PDFVIP

Translation Model Based Cross-Lingual Language Model Adaptation: from Word Models to Phrase Models Shixiang Lu, Wei Wei, Xiaoyin Fu, and Bo Xu Interactive Digital Media Technology Research Center Institute of Automation, Chinese Academy of Sciences 95 Zhongguancun East Road, Haidian District, Beijing 100190, China {shixiang.lu,wei.wei.media,xiaoyin.fu,xubo}@ Abstract this reason, most researchers preferred to select sim- ilar training data from the large training corpus in the In this paper, we propose a novel translation past few years (Eck et al., 2004; Zhao et al., 2004; model (TM) based cross-lingual data selec- Kim, 2005; Masskey and Sethy, 2010; Axelrod et tion model for language model (LM) adapta- al., 2011). This would empirically provide more ac- tion in statistical machine translation (SMT), from word models to phrase models. Given curate lexical probabilities, and thus better match the a source sentence in the translation task, this translation task at hand (Axelrod et al., 2011). model directly estimates the probability that Many previous data selection approaches for LM a sentence in the target LM training corpus adaptation in SMT depend on the ?rst pass transla- is similar. Compared with the traditional ap- tion hypotheses (Eck et al., 2004; Zhao et al., 2004; proaches which utilize the ?rst pass translation Kim, 2005; Masskey and Sethy, 2010), they selec- hypotheses, cross-lingual data selection mod- el avoids the problem of noisy proliferation. t the sentences which are similar to the translation Furthermore, phrase TM based cross-lingual hypotheses. These schemes are overall limited by data selection model is more effective than the quality of the translation hy