Extracting Parallel Sub-Sentential Fragments from Non-Parallel Corpora

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 81–88, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Extracting Parallel Sub-Sentential Fragments from Non-Parallel Corpora Dragos Stefan Munteanu University of Southern California Information Sciences Institute 4676 Admiralty Way, Suite 1001 Marina del Rey, CA, 90292 dragos@isi.edu Daniel Marcu University of Southern California Information Sciences Institute 4676 Admiralty Way, Suite 1001 Marina del Rey, CA, 90292 marcu@isi.edu Abstract We present a novel method for extract- ing parallel sub-sentential fragments from comparable, non-parallel bilingual cor- pora. By analyzing potentially similar sentence pairs using a signal processing- inspired approach, we detect which seg- ments of the source sentence are translated into segments in the target sentence, and which are not. This method enables us to extract useful machine translation train- ing data even from very non-parallel cor- pora, which contain no parallel sentence pairs. We evaluate the quality of the ex- tracted data by showing that it improves the performance of a state-of-the-art sta- tistical machine translation system. 1 Introduction Recently, there has been a surge of interest in the automatic creation of parallel corpora. Sev- eral researchers (Zhao and Vogel, 2002; Vogel, 2003; Resnik and Smith, 2003; Fung and Cheung, 2004a; Wu and Fung, 2005; Munteanu and Marcu, 2005) have shown how fairly good-quality parallel sentence pairs can be automatically extracted from comparable corpora, and used to improve the per- formance of machine translation (MT) systems. This work addresses a major bottleneck in the de- velopment of Statistical MT (SMT) systems: the lack of sufficiently large parallel corpora for most language pairs. Since comparable corpora exist in large quantities and for many languages – tens of thousands of words of news describing the same events are produced daily – the ability to exploit them for parallel data acquisition is highly benefi- cial for the SMT field. Comparable corpora exhibit various degrees of parallelism. Fung and Cheung (2004a) describe corpora ranging from noisy parallel, to compara- ble, and finally to very non-parallel. Corpora from the last category contain “... disparate, very non- parallel bilingual documents that could either be on the same topic (on-topic) or not”. This is the kind of corpora that we are interested to exploit in the context of this paper. Existing methods for exploiting comparable corpora look for parallel data at the sentence level. However, we believe that very non-parallel cor- pora have none or few good sentence pairs; most of their parallel data exists at the sub-sentential level. As an example, consider Figure 1, which presents two news articles from the English and Romanian editions of the BBC. The articles re- port on the same event (the one-year anniversary of Ukraine’s Orange Revolution), have been pub- lished within 25 minutes of each other, and express overlapping content. Although they are “on-topic”, these two docu- ments are non-parallel. In particular, they contain no parallel sentence pairs; methods designed to ex- tract full parallel sentences will not find any use- ful data …

Content truncated to 500 words