Clause-Based Reordering Constraints to Improve Statistical Machine Translation

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1351–1355, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Clause-Based Reordering Constraints to Improve Statistical Machine Translation Ananthakrishnan Ramanathan1, Pushpak Bhattacharyya2, Karthik Visweswariah1, Kushal Ladha2 and Ankur Gandhe1 1IBM Research India 2Department of CSE, IIT Bombay {aramana2,v-karthik,ankugand}@in.ibm.com {pb,kush}@cse.iitb.ac.in Abstract We demonstrate that statistical machine translation (SMT) can be improved sub- stantially by imposing clause-based re- ordering constraints during decoding. Our analysis of clause-wise translation of dif- ferent types of clauses shows that it is ben- eficial to apply these constraints for finite clauses, but not for non-finite clauses. In our experiments in English-Hindi trans- lation with an SMT system (DTM2), on a test corpus containing around 850 sen- tences with manually annotated clause boundaries, BLEU improves to 20.4 from the baseline score of 19.4. This statisti- cally significant improvement is also con- firmed by subjective (human) evaluation. We also report preliminary work on au- tomatically identifying the kind of clause boundaries appropriate for enforcing re- ordering constraints. 1 Introduction It has been recognized widely that reordering is the Achilles’ heel of most SMT models (Birch et al., 2009; Hoang and Koehn, 2009), espe- cially when applied to languages with large syn- tactic differences. In our experiments in English- Hindi SMT, we observe that it is quite frequent in multi-clause sentences for phrases to move out of their respective clauses due to incorrect re- ordering. While such mistakes can be avoided by restricting reordering over clause boundaries, this paper demonstrates that such a strategy works well only for finite clauses, and not for non-finite clauses. Clause-wise or part-by-part translation has been a standard approach in traditional transfer-based systems. In Systran, as described by Hutchins and Somers (1992), conjunct and relative clauses were segmented in a preprocessing step. Simi- lar methods were used in the the Stanford Ma- chine Translation project (Wilks, 1973). Chan- drasekar (1994) applies a sentence simplification method to machine translation, where sentences are split at conjunctions, relative pronouns, etc., before translation. Rao et al. (2000) describe a clause-wise translation strategy within an English- Hindi transfer-based MT system. In the context of SMT, Koehn and Knight (2003) use a dedicated noun phrase (NP) translation subsystem to obtain significant improvements in German-English translation. Other similar work includes (Watanabe et al., 2003) for Japanese-English SMT and (Hewavitha- rana et al., 2007) for Arabic-English. Sudoh et al. (2010) propose methods to perform clause- level alignment of the parallel corpus, and to translate clauses (all clauses identified by a syn- tactic parser) as a unit to improve long-distance reordering in a specialized domain – English- Japanese translation of research paper abstracts in the medical domain. While our approach draws from many of the above, it is novel in the following ways: (i) We provide an analysis of different clause types in translation – we show that only some kinds of clauses benefit from the use of reordering con- straints, and (ii) We demonstrate significant im- provements using this strategy …

Content truncated to 500 words