The TALP-UPC phrase-based translation systems for WMT12: Morphology simplification and domain adaptation

Proceedings of the 7th Workshop on Statistical Machine Translation, pages 275–282, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics The TALP-UPC phrase-based translation systems for WMT12: Morphology simplification and domain adaptation Llu´ıs Formiga, Carlos A. Henr´ıquez Q., Adolfo Hern´andez, Jos´e B. Mari˜no, Enric Monte and Jos´e A. R. Fonollosa TALP Research Centre Universitat Polit`ecnica de Catalunya Barcelona, Spain {lluis.formiga,carlos.henriquez,adolfo.hernandez jose.marino,enric.monte,jose.fonollosa}@upc.edu Abstract This paper describes the UPC participation in the WMT 12 evaluation campaign. All sys- tems presented are based on standard phrase- based Moses systems. Variations adopted sev- eral improvement techniques such as mor- phology simplification and generation and do- main adaptation. The morphology simpli- fication overcomes the data sparsity prob- lem when translating into morphologically- rich languages such as Spanish by translat- ing first to a morphology-simplified language and secondly leave the morphology gener- ation to an independent classification task. The domain adaptation approach improves the SMT system by adding new translation units learned from MT-output and reference align- ment. Results depict an improvement on TER, METEOR, NIST and BLEU scores compared to our baseline system, obtaining on the of- ficial test set more benefits from the domain adaptation approach than from the morpho- logical generalization method. 1 Introduction TALP-UPC (Center of Speech and Language Applications and Technology at the Universitat Polit`ecnica de Catalunya) has participated in the WMT12 shared task translating across two direc- tions: English to Spanish and Spanish to English tasks. For the Spanish to English task we submitted a baseline system that uses all parallel training data and a combination of different target language mod- els (LM) and Part-Of-Speech (POS) language mod- els. A similar configuration was submitted for the English to Spanish task as baseline. Our main ap- proaches enriched the latter baseline in two indepen- dent ways: morphology simplification and domain adaptation by deriving new units into the phrase- table. Furthermore, additional specific strategies have been addressed on all systems to deal with well known linguistic phenomena in Spanish such as cli- tics and contractions. The paper is presented as follows. Section 2 presents the main rationale for the phrase-based sys- tem and the main pipeline of our baseline system. Section 3 presents the approaches taken to improve the baseline system on the English to Spanish task. Section 4 presents the obtained results on internal and official test sets while conclusions and further work are presented in Section 5. 2 Baseline system: Phrase-Based SMT Classically, a phrase-based translation system im- plements a log-linear model in which a foreign lan- guage sentence fj 1 = f1, f2, . . . , fj is translated into another language sentence eI 1 = e1, e2, . . . , eI by searching for the translation hypothesis that max- imizes a log-linear combination of feature models (Brown et al., 1990): ˆeI 1 = arg max eI 1 ( M X m=1 λmhm � eI 1, fJ 1 � ) (1) where the separate feature functions hm refer to the system models and the set of …

Content truncated to 500 words