How to Produce Unseen Teddy Bears: Improved Morphological Processing of Compounds in SMT

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 579–587, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics How to Produce Unseen Teddy Bears: Improved Morphological Processing of Compounds in SMT Fabienne Cap, Alexander Fraser CIS, University of Munich {cap|fraser}@cis.uni-muenchen.de Marion Weller IMS, University of Stuttgart wellermn@ims.uni-stuttgart.de Aoife Cahill Educational Testing Service acahill@ets.org Abstract Compounding in morphologically rich languages is a highly productive process which often causes SMT approaches to fail because of unseen words. We present an approach for translation into a com- pounding language that splits compounds into simple words for training and, due to an underspecified representation, allows for free merging of simple words into compounds after translation. In contrast to previous approaches, we use features pro- jected from the source language to predict compound mergings. We integrate our ap- proach into end-to-end SMT and show that many compounds matching the reference translation are produced which did not ap- pear in the training data. Additional man- ual evaluations support the usefulness of generalizing compound formation in SMT. 1 Introduction Productive processes like compounding or inflec- tion are problematic for traditional phrase-based statistical machine translation (SMT) approaches, because words can only be translated as they have occurred in the parallel training data. As paral- lel training data is limited, it is desirable to ex- tract as much information from it as possible. We present an approach for compound processing in SMT, translating from English to German, that splits compounds prior to training (in order to ac- cess the individual words which together form the compound) and recombines them after translation. While compound splitting is a well-studied task, compound merging has not received as much at- tention in the past. We start from Stymne and Can- cedda (2011), who used sequence models to pre- dict compound merging and Fraser et al. (2012) who, in addition, generalise over German inflec- tion. Our new contributions are: (i) We project features from the source language to support com- pound merging predictions. As the source lan- guage input is fluent, these features are more re- liable than features derived from target language SMT output. (ii) We reduce compound parts to an underspecified representation which allows for maximal generalisation. (iii) We present a detailed manual evaluation methodology which shows that we obtain improved compound translations. We evaluated compound processing both on held-out split data and in end-to-end SMT. We show that using source language features increases the accuracy of compound generation. Moreover, we find more correct compounds than the base- lines, and a considerable number of these com- pounds are unseen in the training data. This is largely due to the underspecified representation we are using. Finally, we show that our approach im- proves upon the previous work. We discuss compound processing in SMT in Section 2, and summarise related work in Sec- tion 3. In Section 4 we present our method for splitting compounds and reducing the component words to an underspecified representation. The …

Content truncated to 500 words