Formemes in English-Czech Deep Syntactic MT

Proceedings of the 7th Workshop on Statistical Machine Translation, pages 267–274, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics Formemes in English-Czech Deep Syntactic MT ∗ Ondˇrej Dušek, Zdenˇek Žabokrtský, Martin Popel, Martin Majliš, Michal Novák, and David Mareˇcek Charles University in Prague, Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics Malostranské námˇestí 25, Prague {odusek,zabokrtsky,popel,majlis,mnovak,marecek}@ufal.mff.cuni.cz Abstract One of the most notable recent improve- ments of the TectoMT English-to-Czech trans- lation is a systematic and theoretically sup- ported revision of formemes—the annotation of morpho-syntactic features of content words in deep dependency syntactic structures based on the Prague tectogrammatics theory. Our modifications aim at reducing data sparsity, increasing consistency across languages and widening the usage area of this markup. Formemes can be used not only in MT, but in various other NLP tasks. 1 Introduction The cornerstone of the TectoMT tree-to-tree ma- chine translation system is the deep-syntactic lan- guage representation following the Prague tec- togrammatics theory (Sgall et al., 1986), and its ap- plication in the Prague Dependency Treebank (PDT) 2.01 (Hajiˇc et al., 2006), where each sentence is analyzed to a dependency tree whose nodes corre- spond to content words. Each node has a number of attributes, but the most important (and difficult) for the transfer phase are lemma—lexical informa- tion, and formeme—surface morpho-syntactic infor- ∗ This research has been supported by the grants FP7-ICT-2009-4-247762 (FAUST), FP7-ICT-2009-4-249119 (Metanet), LH12093 (Kontakt II), DF12P01OVV022 (NAKI), 201/09/H057 (Czech Science Foundation), GAUK 116310, and SVV 265 314. This work has been using language resources de- veloped and/or stored and/or distributed by the LINDAT-Clarin project of the Ministry of Education of the Czech Republic (project LM2010013). 1http://ufal.mff.cuni.cz/pdt2.0 mation, including selected auxiliary words (Ptáˇcek and Žabokrtský, 2006; Žabokrtský et al., 2008). This paper focuses on formemes—their definition and recent improvements of the annotation, which has been thoroughly revised in the course of prepa- ration of the CzEng 1.0 parallel corpus (Bojar et al., 2012b), whose utilization in TectoMT along with the new formemes version has brought the greatest ben- efit to our English-Czech MT system in the recent year. However, the area of possible application of formemes is not limited to MT only or to the lan- guage pair used in our system; the underlying ideas are language-independent. We summarize the development of morpho- syntactic annotations related to formemes (Sec- tion 2), provide an overview of the whole TectoMT system (Section 3), then describe the formeme an- notation (Section 4) and our recent improvements (Section 5), as well as experimental applications, in- cluding English-Czech MT (Section 6). The main asset of the formeme revision is a first systematic re- organization of the existing practical aid, providing it with a solid theoretical base, but still bearing its intended applications in mind. 2 Related Work Numerous theoretical approaches had been made to morpho-syntactic description, mainly within va- lency lexicons, starting probably with the work by Helbig and Schenkel (1969). Perhaps the best one for Czech is PDT-VALLEX (Hajiˇc et al., 2003), list- …

Content truncated to 500 words