Prague Czech-English dependency treebank: resource for structure-based MT

Prague Czech-English Dependency Treebank Resource for Structure-based MT Martin ˇCmejrek, Jan Cuˇr´ın, Jan Hajiˇc, Jiˇr´ı Havelka Institute of Formal and Applied Linguistics Faculty of Mathematics and Physics, Charles University in Prague Malostransk´e n´am. 25, Praha 1, Czech Republic {cmejrek,curin,hajic,havelka}@ufal.mff.cuni.cz Abstract. The Prague Czech-English Dependency Treebank (PCEDT) is a syntactically annotated Czech-English parallel corpus. The Penn Treebank has been translated to Czech, and its annotation automatically transformed into dependency annotation scheme. The dependency annotation of Czech is done from plain text by automatic procedures. A small subset of corresponding Czech and English sentences has been annotated by humans. First experiments in Czech-English machine translation using these data have already been carried out. The resources have been created at Charles University in Prague and released by Linguistic Data Consortium in 2004. 1. Introduction The Prague Czech-English Dependency Treebank – PCEDT (Linguistic Data Consortium, 2004) is a project of creating a Czech-English syntactically anno- tated parallel corpus motivated by research in the field of machine translation. Parallel data are needed for de- signing, training, and evaluation of both statistical and rule-based machine translation systems. When starting the PCEDT project, we decided to translate and annotate an existing syntactically anno- tated corpus, rather than to annotate in parallel an al- ready existing parallel corpus of raw texts, since the latter option would have been more money and time consuming. The choice of the Penn Treebank as the source corpus was pragmatically motivated: firstly it is a widely recognized linguistic resource, and secondly the translators were native speakers of Czech, capable of high quality translation into their native language. Since Czech is a language with a relatively high degree of word-order freedom, and its sentences con- tain certain syntactic phenomena, such as discontinu- ous constituents (non-projective constructions), which cannot be straightforwardly handled using the annota- tion scheme of Penn Treebank (Marcus et al., 1993; Linguistic Data Consortium, 1999), based on phrase- structure trees, we decided to adopt for the PCEDT the dependency-based annotation scheme of the Prague Dependency Treebank – PDT (Linguistic Data Consor- tium, 2001; Sgall et al., 1986), which is described in Section 3. In Section 2., we describe the process of translating the Penn Treebank into Czech and reference retransla- tions. Section 4. presents manual tectogrammatical an- notation for both languages. The automatic process of transformation of Penn Treebank annotation of English into both representations, analytical and tectogrammat- ical, is described in Section 5., the automatic anno- tation of Czech is described in Section 6. Section 7. gives an overview of additional resources included in the PCEDT corpus. Section 8. mentions two experi- ments that have been carried out on the data collection. 2. English to Czech Translation of Penn Treebank Since the PCEDT is aimed as a resource for the purpose of MT, the translators were asked to trans- late each English sentence as a single Czech sentence and to avoid unnecessary stylistic changes of trans- lated sentences. About half of the Penn Treebank has been translated so far (currently 21,628 …

Content truncated to 500 words