Coreference in Prague Czech-English Dependency Treebank

Coreference in Prague Czech-English Dependency Treebank Anna Nedoluzhko, Michal Nov´ak, Silvie Cinkov´a, Marie Mikulov´a, Jiˇr´ı M´ırovsk´y Charles University in Prague, Faculty of Mathematics and Physics, Institute of Formal and Applied Linguistics Malostransk´e n´amˇest´ı 25, CZ-11800 Prague 1 {nedoluzko,mnovak,cinkova,mikulova,mirovsky}@ufal.mff.cuni.cz Abstract We present coreference annotation on parallel Czech-English texts of the Prague Czech-English Dependency Treebank (PCEDT). The paper describes innovations made to PCEDT 2.0 concerning coreference, as well as the coreference information already present there. We characterize the coreference annotation scheme, give the statistics and compare our annotation with the coreference annotation in Ontonotes and Prague Dependency Treebank for Czech. We also present the experiments made using this corpus to improve the alignment of coreferential expressions, which helps us to collect better statistics of correspondences between types of coreferential relations in Czech and English. The corpus released as PCEDT 2.0 Coref is publicly available. Keywords: parallel corpus, bilingual coreference, alignment, Czech, English 1. Introduction Over the last years, cross-lingual studies have been attract- ing a great deal of attention. Cross-lingual studies on dis- course phenomena are no exception, mainly motivated by the task of machine translation, which still often overlooks phenomena beyond the sentence span. Corpora annotated with coreference are arguably a valuable resource for such studies. However, they are mostly monolingual. To our best knowledge, coreference-annotated parallel corpora are very rare, with ParCor 1.0 (Guillou et al., 2014), and the original release of Prague Czech-English Dependency Treebank 2.0 (Hajiˇc et al., 2012, PCEDT) being the main representatives. In this paper, we present Prague Czech-English Depen- dency Treebank 2.0 Coref (Nedoluzhko et al., 2016, PCEDT 2.0 Coref), which comprises large-scale manual annotation of coreference links in a parallel corpus of Czech and English texts. Furthermore, it includes improve- ments to alignment of coreferential expressions. Both the coreference and the alignment annotation aim at serving as high-quality data for coreference-related studies on these languages. The annotation has been built upon the original release of PCEDT 2.0, a Czech-English parallel treebank sized over 1.2 million tokens in almost 50,000 sentence pairs. Its English part consists of the Wall Street Journal section of the Penn Treebank (Marcus et al., 1999), while the Czech part was manually translated from the English source sentence by sentence. PCEDT is the second Praguian corpus with manual coref- erence annotation. It was preceded by the monolingual Prague Dependency Treebank (PDT), which pioneered the coreference annotation on Czech texts in its version 2.0 (Hajiˇc et al., 2006), later enriched by other types of coref- erence and discourse relations in version 3.0 (Bejˇcek et al., 2013). The differences between the PCEDT 2.0 Coref and PDT 3.0 coreference annotation are addressed in Sec- tion 3.5. The annotation of coreference relations in PCEDT has pro- ceeded in multiple stages. While the version 2.0 intro- duced by Hajiˇc et al. (2012) contained the annotation of the so-called grammatical coreference and pronominal coref- erence, it has been recently enhanced by adding nominal coreference. Both annotation stages are elaborated in Sec- tions 3.1. and 3.3. The coreference …

Content truncated to 500 words