ParCorFull: a Parallel Corpus Annotated with Full Coreference

ParCorFull: a Parallel Corpus Annotated with Full Coreference Ekaterina Lapshinova-Koltunski, Christian Hardmeier, Pauline Krielke Saarland University, Uppsala University, Saarland University e.lapshinova@mx.uni-saarland.de, christian.hardmeier@lingfil.uu.se, pauline.krielke@gmail.com Abstract In this paper, we describe a parallel corpus annotated with full coreference chains that has been created to address an important problem that machine translation and other multilingual natural language processing (NLP) technologies face – translation of coreference across languages. Recent research in multilingual coreference and automatic pronoun translation has led to important insights into the problem and some promising results. However, its scope has been restricted to pronouns, whereas the phenomenon is not limited to anaphoric pronouns. Our corpus contains parallel texts for the language pair English-German, two major European languages. Despite being typologically very close, these languages still have systemic differences in the realisation of coreference, and thus pose problems for multilingual coreference resolution and machine translation. Our parallel corpus with full annotation of coreference will be a valuable resource with a variety of uses not only for NLP applications, but also for contrastive linguists and researchers in translation studies. This resource supports research on the mechanisms involved in coreference translation in order to develop a better understanding of the phenomenon. Keywords: coreference, full coreference, cross-lingual coreference resolution, coreference annotation, linguistic annotation, ma- chine translation, multilingual NLP 1. Introduction We present a parallel corpus containing full corefer- ence annotation that has been created to address an im- portant problem affecting machine translation (MT) and multilingual NLP technologies: translation of corefer- ence across languages. The corpus is available from the LINDAT repository at http://hdl.handle.net/ 11372/LRT-2614. Texts of various genres often contain recurring references to objects and other discourse entities, realised with a vari- ety of linguistic devices such as noun phrases (NPs), pro- nouns or other linguistic means. Devices referring to the same entity are said to corefer. The coreference relation is shared across all languages. However, languages differ considerably in the range of linguistic means triggering this relation (Kunz and Steiner, 2012; Kunz and Lapshinova- Koltunski, 2015; Nov´ak and Nedoluzhko, 2015). The choice between these referring expressions is governed by language-specific constraints. Differences in their realisa- tion give rise to transformation patterns used to create co- herent translations. In translation, references in the source language (SL) must be rendered with appropriate linguistic devices from the repertoire of the target language (TL), with different constraints. For instance, pronouns and ad- jectives in German (DE) are subject to grammatical gender agreement, whereas in English (EN), only person pronouns have this marking and adjectives are unmarked. Recent research in multilingual coreference and automatic pronoun translation has led to important insights into the problem and some promising results, but a working solu- tion for coreference translation in an end-to-end MT has not been demonstrated yet. Research on automatic core- ference translation has been restricted to pronouns, but the phenomenon is by no means limited to anaphoric pronouns. Example (1) illustrates a coreference chain expressing a re- lation of comparison, where we have a closed class of expli- cit …

Content truncated to 500 words