IMS HotCoref DE: A Data-driven Co-reference Resolver for German

IMS HotCoref DE: A Data-Driven Co-Reference Resolver for German Ina R¨osiger and Jonas Kuhn Institute for Natural Language Processing University of Stuttgart Germany [roesigia|kuhn]@ims.uni-stuttgart.de Abstract This paper presents a data-driven co-reference resolution system for German that has been adapted from IMS HotCoref, a co-reference resolver for English. It describes the difficulties when resolving co-reference in German text, the adaptation process and the features designed to address linguistic challenges brought forth by German. We report performance on the reference dataset T¨uBa-D/Z and include a post-task SemEval 2010 evaluation, showing that the resolver achieves state-of-the-art performance. We also include ablation experiments that indicate that integrating linguistic features increases results. The paper also describes the steps and the format necessary to use the resolver on new texts. The tool is freely available for download. Keywords: Co-reference Resolution, German, Adaptation, Tool 1. Introduction Noun phrase co-reference resolution is the task of deter- mining which noun phrases (NPs) in a text or dialogue re- fer to the same discourse entities (Ng, 2010). Coreference resolution has been extensively addressed in NLP research, e.g. in the CoNLL shared task 2012 and 2011 (Pradhan et al., 2012; Pradhan et al., 2011) or in the SemEval shared task 2010 (Recasens et al., 2010). A lot of research focuses on English co-reference, resulting in a number of high performing English co- reference systems, e.g. Clark and Manning (2015), Durrett and Klein (2014) or Bj¨orkelund and Kuhn (2014). However, there has been less work on German co- reference resolution. Since the SemEval shared task 2010, only a few systems have been improved or developed, such as the rule-based CorZu system (Klenner and Tuggener, 2011; Tuggener and Klenner, 2014) or Krug et al. (2015)’s system which is tailored to the domain of historic novels. This paper presents a data-driven co-reference resolu- tion system that is based on the English IMS HotCoref sys- tem (Bj¨orkelund and Kuhn, 2014). It describes the adap- tation process, the specific requirements for co-reference resolution in German text as well as the tool that is freely available for download1. 2. Noun Phrase Coreference Resolution Coreferent links exist between two NPs if the first NP refers back to a discourse entity that has already been introduced in the discourse and is thereby known to the reader. The referring entity in the text is called an anaphor while the entity to which the anaphor refers back is called the an- tecedent. Coreferent entities include pronominal NPs (1), nominal NPs (2) and named entities (3). (1) Pronominal: DE: Peter ging in den Supermarkt. Er kaufte eine Pizza.2 1http://www.ims.uni-stuttgart.de/ forschung/ressourcen/werkzeuge/HotCorefDe 2Anaphors are typed in bold face, their antecedents are under- lined. EN: Peter went into the supermarket. He bought a pizza. (2) Nominal: DE: Peter kaufte gestern ein neues Buch. Der Roman war sehr unterhaltsam. EN: Peter bought a new book yesterday. The novel turned out to be very entertaining. (3) Named entities: DE: Barack Hussein Obama ist der 44. Pr¨asident der Vereinigten Staaten. Obama erhielt 2009 den Friedensnobelpreis. EN: Barack Hussein …

Content truncated to 500 words