Coreference in Wikipedia: Main Concept Resolution

Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning (CoNLL), pages 229–238, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Coreference in Wikipedia: Main Concept Resolution Abbas Ghaddar RALI-DIRO Universit´e de Montr´eal Montr´eal, Canada abbas.ghaddar@umontreal.ca Philippe Langlais RALI-DIRO Universit´e de Montr´eal Montr´eal, Canada felipe@iro.umontreal.ca Abstract Wikipedia is a resource of choice ex- ploited in many NLP applications, yet we are not aware of recent attempts to adapt coreference resolution to this resource. In this work, we revisit a seldom studied task which consists in identifying in a Wikipedia article all the mentions of the main concept being described. We show that by exploiting the Wikipedia markup of a document, as well as links to external knowledge bases such as Freebase, we can acquire useful information on entities that helps to classify mentions as coreferent or not. We designed a classifier which dras- tically outperforms fair baselines built on top of state-of-the-art coreference resolu- tion systems. We also measure the benefits of this classifier in a full coreference reso- lution pipeline applied to Wikipedia texts. 1 Introduction Coreference Resolution (CR) is the task of iden- tifying all mentions of entities in a document and grouping them into equivalence classes. CR is a prerequisite for many NLP tasks. For example, in Open Information Extraction (OIE) (Yates et al., 2007), one acquires subject-predicate-object rela- tions, many of which (e.g., <the foundation stone, was laid by, the Queen s daughter>) are useless because the subject or the object contains mate- rial coreferring to other mentions in the text being mined. Most CR systems, including state-of-the-art ones (Durrett and Klein, 2014; Martschat and Strube, 2015; Clark and Manning, 2015) are es- sentially adapted to news-like texts. This is basi- cally imputable to the availability of large datasets where this text genre is dominant. This includes resources developed within the Message Under- standing Conferences (Hirshman and Chinchor, 1998) or the Automatic Content Extraction (ACE) program (Doddington et al., 2004), as well as re- sources developed within the collaborative anno- tation project OntoNotes (Pradhan et al., 2007). It is now widely accepted that coreference reso- lution systems trained on newswire data performs poorly when tested on other text genres (Hen- drickx and Hoste, 2009; Sch¨afer et al., 2012), in- cluding Wikipedia texts, as we shall see in our ex- periments. Wikipedia is a large, multilingual, highly struc- tured, multi-domain encyclopedia, providing an increasingly large wealth of knowledge. It is known to contain well-formed, grammatical and meaningful sentences, compared to say, ordinary internet documents. It is therefore a resource of choice in many NLP systems, see (Medelyan et al., 2009) for a review of some pioneering works. While being a ubiquitous resource in the NLP community, we are not aware of much work con- ducted to adapt CR to this text genre. Two no- table exceptions are (Nguyen et al., 2007) and (Nakayama, 2008), two studies dedicated to ex- tract tuples from Wikipedia articles. Both studies demonstrate that the design of a dedicated rule- …

Content truncated to 500 words