PARC 3.0: A Corpus of Attribution Relations

PARC 3.0: A Corpus of Attribution Relations Silvia Pareti+ School of Informatics The University of Edinburgh, UK S.Pareti@sms.ed.ac.uk Abstract Quotation and opinion extraction, discourse and factuality have all partly addressed the annotation and identification of Attribution Relations. However, disjoint efforts have provided a partial and partly inaccurate picture of attribution and generated small or incomplete resources, thus limiting the applicability of machine learning approaches. This paper presents PARC 3.0, a large corpus fully annotated with attribution relations (ARs). The annotation scheme was tested with an inter-annotator agreement study showing satisfactory results for the identification of ARs and high agreement on the selection of the text spans corresponding to its constitutive elements: source, cue and content. The corpus, which comprises around 20k ARs, was used to investigate the range of structures that can express attribution. The results show a complex and varied relation of which the literature has addressed only a portion. PARC 3.0 is available for research use and can be used in a range of different studies to analyse attribution and validate assumptions as well as to develop supervised attribution extraction models. Keywords: attribution, corpora, quotations 1. Introduction With a vast amount of data being available, in particular through the world wide web, more than ever before users have the chance to access an enormous amount of informa- tion. While information per se is a resource, this informa- tion overload can hinder our ability to process it and use it to understand issues or make decisions. To manage the vast amount of information available today requires ways to organise, filter and select it. It therefore becomes impor- tant to recognise different point of views (e.g. to make a medical or financial decision), monitor the statements of a specific person (e.g. a politician) and identify truthful and reliable information. These tasks require the identification of attribution relations, thus allowing to link the attributed material to the entity representing its source. Attribution allows to identify what has been attributed to a specific source but also affects how the text itself is per- ceived. Changes in the source or attributional verb (e.g. say vs suspect vs joke) can affect our perception of the quoted statement. While research and commercial systems for the automatic identification and extraction of attribution relations have multiplied in recent years (e.g. NewsExplorer1 (Pouliquen et al., 2007)), several issues are still to be addressed. The applications of such systems are severely limited by low precision and low recall. The reason for this relatively poor performance is partly to be found in the limited scope of such approaches. Stud- ies on attribution focused either on its overlap and interac- tion with other linguistic phenomena, such as discourse re- lation (Carlson and Marcu, 2001; Wolf and Gibson, 2005) and factuality (Saur´ı and Pustejovsky, 2009), or on specific types of attributions such as direct quotations (Elson and McKeown, 2010) or opinions (Wiebe, 2002). While these studies show that attribution is relevant for different linguist +Current affiliation: Google Inc., spareti@google.com 1http://emm.newsexplorer.eu/ fields, their …

Content truncated to 500 words