Automatically Detecting and Attributing Indirect Quotations

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 989–999, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Automatically Detecting and Attributing Indirect Quotations Silvia Pareti⋄∗ Tim O’Keefe†∗ Ioannis Konstas⋄ James R. Curran† Irena Koprinska† ⋄ILCC, School of Informatics † e -lab, School of IT University of Edinburgh University of Sydney United Kingdom NSW 2006, Australia {s.pareti,i.konstas}@sms.ed.ac.uk {tokeefe,james,irena}@it.usyd.edu.au Abstract Direct quotations are used for opinion min- ing and information extraction as they have an easy to extract span and they can be attributed to a speaker with high accuracy. However, simply focusing on direct quotations ignores around half of all reported speech, which is in the form of indirect or mixed speech. This work presents the first large-scale experiments in indirect and mixed quotation extraction and attribution. We propose two methods of ex- tracting all quote types from news articles and evaluate them on two large annotated corpora, one of which is a contribution of this work. We further show that direct quotation attribu- tion methods can be successfully applied to in- direct and mixed quotation attribution. 1 Introduction Quotations are crucial carriers of information, par- ticularly in news texts, with up to 90% of sentences in some articles being reported speech (Bergler et al., 2004). Reported speech is a carrier of evi- dence and factuality (Bergler, 1992; Saur´ı and Puste- jovsky, 2009), and as such, text mining applications use quotations to summarise, organise and validate information. Extraction of quotations is also rele- vant to researchers interested in media monitoring. Most quotation attribution studies (Pouliquen et al., 2007; Glass and Bangay, 2007; Elson and McKeown, 2010) thus far have limited their scope to direct quotations (Ex.1a), as they are delimited ∗*These authors contributed equally to this work. by quotation marks, which makes them easy to ex- tract. However, annotated resources suggest that di- rect quotations represent only a limited portion of all quotations, i.e., around 30% in the Penn Attribution Relation Corpus (PARC), which covers Wall Street Journal articles, and 52% in the Sydney Morning Herald Corpus (SMHC), with the remainder being in- direct (Ex.1c) or mixed (Ex.1b) quotations. Retriev- ing only direct quotations can miss key content that can change the interpretation of the quotation (Ex. 1b) and will entirely miss indirect quotations. (1) a. “For 10 million, you can move $100 mil- lion of stocks,” a specialist on the Big Board gripes. “That gives futures traders a lot more power.” b. Police would only apply for the restrictions when “we have a lot of evidence that late- night noise...is disturbing the residents of that neighbourhood”, Superintendent Tony Cooke said. c. Mr Walsh said Rio was continuing to hold discussions with its customers to arrive at a mutually agreed price. Previous work on extracting indirect and mixed quotations has suffered from a lack of large-scale data, and has instead used hand-crafted lexica of re- porting verbs with rule-based approaches. The lack of data has also made comparing the relative merit of these …

Content truncated to 500 words