Introducing a Corpus of Human-Authored Dialogue Summaries in Portuguese

Proceedings of Recent Advances in Natural Language Processing, pages 692–701, Hissar, Bulgaria, 7-13 September 2013. Introducing a Corpus of Human-Authored Dialogue Summaries in Portuguese Norton Trevisan Roman School of Arts, Sciences and Humanities University of S˜ao Paulo S˜ao Paulo – SP, Brazil norton@usp.br Paul Piwek Centre for Research in Computing The Open University Milton Keynes, UK p.piwek@open.ac.uk Ariadne M. B. Rizzoni Carvalho Institute of Computing University of Campinas Campinas – SP, Brazil ariadne@ic.unicamp.br Alexandre Rossi Alvares School of Arts, Sciences and Humanities University of S˜ao Paulo S˜ao Paulo – SP, Brazil alexandre.alvaresl@usp.br Abstract In this paper, we introduce a corpus of human-authored dialogue summaries col- lected through a web-experiment. The corpus features (i) one of the few ex- isting corpora of written dialogue sum- maries; (ii) the only corpus available for dialogue summaries in Portuguese; and (iii) the only available corpus of sum- maries produced for dialogues whose par- ticipants’ politeness alignment was sys- tematically varied. Comprising 1,808 human-authored summaries, produced by 452 summarisers, for four different dia- logues, this is, to the best of our knowl- edge, the largest individual corpus avail- able for dialogue summaries, with the highest number of participants involved. 1 Introduction As an important part of current mainstream re- search on automatic summarisation, corpora are used for a vast range of applications, from the construction of tutoring systems (e.g., (Callaway et al., 2005)) to abstract production from extracts (e.g., (Hasler, 2007)), to multi-document sum- marisation (e.g., (Atkinson and Munoz, 2013)). Still, most corpora are available in English only, which may have an impact on the performance of automatic summarisation methods when applied to other languages (de Loupy et al., 2010). Also, there seems to be a preference for newswire (e.g., (Amini, 2000; Copeck and Szpakowicz, 2004; Hasler, 2007; de Loupy et al., 2010)) and aca- demic texts summaries (e.g., (Teufel and Moens, 1997)), with fewer sources available for dialogue summaries, and those available mostly restricted to spoken dialogues (e.g., (Murray et al., 2005; Carletta et al., 2006; Liu and Liu, 2008)). In this paper we introduce a corpus of human- authored dialogue summaries, which we have re- leased for use by the research community.1 The corpus comprises 1,808 summaries, produced by 452 summarisers, for four different dialogues (each summariser produced a summary for each dialogue). To the best of our knowledge, this is the largest individual corpus available for dialogue summaries, with the largest number of participants involved. Collected through a web-experiment, where participants had to summarise a set of writ- ten dialogues, the corpus has the additional char- acteristic of being written in Portuguese (a lan- guage spoken by over 200 million people2, if one accounts only for Brazil and Portugal), thereby helping reduce the dearth of corpora for written dialogue summaries in languages other than En- glish. Additionally, source dialogues were carefully chosen so they portray interactions with different degree of politeness, as measured in an experiment carried out by Roman et al. (2006b). Resulting summaries may therefore be used for a range …

Content truncated to 500 words