Tweeting and Being Ironic in the Debate about a Political Reform: the French Annotated Corpus TWitter-MariagePourTous

Tweeting and being ironic in the debate about a political reform: the French annotated corpus TWitter-MariagePourTous Cristina Bosco1, Mirko Lai1, Viviana Patti1, Daniela Virone2 1Dipartimento di Informatica – Universit`a degli Studi di Torino, Corso Svizzera 185 – 10149 Torino (Italy), 2Dipartimento di Studi Umanistici – Universit`a degli Studi di Torino Via Verdi 10 – 10124 Torino (Italy) {bosco,lai,patti}@di.unito.it, dvirone@unito.it Abstract The paper introduces a new annotated French data set for Sentiment Analysis, which is a currently missing resource. It focuses on the collection from Twitter of data related to the socio-political debate about the reform of the bill for wedding in France. The design of the annotation scheme is described, which extends a polarity label set by making available tags for marking target semantic areas and figurative language devices. The annotation process is presented and the disagreement discussed, in particular, in the perspective of figurative language use and in that of the semantic oriented annotation, which are open challenges for NLP systems. Keywords: sentiment analysis, French, political debates, Twitter, figurative language 1. Introduction Texts generated by users within the context of social media can be a great opportunity for moving onward the develop- ment of corpus-based techniques for Sentiment Analysis and Opinion Mining (SA&OM). But their annotation raises several challenging issues. Even more attention than in classical tasks must indeed be paid to the development of such kind of data sets. First of all, it should be observed that we are still far from sharing standardized annotation schemes and tag sets, and the big effort devoted to the development of a resource may also not result in a proportional advancement for the area e.g. if is not enough reusable of comparable with others. A suitable solution is adopting schemes which includes comparable annotations, which can be matched in particular with those exploited in evaluation campaigns (see e.g. SemEval1, or Evalita2), but also, considering the relative recentness of the area, enriching the annotation with new tags for describing aspects that potentially have impact on the results of sentiment engines, even if not previously annotated. Second, the topics of texts of interest for SA&OM can be several, and we must take into account also the coverage of domains when we design a resource. The same problem of coverage can be related to languages, since several resources are available for a few languages when they are almost missing at all for several others. Finally, a variety of phenomena features the language of social media, e.g. figurative devices, which can worst the performance of SA systems (Ghosh et al., 2015b); by contrast, the availability of data where they are annotated can positively influence results, but it is very difficult to evaluate the suitability of this annotation task. 1http://aclweb.org/aclwiki/index.php? title=SemEval_Portal 2http://www.evalita.it/ In this paper, we present a new annotated corpus for French SA&OM called TWitter-MariagePourTous (henceforth TW-MPT), collecting texts from Twitter about the debate in France on the homosexual wedding and the related reform. This resource has been developed within the context of a project …

Content truncated to 500 words