Treebanking User-Generated Content: A Proposal for a Unified Representation in Universal Dependencies

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5240–5250 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5240 Treebanking User-Generated Content: A Proposal for a Unified Representation in Universal Dependencies Manuela Sanguinetti1, Cristina Bosco1, Lauren Cassidy2, Özlem Çetino˘glu3, Alessandra Teresa Cignarella1,4, Teresa Lynn2, Ines Rehbein5 Josef Ruppenhofer6, Djamé Seddah7, Amir Zeldes8 1. Dipartimento di Informatica, Università degli Studi di Torino, Italy 2. ADAPT Centre, Dublin City University, Ireland 3. IMS, University of Stuttgart, Germany 4. PRHLT Research Center, Universitat Politècnica de València, Spain 5. University of Mannheim, Germany 6. Leibniz-Institut für Deutsche Sprache Mannheim, Germany 7. INRIA Paris, France 8. Georgetown University, USA {msanguin|bosco|cigna}@di.unito.it, ozlem.cetinoglu@ims.uni-stuttgart.de, {teresa.lynn|lauren.cassidy}@adaptcentre.ie, ines@informatik.uni-mannheim.de, ruppenhofer@ids-mannheim.de, djame.seddah@inria.fr, amir.zeldes@georgetown.edu Abstract The paper presents a discussion on the main linguistic phenomena of user-generated texts found in web and social media, and proposes a set of annotation guidelines for their treatment within the Universal Dependencies (UD) framework. Given on the one hand the increasing number of treebanks featuring user-generated content, and its somewhat inconsistent treatment in these resources on the other, the aim of this paper is twofold: (1) to provide a short, though comprehensive, overview of such treebanks - based on available literature - along with their main features and a comparative analysis of their annotation criteria, and (2) to propose a set of tentative UD-based annotation guidelines, to promote consistent treatment of the particular phenomena found in these types of texts. The main goal of this paper is to provide a common framework for those teams interested in developing similar resources in UD, thus enabling cross-linguistic consistency, which is a principle that has always been in the spirit of UD. Keywords: Web, social media, treebanks, Universal Dependencies, annotation guidelines, UGC 1. Introduction The immense popularity gained by social media in the last decade has made it an eligible source of data for a large number of research fields and applications, especially for sentiment analysis and opinion mining. In order to success- fully process the data available from such sources, linguistic analysis is often helpful, which in turn prompts the use of NLP tools to that end. Despite the ever increasing number of contributions, especially on Part-of-Speech tagging (Gimpel et al., 2011; Owoputi et al., 2013; Lynn et al., 2015; Bosco et al., 2016; Çetino˘glu and Çöltekin, 2016; Proisl, 2018) and parsing (Foster, 2010; Petrov and McDonald, 2012; Kong et al., 2014; Liu et al., 2018), automatic processing of user- generated content (UGC) still represents a challenging task, as is shown by the workshop series on noisy user-generated text (W-NUT)1. UGC is a continuum of text sub-genres that may considerably vary according to the specific conventions and limitations posed by the medium used (blog, discussion forum, online chat, microblog, etc.), its degree of "canonical- ness" with respect to a more standard language, as well as the linguistic devices2 adopted to convey a message. Over- all, however, there are some well-recognized phenomena that characterize UGC as a whole (Foster, 2010; Seddah …

Content truncated to 500 words