Automatic Detection of Irony

3A Multilevel Scheme for Irony Annotation in Social Network Content

3.1. Introduction

The aim of this chapter is to propose an annotation scheme for irony in a specific type of text, i.e. tweets. In Chapter 2 (section 2.2), we provided an overview of the different schemes that have been put forward for annotating tweets in Italian and English (Gianti et al. 2012, Shutova et al. 2013, Van Hee et al. 2016). These schemes are similar in that they all take a global approach to characterize irony, without considering linguistic or extra-linguistic cues at message level. The majority only include one level of annotation, characterizing tweets by figurative type (ironic/non-ironic), polarity (positive, negative or neutral) or, more rarely, the pragmatic device used to create irony (polarity reversal, hyperbole or euphemism).

Considering the work carried out in the field of linguistics on verbal irony markers in poems, novels, etc. (see section 3.3), we see that work on irony in social networks from a computational perspective has barely scratched the surface of the problem, without going into specifics. Our objective here is to discuss in detail, providing a fine-grained study of different markers and responding to the following questions:

• – can the different types of irony identified in the field of linguistics be found in a specific corpus harvested from social networks, such as Twitter?
• – if so, which types are encountered most frequently?
• – are these types marked explicitly? ...