A Contextualized Word Representation Approach for Irony Detection

IroSvA (Irony Detection in Spanish variants) is a contest dedicated to identify the presence of irony in a given context in short messages written in Spanish, specifically tweets and news comments. In this case, it is necessary to consider that irony may be expressed in a different way according to the Spanish variant, which makes this task more complex to tackle it. Taking into account that irony detection is a very important task in many applications, we proposed a system based on a distributed representation of the texts, using the ELMo approach. We did not use any handcrafted features, lexicons or external datasets as prior information.