IronyTR: Irony Detection in Turkish Informal Texts

Introduction

Together with the rapid growth of online services, there occurred an increasing amount of textual data produced by users every day. There is a need to analyze this huge textual data to understand users' demands, and make deductions for a wide set of applications such as product improvements for e-commerce. In order to meet these needs, sentiment analysis methods, which are useful to extract emotions from textual data, are used (Chakraborty et al., 2020; Yadav et al., 2020).

There is a rich variety of methods to extract emotional information from texts, however these methods lack the ability to analyze irony, especially in Turkish texts. Oxford Dictionary defines irony as the expression of one's meaning by using language that normally signifies the opposite, typically for humorous or emphatic effect (“Irony”, 2020). From the definition, one can understand why irony is particularly hard to detect: opposition of the meaning is mostly implicit, and automated emotion detection methods lack the understanding of common sense that we humans share.

Irony can be in various forms, which further makes it a complex task to model and detect irony. On this issue, Van Hee and others focus on two irony types: situational irony and irony by means of a polar clash. Any ironic text that does not fit in to these two categories are referred to as other textual irony. They define situational irony as the type of irony where the text is a written description of an ironic situation, and claim that this type of irony is harder to detect and model since it needs an understanding of the context (Van Hee et al., 2016a). On the other hand, Carvalho and others focus on this type of irony in their study conducted in Portuguese (Carvalho et al., 2020). Following is a good example of situational irony:

As the second type, Van Hee and others define irony by means of a polar clash as the type of irony where two opposite sentiments can be extracted from the same text (Van Hee et al., 2016a). Since this type of irony can be modeled by sentiment scoring of the tokens, several studies conducted on English use sentiment scores of words to improve the performance of their models (Ahmed et al., 2018; Van Hee et al., 2016b; Xu et al., 2015). Following sentence clearly illustrates irony by means of a polar clash: