Team TheSyllogist at SemEval-2023 Task 3: Language-Agnostic Framing Detection in Multi-Lingual Online News: A Zero-Shot Transfer Approach

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 2058–2061 July 13-14, 2023 ©2023 Association for Computational Linguistics Team TheSyllogist at SemEval-2023 Task 3: Language-Agnostic Framing Detection in Multi-Lingual Online News: A Zero-Shot Transfer Approach Osama Mohammed Afzal, Preslav Nakov Mohamed bin Zayed University of Artificial Intelligence {osama.afzal, preslav.nakov}@mbzuai.ac.ae Abstract We describe our system for SemEval-2022 Task 3 subtask 2 which on detecting the frames used in a news article in a multi-lingual setup. We propose a multi-lingual approach based on machine translation of the input, followed by an English prediction model. Our sys- tem demonstrated good zero-shot transfer ca- pability, achieving micro-F1 scores of 53% for Greek (4th on the leaderboard) and 56.1% for Georgian (3rd on the leaderboard), without any prior training on translated data for these lan- guages. Moreover, our system achieved com- parable performance on seven other languages, including German, English, French, Russian, Italian, Polish, and Spanish. Our results demon- strate the feasibility of creating a language- agnostic model for automatic framing detection in online news. 1 Introduction Framing is an advanced form of discourse through which the communicator attempts to generate a cognitive bias by consistently associating a topic with a specific context, commonly referred to as a frame. It serves as a commonly used, albeit pow- erful, political tool and is widely credited as a sig- nificant factor for influencing the attitude of the audience towards important policy matters (Chong and Druckman, 2007). In the era of digitalization, as media articles wield more power than ever to sway and to mould public opinion, we believe it is worth an attempt to foray into how multilingual news articles frame prominent topics for consumption by their audi- ences. In essence, in the media context, frame is the perspective under which a piece of news is pre- sented. How a piece of information is framed is critical for determining how the information will be interpreted by the audience. Analyzing the frames used by the media and their impact on public per- ception is challenging due to language dynamics and the nuances of wording. The goal of SemEval-2023 task 3 subtask 2 (Piskorski et al., 2023) is to predict the frames used in a given news article, drawing from a set of 14 possible frames from (Card et al., 2015). This is a multi-label classification task at the article level. Our hypothesis is that despite linguistic differences, different languages use similar framing in news ar- ticles, and that framingremains consistent despite information loss during translation. Our results pro- vide evidence in support of this hypothesis, in the form of zero-shot transfer for Greek and Georgian. This suggests that the frames used in online news articles remain intact, despite information loss dur- ing translation, and that these techniques may be similar across languages. Our system first translates the input article to English, and then applies an English model on the translated input. On the official test set, it achieved micro-D1 scores of 53% for Greek and …

Content truncated to 500 words