Robust Detection of Persuasion Techniques in Slavic Languages via Multitask Debiasing and Walking Embeddings

Our approach to Subtask 1 integrates fine-tuned multilingual transformer models with two complementary robustness-oriented strategies: Walking Embeddings and Content-Debiasing.With the first, we tried to understand the change in embeddings when various manipulation techniques were applied.The latter leverages a supervised contrastive objective over semantically equivalent yet stylistically divergent text pairs, generated via GPT-4.We conduct extensive experiments, including 5-fold crossvalidation and out-of-domain evaluation, and explore the impact of contrastive loss weighting.