This paper describes our system for POLAR Subtask 1 on multilingual polarization detection. The task involves binary sequence classification over 22 languages, where the model aims to predict whether a given text exhibits polarized discourse. To deal with the multilingual and resource-imbalanced nature of the dataset, we fine-tune the XLM-R, a pre-trained multilingual transformer encoder, using a language-aware sampling strategy that combines all available training data into a unified multilingual corpus. Our system achieves an overall macro-F1 of 0.781 and an average accuracy of 0.823 on the official test set. Results show strong performance in low-resource languages, though some discrepancies indicate remaining class imbalance.

Sarcasm detection is a significant challenge in sentiment analysis due to the nuanced and context-dependent nature of verbiage. We introduce Pragmatic Metacognitive Prompting (PMP) to improve the performance of Large Language Models (LLMs) in sarcasm detection, which leverages principles from pragmatics and reflection helping LLMs interpret implied meanings, consider contextual cues, and reflect on discrepancies to identify sarcasm. Using state-of-the-art LLMs such as LLaMA-3-8B, GPT-4o, and Claude 3.5 Sonnet, PMP achieves state-of-the-art performance on GPT-4o on MUStARD and SemEval2018. This study demonstrates that integrating pragmatic reasoning and metacognitive strategies into prompting significantly enhances LLMs’ ability to detect sarcasm, offering a promising direction for future research in sentiment analysis.