Title: Lingo_Research_Group’s Approach to SemEval-2026 Task 9: Assessing Prompt Variations in Polarization Detection

Abstract:

This paper details our entry for SemEval-2026 Task 9, the Multilingual Text Classification Challenge focused on Polarization Detection. Our submission addresses all three designated subtasks: binary polarization detection, polarization type classification, and polarization manifestation identification. We implemented a rigorous research methodology centered on the evaluation of twelve specifically crafted short prompts. These prompts varied according to several criteria, including the clarity of terminology, the depth of definition provided, the extent of reasoning guidance, and the inclusion of in-context examples.

For our experiments, we utilized the aya-101 model and Gemma3-27B. Based on performance metrics observed during the development phase, we selected Gemma3-27B for our final submission. On the official test set, which encompassed 22 languages, our system achieved the following average scores: an F1-score of 0.762 for Subtask 1, 0.587 for Subtask 2, and 0.444 for Subtask 3. Correspondingly, the average accuracy rates were 0.819, 0.678, and 0.498, respectively. Through cross-task and cross-lingual analysis, we illustrate that while prompt-based methods are effective for identifying coarse-grained polarization, they face increasing challenges when applied to fine-grained, multi-label sociolinguistic classification.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC