XplaiNLP at CheckThat! 2025: Multilingual Subjectivity Detection with Finetuned Transformers and Prompt-Based Inference with Large Language Models
This work addresses subjectivity detection for multilingual content analysis, but it is incremental as it applies existing methods to a shared task.
The paper tackled multilingual subjectivity detection by evaluating fine-tuned transformers and prompt-based LLMs, achieving top results in Italian (F1 0.8104 vs. baseline 0.6941) and competitive scores in Romanian (F1 0.7917 vs. baseline 0.6461), but struggled in low-resource languages like Ukrainian and Polish.
This notebook reports the XplaiNLP submission to the CheckThat! 2025 shared task on multilingual subjectivity detection. We evaluate two approaches: (1) supervised fine-tuning of transformer encoders, EuroBERT, XLM-RoBERTa, and German-BERT, on monolingual and machine-translated training data; and (2) zero-shot prompting using two LLMs: o3-mini for Annotation (rule-based labelling) and gpt-4.1-mini for DoubleDown (contrastive rewriting) and Perspective (comparative reasoning). The Annotation Approach achieves 1st place in the Italian monolingual subtask with an F_1 score of 0.8104, outperforming the baseline of 0.6941. In the Romanian zero-shot setting, the fine-tuned XLM-RoBERTa model obtains an F_1 score of 0.7917, ranking 3rd and exceeding the baseline of 0.6461. The same model also performs reliably in the multilingual task and improves over the baseline in Greek. For German, a German-BERT model fine-tuned on translated training data from typologically related languages yields competitive performance over the baseline. In contrast, performance in the Ukrainian and Polish zero-shot settings falls slightly below the respective baselines, reflecting the challenge of generalization in low-resource cross-lingual scenarios.