PAST-TIDE: Prototype-Anchored Statement Tuning with Topic-Invariant Normalization for Stance Detection

2026-07-06

AI Summary

This work addresses the low-resource challenge in cross-topic stance detection for Arabic by reframing stance identification and target linking as a unified fill-in-the-blank masked language modeling (MLM) task, thereby eliminating the need for additional classification heads. The approach introduces two key innovations: a contrastive learning mechanism based on learnable class prototypes to enhance semantic discriminability, and a topic-conditional layer normalization strategy to improve generalization across topics. By fine-tuning only the MLM head of a pretrained language model to map labels into natural language prompts, the method achieves macro F1 scores of 0.75 and 0.74 on Subtasks A and B, respectively, in the StanceNakba shared task—substantially outperforming existing low-resource baselines.

Abstract

We introduce PAST-TIDE, our stance detection system addressing both subtasks of the StanceNakba Shared Task at NakbaNLP@LREC-COLING 2026. The main idea is statement tuning. We redefine stance as cloze-style masked language modeling (MLM), letting a verbalizer map label words to stance categories through the pre-trained MLM head rather than appending a randomly initialized classification head. We complement this with prototypical contrastive learning, which uses learnable class prototypes for batch-size independent contrastive training, and topic-conditional layer normalization for cross-topic Arabic stance detection. PAST-TIDE achieves macro-F1 scores of 0.75 for Subtask A and 0.74 for Subtask B on the official leaderboard, indicating that minimal architectural additions to a pre-trained model can remain competitive in low-resource settings.

Problem

• stance detection
• cross-topic
• Arabic
• low-resource
• shared task

Innovation

• statement tuning
• prototypical contrastive learning
• topic-invariant normalization
• masked language modeling
• stance detection