Rationale-Guided Knowledge Distillation for Cross-Lingual Stance Detection

2026-07-21

AI Summary

This work addresses the challenge of stance detection in low-resource languages, where scarce annotated data hinders the training of reliable models, and existing cross-lingual approaches often neglect the reasoning process, limiting performance. To overcome this, the authors propose a rationale-guided knowledge distillation framework that leverages large language models with chain-of-thought prompting to generate explicit reasoning rationales for stance predictions. These rationales are then effectively transferred to a lightweight student model through dual-path representation alignment and two contrastive learning strategies. Evaluated on multilingual benchmarks, the proposed method significantly outperforms strong baselines and substantially improves cross-lingual stance detection performance for low-resource languages.

Abstract

Stance detection aims to identify whether a text expresses a favorable or opposing attitude toward a given target, and serves as an important task for various downstream applications. Although existing studies have achieved strong performance in monolingual settings, especially in English, many low-resource languages such as Catalan still lack sufficient annotated data for training effective models. Cross-lingual stance detection alleviates this problem by transferring stance knowledge from resource-rich languages to low-resource languages. However, most existing methods mainly rely on semantic alignment between texts and targets, while ignoring the reasoning process required for reliable stance inference. Although Large Language Models provide strong reasoning ability, their high computational cost and inference latency limit practical deployment. To address these limitations, we propose a rationale-guided knowledge distillation framework for cross-lingual stance detection. Specifically, we use Chain-of-Thought prompting to guide Large Language Models in generating informative rationales, and distill the resulting reasoning knowledge into a compact student model. We further design a dual-path distillation mechanism to align rationale-enhanced and rationale-free representations, together with their prediction distributions. In addition, two contrastive learning strategies are introduced to improve stance discrimination. Experiments on multilingual benchmarks demonstrate that our method consistently outperforms competitive baselines.

Problem

• cross-lingual stance detection
• low-resource languages
• reasoning process
• knowledge distillation
• stance detection

Innovation

• Rationale-Guided Knowledge Distillation
• Chain-of-Thought Prompting
• Cross-Lingual Stance Detection
• Dual-Path Distillation
• Contrastive Learning