LocuPrompt at SemEval-2026 Task 7: A Multilingual Prompting Framework for Cross-Cultural Everyday Knowledge in LLMs

AbstractUnderstanding everyday cultural knowledge remains a fundamental challenge for large language models (LLMs). This paper presents LocuPrompt, a multilingual framework for SemEval-2026 Task 7: Everyday Knowledge Across Diverse Languages and Cultures. To address Short Answer Questions (SAQ), we employ an English-pivot generation strategy with back-translation, combined with empirical locale-specific routing that dynamically assigns the optimal LLM to each target region. For Multiple-Choice Questions (MCQ), we apply parameter-efficient fine-tuning to a robust multilingual base model, utilizing locale-aware instructions that frame the LLM as a “local resident.” By integrating strategic model selection with resource-efficient adaptation, LocuPrompt effectively bridges cross-lingual cultural gaps while maintaining a fully reproducible pipeline.