This project investigates the cross-cultural dynamics of sarcasm in news headlines by comparing Western and Singaporean media. We aim to assess the generalizability of sarcasm detection models—particularly those trained on Western datasets—when applied to Singaporean headlines.

By examining linguistic patterns, contextual cues, and cultural references, we explore whether existing models can accurately detect sarcasm across different socio-cultural environments. Additionally, we conduct transformation-based testing to uncover biases and misclassifications rooted in location-specific language.

Through this comparative analysis, the study contributes to the advancement of cross-cultural natural language processing and promotes AI fairness by identifying limitations in current sarcasm detection systems and proposing paths for more inclusive, culturally aware models.