Search papers, labs, and topics across Lattice.
This paper benchmarks cross-lingual sentiment alignment between Bengali and English using four transformer architectures, revealing significant failures in current alignment paradigms. It identifies issues such as sentiment inversion, asymmetric empathy (differential affective weighting), and modern bias (dialectal sensitivity). The study highlights the need for culturally grounded alignment that respects language and dialectal diversity to ensure reciprocal human-AI trust.
Current cross-lingual models can disastrously invert sentiment when translating between English and Bengali, with one model flipping positive intent to negative nearly 30% of the time.
The core theme of bidirectional alignment is ensuring that AI systems accurately understand human intent and that humans can trust AI behavior. However, this loop fractures significantly across language barriers. Our research addresses Cross-Lingual Sentiment Misalignment between Bengali and English by benchmarking four transformer architectures. We reveal severe safety and representational failures in current alignment paradigms. We demonstrate that compressed model (mDistilBERT) exhibits 28.7% "Sentiment Inversion Rate," fundamentally misinterpreting positive user intent as negative (or vice versa). Furthermore, we identify systemic nuances affecting human-AI trust, including "Asymmetric Empathy" where some models systematically dampen and others amplify the affective weight of Bengali text relative to its English counterpart. Finally, we reveal a "Modern Bias" in the regional model (IndicBERT), which shows a 57% increase in alignment error when processing formal (Sadhu) Bengali. We argue that equitable human-AI co-evolution requires pluralistic, culturally grounded alignment that respects language and dialectal diversity over universal compression, which fails to preserve the emotional fidelity required for reciprocal human-AI trust. We recommend that alignment benchmarks incorporate "Affective Stability" metrics that explicitly penalize polarity inversions in low-resource and dialectal contexts.
