Auditing Reciprocal Sentiment Alignment: Inversion Risk, Dialect Representation and Intent Misalignment in Transformers

2026-02-19

AI Summary

This study addresses the risk of misaligned sentiment representations in cross-lingual (Bengali–English) AI systems, which can lead to erroneous intent interpretation and erode human–AI trust. The authors construct an evaluation framework based on multilingual Transformer models—including mDistilBERT and IndicBERT—to quantify, for the first time, phenomena such as “sentiment reversal rate” and “asymmetric empathy.” Their analysis reveals significant representational gaps and safety risks in current alignment methods when applied to low-resource languages and dialectal variants. Notably, compressed models exhibit a sentiment reversal rate of 28.7%, while IndicBERT shows a 57% increase in alignment error on formal Bengali, underscoring how generic compression strategies compromise affective fidelity. The work advocates incorporating “sentiment stability” into alignment evaluation protocols and calls for culturally attuned sentiment alignment mechanisms.

Technology Category

• Natural Language Processing: Safety and Robustness
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

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.

Problem

• Cross-Lingual Sentiment Misalignment
• Sentiment Inversion
• Dialect Representation
• Affective Stability
• Reciprocal Trust

Innovation

• Sentiment Inversion Rate
• Asymmetric Empathy
• Modern Bias
• Affective Stability
• Cross-Lingual Alignment