arXiv:2605.30481v1 Announce Type: new Abstract: Large language models (LLMs) are widely used as cross-lingual knowledge interfaces. However, culturally grounded questions often reflect globally dominant narratives rather than local contexts. We study this failure mode as \textit{global narrative dominance} in Bangla, a low-resource cultural context. We introduce \texttt{CulturalNB}, a dataset of 717 manually curated Bengali cultural instances with parallel Bangla--English question--answer pairs and supporting evidence, metadata, and sociocultural annotations. Using question-only and evidence-b
The proliferation of LLMs as global knowledge interfaces necessitates a deeper examination of their inherent biases and the dominance of specific cultural narratives, especially as their use expands into diverse linguistic and cultural contexts.
This research highlights a critical flaw in current large language models, where culturally specific knowledge is overlooked or misrepresented, impacting their reliability and equity in a global context.
The understanding of LLM limitations is deepened, pushing for more culturally sensitive data curation and model development, and potentially influencing how multi-lingual and multi-cultural AI is approached.
- · Ethical AI researchers
- · Low-resource language communities
- · Developers of culturally-aware AI models
- · Uncritically deployed LLMs
- · Monolingual AI development methodologies
- · Users relying solely on LLMs for local knowledge
Increased demand for culturally representative training data and fine-tuning techniques for LLMs.
Development of specialized LLMs for specific cultural and linguistic contexts to counter global narrative dominance.
Potential for a 'digital divide' based on cultural and linguistic representation within powerful AI models, leading to new forms of information inequality.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL