arXiv:2602.00945v2 Announce Type: replace Abstract: LLMs are multilingual by training, yet their lingua franca is often English, reflecting English language dominance in pretraining. Other languages remain in parametric memory but are systematically suppressed. We argue that language defaultness is governed by a sparse, low-rank control circuit, language neurons, that can be mechanistically isolated and safely steered. We introduce Neural FOXP2, that makes a chosen language (Hindi or Spanish) primary in a model by steering language-specific neurons. Neural FOXP2 proceeds in three stages: (i) L
The increasing multilingual deployment of LLMs and recognition of English dominance necessitates methods to balance language representation effectively.
This development allows for targeted linguistic steering in AI models, potentially reducing bias and increasing utility for non-English speakers and markets.
LLMs can now be more explicitly reconfigured to prioritize specific languages, moving beyond default English dominance without extensive retraining.
- · Non-English language users
- · Multilingual AI developers
- · Generative AI sector
- · Governments promoting local languages
- · English-centric AI products
LLMs will become more linguistically adaptable to diverse global markets.
This could lead to a proliferation of AI tools optimized for specific languages, fostering digital inclusivity.
Nations might invest more in language-specific AI development to preserve cultural identity and improve access to AI services.
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