LLMs' multilingualism is incidental, leading to unequal and brittle performance

A new paper argues that current large language models (LLMs) achieve multilingual capabilities incidentally through massive, uneven web data, rather than through intentional design for multilingual competence. This "incidental multilingualism" leads to unequal, brittle, and opaque performance across languages, posing risks in real-world applications. The authors propose a shift towards "multilingualism by design," prioritizing equitable performance, cultural grounding, and cross-lingual understanding as core objectives. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Highlights potential risks in agentic deployments due to incidental multilingualism, calling for a more deliberate approach to cross-lingual AI development.

RANK_REASON This is a research paper published on arXiv discussing the limitations of current LLMs regarding multilingualism. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• LLMs
• arXiv
• Anjishnu Mukherjee

• paper
• safety