Reproduce “Lost in the Tower of Babel: The Adverse Effects of Incidental Multilingualism in LLMs” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Anjishnu Mukherjee, Chutong Meng, Antonios Anastasopoulos

The authors examine whether frontier language models provide dependable multilingual support, distinguishing stated support, checked language output, and task performance. Across five models, they test paraphrased support questions and multilingual tasks; Llama 4 Maverick’s verified language union spans 134 languages, but only English appears across all 20 variants and reruns. A seven-agent language-switching test lowers accuracy without isolating coordination effects.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.