Which Language Should a Skeleton Speak? Language Choices in Multilingual Reasoning
LASEF finds English reasoning skeletons help smaller models slightly, but language choice is context-dependent.
LASEF studies which language to use for skeleton-based reasoning prompts in multilingual mathematical reasoning, a training-free method previously evaluated mainly in English. Across math benchmarks, model scales, and languages, English skeletons show a small average benefit, most visible for smaller models and low-resource languages. Few language-level gains remain significant after correction, and English is not universally optimal. The authors report directionally consistent, evaluation-dependent, and asymmetric negative effects that generation quality alone does not explain, and they release resources on GitHub.
- English skeletons show a small average benefit for smaller models
- Few language-level gains stay significant after statistical correction
- English is not universally optimal across languages and benchmarks
- Observed effects are consistent, benchmark-dependent, or asymmetrically negative
From the original article
1 min read at huggingface.co
Skeleton-based reasoning prompting is a promising training-free approach for structuring LLM reasoning, but prior work largely assumes an English-centric setting. We propose the Language-Aware Skeleton Exploration Framework (LASEF) to study skeleton-language choice in multilingual mathematical reasoning.
By HyeonSeok Lim, SeungWoo Song, Inho Won · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.