AI

Multilingual GSM-Symbolic: What determines capability transfer across languages?

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 12:18:24 UTC.

Briefings

Multilingual GSM-Symbolic is a new dataset of 30,000 question-answer pairs across 15 languages designed to study cross-lingual capability transfer. Researchers identified four key determinants of transfer: model size (beta=1.77), language resource level (beta=0.77), reasoning ability (beta=0.67), and typological distance (beta=-0.25). A 32B model in Marathi performs similarly to a 10B model in English, suggesting model size and reasoning are critical for briding resource gaps. The framework explains 92% of between-language variation but only 23% of model-by-language variation, with performance predictions within 6.0pp on unseen languages.

Key points

• Multilingual GSM-Symbolic is an extensible multilingual mathematical dataset containing 30,000 item-matched question-answer pairs spanning 15 languages.
• Symbolic templates prevent overfitting by enabling the generation of millions of high-quality variations from a single sample.

Context and limitations

Evaluation relies on incomparable, saturation-prone datasets that rarely examine determinants jointly. Coverage is limited to 15 languages and 30,000 items, with no downstream dataset required for estimates.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 07/10/2026, 01:36:52 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeigkkg64n3qn3ybnjd44tfqc5vx3hpdy64zsufygqq36nfu4dgtr4a