Tag
This paper investigates the structural sensitivity of multilingual large language models to semantics-preserving perturbations in Hindi and Malayalam, showing significant degradation in mathematical reasoning performance and introducing the IndicReStruct benchmark for evaluation.
This paper investigates whether text generated by multilingual large language models exhibits signs of translationese, and compares it to human-written translations to assess linguistic naturalness.
HOTFIXR is a data generation framework that targets multilingual reasoning weaknesses in LLMs to improve cross-lingual performance using synthetic data without sacrificing overall capability.
This paper explores cross-lingual transfer of internal representations for figurative language generation in multilingual LLMs, showing that activation directions learned in one language can effectively steer generation in other languages.
Benchmarks seven foundation models on Ukrainian legal text, finding tokenizer fertility varies 1.6×, few-shot prompting degrades performance, and cost-performance analysis shows NVIDIA Nemotron Super 3 outperforms larger models.
Researchers from National Taiwan University propose replacing fixed translation-based prompting strategies in multilingual LLMs with lightweight learned classifiers that route each instance to either native or translation-based prompting. Their analysis across 10 languages and 4 benchmarks shows no single strategy is universally optimal, with translation benefiting low-resource languages most, and the learned routing achieving statistically significant improvements over fixed strategies.