arXiv NLP
techCenter
Lost in Reordering: Structural Sensitivity of Multilingual LLMs under Semantics-Preserving Perturbationstranslating…
arXiv:2609.03511v1 Announce Type: new
Abstract: Large Language Models (LLMs) demonstrate strong multilingual reasoning performance, yet their robustness to semantics-preserving structural variation remains underexplored, particularly for relatively free word-order languages. We…
Keywords#LLMs#Multilingual#Semantics-Preserving#Structural#Lost
Comments
Sign in to join the discussion.
Loading comments…