@inproceedings{bibcite_1091,
  author = {Supantho Rakshit and Adele Goldberg and Henry Conklin},
  title = {Similarity All The Way Up: Multilingual Generalization in LLMs Relies on Language-Level Similarity Structures},
  abstract = {<p>As Large Language Models (LLMs) become more and more successful at a wide range of tasks, their (in)ability to generalize remains difficult to quantify and poorly understood beyond limited domains. In particular, LLMs are known to struggle generalizing multilingually, to languages \&nbsp;<br>\&nbsp;poorly attested in their training data. To understand why this may be, and what enables some models to perform better than others, we turn to a long history of work across the cognitive sciences, arguing that successful generalization derives from appropriate representations in similarity space. We look at how well LLMs{\textquoteright} representations capture the hierarchical similarity among distinct languages. Strikingly, we show LLMs{\textquoteright} representations largely recover the hierarchical structure of the Indo-European language family tree -- grouping languages that are members of the same subfamily closely together in representation space. Furthermore, we show that the degree to which models reflect the similarity structure of \&nbsp;languages correlates with their performance on XNLI, a multilingual natural language inference benchmark. This extends classic work on similarity-driven generalization at scale, showing how models that represent similar languages similarly generalize better from one language to another.</p>},
  year = {2026},
  journal = {Proceedings of the Cognitive Science Society},
}