Word Recovery in Large Language Models Enables Character-Level Tokenization Robustness

ArXi:2603.10771v1 Announce Type: new Large language models (LLMs) trained with canonical tokenization exhibit surprising robustness to non-canonical inputs such as character-level tokenization, yet the mechanisms underlying this robustness remain unclear. We study this phenomenon through mechanistic interpretability and identify a core process we term word recovery. We first