Guardado en:
Detalles Bibliográficos
Autores principales: Yang, Zhipeng, Yang, Shu, Hu, Lijie, Wang, Di
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2603.10771
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918383429091328
author Yang, Zhipeng
Yang, Shu
Hu, Lijie
Wang, Di
author_facet Yang, Zhipeng
Yang, Shu
Hu, Lijie
Wang, Di
contents 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 introduce a decoding-based method to detect word recovery, showing that hidden states reconstruct canonical word-level token identities from character-level inputs. We then provide causal evidence by removing the corresponding subspace from hidden states, which consistently degrades downstream task performance. Finally, we conduct a fine-grained attention analysis and show that in-group attention among characters belonging to the same canonical token is critical for word recovery: masking such attention in early layers substantially reduces both recovery scores and task performance. Together, our findings provide a mechanistic explanation for tokenization robustness and identify word recovery as a key mechanism enabling LLMs to process character-level inputs.
format Preprint
id arxiv_https___arxiv_org_abs_2603_10771
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Word Recovery in Large Language Models Enables Character-Level Tokenization Robustness
Yang, Zhipeng
Yang, Shu
Hu, Lijie
Wang, Di
Computation and Language
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 introduce a decoding-based method to detect word recovery, showing that hidden states reconstruct canonical word-level token identities from character-level inputs. We then provide causal evidence by removing the corresponding subspace from hidden states, which consistently degrades downstream task performance. Finally, we conduct a fine-grained attention analysis and show that in-group attention among characters belonging to the same canonical token is critical for word recovery: masking such attention in early layers substantially reduces both recovery scores and task performance. Together, our findings provide a mechanistic explanation for tokenization robustness and identify word recovery as a key mechanism enabling LLMs to process character-level inputs.
title Word Recovery in Large Language Models Enables Character-Level Tokenization Robustness
topic Computation and Language
url https://arxiv.org/abs/2603.10771