# Spelling-out is not Straightforward: LLMs' Capability of Tokenization from Token to Characters

arXiv: https://arxiv.org/abs/2506.10641 · HTML: https://mlsanity.com/paper/2506.10641/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 34 references that are in the map.
- First posted on arXiv 12 Jun 2025.

## Builds on (most cited first, 25 of 25)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Gemma: Open Models Based on Gemini Research and Technology](https://mlsanity.com/paper/2403.08295/) (2024)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [The Remarkable Robustness of LLMs: Stages of Inference?](https://mlsanity.com/paper/2406.19384/) (2024)
- [LLM360: Towards Fully Transparent Open-Source LLMs](https://mlsanity.com/paper/2312.06550/) (2023)
- [Language Models Implement Simple Word2Vec-style Vector Arithmetic](https://mlsanity.com/paper/2305.16130/) (2023)
- [Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristics](https://mlsanity.com/paper/2410.21272/) (2024)
- [Neurons in Large Language Models: Dead, N-gram, Positional](https://mlsanity.com/paper/2309.04827/) (2023)
- [What does the Knowledge Neuron Thesis Have to do with Knowledge?](https://mlsanity.com/paper/2405.02421/) (2024)
- [From Tokens to Words: On the Inner Lexicon of LLMs](https://mlsanity.com/paper/2410.05864/) (2024)
- [Probing Classifiers are Unreliable for Concept Removal and Detection](https://mlsanity.com/paper/2207.04153/) (2022)
- [Why Do Large Language Models (LLMs) Struggle to Count Letters?](https://mlsanity.com/paper/2412.18626/) (2024)
- [What do tokens know about their characters and how do they know it?](https://mlsanity.com/paper/2206.02608/) (2022)
- [CUTE: Measuring LLMs' Understanding of Their Tokens](https://mlsanity.com/paper/2409.15452/) (2024)
- [Models In a Spelling Bee: Language Models Implicitly Learn the Character Composition of Tokens](https://mlsanity.com/paper/2108.11193/) (2021)
- [Large Language Models Lack Understanding of Character Composition of Words](https://mlsanity.com/paper/2405.11357/) (2024)
- [The Impact of Visual Information in Chinese Characters: Evaluating Large Models' Ability to Recognize and Utilize Radicals](https://mlsanity.com/paper/2410.09013/) (2024)
- [Word Form Matters: LLMs' Semantic Reconstruction under Typoglycemia](https://mlsanity.com/paper/2503.01714/) (2025)
- [Knowledge of Pretrained Language Models on Surface Information of Tokens](https://mlsanity.com/paper/2402.09808/) (2024)
- [Weight-based Analysis of Detokenization in Language Models: Understanding the First Stage of Inference Without Inference](https://mlsanity.com/paper/2501.15754/) (2025)
- [Enhancing Character-Level Understanding in LLMs through Token Internal Structure Learning](https://mlsanity.com/paper/2411.17679/) (2024)

## Built on by (most cited first, 4 of 4)

- [Stochasticity in Tokenisation Improves Robustness](https://mlsanity.com/paper/2604.16037/) (2026)
- [TASE: Token Awareness and Structured Evaluation for Multilingual Language Models](https://mlsanity.com/paper/2508.05468/) (2025)
- [SubTokenTest: A Practical Benchmark for Real-World Sub-token Understanding](https://mlsanity.com/paper/2601.09089/) (2026)
- [Correct Choices, Failed Sequences: Choice2Seq Benchmarks Correctness Retention Across Output Interfaces](https://mlsanity.com/paper/2610.16191/) (2026)

## Abstract (arXiv)

Large language models (LLMs) can spell out tokens character by character with high accuracy, yet they struggle with more complex character-level tasks, such as identifying compositional subcomponents within tokens. In this work, we investigate how LLMs internally represent and utilize character-level information during the spelling-out process. Our analysis reveals that, although spelling out is a simple task for humans, it is not handled in a straightforward manner by LLMs. Specifically, we show that the embedding layer does not fully encode character-level information, particularly beyond the first character. As a result, LLMs rely on intermediate and higher Transformer layers to reconstruct character-level knowledge, where we observe a distinct "breakthrough" in their spelling behavior. We validate this mechanism through three complementary analyses: probing classifiers, identification of knowledge neurons, and inspection of attention weights.