# Writing-System-Level Tokenizer Adaptation for Byte-Level BPE

arXiv: https://arxiv.org/abs/2608.00582 · HTML: https://mlsanity.com/paper/2608.00582/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 17 of its 27 references that are in the map.
- First posted on arXiv 1 Aug 2026.

## Builds on (most cited first, 17 of 17)

- [How Good is Your Tokenizer? On the Monolingual Performance of Multilingual Language Models](https://mlsanity.com/paper/2012.15613/) (2020)
- [Language Model Tokenizers Introduce Unfairness Between Languages](https://mlsanity.com/paper/2305.15425/) (2023)
- [Adapting Pre-trained Language Models to African Languages via Multilingual Adaptive Fine-Tuning](https://mlsanity.com/paper/2204.06487/) (2022)
- [Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontier](https://mlsanity.com/paper/2412.04261/) (2024)
- [Extending Multilingual BERT to Low-Resource Languages](https://mlsanity.com/paper/2004.13640/) (2020)
- [XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models](https://mlsanity.com/paper/2301.10472/) (2023)
- [Fishing for Magikarp: Automatically Detecting Under-trained Tokens in Large Language Models](https://mlsanity.com/paper/2405.05417/) (2024)
- [Overlap-based Vocabulary Generation Improves Cross-lingual Transfer Among Related Languages](https://mlsanity.com/paper/2203.01976/) (2022)
- [Embedding structure matters: Comparing methods to adapt multilingual vocabularies to new languages](https://mlsanity.com/paper/2309.04679/) (2023)
- [Efficient and Effective Vocabulary Expansion Towards Multilingual Large Language Models](https://mlsanity.com/paper/2402.14714/) (2024)
- [Bielik 7B v0.1: A Polish Language Model -- Development, Insights, and Evaluation](https://mlsanity.com/paper/2410.18565/) (2024)
- [Tokenization is Sensitive to Language Variation](https://mlsanity.com/paper/2502.15343/) (2025)
- [Tik-to-Tok: Translating Language Models One Token at a Time: An Embedding Initialization Strategy for Efficient Language Adaptation](https://mlsanity.com/paper/2310.03477/) (2023)
- [Beyond Text Compression: Evaluating Tokenizers Across Scales](https://mlsanity.com/paper/2506.03101/) (2025)
- [Model-Aware Tokenizer Transfer](https://mlsanity.com/paper/2510.21954/) (2025)
- [From English-Centric to Effective Bilingual: LLMs with Custom Tokenizers for Underrepresented Languages](https://mlsanity.com/paper/2410.18836/) (2024)
- [Achieving Tokenizer Flexibility in Language Models through Heuristic Adaptation and Supertoken Learning](https://mlsanity.com/paper/2505.09738/) (2025)

## Built on by (most cited first, 1 of 1)

- [Dynamic-Programming-Guided Hierarchical BPE and Empirical Analysis of Vocabulary Pruning](https://mlsanity.com/paper/2609.06898/) (2026)

## Abstract (arXiv)

Pretrained byte-level BPE tokenizers can segment underrepresented languages inefficiently. Replacing a tokenizer changes the meaning of nearly every token ID, while vocabulary expansion enlarges the model's embedding and output matrices. We study post-hoc adaptation that keeps the model-vocabulary size fixed and preserves most existing token-to-ID assignments as a construction-time compatibility property. Directly transferring tokens from a language-specific tokenizer does not guarantee derivability through the target BPE merge graph: an inserted entry can conflict with the target's greedy merge ranks. We formalize this failure as the merge ordering problem and introduce BPE-guided insertion, which builds each transferred token through a target-reachable decomposition. Our pipeline uses script-aware row selection to limit collateral fragmentation, reconstructs target-script byte-level prerequisites, and applies guided insertion to maintain merge-graph reachability. On Ukrainian adaptations of Nemotron and GPT-OSS, it reduces token counts by 33.5% and 36.6%, keeps changes on English and the evaluated four-language European aggregate within 0.05%, and retains 78.5%/77.3% of original model-vocabulary rows at the same IDs. Constraint-matched global and frequency-based removal achieve similar Ukrainian compression but increase English/European token counts by 0.7-2.2%; fresh same-size retraining compresses Ukrainian slightly more but retains effectively no same-ID rows and increases English token counts by 7.6-8.6%. The reallocation increases token counts on the evaluated three-language Cyrillic micro-aggregate by 6.7%/10.1%. Structural audits find all 28,134/45,398 inserted BPE nodes reachable under ordinary rank-ordered merging and no retained same-ID model-vocabulary entry newly broken. We release all tokenizers and code.