@inproceedings{al-hakim-etal-2026-mitigating,
title = "Mitigating the Impact of Glitch Tokens via Targeted Retokenization",
author = "Al Hakim, Muhammad Alif and
Kuzmin, Dmitrii and
Zavala, Carlos Javier Gal{\'a}n and
Hiraoka, Tatsuya and
Nwadike, Munachiso Samuel and
Niwa, Ayana and
Cognetta, Marco and
Pohl, David and
Inui, Kentaro",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 2: Short Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-short.13/",
pages = "155--169",
ISBN = "979-8-89176-440-8",
abstract = "This paper examines whether avoiding undertrained or ``glitch'' tokens and maximizing tokenization quality during inference can improve downstream task performance without retraining the model. We evaluate this question using a retokenization method that selects the highest-quality tokenization based on the trainedness of tokens, which is defined through the norm or entropy of token embeddings. We compare the performance difference between the tokenization that includes undertrained tokens and the one that avoids such tokens on multiple benchmarks. Our findings reveal a key nuance: undertrained tokens degrade performance primarily when they occupy semantically salient positions that are critical for solving the task. Consequently, retokenization preserves performance on ordinary inputs, while yielding consistent performance improvements when such tokens appear in high-importance contexts."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="al-hakim-etal-2026-mitigating">
<titleInfo>
<title>Mitigating the Impact of Glitch Tokens via Targeted Retokenization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Muhammad</namePart>
<namePart type="given">Alif</namePart>
<namePart type="family">Al Hakim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dmitrii</namePart>
<namePart type="family">Kuzmin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carlos</namePart>
<namePart type="given">Javier</namePart>
<namePart type="given">Galán</namePart>
<namePart type="family">Zavala</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tatsuya</namePart>
<namePart type="family">Hiraoka</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Munachiso</namePart>
<namePart type="given">Samuel</namePart>
<namePart type="family">Nwadike</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayana</namePart>
<namePart type="family">Niwa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Cognetta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Pohl</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kentaro</namePart>
<namePart type="family">Inui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-440-8</identifier>
</relatedItem>
<abstract>This paper examines whether avoiding undertrained or “glitch” tokens and maximizing tokenization quality during inference can improve downstream task performance without retraining the model. We evaluate this question using a retokenization method that selects the highest-quality tokenization based on the trainedness of tokens, which is defined through the norm or entropy of token embeddings. We compare the performance difference between the tokenization that includes undertrained tokens and the one that avoids such tokens on multiple benchmarks. Our findings reveal a key nuance: undertrained tokens degrade performance primarily when they occupy semantically salient positions that are critical for solving the task. Consequently, retokenization preserves performance on ordinary inputs, while yielding consistent performance improvements when such tokens appear in high-importance contexts.</abstract>
<identifier type="citekey">al-hakim-etal-2026-mitigating</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-short.13/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>155</start>
<end>169</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mitigating the Impact of Glitch Tokens via Targeted Retokenization
%A Al Hakim, Muhammad Alif
%A Kuzmin, Dmitrii
%A Zavala, Carlos Javier Galán
%A Hiraoka, Tatsuya
%A Nwadike, Munachiso Samuel
%A Niwa, Ayana
%A Cognetta, Marco
%A Pohl, David
%A Inui, Kentaro
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-440-8
%F al-hakim-etal-2026-mitigating
%X This paper examines whether avoiding undertrained or “glitch” tokens and maximizing tokenization quality during inference can improve downstream task performance without retraining the model. We evaluate this question using a retokenization method that selects the highest-quality tokenization based on the trainedness of tokens, which is defined through the norm or entropy of token embeddings. We compare the performance difference between the tokenization that includes undertrained tokens and the one that avoids such tokens on multiple benchmarks. Our findings reveal a key nuance: undertrained tokens degrade performance primarily when they occupy semantically salient positions that are critical for solving the task. Consequently, retokenization preserves performance on ordinary inputs, while yielding consistent performance improvements when such tokens appear in high-importance contexts.
%U https://aclanthology.org/2026.aacl-short.13/
%P 155-169
Markdown (Informal)
[Mitigating the Impact of Glitch Tokens via Targeted Retokenization](https://aclanthology.org/2026.aacl-short.13/) (Al Hakim et al., AACL-IJCNLP 2026)
ACL
- Muhammad Alif Al Hakim, Dmitrii Kuzmin, Carlos Javier Galán Zavala, Tatsuya Hiraoka, Munachiso Samuel Nwadike, Ayana Niwa, Marco Cognetta, David Pohl, and Kentaro Inui. 2026. Mitigating the Impact of Glitch Tokens via Targeted Retokenization. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers), pages 155–169, Hengqin, China. Association for Computational Linguistics.