@inproceedings{lin-etal-2026-identity,
title = "Identity-Preserving Detoxification via Local Causal Effect Ranking on Masked Diffusion Language Models",
author = "Lin, Yucheng and
Liu, Yue and
Jiang, Jiaojiao",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.234/",
pages = "5360--5378",
ISBN = "979-8-89176-459-0",
abstract = "Text detoxification should produce safer rewrites while retaining identity mentions from the original text. However, classifier-guided methods can remove identity descriptors alongside genuinely harmful content when toxicity classifiers associate identity mentions with harmfulness and guidance relies on absolute scores. We propose CIDER (Contrastive Identity-preserving Detoxification via Explicit Reference), a default-relative, margin-gated selective-guidance method for masked diffusion language models. At each selected masked position, CIDER evaluates every candidate in the admissible top-k set relative to the model-default token in the same partially decoded state and commits a non-default candidate only when the best candidate lowers the one-step proxy score by more than a preset margin. On the primary LLaDA-8B CivilComments{\texttimes}Jigsaw setting, CIDER improves retention by +0.06 to +0.16 over the interpolated absolute-score frontier at matched residual-toxicity levels. At the reported aggressive endpoints, the retention advantage recurs across four data pairings on LLaDA-8B and three bases on CivilComments{\texttimes}Jigsaw. Controlled ablations indicate that the selection rule is an important contributor to the improvement."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lin-etal-2026-identity">
<titleInfo>
<title>Identity-Preserving Detoxification via Local Causal Effect Ranking on Masked Diffusion Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yucheng</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaojiao</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Text detoxification should produce safer rewrites while retaining identity mentions from the original text. However, classifier-guided methods can remove identity descriptors alongside genuinely harmful content when toxicity classifiers associate identity mentions with harmfulness and guidance relies on absolute scores. We propose CIDER (Contrastive Identity-preserving Detoxification via Explicit Reference), a default-relative, margin-gated selective-guidance method for masked diffusion language models. At each selected masked position, CIDER evaluates every candidate in the admissible top-k set relative to the model-default token in the same partially decoded state and commits a non-default candidate only when the best candidate lowers the one-step proxy score by more than a preset margin. On the primary LLaDA-8B CivilComments×Jigsaw setting, CIDER improves retention by +0.06 to +0.16 over the interpolated absolute-score frontier at matched residual-toxicity levels. At the reported aggressive endpoints, the retention advantage recurs across four data pairings on LLaDA-8B and three bases on CivilComments×Jigsaw. Controlled ablations indicate that the selection rule is an important contributor to the improvement.</abstract>
<identifier type="citekey">lin-etal-2026-identity</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.234/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>5360</start>
<end>5378</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Identity-Preserving Detoxification via Local Causal Effect Ranking on Masked Diffusion Language Models
%A Lin, Yucheng
%A Liu, Yue
%A Jiang, Jiaojiao
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F lin-etal-2026-identity
%X Text detoxification should produce safer rewrites while retaining identity mentions from the original text. However, classifier-guided methods can remove identity descriptors alongside genuinely harmful content when toxicity classifiers associate identity mentions with harmfulness and guidance relies on absolute scores. We propose CIDER (Contrastive Identity-preserving Detoxification via Explicit Reference), a default-relative, margin-gated selective-guidance method for masked diffusion language models. At each selected masked position, CIDER evaluates every candidate in the admissible top-k set relative to the model-default token in the same partially decoded state and commits a non-default candidate only when the best candidate lowers the one-step proxy score by more than a preset margin. On the primary LLaDA-8B CivilComments×Jigsaw setting, CIDER improves retention by +0.06 to +0.16 over the interpolated absolute-score frontier at matched residual-toxicity levels. At the reported aggressive endpoints, the retention advantage recurs across four data pairings on LLaDA-8B and three bases on CivilComments×Jigsaw. Controlled ablations indicate that the selection rule is an important contributor to the improvement.
%U https://aclanthology.org/2026.emnlp-main.234/
%P 5360-5378
Markdown (Informal)
[Identity-Preserving Detoxification via Local Causal Effect Ranking on Masked Diffusion Language Models](https://aclanthology.org/2026.emnlp-main.234/) (Lin et al., EMNLP 2026)
ACL