@inproceedings{duan-liu-2026-rule,
title = "Rule-Based Perturbation, Real Evaluation: How Far Do Fine-Tuned Encoders Go on {C}hinese Phonetically Cloaked Toxic Language?",
author = "Duan, Minggui and
Liu, Shijie",
editor = "Calabrese, Agostina and
Davidson, Thomas and
de Kock, Christine and
Khurana, Urja and
Marchiori Manerba, Marta and
Piot, Paloma and
Talat, Zeerak",
booktitle = "Proceedings of the 10th Workshop on Online Abuse and Harms ({WOAH})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.woah-1.29/",
pages = "338--345",
ISBN = "979-8-89176-487-3",
abstract = "Phonetic cloaking, the deliberate use of homophones, Pinyin, and numeric substitutions to disguise toxic content, poses a persistent challenge to Chinese toxic language detection. Existing work mainly relies on rule-based synthetic perturbations for benchmark evaluation, leaving uncertain how well such results transfer to naturally occurring cloaking. We present a controlled evaluation of fine-tuned encoders on PCR-ToxiCN, a dataset of naturally occurring, phonetically cloaked Chinese posts. Fine-tuning ChineseBERT and RoBERTa-wwm-ext on the CNTP synthetic perturbation dataset, we quantify synthetic-to-real performance gaps across four cloaking strategies. Synthetic perturbation training produces consistent but limited improvements, with diminishing returns as perturbation coverage expands. Hanzi (Chinese-character) replacement remains consistently difficult for both encoders, and few-shot fine-tuning on roughly 50 real examples provides only marginal improvement. The few-shot experiment, however, cannot distinguish representation limits from data scarcity or sampling effects. Overall, our findings caution against equating synthetic benchmark performance with real-world robustness."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="duan-liu-2026-rule">
<titleInfo>
<title>Rule-Based Perturbation, Real Evaluation: How Far Do Fine-Tuned Encoders Go on Chinese Phonetically Cloaked Toxic Language?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Minggui</namePart>
<namePart type="family">Duan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shijie</namePart>
<namePart type="family">Liu</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 10th Workshop on Online Abuse and Harms (WOAH)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Agostina</namePart>
<namePart type="family">Calabrese</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Davidson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christine</namePart>
<namePart type="family">de Kock</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Urja</namePart>
<namePart type="family">Khurana</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marta</namePart>
<namePart type="family">Marchiori Manerba</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paloma</namePart>
<namePart type="family">Piot</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zeerak</namePart>
<namePart type="family">Talat</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">979-8-89176-487-3</identifier>
</relatedItem>
<abstract>Phonetic cloaking, the deliberate use of homophones, Pinyin, and numeric substitutions to disguise toxic content, poses a persistent challenge to Chinese toxic language detection. Existing work mainly relies on rule-based synthetic perturbations for benchmark evaluation, leaving uncertain how well such results transfer to naturally occurring cloaking. We present a controlled evaluation of fine-tuned encoders on PCR-ToxiCN, a dataset of naturally occurring, phonetically cloaked Chinese posts. Fine-tuning ChineseBERT and RoBERTa-wwm-ext on the CNTP synthetic perturbation dataset, we quantify synthetic-to-real performance gaps across four cloaking strategies. Synthetic perturbation training produces consistent but limited improvements, with diminishing returns as perturbation coverage expands. Hanzi (Chinese-character) replacement remains consistently difficult for both encoders, and few-shot fine-tuning on roughly 50 real examples provides only marginal improvement. The few-shot experiment, however, cannot distinguish representation limits from data scarcity or sampling effects. Overall, our findings caution against equating synthetic benchmark performance with real-world robustness.</abstract>
<identifier type="citekey">duan-liu-2026-rule</identifier>
<location>
<url>https://aclanthology.org/2026.woah-1.29/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>338</start>
<end>345</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Rule-Based Perturbation, Real Evaluation: How Far Do Fine-Tuned Encoders Go on Chinese Phonetically Cloaked Toxic Language?
%A Duan, Minggui
%A Liu, Shijie
%Y Calabrese, Agostina
%Y Davidson, Thomas
%Y de Kock, Christine
%Y Khurana, Urja
%Y Marchiori Manerba, Marta
%Y Piot, Paloma
%Y Talat, Zeerak
%S Proceedings of the 10th Workshop on Online Abuse and Harms (WOAH)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-487-3
%F duan-liu-2026-rule
%X Phonetic cloaking, the deliberate use of homophones, Pinyin, and numeric substitutions to disguise toxic content, poses a persistent challenge to Chinese toxic language detection. Existing work mainly relies on rule-based synthetic perturbations for benchmark evaluation, leaving uncertain how well such results transfer to naturally occurring cloaking. We present a controlled evaluation of fine-tuned encoders on PCR-ToxiCN, a dataset of naturally occurring, phonetically cloaked Chinese posts. Fine-tuning ChineseBERT and RoBERTa-wwm-ext on the CNTP synthetic perturbation dataset, we quantify synthetic-to-real performance gaps across four cloaking strategies. Synthetic perturbation training produces consistent but limited improvements, with diminishing returns as perturbation coverage expands. Hanzi (Chinese-character) replacement remains consistently difficult for both encoders, and few-shot fine-tuning on roughly 50 real examples provides only marginal improvement. The few-shot experiment, however, cannot distinguish representation limits from data scarcity or sampling effects. Overall, our findings caution against equating synthetic benchmark performance with real-world robustness.
%U https://aclanthology.org/2026.woah-1.29/
%P 338-345
Markdown (Informal)
[Rule-Based Perturbation, Real Evaluation: How Far Do Fine-Tuned Encoders Go on Chinese Phonetically Cloaked Toxic Language?](https://aclanthology.org/2026.woah-1.29/) (Duan & Liu, WOAH 2026)
ACL