@inproceedings{park-etal-2026-typology,
title = "Typology-Aware Dual-Track: Detecting Hybrid {AI} Text Across Languages",
author = "Park, Kihoon and
Lim, Seungwoo and
Kim, Jihie",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.809/",
doi = "10.18653/2026.findings-emnlp.809",
pages = "17150--17167",
ISBN = "979-8-89176-462-0",
abstract = "Existing generalist detectors, such as mDeBERTa based classifiers, struggle with fine-grained AI-text detection in multilingual settings, particularly when hybrid authorship and morphologically rich languages obscure machine-generated artifacts. We propose a typology-aware Smart Routing Late Fusion Framework for 4-class detection: Human-Written (HW), Machine-Generated (MG), Machine-Humanized (MH), and Human-Polished (HM). The framework consists of two complementary tracks. Branch A targets agglutinative languages, including Korean, Turkish, and Hungarian, using error focused hard-negative refinement and Dependency Distance Minimization(DDM) / Uniform Information Density(UID) based pairwise calibration to recover weak generative signals hidden by dense morphology. Branch B applies RADAR-inspired invariant learning to capture machine-originated artifacts that persist after adversarial rewriting in languages with clearer segmentation cues. Experiments across eight languages show that typology-aware routing improves detection in agglutinative languages while maintaining strong multilingual performance. Our results highlight the importance of modeling both linguistic typology and hybrid authorship for robust multilingual AI-text detection."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="park-etal-2026-typology">
<titleInfo>
<title>Typology-Aware Dual-Track: Detecting Hybrid AI Text Across Languages</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kihoon</namePart>
<namePart type="family">Park</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seungwoo</namePart>
<namePart type="family">Lim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jihie</namePart>
<namePart type="family">Kim</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Existing generalist detectors, such as mDeBERTa based classifiers, struggle with fine-grained AI-text detection in multilingual settings, particularly when hybrid authorship and morphologically rich languages obscure machine-generated artifacts. We propose a typology-aware Smart Routing Late Fusion Framework for 4-class detection: Human-Written (HW), Machine-Generated (MG), Machine-Humanized (MH), and Human-Polished (HM). The framework consists of two complementary tracks. Branch A targets agglutinative languages, including Korean, Turkish, and Hungarian, using error focused hard-negative refinement and Dependency Distance Minimization(DDM) / Uniform Information Density(UID) based pairwise calibration to recover weak generative signals hidden by dense morphology. Branch B applies RADAR-inspired invariant learning to capture machine-originated artifacts that persist after adversarial rewriting in languages with clearer segmentation cues. Experiments across eight languages show that typology-aware routing improves detection in agglutinative languages while maintaining strong multilingual performance. Our results highlight the importance of modeling both linguistic typology and hybrid authorship for robust multilingual AI-text detection.</abstract>
<identifier type="citekey">park-etal-2026-typology</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.809</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.809/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>17150</start>
<end>17167</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Typology-Aware Dual-Track: Detecting Hybrid AI Text Across Languages
%A Park, Kihoon
%A Lim, Seungwoo
%A Kim, Jihie
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F park-etal-2026-typology
%X Existing generalist detectors, such as mDeBERTa based classifiers, struggle with fine-grained AI-text detection in multilingual settings, particularly when hybrid authorship and morphologically rich languages obscure machine-generated artifacts. We propose a typology-aware Smart Routing Late Fusion Framework for 4-class detection: Human-Written (HW), Machine-Generated (MG), Machine-Humanized (MH), and Human-Polished (HM). The framework consists of two complementary tracks. Branch A targets agglutinative languages, including Korean, Turkish, and Hungarian, using error focused hard-negative refinement and Dependency Distance Minimization(DDM) / Uniform Information Density(UID) based pairwise calibration to recover weak generative signals hidden by dense morphology. Branch B applies RADAR-inspired invariant learning to capture machine-originated artifacts that persist after adversarial rewriting in languages with clearer segmentation cues. Experiments across eight languages show that typology-aware routing improves detection in agglutinative languages while maintaining strong multilingual performance. Our results highlight the importance of modeling both linguistic typology and hybrid authorship for robust multilingual AI-text detection.
%R 10.18653/2026.findings-emnlp.809
%U https://aclanthology.org/2026.findings-emnlp.809/
%U https://doi.org/10.18653/2026.findings-emnlp.809
%P 17150-17167
Markdown (Informal)
[Typology-Aware Dual-Track: Detecting Hybrid AI Text Across Languages](https://aclanthology.org/2026.findings-emnlp.809/) (Park et al., Findings 2026)
ACL