@inproceedings{agrahari-etal-2026-bhaid,
title = "{BHAID}: A Dataset of {LLM}-Generated Text in {I}ndian Languages with a Dual Contrastive Learning Framework for Detection",
author = "Agrahari, Shifali and
Kumar, Sujit and
Sanasam, Ranbir Singh",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.92/",
pages = "1850--1871",
ISBN = "979-8-89176-444-6",
abstract = "The rapid growth of multilingual Large Language Models (LLMs) has enabled the generation of fluent and culturally nuanced text across Indian languages, posing new challenges in distinguishing LLM-generated from human-written content. Existing datasets and methodologies have primarily focused on AI-generated text detection in the English language. Existing datasets and methodologies, which are predominantly trained on English data, exhibit limited generalisation to low-resource and morphologically rich Indian languages, underscoring the need for linguistically diverse and culturally grounded benchmark datasets and methodologies. To address this gap, we introduce BHAID, BHaratiya AI-generated text Detection, the first large-scale multilingual benchmark for AI-text detection spanning nine Indian languages: Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Odia, and Assamese. Furthermore, we propose a Dual Hierarchical Contrastive Learning framework, DualCLF, that integrates self-supervised and supervised contrastive objectives with focal loss to jointly enhance semantic robustness and inter-class discriminability. Observations from our experiments on the BHAID and benchmark multilingual datasets from the literature reveal that our method consistently outperforms recent state-of-the-art baseline approaches from the literature in detecting AI-generated text."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="agrahari-etal-2026-bhaid">
<titleInfo>
<title>BHAID: A Dataset of LLM-Generated Text in Indian Languages with a Dual Contrastive Learning Framework for Detection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shifali</namePart>
<namePart type="family">Agrahari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sujit</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ranbir</namePart>
<namePart type="given">Singh</namePart>
<namePart type="family">Sanasam</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>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</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">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>The rapid growth of multilingual Large Language Models (LLMs) has enabled the generation of fluent and culturally nuanced text across Indian languages, posing new challenges in distinguishing LLM-generated from human-written content. Existing datasets and methodologies have primarily focused on AI-generated text detection in the English language. Existing datasets and methodologies, which are predominantly trained on English data, exhibit limited generalisation to low-resource and morphologically rich Indian languages, underscoring the need for linguistically diverse and culturally grounded benchmark datasets and methodologies. To address this gap, we introduce BHAID, BHaratiya AI-generated text Detection, the first large-scale multilingual benchmark for AI-text detection spanning nine Indian languages: Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Odia, and Assamese. Furthermore, we propose a Dual Hierarchical Contrastive Learning framework, DualCLF, that integrates self-supervised and supervised contrastive objectives with focal loss to jointly enhance semantic robustness and inter-class discriminability. Observations from our experiments on the BHAID and benchmark multilingual datasets from the literature reveal that our method consistently outperforms recent state-of-the-art baseline approaches from the literature in detecting AI-generated text.</abstract>
<identifier type="citekey">agrahari-etal-2026-bhaid</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.92/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>1850</start>
<end>1871</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T BHAID: A Dataset of LLM-Generated Text in Indian Languages with a Dual Contrastive Learning Framework for Detection
%A Agrahari, Shifali
%A Kumar, Sujit
%A Sanasam, Ranbir Singh
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F agrahari-etal-2026-bhaid
%X The rapid growth of multilingual Large Language Models (LLMs) has enabled the generation of fluent and culturally nuanced text across Indian languages, posing new challenges in distinguishing LLM-generated from human-written content. Existing datasets and methodologies have primarily focused on AI-generated text detection in the English language. Existing datasets and methodologies, which are predominantly trained on English data, exhibit limited generalisation to low-resource and morphologically rich Indian languages, underscoring the need for linguistically diverse and culturally grounded benchmark datasets and methodologies. To address this gap, we introduce BHAID, BHaratiya AI-generated text Detection, the first large-scale multilingual benchmark for AI-text detection spanning nine Indian languages: Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Gujarati, Odia, and Assamese. Furthermore, we propose a Dual Hierarchical Contrastive Learning framework, DualCLF, that integrates self-supervised and supervised contrastive objectives with focal loss to jointly enhance semantic robustness and inter-class discriminability. Observations from our experiments on the BHAID and benchmark multilingual datasets from the literature reveal that our method consistently outperforms recent state-of-the-art baseline approaches from the literature in detecting AI-generated text.
%U https://aclanthology.org/2026.findings-aacl.92/
%P 1850-1871
Markdown (Informal)
[BHAID: A Dataset of LLM-Generated Text in Indian Languages with a Dual Contrastive Learning Framework for Detection](https://aclanthology.org/2026.findings-aacl.92/) (Agrahari et al., Findings 2026)
ACL