@inproceedings{abedin-khan-2026-machine,
title = "Machine-Translated {B}engali in the Wild: A Provenance Benchmark and a Detector-Flagged Shift Toward Translation in {C}ommon {C}rawl",
author = "Abedin, Mantaqa and
Khan, Jeba Afia",
editor = "Nakazawa, Toshiaki and
Goto, Isao",
booktitle = "Proceedings of the Thirteenth Workshop on {A}sian Translation ({WAT} 2026)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.wat-1.7/",
pages = "104--118",
ISBN = "979-8-89176-446-0",
abstract = "Low-resource Asian machine translation draws on monolingual web text from Common Crawl. We ask what share of Bengali Common Crawl a detector flags as machine output, and whether it moves with translation or generation. We introduce BanglaGen-Bench, a 2,700-document Bengali benchmark with human-written, English-to-Bengali machine-translated and LLM-generated classes, and a stylometric detector that needs no Bengali-specific tooling. Translatedness did not transfer between the two NMT systems we tested: a detector trained only on NLLB-200 output recognises mBART-50 output 5.67{\%} of the time, and the reverse 0.00{\%}, while mixed-system training reaches 99.3{\%} on held-out documents of a seen system. A detector trained on dedicated NMT catches an LLM prompted to translate 20 times in 100, against 97 when the same model generates. Across four snapshots (2018{--}2026), the detector-flagged non-human share rises from 2.13{\%} to 6.20{\%}, and after 2023 the flagged movement is on the translation side: the MT-flagged share climbs from 0.30{\%} to 3.48{\%}, while the LLM-flagged share peaks and falls. These are flagged shares, not validated prevalence: on the only independent Bengali corpus available, which has no translated text, the detector scores 62.7{\%} against 97.3{\%} in-benchmark. The benchmark, checkpoints and scripts will be released."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="abedin-khan-2026-machine">
<titleInfo>
<title>Machine-Translated Bengali in the Wild: A Provenance Benchmark and a Detector-Flagged Shift Toward Translation in Common Crawl</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mantaqa</namePart>
<namePart type="family">Abedin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeba</namePart>
<namePart type="given">Afia</namePart>
<namePart type="family">Khan</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 Thirteenth Workshop on Asian Translation (WAT 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Toshiaki</namePart>
<namePart type="family">Nakazawa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Isao</namePart>
<namePart type="family">Goto</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">979-8-89176-446-0</identifier>
</relatedItem>
<abstract>Low-resource Asian machine translation draws on monolingual web text from Common Crawl. We ask what share of Bengali Common Crawl a detector flags as machine output, and whether it moves with translation or generation. We introduce BanglaGen-Bench, a 2,700-document Bengali benchmark with human-written, English-to-Bengali machine-translated and LLM-generated classes, and a stylometric detector that needs no Bengali-specific tooling. Translatedness did not transfer between the two NMT systems we tested: a detector trained only on NLLB-200 output recognises mBART-50 output 5.67% of the time, and the reverse 0.00%, while mixed-system training reaches 99.3% on held-out documents of a seen system. A detector trained on dedicated NMT catches an LLM prompted to translate 20 times in 100, against 97 when the same model generates. Across four snapshots (2018–2026), the detector-flagged non-human share rises from 2.13% to 6.20%, and after 2023 the flagged movement is on the translation side: the MT-flagged share climbs from 0.30% to 3.48%, while the LLM-flagged share peaks and falls. These are flagged shares, not validated prevalence: on the only independent Bengali corpus available, which has no translated text, the detector scores 62.7% against 97.3% in-benchmark. The benchmark, checkpoints and scripts will be released.</abstract>
<identifier type="citekey">abedin-khan-2026-machine</identifier>
<location>
<url>https://aclanthology.org/2026.wat-1.7/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>104</start>
<end>118</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Machine-Translated Bengali in the Wild: A Provenance Benchmark and a Detector-Flagged Shift Toward Translation in Common Crawl
%A Abedin, Mantaqa
%A Khan, Jeba Afia
%Y Nakazawa, Toshiaki
%Y Goto, Isao
%S Proceedings of the Thirteenth Workshop on Asian Translation (WAT 2026)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-446-0
%F abedin-khan-2026-machine
%X Low-resource Asian machine translation draws on monolingual web text from Common Crawl. We ask what share of Bengali Common Crawl a detector flags as machine output, and whether it moves with translation or generation. We introduce BanglaGen-Bench, a 2,700-document Bengali benchmark with human-written, English-to-Bengali machine-translated and LLM-generated classes, and a stylometric detector that needs no Bengali-specific tooling. Translatedness did not transfer between the two NMT systems we tested: a detector trained only on NLLB-200 output recognises mBART-50 output 5.67% of the time, and the reverse 0.00%, while mixed-system training reaches 99.3% on held-out documents of a seen system. A detector trained on dedicated NMT catches an LLM prompted to translate 20 times in 100, against 97 when the same model generates. Across four snapshots (2018–2026), the detector-flagged non-human share rises from 2.13% to 6.20%, and after 2023 the flagged movement is on the translation side: the MT-flagged share climbs from 0.30% to 3.48%, while the LLM-flagged share peaks and falls. These are flagged shares, not validated prevalence: on the only independent Bengali corpus available, which has no translated text, the detector scores 62.7% against 97.3% in-benchmark. The benchmark, checkpoints and scripts will be released.
%U https://aclanthology.org/2026.wat-1.7/
%P 104-118
Markdown (Informal)
[Machine-Translated Bengali in the Wild: A Provenance Benchmark and a Detector-Flagged Shift Toward Translation in Common Crawl](https://aclanthology.org/2026.wat-1.7/) (Abedin & Khan, WAT 2026)
ACL