@inproceedings{sayeedi-etal-2026-many,
title = "Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual {VLM}s for {B}engali Culture Understanding Across Historically Linked Languages and Regional Dialects",
author = "Sayeedi, Nurul Labib and
Sayeedi, Md. Faiyaz Abdullah and
Roy Dipta, Shubhashis and
E Sobhani, Mahbub and
Tabassum, Rubaya and
Hridoy, Ariful Ekraj and
Mahmood, Mehraj and
Hasan, Md. Tarek and
Shatabda, Swakkhar",
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.2129/",
pages = "45421--45454",
ISBN = "979-8-89176-462-0",
abstract = "Bangla culture is richly expressed through region, dialect, history, food, politics, media, and everyday visual life, yet it remains underrepresented in multimodal evaluation. To address this gap, we introduce **BanglaVerse**, a culturally grounded benchmark for evaluating multilingual vision{--}language models (VLMs) on Bengali culture across historically linked languages and regional dialects. Built from 1,152 manually curated images across nine domains, the benchmark supports visual question answering and captioning, and is expanded into four languages and five Bangla dialects, yielding {\ensuremath{\sim}}32.2K artifacts. Our experiments show that evaluating only standard Bangla overestimates true model capability: performance drops under dialectal variation, especially for caption generation, while historically linked languages such as Hindi and Urdu retain some cultural meaning but remain weaker for structured reasoning. Across domains, the main bottleneck is missing cultural knowledge rather than visual grounding alone, with knowledge-intensive categories. These findings position **BanglaVerse** as a more realistic test bed for measuring culturally grounded multimodal understanding under linguistic variation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sayeedi-etal-2026-many">
<titleInfo>
<title>Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual VLMs for Bengali Culture Understanding Across Historically Linked Languages and Regional Dialects</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nurul</namePart>
<namePart type="given">Labib</namePart>
<namePart type="family">Sayeedi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md.</namePart>
<namePart type="given">Faiyaz</namePart>
<namePart type="given">Abdullah</namePart>
<namePart type="family">Sayeedi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shubhashis</namePart>
<namePart type="family">Roy Dipta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mahbub</namePart>
<namePart type="family">E Sobhani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rubaya</namePart>
<namePart type="family">Tabassum</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ariful</namePart>
<namePart type="given">Ekraj</namePart>
<namePart type="family">Hridoy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mehraj</namePart>
<namePart type="family">Mahmood</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md.</namePart>
<namePart type="given">Tarek</namePart>
<namePart type="family">Hasan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Swakkhar</namePart>
<namePart type="family">Shatabda</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>Bangla culture is richly expressed through region, dialect, history, food, politics, media, and everyday visual life, yet it remains underrepresented in multimodal evaluation. To address this gap, we introduce **BanglaVerse**, a culturally grounded benchmark for evaluating multilingual vision–language models (VLMs) on Bengali culture across historically linked languages and regional dialects. Built from 1,152 manually curated images across nine domains, the benchmark supports visual question answering and captioning, and is expanded into four languages and five Bangla dialects, yielding \ensuremath\sim32.2K artifacts. Our experiments show that evaluating only standard Bangla overestimates true model capability: performance drops under dialectal variation, especially for caption generation, while historically linked languages such as Hindi and Urdu retain some cultural meaning but remain weaker for structured reasoning. Across domains, the main bottleneck is missing cultural knowledge rather than visual grounding alone, with knowledge-intensive categories. These findings position **BanglaVerse** as a more realistic test bed for measuring culturally grounded multimodal understanding under linguistic variation.</abstract>
<identifier type="citekey">sayeedi-etal-2026-many</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2129/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>45421</start>
<end>45454</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual VLMs for Bengali Culture Understanding Across Historically Linked Languages and Regional Dialects
%A Sayeedi, Nurul Labib
%A Sayeedi, Md. Faiyaz Abdullah
%A Roy Dipta, Shubhashis
%A E Sobhani, Mahbub
%A Tabassum, Rubaya
%A Hridoy, Ariful Ekraj
%A Mahmood, Mehraj
%A Hasan, Md. Tarek
%A Shatabda, Swakkhar
%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 sayeedi-etal-2026-many
%X Bangla culture is richly expressed through region, dialect, history, food, politics, media, and everyday visual life, yet it remains underrepresented in multimodal evaluation. To address this gap, we introduce **BanglaVerse**, a culturally grounded benchmark for evaluating multilingual vision–language models (VLMs) on Bengali culture across historically linked languages and regional dialects. Built from 1,152 manually curated images across nine domains, the benchmark supports visual question answering and captioning, and is expanded into four languages and five Bangla dialects, yielding \ensuremath\sim32.2K artifacts. Our experiments show that evaluating only standard Bangla overestimates true model capability: performance drops under dialectal variation, especially for caption generation, while historically linked languages such as Hindi and Urdu retain some cultural meaning but remain weaker for structured reasoning. Across domains, the main bottleneck is missing cultural knowledge rather than visual grounding alone, with knowledge-intensive categories. These findings position **BanglaVerse** as a more realistic test bed for measuring culturally grounded multimodal understanding under linguistic variation.
%U https://aclanthology.org/2026.findings-emnlp.2129/
%P 45421-45454
Markdown (Informal)
[Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual VLMs for Bengali Culture Understanding Across Historically Linked Languages and Regional Dialects](https://aclanthology.org/2026.findings-emnlp.2129/) (Sayeedi et al., Findings 2026)
ACL
- Nurul Labib Sayeedi, Md. Faiyaz Abdullah Sayeedi, Shubhashis Roy Dipta, Mahbub E Sobhani, Rubaya Tabassum, Ariful Ekraj Hridoy, Mehraj Mahmood, Md. Tarek Hasan, and Swakkhar Shatabda. 2026. Many Dialects, Many Languages, One Cultural Lens: Evaluating Multilingual VLMs for Bengali Culture Understanding Across Historically Linked Languages and Regional Dialects. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 45421–45454, Budapest, Hungary. Association for Computational Linguistics.