@inproceedings{alizade-etal-2026-parsbench,
title = "{P}ars{B}ench: A Benchmark for Evaluating Fine-Grained Culturally Grounded Commonsense Reasoning in {P}ersian",
author = "Alizade, Mehrnoush and
Davoodi, Farbod and
Maity, Suman Kalyan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2635/",
pages = "59892--59913",
ISBN = "979-8-89176-459-0",
abstract = "Large Language Models (LLMs) have achieved remarkable progress in natural language understanding and generation, yet their evaluation remains heavily centered on English and Western cultural contexts, limiting their reliability in culturally diverse and low-resource settings. We introduce \textit{ParsBench}, a comprehensive benchmark for evaluating fine-grained culturally grounded commonsense reasoning in Persian. ParsBench consists of three complementary components: Expert-QA, Multimodal-QA, and Long-Form QA, collectively covering structured reasoning, multimodal cultural understanding, and open-ended generation across diverse cultural categories. We evaluate a broad range of multilingual and multimodal LLMs and observe substantial variation across models and cultural categories. Our results reveal persistent weaknesses in culturally grounded non-English reasoning and significant disparities between textual and multimodal understanding. Overall, the findings demonstrate that existing English-centric evaluations fail to capture important dimensions of cultural competence and highlight the need for culturally grounded benchmarks for more inclusive evaluation. Our code and dataset are available at https://github.com/ParsBench-CCS/ParsBench"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="alizade-etal-2026-parsbench">
<titleInfo>
<title>ParsBench: A Benchmark for Evaluating Fine-Grained Culturally Grounded Commonsense Reasoning in Persian</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mehrnoush</namePart>
<namePart type="family">Alizade</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Farbod</namePart>
<namePart type="family">Davoodi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Suman</namePart>
<namePart type="given">Kalyan</namePart>
<namePart type="family">Maity</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 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) have achieved remarkable progress in natural language understanding and generation, yet their evaluation remains heavily centered on English and Western cultural contexts, limiting their reliability in culturally diverse and low-resource settings. We introduce ParsBench, a comprehensive benchmark for evaluating fine-grained culturally grounded commonsense reasoning in Persian. ParsBench consists of three complementary components: Expert-QA, Multimodal-QA, and Long-Form QA, collectively covering structured reasoning, multimodal cultural understanding, and open-ended generation across diverse cultural categories. We evaluate a broad range of multilingual and multimodal LLMs and observe substantial variation across models and cultural categories. Our results reveal persistent weaknesses in culturally grounded non-English reasoning and significant disparities between textual and multimodal understanding. Overall, the findings demonstrate that existing English-centric evaluations fail to capture important dimensions of cultural competence and highlight the need for culturally grounded benchmarks for more inclusive evaluation. Our code and dataset are available at https://github.com/ParsBench-CCS/ParsBench</abstract>
<identifier type="citekey">alizade-etal-2026-parsbench</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2635/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>59892</start>
<end>59913</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ParsBench: A Benchmark for Evaluating Fine-Grained Culturally Grounded Commonsense Reasoning in Persian
%A Alizade, Mehrnoush
%A Davoodi, Farbod
%A Maity, Suman Kalyan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F alizade-etal-2026-parsbench
%X Large Language Models (LLMs) have achieved remarkable progress in natural language understanding and generation, yet their evaluation remains heavily centered on English and Western cultural contexts, limiting their reliability in culturally diverse and low-resource settings. We introduce ParsBench, a comprehensive benchmark for evaluating fine-grained culturally grounded commonsense reasoning in Persian. ParsBench consists of three complementary components: Expert-QA, Multimodal-QA, and Long-Form QA, collectively covering structured reasoning, multimodal cultural understanding, and open-ended generation across diverse cultural categories. We evaluate a broad range of multilingual and multimodal LLMs and observe substantial variation across models and cultural categories. Our results reveal persistent weaknesses in culturally grounded non-English reasoning and significant disparities between textual and multimodal understanding. Overall, the findings demonstrate that existing English-centric evaluations fail to capture important dimensions of cultural competence and highlight the need for culturally grounded benchmarks for more inclusive evaluation. Our code and dataset are available at https://github.com/ParsBench-CCS/ParsBench
%U https://aclanthology.org/2026.emnlp-main.2635/
%P 59892-59913
Markdown (Informal)
[ParsBench: A Benchmark for Evaluating Fine-Grained Culturally Grounded Commonsense Reasoning in Persian](https://aclanthology.org/2026.emnlp-main.2635/) (Alizade et al., EMNLP 2026)
ACL