@inproceedings{rashidov-etal-2026-talim,
title = "{TALIM}: A Benchmark Dataset for {T}ajik Academic Language Intelligence Measurement",
author = "Rashidov, Azamat and
Rashidov, Anis and
Narumanchi, Archit and
Park, Jong C.",
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.1646/",
doi = "10.18653/2026.findings-emnlp.1646",
pages = "35042--35056",
ISBN = "979-8-89176-462-0",
abstract = "We introduce TALIM (Tajik Academic Language Intelligence Measurement), a large-scale academic benchmark for evaluating academic reasoning and language understanding in Tajik. TALIM spans nine subjects and three question formats, multiple-choice, matching, and free-response, sourced natively from the Tajik national curriculum rather than translated from existing benchmarks. We evaluate nine LLMs, both proprietary and open-weight models, finding that proprietary models substantially outperform open-weight models across all subjects and question types. Grammar is the weakest subject across all models, while physics is consistently the strongest. English translation substantially improves open-weight performance while Farsi transliteration yields only marginal gains, indicating that failures stem from data scarcity rather than script unfamiliarity. Chain-of-thought analysis reveals several systematic failure modes: hallucination, reasoning process failures, and grammatical category confusion. We hope TALIM will serve as a resource for assessing Tajik language capabilities of frontier models and motivate similar native benchmark construction efforts for other low-resource languages."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rashidov-etal-2026-talim">
<titleInfo>
<title>TALIM: A Benchmark Dataset for Tajik Academic Language Intelligence Measurement</title>
</titleInfo>
<name type="personal">
<namePart type="given">Azamat</namePart>
<namePart type="family">Rashidov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anis</namePart>
<namePart type="family">Rashidov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Archit</namePart>
<namePart type="family">Narumanchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jong</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Park</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>We introduce TALIM (Tajik Academic Language Intelligence Measurement), a large-scale academic benchmark for evaluating academic reasoning and language understanding in Tajik. TALIM spans nine subjects and three question formats, multiple-choice, matching, and free-response, sourced natively from the Tajik national curriculum rather than translated from existing benchmarks. We evaluate nine LLMs, both proprietary and open-weight models, finding that proprietary models substantially outperform open-weight models across all subjects and question types. Grammar is the weakest subject across all models, while physics is consistently the strongest. English translation substantially improves open-weight performance while Farsi transliteration yields only marginal gains, indicating that failures stem from data scarcity rather than script unfamiliarity. Chain-of-thought analysis reveals several systematic failure modes: hallucination, reasoning process failures, and grammatical category confusion. We hope TALIM will serve as a resource for assessing Tajik language capabilities of frontier models and motivate similar native benchmark construction efforts for other low-resource languages.</abstract>
<identifier type="citekey">rashidov-etal-2026-talim</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1646</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1646/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>35042</start>
<end>35056</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T TALIM: A Benchmark Dataset for Tajik Academic Language Intelligence Measurement
%A Rashidov, Azamat
%A Rashidov, Anis
%A Narumanchi, Archit
%A Park, Jong C.
%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 rashidov-etal-2026-talim
%X We introduce TALIM (Tajik Academic Language Intelligence Measurement), a large-scale academic benchmark for evaluating academic reasoning and language understanding in Tajik. TALIM spans nine subjects and three question formats, multiple-choice, matching, and free-response, sourced natively from the Tajik national curriculum rather than translated from existing benchmarks. We evaluate nine LLMs, both proprietary and open-weight models, finding that proprietary models substantially outperform open-weight models across all subjects and question types. Grammar is the weakest subject across all models, while physics is consistently the strongest. English translation substantially improves open-weight performance while Farsi transliteration yields only marginal gains, indicating that failures stem from data scarcity rather than script unfamiliarity. Chain-of-thought analysis reveals several systematic failure modes: hallucination, reasoning process failures, and grammatical category confusion. We hope TALIM will serve as a resource for assessing Tajik language capabilities of frontier models and motivate similar native benchmark construction efforts for other low-resource languages.
%R 10.18653/2026.findings-emnlp.1646
%U https://aclanthology.org/2026.findings-emnlp.1646/
%U https://doi.org/10.18653/2026.findings-emnlp.1646
%P 35042-35056
Markdown (Informal)
[TALIM: A Benchmark Dataset for Tajik Academic Language Intelligence Measurement](https://aclanthology.org/2026.findings-emnlp.1646/) (Rashidov et al., Findings 2026)
ACL