@inproceedings{pournemat-etal-2026-reasoning,
title = "Reasoning Under Uncertainty: Exploring Probabilistic Reasoning Capabilities of {LLM}s",
author = "Pournemat, Mobina and
Rezaei, Keivan and
Sriramanan, Gaurang and
Zarei, Arman and
Eghbalzadeh, Hamid and
Wang, Yang and
Fu, Jiaxiang and
Peng, Fuchun and
Wang, Zhigang and
Feng, Xue and
Noorshams, Nima and
Feizi, Soheil",
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.37/",
pages = "811--829",
ISBN = "979-8-89176-459-0",
abstract = "Large language models (LLMs) have demonstrated strong performance across a wide range of reasoning tasks, yet their understanding of joint and conditional probability distributions remains largely unexplored. In this work, we present a comprehensive evaluation of LLMs' ability to reason about discrete joint and conditional probability distributions involving multiple random variables. Given observations drawn from a distribution, we evaluate models on three carefully designed tasks{---}mode identification, maximum likelihood estimation, and sample generation{---}which together capture fundamental skills required for probabilistic reasoning. Our study reveals both capabilities and limitations of current LLMs. In sample generation task, some models produce empirical distributions that closely match the target distribution despite exhibiting limited sample independence. We further observe that model performance is highly sensitive to representation: replacing explicit frequency information with raw samples leads to large accuracy drops, and even minor changes to outcome labels can further degrade performance. Together, these findings provide new insights into how LLMs reason about discrete joint and conditional probability distributions and highlight their strengths and limitations in certain probabilistic reasoning settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="pournemat-etal-2026-reasoning">
<titleInfo>
<title>Reasoning Under Uncertainty: Exploring Probabilistic Reasoning Capabilities of LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mobina</namePart>
<namePart type="family">Pournemat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keivan</namePart>
<namePart type="family">Rezaei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gaurang</namePart>
<namePart type="family">Sriramanan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Arman</namePart>
<namePart type="family">Zarei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hamid</namePart>
<namePart type="family">Eghbalzadeh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxiang</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fuchun</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhigang</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xue</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nima</namePart>
<namePart type="family">Noorshams</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Soheil</namePart>
<namePart type="family">Feizi</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 demonstrated strong performance across a wide range of reasoning tasks, yet their understanding of joint and conditional probability distributions remains largely unexplored. In this work, we present a comprehensive evaluation of LLMs’ ability to reason about discrete joint and conditional probability distributions involving multiple random variables. Given observations drawn from a distribution, we evaluate models on three carefully designed tasks—mode identification, maximum likelihood estimation, and sample generation—which together capture fundamental skills required for probabilistic reasoning. Our study reveals both capabilities and limitations of current LLMs. In sample generation task, some models produce empirical distributions that closely match the target distribution despite exhibiting limited sample independence. We further observe that model performance is highly sensitive to representation: replacing explicit frequency information with raw samples leads to large accuracy drops, and even minor changes to outcome labels can further degrade performance. Together, these findings provide new insights into how LLMs reason about discrete joint and conditional probability distributions and highlight their strengths and limitations in certain probabilistic reasoning settings.</abstract>
<identifier type="citekey">pournemat-etal-2026-reasoning</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.37/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>811</start>
<end>829</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Reasoning Under Uncertainty: Exploring Probabilistic Reasoning Capabilities of LLMs
%A Pournemat, Mobina
%A Rezaei, Keivan
%A Sriramanan, Gaurang
%A Zarei, Arman
%A Eghbalzadeh, Hamid
%A Wang, Yang
%A Fu, Jiaxiang
%A Peng, Fuchun
%A Wang, Zhigang
%A Feng, Xue
%A Noorshams, Nima
%A Feizi, Soheil
%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 pournemat-etal-2026-reasoning
%X Large language models (LLMs) have demonstrated strong performance across a wide range of reasoning tasks, yet their understanding of joint and conditional probability distributions remains largely unexplored. In this work, we present a comprehensive evaluation of LLMs’ ability to reason about discrete joint and conditional probability distributions involving multiple random variables. Given observations drawn from a distribution, we evaluate models on three carefully designed tasks—mode identification, maximum likelihood estimation, and sample generation—which together capture fundamental skills required for probabilistic reasoning. Our study reveals both capabilities and limitations of current LLMs. In sample generation task, some models produce empirical distributions that closely match the target distribution despite exhibiting limited sample independence. We further observe that model performance is highly sensitive to representation: replacing explicit frequency information with raw samples leads to large accuracy drops, and even minor changes to outcome labels can further degrade performance. Together, these findings provide new insights into how LLMs reason about discrete joint and conditional probability distributions and highlight their strengths and limitations in certain probabilistic reasoning settings.
%U https://aclanthology.org/2026.emnlp-main.37/
%P 811-829
Markdown (Informal)
[Reasoning Under Uncertainty: Exploring Probabilistic Reasoning Capabilities of LLMs](https://aclanthology.org/2026.emnlp-main.37/) (Pournemat et al., EMNLP 2026)
ACL
- Mobina Pournemat, Keivan Rezaei, Gaurang Sriramanan, Arman Zarei, Hamid Eghbalzadeh, Yang Wang, Jiaxiang Fu, Fuchun Peng, Zhigang Wang, Xue Feng, Nima Noorshams, and Soheil Feizi. 2026. Reasoning Under Uncertainty: Exploring Probabilistic Reasoning Capabilities of LLMs. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 811–829, Budapest, Hungary. Association for Computational Linguistics.