@inproceedings{das-etal-2026-rethink,
title = "{R}e{T}hink: Adversarially Robust Arithmetic Reasoning via Neuro-Symbolic Soft Constraints",
author = "Das, Soikat and
Mittal, Sparsh and
Singhal, Rekha",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.66/",
pages = "1303--1313",
ISBN = "979-8-89176-444-6",
abstract = "While Large Language Models (LLMs) perform well on arithmetic word problems, they remain brittle to adversarial perturbations like irrelevant distractors or contradictory facts. To address this, we introduce \textsc{ReThink}, a training-free neuro-symbolic framework that formulates adversarial arithmetic reasoning as a Weighted MaxSAT/MaxSMT problem. \textsc{ReThink} uses a four-stage pipeline: it explicitly decomposes problems to strip noise, encodes the mathematical core as confidence-weighted soft constraints in the Z3 theorem prover, and cross-verifies the formal solution against a parallel Chain-of-Thought (CoT) path. This formulation can sacrifice lower-weight conflicting statements while satisfying higher-weight constraints. Evaluated on a balanced 2,400-example subset of GSM-Plus, \textsc{ReThink} with GPT-4.1 achieves 81.62{\%} adversarial accuracy, outperforming CoT, PAL, and Self-Verification. It narrows the robustness gap to 14.69 percentage points while achieving 96.31{\%} clean accuracy."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="das-etal-2026-rethink">
<titleInfo>
<title>ReThink: Adversarially Robust Arithmetic Reasoning via Neuro-Symbolic Soft Constraints</title>
</titleInfo>
<name type="personal">
<namePart type="given">Soikat</namePart>
<namePart type="family">Das</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sparsh</namePart>
<namePart type="family">Mittal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rekha</namePart>
<namePart type="family">Singhal</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>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</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">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>While Large Language Models (LLMs) perform well on arithmetic word problems, they remain brittle to adversarial perturbations like irrelevant distractors or contradictory facts. To address this, we introduce ReThink, a training-free neuro-symbolic framework that formulates adversarial arithmetic reasoning as a Weighted MaxSAT/MaxSMT problem. ReThink uses a four-stage pipeline: it explicitly decomposes problems to strip noise, encodes the mathematical core as confidence-weighted soft constraints in the Z3 theorem prover, and cross-verifies the formal solution against a parallel Chain-of-Thought (CoT) path. This formulation can sacrifice lower-weight conflicting statements while satisfying higher-weight constraints. Evaluated on a balanced 2,400-example subset of GSM-Plus, ReThink with GPT-4.1 achieves 81.62% adversarial accuracy, outperforming CoT, PAL, and Self-Verification. It narrows the robustness gap to 14.69 percentage points while achieving 96.31% clean accuracy.</abstract>
<identifier type="citekey">das-etal-2026-rethink</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.66/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>1303</start>
<end>1313</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ReThink: Adversarially Robust Arithmetic Reasoning via Neuro-Symbolic Soft Constraints
%A Das, Soikat
%A Mittal, Sparsh
%A Singhal, Rekha
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F das-etal-2026-rethink
%X While Large Language Models (LLMs) perform well on arithmetic word problems, they remain brittle to adversarial perturbations like irrelevant distractors or contradictory facts. To address this, we introduce ReThink, a training-free neuro-symbolic framework that formulates adversarial arithmetic reasoning as a Weighted MaxSAT/MaxSMT problem. ReThink uses a four-stage pipeline: it explicitly decomposes problems to strip noise, encodes the mathematical core as confidence-weighted soft constraints in the Z3 theorem prover, and cross-verifies the formal solution against a parallel Chain-of-Thought (CoT) path. This formulation can sacrifice lower-weight conflicting statements while satisfying higher-weight constraints. Evaluated on a balanced 2,400-example subset of GSM-Plus, ReThink with GPT-4.1 achieves 81.62% adversarial accuracy, outperforming CoT, PAL, and Self-Verification. It narrows the robustness gap to 14.69 percentage points while achieving 96.31% clean accuracy.
%U https://aclanthology.org/2026.findings-aacl.66/
%P 1303-1313
Markdown (Informal)
[ReThink: Adversarially Robust Arithmetic Reasoning via Neuro-Symbolic Soft Constraints](https://aclanthology.org/2026.findings-aacl.66/) (Das et al., Findings 2026)
ACL