@inproceedings{shafiei-etal-2026-directional,
title = "Directional Drift in Arithmetic Reasoning: How Comparative Cues Steer {LLM} Predictions",
author = "Shafiei, Mohammadamin and
Saffari, Hamidreza and
Moosavi, Nafise Sadat",
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.1989/",
doi = "10.18653/2026.findings-emnlp.1989",
pages = "42392--42406",
ISBN = "979-8-89176-462-0",
abstract = "Arithmetic tasks are widely used to evaluate LLM reasoning because they provide objective truth. Yet accuracy alone cannot distinguish whether prompt sensitivity reflects random prediction instability or systematic semantic bias. We study this question in comparative arithmetic, where semantically equivalent framings can cue directional outcomes such as more, less, or equal. We introduce MathComp, a balanced benchmark of 10,800 controlled comparison problems spanning multiple framing conditions and answer formats. Across ten LLMs, semantically equivalent comparative cues systematically bias predictions toward cued outcomes even when it is numerically incorrect. Notably, this effect persists even when cue words are removed from the answer options, suggesting that framing bias reflects broader semantic steering rather than simple label copying. Chain-of-thought reasoning mitigates the effect substantially, but residual directional drift remains. These findings suggest that prompt sensitivity in arithmetic reasoning can exhibit structured semantic bias toward linguistically cued outcomes, with implications for robustness evaluation and benchmark design."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shafiei-etal-2026-directional">
<titleInfo>
<title>Directional Drift in Arithmetic Reasoning: How Comparative Cues Steer LLM Predictions</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mohammadamin</namePart>
<namePart type="family">Shafiei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hamidreza</namePart>
<namePart type="family">Saffari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nafise</namePart>
<namePart type="given">Sadat</namePart>
<namePart type="family">Moosavi</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>Arithmetic tasks are widely used to evaluate LLM reasoning because they provide objective truth. Yet accuracy alone cannot distinguish whether prompt sensitivity reflects random prediction instability or systematic semantic bias. We study this question in comparative arithmetic, where semantically equivalent framings can cue directional outcomes such as more, less, or equal. We introduce MathComp, a balanced benchmark of 10,800 controlled comparison problems spanning multiple framing conditions and answer formats. Across ten LLMs, semantically equivalent comparative cues systematically bias predictions toward cued outcomes even when it is numerically incorrect. Notably, this effect persists even when cue words are removed from the answer options, suggesting that framing bias reflects broader semantic steering rather than simple label copying. Chain-of-thought reasoning mitigates the effect substantially, but residual directional drift remains. These findings suggest that prompt sensitivity in arithmetic reasoning can exhibit structured semantic bias toward linguistically cued outcomes, with implications for robustness evaluation and benchmark design.</abstract>
<identifier type="citekey">shafiei-etal-2026-directional</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1989</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1989/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>42392</start>
<end>42406</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Directional Drift in Arithmetic Reasoning: How Comparative Cues Steer LLM Predictions
%A Shafiei, Mohammadamin
%A Saffari, Hamidreza
%A Moosavi, Nafise Sadat
%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 shafiei-etal-2026-directional
%X Arithmetic tasks are widely used to evaluate LLM reasoning because they provide objective truth. Yet accuracy alone cannot distinguish whether prompt sensitivity reflects random prediction instability or systematic semantic bias. We study this question in comparative arithmetic, where semantically equivalent framings can cue directional outcomes such as more, less, or equal. We introduce MathComp, a balanced benchmark of 10,800 controlled comparison problems spanning multiple framing conditions and answer formats. Across ten LLMs, semantically equivalent comparative cues systematically bias predictions toward cued outcomes even when it is numerically incorrect. Notably, this effect persists even when cue words are removed from the answer options, suggesting that framing bias reflects broader semantic steering rather than simple label copying. Chain-of-thought reasoning mitigates the effect substantially, but residual directional drift remains. These findings suggest that prompt sensitivity in arithmetic reasoning can exhibit structured semantic bias toward linguistically cued outcomes, with implications for robustness evaluation and benchmark design.
%R 10.18653/2026.findings-emnlp.1989
%U https://aclanthology.org/2026.findings-emnlp.1989/
%U https://doi.org/10.18653/2026.findings-emnlp.1989
%P 42392-42406
Markdown (Informal)
[Directional Drift in Arithmetic Reasoning: How Comparative Cues Steer LLM Predictions](https://aclanthology.org/2026.findings-emnlp.1989/) (Shafiei et al., Findings 2026)
ACL