@inproceedings{peng-etal-2026-timeclaimbench,
title = "{T}ime{C}laim{B}ench: A Claim-Level Benchmark for Grounded {LLM} Explanations of Time Series",
author = "Peng, Sijia and
Zhang, Fan and
Zhou, Yixi and
Li, Changlun and
Xiong, Yun and
Zhu, Yangyong and
Chen, Xi and
Yu, Yanwei and
Tang, Nan",
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.1042/",
pages = "22071--22094",
ISBN = "979-8-89176-462-0",
abstract = "Large language models (LLMs) are increasingly used to explain numerical time series, but fluent explanations can misstate when a change occurs, overstate a trend, or introduce unsupported patterns. Existing evaluations often score forecasts, answers, or whole explanations, leaving claim-level temporal grounding undermeasured. We introduce TimeClaimBench, a benchmark that parses explanations into atomic temporal claims and verifies them against a deterministic extracted event graph. Across five LLMs on controlled synthetic series, event-grounded prompting raises the strict graph-supported rate by 18.3 points and reduces graph-unsupported claims by 27.9 points under this fixed operational criterion. Calibration analyses show the boundary of that result. The extractor recovers the designed primary event in 394 of 400 series, but this does not estimate the precision of additional extracted events. A blinded two-annotator comparison on one model and 40 series reverses the automatic ordering: holistic ratings favor Direct and chain-of-thought outputs over event-grounded outputs. therefore distinguishes four axes that should not be conflated: designed-event recovery, the unresolved precision of additional events, claim-to-graph consistency, and whole-explanation human judgment. Code: https://github.com/lunaaa95/TimeClaimBench"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="peng-etal-2026-timeclaimbench">
<titleInfo>
<title>TimeClaimBench: A Claim-Level Benchmark for Grounded LLM Explanations of Time Series</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sijia</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixi</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Changlun</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yun</namePart>
<namePart type="family">Xiong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yangyong</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xi</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanwei</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nan</namePart>
<namePart type="family">Tang</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>Large language models (LLMs) are increasingly used to explain numerical time series, but fluent explanations can misstate when a change occurs, overstate a trend, or introduce unsupported patterns. Existing evaluations often score forecasts, answers, or whole explanations, leaving claim-level temporal grounding undermeasured. We introduce TimeClaimBench, a benchmark that parses explanations into atomic temporal claims and verifies them against a deterministic extracted event graph. Across five LLMs on controlled synthetic series, event-grounded prompting raises the strict graph-supported rate by 18.3 points and reduces graph-unsupported claims by 27.9 points under this fixed operational criterion. Calibration analyses show the boundary of that result. The extractor recovers the designed primary event in 394 of 400 series, but this does not estimate the precision of additional extracted events. A blinded two-annotator comparison on one model and 40 series reverses the automatic ordering: holistic ratings favor Direct and chain-of-thought outputs over event-grounded outputs. therefore distinguishes four axes that should not be conflated: designed-event recovery, the unresolved precision of additional events, claim-to-graph consistency, and whole-explanation human judgment. Code: https://github.com/lunaaa95/TimeClaimBench</abstract>
<identifier type="citekey">peng-etal-2026-timeclaimbench</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1042/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>22071</start>
<end>22094</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T TimeClaimBench: A Claim-Level Benchmark for Grounded LLM Explanations of Time Series
%A Peng, Sijia
%A Zhang, Fan
%A Zhou, Yixi
%A Li, Changlun
%A Xiong, Yun
%A Zhu, Yangyong
%A Chen, Xi
%A Yu, Yanwei
%A Tang, Nan
%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 peng-etal-2026-timeclaimbench
%X Large language models (LLMs) are increasingly used to explain numerical time series, but fluent explanations can misstate when a change occurs, overstate a trend, or introduce unsupported patterns. Existing evaluations often score forecasts, answers, or whole explanations, leaving claim-level temporal grounding undermeasured. We introduce TimeClaimBench, a benchmark that parses explanations into atomic temporal claims and verifies them against a deterministic extracted event graph. Across five LLMs on controlled synthetic series, event-grounded prompting raises the strict graph-supported rate by 18.3 points and reduces graph-unsupported claims by 27.9 points under this fixed operational criterion. Calibration analyses show the boundary of that result. The extractor recovers the designed primary event in 394 of 400 series, but this does not estimate the precision of additional extracted events. A blinded two-annotator comparison on one model and 40 series reverses the automatic ordering: holistic ratings favor Direct and chain-of-thought outputs over event-grounded outputs. therefore distinguishes four axes that should not be conflated: designed-event recovery, the unresolved precision of additional events, claim-to-graph consistency, and whole-explanation human judgment. Code: https://github.com/lunaaa95/TimeClaimBench
%U https://aclanthology.org/2026.findings-emnlp.1042/
%P 22071-22094
Markdown (Informal)
[TimeClaimBench: A Claim-Level Benchmark for Grounded LLM Explanations of Time Series](https://aclanthology.org/2026.findings-emnlp.1042/) (Peng et al., Findings 2026)
ACL
- Sijia Peng, Fan Zhang, Yixi Zhou, Changlun Li, Yun Xiong, Yangyong Zhu, Xi Chen, Yanwei Yu, and Nan Tang. 2026. TimeClaimBench: A Claim-Level Benchmark for Grounded LLM Explanations of Time Series. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 22071–22094, Budapest, Hungary. Association for Computational Linguistics.