<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="arai-etal-2025-evaluating">
    <titleInfo>
        <title>Evaluating LLMs’ Ability to Understand Numerical Time Series for Text Generation</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Mizuki</namePart>
        <namePart type="family">Arai</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tatsuya</namePart>
        <namePart type="family">Ishigaki</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Masayuki</namePart>
        <namePart type="family">Kawarada</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yusuke</namePart>
        <namePart type="family">Miyao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hiroya</namePart>
        <namePart type="family">Takamura</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ichiro</namePart>
        <namePart type="family">Kobayashi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2025-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 18th International Natural Language Generation Conference</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Lucie</namePart>
            <namePart type="family">Flek</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Shashi</namePart>
            <namePart type="family">Narayan</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Lê</namePart>
            <namePart type="given">Hồng</namePart>
            <namePart type="family">Phương</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jiahuan</namePart>
            <namePart type="family">Pei</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hanoi, Vietnam</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>Data-to-text generation tasks often involve processing numerical time-series as input such as financial statistics or meteorological data. Although large language models (LLMs) are a powerful approach to data-to-text, we still lack a comprehensive understanding of how well they actually understand time-series data. We therefore introduce a benchmark with 18 evaluation tasks to assess LLMs’ abilities of interpreting numerical time-series, which are categorized into: 1) event detection—identifying maxima and minima; 2) computation—averaging and summation; 3) pairwise comparison—comparing values over time; and 4) inference—imputation and forecasting. Our experiments reveal five key findings: 1) even state-of-the-art LLMs struggle with complex multi-step reasoning; 2) tasks that require extracting values or performing computations within a specified range of the time-series significantly reduce accuracy; 3) instruction tuning offers inconsistent improvements for numerical interpretation; 4) reasoning-based models outperform standard LLMs in complex numerical tasks; and 5) LLMs perform interpolation better than forecasting. These results establish a clear baseline and serve as a wake-up call for anyone aiming to blend fluent language with trustworthy numeric precision in time-series scenarios.</abstract>
    <identifier type="citekey">arai-etal-2025-evaluating</identifier>
    <location>
        <url>https://aclanthology.org/2025.inlg-main.16/</url>
    </location>
    <part>
        <date>2025-10</date>
        <extent unit="page">
            <start>232</start>
            <end>248</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Evaluating LLMs’ Ability to Understand Numerical Time Series for Text Generation
%A Arai, Mizuki
%A Ishigaki, Tatsuya
%A Kawarada, Masayuki
%A Miyao, Yusuke
%A Takamura, Hiroya
%A Kobayashi, Ichiro
%Y Flek, Lucie
%Y Narayan, Shashi
%Y Phương, Lê Hồng
%Y Pei, Jiahuan
%S Proceedings of the 18th International Natural Language Generation Conference
%D 2025
%8 October
%I Association for Computational Linguistics
%C Hanoi, Vietnam
%F arai-etal-2025-evaluating
%X Data-to-text generation tasks often involve processing numerical time-series as input such as financial statistics or meteorological data. Although large language models (LLMs) are a powerful approach to data-to-text, we still lack a comprehensive understanding of how well they actually understand time-series data. We therefore introduce a benchmark with 18 evaluation tasks to assess LLMs’ abilities of interpreting numerical time-series, which are categorized into: 1) event detection—identifying maxima and minima; 2) computation—averaging and summation; 3) pairwise comparison—comparing values over time; and 4) inference—imputation and forecasting. Our experiments reveal five key findings: 1) even state-of-the-art LLMs struggle with complex multi-step reasoning; 2) tasks that require extracting values or performing computations within a specified range of the time-series significantly reduce accuracy; 3) instruction tuning offers inconsistent improvements for numerical interpretation; 4) reasoning-based models outperform standard LLMs in complex numerical tasks; and 5) LLMs perform interpolation better than forecasting. These results establish a clear baseline and serve as a wake-up call for anyone aiming to blend fluent language with trustworthy numeric precision in time-series scenarios.
%U https://aclanthology.org/2025.inlg-main.16/
%P 232-248

Markdown (Informal)

[Evaluating LLMs’ Ability to Understand Numerical Time Series for Text Generation](https://aclanthology.org/2025.inlg-main.16/) (Arai et al., INLG 2025)

• Evaluating LLMs’ Ability to Understand Numerical Time Series for Text Generation (Arai et al., INLG 2025)

ACL

• Mizuki Arai, Tatsuya Ishigaki, Masayuki Kawarada, Yusuke Miyao, Hiroya Takamura, and Ichiro Kobayashi. 2025. Evaluating LLMs’ Ability to Understand Numerical Time Series for Text Generation. In Proceedings of the 18th International Natural Language Generation Conference, pages 232–248, Hanoi, Vietnam. Association for Computational Linguistics.