@inproceedings{gustafsson-etal-2026-text,
title = "Text-Based Veracity Assessment of Verbal Statements in {S}wedish",
author = "Gustafsson, Clara and
Lid{\'e}n, Moa and
Nivre, Joakim",
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.2438/",
doi = "10.18653/2026.findings-emnlp.2438",
pages = "52303--52320",
ISBN = "979-8-89176-462-0",
abstract = "Determining the veracity of a verbal statement is a frequent and challenging task for legal practitioners. The Swedish Supreme Court has proposed a set of truth-indicative reliability criteria for assessing verbal statements in the Swedish legal system. LLMs' growing ability to reason and handle complex tasks raises questions about how they perform veracity judgments of verbal statements. In this study, we explore this question on a novel Swedish mock-crime dataset. We compare simple prompting to a two-step method in which the LLM is first asked to assess the reliability criteria and then perform a criteria-informed veracity judgment. We find that accuracy is at chance level under both conditions, but that the two-step method often leads to a more reasonable truth bias and reduces models' overconfidence. Further analysis indicates that criteria assessments predict final judgments, but not ground truth. We also find that the statement category (true or false and confession or denial) matters to overall performance and the effect of two-step prompting."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gustafsson-etal-2026-text">
<titleInfo>
<title>Text-Based Veracity Assessment of Verbal Statements in Swedish</title>
</titleInfo>
<name type="personal">
<namePart type="given">Clara</namePart>
<namePart type="family">Gustafsson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Moa</namePart>
<namePart type="family">Lidén</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joakim</namePart>
<namePart type="family">Nivre</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>Determining the veracity of a verbal statement is a frequent and challenging task for legal practitioners. The Swedish Supreme Court has proposed a set of truth-indicative reliability criteria for assessing verbal statements in the Swedish legal system. LLMs’ growing ability to reason and handle complex tasks raises questions about how they perform veracity judgments of verbal statements. In this study, we explore this question on a novel Swedish mock-crime dataset. We compare simple prompting to a two-step method in which the LLM is first asked to assess the reliability criteria and then perform a criteria-informed veracity judgment. We find that accuracy is at chance level under both conditions, but that the two-step method often leads to a more reasonable truth bias and reduces models’ overconfidence. Further analysis indicates that criteria assessments predict final judgments, but not ground truth. We also find that the statement category (true or false and confession or denial) matters to overall performance and the effect of two-step prompting.</abstract>
<identifier type="citekey">gustafsson-etal-2026-text</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.2438</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2438/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>52303</start>
<end>52320</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Text-Based Veracity Assessment of Verbal Statements in Swedish
%A Gustafsson, Clara
%A Lidén, Moa
%A Nivre, Joakim
%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 gustafsson-etal-2026-text
%X Determining the veracity of a verbal statement is a frequent and challenging task for legal practitioners. The Swedish Supreme Court has proposed a set of truth-indicative reliability criteria for assessing verbal statements in the Swedish legal system. LLMs’ growing ability to reason and handle complex tasks raises questions about how they perform veracity judgments of verbal statements. In this study, we explore this question on a novel Swedish mock-crime dataset. We compare simple prompting to a two-step method in which the LLM is first asked to assess the reliability criteria and then perform a criteria-informed veracity judgment. We find that accuracy is at chance level under both conditions, but that the two-step method often leads to a more reasonable truth bias and reduces models’ overconfidence. Further analysis indicates that criteria assessments predict final judgments, but not ground truth. We also find that the statement category (true or false and confession or denial) matters to overall performance and the effect of two-step prompting.
%R 10.18653/2026.findings-emnlp.2438
%U https://aclanthology.org/2026.findings-emnlp.2438/
%U https://doi.org/10.18653/2026.findings-emnlp.2438
%P 52303-52320
Markdown (Informal)
[Text-Based Veracity Assessment of Verbal Statements in Swedish](https://aclanthology.org/2026.findings-emnlp.2438/) (Gustafsson et al., Findings 2026)
ACL