@inproceedings{otsuka-etal-2026-multi,
title = "Multi-dimensional Evaluation of Character-Authentic Dialogue Models Learned from Question-Answer Data",
author = "Otsuka, Atsushi and
Matsuo, Kazuya and
Hama, Kenta and
Mizukami, Masahiro and
Arimoto, Tsunehiro and
Sugiyama, Hiroaki and
Nakatsuji, Makoto and
Nomoto, Narichika",
editor = "Piperidis, Stelios and
Bel, N{\'u}ria and
van den Heuvel, Henk and
Ide, Nancy and
Krek, Simon and
Toral, Antonio",
booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
month = may,
year = "2026",
address = "Palma de Mallorca, Spain",
publisher = "ELRA Language Resource Association",
url = "https://aclanthology.org/2026.lrec-1.209/",
doi = "10.63317/4ixiaqyfsd93",
pages = "2667--2681",
abstract = "Character-authentic dialogue remains challenging for large language models (LLMs) due to limited character-specific data, generic-style collapse, and hallucinations regarding persona facts. Our work presents a comparative evaluation of several learning strategies for character dialogue grounded in question{--}answer (QA) data, comparing zero/few-shot prompting, supervised fine-tuning (SFT), direct preference optimization (DPO), and a hybrid approach that integrates retrieval-augmented character profiles and knowledge with policy optimization. Using both single-turn and multi-turn settings, we assess multiple dimensions central to character dialogue quality: reproducibility, diversity, hallucination, and character authenticity. Results show that SFT excels in reproducibility and hallucination reduction but tends to shorten and simplify outputs, thereby reducing diversity and authenticity. DPO improves stylistic fidelity and authenticity but depends strongly on externalized character knowledge to limit hallucinations. The hybrid variant that combines character-knowledge retrieval with DPO achieves the best overall balance, delivering strong authenticity while maintaining factual consistency and competitive reproducibility in both single- and multi-turn dialogues. We further analyze the model{'}s sensitivity to knowledge retrieval and response-length effects and discuss trade-offs among optimization targets that inform practical design choices for developing faithful and engaging character agents trained from scalable QA resources."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="otsuka-etal-2026-multi">
<titleInfo>
<title>Multi-dimensional Evaluation of Character-Authentic Dialogue Models Learned from Question-Answer Data</title>
</titleInfo>
<name type="personal">
<namePart type="given">Atsushi</namePart>
<namePart type="family">Otsuka</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kazuya</namePart>
<namePart type="family">Matsuo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kenta</namePart>
<namePart type="family">Hama</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Masahiro</namePart>
<namePart type="family">Mizukami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tsunehiro</namePart>
<namePart type="family">Arimoto</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hiroaki</namePart>
<namePart type="family">Sugiyama</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Makoto</namePart>
<namePart type="family">Nakatsuji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Narichika</namePart>
<namePart type="family">Nomoto</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Fifteenth Language Resources and Evaluation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Stelios</namePart>
<namePart type="family">Piperidis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Núria</namePart>
<namePart type="family">Bel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Henk</namePart>
<namePart type="family">van den Heuvel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nancy</namePart>
<namePart type="family">Ide</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Simon</namePart>
<namePart type="family">Krek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Antonio</namePart>
<namePart type="family">Toral</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resource Association</publisher>
<place>
<placeTerm type="text">Palma de Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Character-authentic dialogue remains challenging for large language models (LLMs) due to limited character-specific data, generic-style collapse, and hallucinations regarding persona facts. Our work presents a comparative evaluation of several learning strategies for character dialogue grounded in question–answer (QA) data, comparing zero/few-shot prompting, supervised fine-tuning (SFT), direct preference optimization (DPO), and a hybrid approach that integrates retrieval-augmented character profiles and knowledge with policy optimization. Using both single-turn and multi-turn settings, we assess multiple dimensions central to character dialogue quality: reproducibility, diversity, hallucination, and character authenticity. Results show that SFT excels in reproducibility and hallucination reduction but tends to shorten and simplify outputs, thereby reducing diversity and authenticity. DPO improves stylistic fidelity and authenticity but depends strongly on externalized character knowledge to limit hallucinations. The hybrid variant that combines character-knowledge retrieval with DPO achieves the best overall balance, delivering strong authenticity while maintaining factual consistency and competitive reproducibility in both single- and multi-turn dialogues. We further analyze the model’s sensitivity to knowledge retrieval and response-length effects and discuss trade-offs among optimization targets that inform practical design choices for developing faithful and engaging character agents trained from scalable QA resources.</abstract>
<identifier type="citekey">otsuka-etal-2026-multi</identifier>
<identifier type="doi">10.63317/4ixiaqyfsd93</identifier>
<location>
<url>https://aclanthology.org/2026.lrec-1.209/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>2667</start>
<end>2681</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Multi-dimensional Evaluation of Character-Authentic Dialogue Models Learned from Question-Answer Data
%A Otsuka, Atsushi
%A Matsuo, Kazuya
%A Hama, Kenta
%A Mizukami, Masahiro
%A Arimoto, Tsunehiro
%A Sugiyama, Hiroaki
%A Nakatsuji, Makoto
%A Nomoto, Narichika
%Y Piperidis, Stelios
%Y Bel, Núria
%Y van den Heuvel, Henk
%Y Ide, Nancy
%Y Krek, Simon
%Y Toral, Antonio
%S Proceedings of the Fifteenth Language Resources and Evaluation Conference
%D 2026
%8 May
%I ELRA Language Resource Association
%C Palma de Mallorca, Spain
%F otsuka-etal-2026-multi
%X Character-authentic dialogue remains challenging for large language models (LLMs) due to limited character-specific data, generic-style collapse, and hallucinations regarding persona facts. Our work presents a comparative evaluation of several learning strategies for character dialogue grounded in question–answer (QA) data, comparing zero/few-shot prompting, supervised fine-tuning (SFT), direct preference optimization (DPO), and a hybrid approach that integrates retrieval-augmented character profiles and knowledge with policy optimization. Using both single-turn and multi-turn settings, we assess multiple dimensions central to character dialogue quality: reproducibility, diversity, hallucination, and character authenticity. Results show that SFT excels in reproducibility and hallucination reduction but tends to shorten and simplify outputs, thereby reducing diversity and authenticity. DPO improves stylistic fidelity and authenticity but depends strongly on externalized character knowledge to limit hallucinations. The hybrid variant that combines character-knowledge retrieval with DPO achieves the best overall balance, delivering strong authenticity while maintaining factual consistency and competitive reproducibility in both single- and multi-turn dialogues. We further analyze the model’s sensitivity to knowledge retrieval and response-length effects and discuss trade-offs among optimization targets that inform practical design choices for developing faithful and engaging character agents trained from scalable QA resources.
%R 10.63317/4ixiaqyfsd93
%U https://aclanthology.org/2026.lrec-1.209/
%U https://doi.org/10.63317/4ixiaqyfsd93
%P 2667-2681
Markdown (Informal)
[Multi-dimensional Evaluation of Character-Authentic Dialogue Models Learned from Question-Answer Data](https://aclanthology.org/2026.lrec-1.209/) (Otsuka et al., LREC 2026)
ACL
- Atsushi Otsuka, Kazuya Matsuo, Kenta Hama, Masahiro Mizukami, Tsunehiro Arimoto, Hiroaki Sugiyama, Makoto Nakatsuji, and Narichika Nomoto. 2026. Multi-dimensional Evaluation of Character-Authentic Dialogue Models Learned from Question-Answer Data. In Proceedings of the Fifteenth Language Resources and Evaluation Conference, pages 2667–2681, Palma de Mallorca, Spain. ELRA Language Resource Association.