@inproceedings{prud-hommeaux-etal-2026-family,
title = "A Family Affair: Counterintuitive transfer effects in {ASR} for endangered languages",
author = "Prud{'}hommeaux, Emily and
Scheppat, Hunter and
Hernandez, Nicholas and
Le Ferrand, Eric",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1246/",
doi = "10.18653/2026.emnlp-main.1246",
pages = "28389--28397",
ISBN = "979-8-89176-459-0",
abstract = "Recent work on ASR for endangered languages has focused on fine-tuning a pretrained multilingual model to a small amount of data in the target language, but even with large and diverse pretrained models, error rates can be stubbornly high. Including additional data from the target language can yield improvements, but acquiring this kind of data for endangered languages is challenging. One commonly proposed alternative is to leverage data in a more highly resourced language from the same language family, but there has been remarkably little work evaluating the real-world utility of this practice. In this short paper, we explore the relevance of phylogenetic relatedness of transfer data for improving ASR for endangered languages with minimal training resources. We find, unexpectedly, that augmenting with unrelated donor language data improves word error rate (WER) more than related language data. In addition, the performance improvement observed via transfer of unrelated data is larger than that observed when nearly doubling the quantity of in-language training data. These surprising results hold even when controlling for factors such as acoustic similarity, audio quality, and subword overlap. Our findings point the way to practical new approaches for improving ASR for endangered languages."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="prud-hommeaux-etal-2026-family">
<titleInfo>
<title>A Family Affair: Counterintuitive transfer effects in ASR for endangered languages</title>
</titleInfo>
<name type="personal">
<namePart type="given">Emily</namePart>
<namePart type="family">Prud’hommeaux</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hunter</namePart>
<namePart type="family">Scheppat</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nicholas</namePart>
<namePart type="family">Hernandez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eric</namePart>
<namePart type="family">Le Ferrand</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>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Recent work on ASR for endangered languages has focused on fine-tuning a pretrained multilingual model to a small amount of data in the target language, but even with large and diverse pretrained models, error rates can be stubbornly high. Including additional data from the target language can yield improvements, but acquiring this kind of data for endangered languages is challenging. One commonly proposed alternative is to leverage data in a more highly resourced language from the same language family, but there has been remarkably little work evaluating the real-world utility of this practice. In this short paper, we explore the relevance of phylogenetic relatedness of transfer data for improving ASR for endangered languages with minimal training resources. We find, unexpectedly, that augmenting with unrelated donor language data improves word error rate (WER) more than related language data. In addition, the performance improvement observed via transfer of unrelated data is larger than that observed when nearly doubling the quantity of in-language training data. These surprising results hold even when controlling for factors such as acoustic similarity, audio quality, and subword overlap. Our findings point the way to practical new approaches for improving ASR for endangered languages.</abstract>
<identifier type="citekey">prud-hommeaux-etal-2026-family</identifier>
<identifier type="doi">10.18653/2026.emnlp-main.1246</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1246/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>28389</start>
<end>28397</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T A Family Affair: Counterintuitive transfer effects in ASR for endangered languages
%A Prud’hommeaux, Emily
%A Scheppat, Hunter
%A Hernandez, Nicholas
%A Le Ferrand, Eric
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F prud-hommeaux-etal-2026-family
%X Recent work on ASR for endangered languages has focused on fine-tuning a pretrained multilingual model to a small amount of data in the target language, but even with large and diverse pretrained models, error rates can be stubbornly high. Including additional data from the target language can yield improvements, but acquiring this kind of data for endangered languages is challenging. One commonly proposed alternative is to leverage data in a more highly resourced language from the same language family, but there has been remarkably little work evaluating the real-world utility of this practice. In this short paper, we explore the relevance of phylogenetic relatedness of transfer data for improving ASR for endangered languages with minimal training resources. We find, unexpectedly, that augmenting with unrelated donor language data improves word error rate (WER) more than related language data. In addition, the performance improvement observed via transfer of unrelated data is larger than that observed when nearly doubling the quantity of in-language training data. These surprising results hold even when controlling for factors such as acoustic similarity, audio quality, and subword overlap. Our findings point the way to practical new approaches for improving ASR for endangered languages.
%R 10.18653/2026.emnlp-main.1246
%U https://aclanthology.org/2026.emnlp-main.1246/
%U https://doi.org/10.18653/2026.emnlp-main.1246
%P 28389-28397
Markdown (Informal)
[A Family Affair: Counterintuitive transfer effects in ASR for endangered languages](https://aclanthology.org/2026.emnlp-main.1246/) (Prud’hommeaux et al., EMNLP 2026)
ACL