@inproceedings{ni-dheorain-etal-2026-less,
title = "Doing More with Less: Determining Optimal Pre-training Model for {I}rish Automatic Speech Recognition through Multi-step Fine-tuning",
author = "N{\'i} Dheor{\'a}in, Caoilfhionn and
Holmes, Ruth and
Evans, Nicholas and
Laurent, Thomas and
Ventresque, Anthony and
Rushe, Ellen",
editor = "Hosseini-Kivanani, Nina and
Brutti, Alessio and
Matassoni, Marco and
Dowerah, Sandipana and
Liga, Davide and
Schommer, Christoph",
booktitle = "Proceedings of Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis ({SPEAKABLE}) @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.speakable-1.18/",
doi = "10.63317/43cqo4vswrry",
pages = "162--173",
abstract = "In recent years, there has been an upsurge in research on automatic speech recognition (ASR) for low-resource languages. Particularly, transfer learning using multi-lingual models has become a popular remedy for the lack of available datasets for target languages. However, given the complexities associated with each individual language, we argue it is unlikely that a single multi-lingual pre-training model will provide equal performance gains across all languages. We also recognise the important, and insufficiently studied influence that the specific pre-training dataset has on the performance of the model. In this paper, using the Irish language as a case study, we propose a more directed, incremental form of pre-training which we term multi-step fine-tuning. This method accounts for the complex relationships between the language and dataset features of the source pre-training and target datasets. We show multi-step fine-tuning improves performance over simple multi-lingual fine-tuning alone, and we investigate factors leading to certain pre-trained models achieving better results through linguistic and dataset similarity measures. This research also investigates the uniformity of the performance gains across different demographics. We show that the optimal pre-training strategy can differ between demographics suggesting that more careful pre-training dataset selection is necessary to ensure equitable outcomes in practice."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ni-dheorain-etal-2026-less">
<titleInfo>
<title>Doing More with Less: Determining Optimal Pre-training Model for Irish Automatic Speech Recognition through Multi-step Fine-tuning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Caoilfhionn</namePart>
<namePart type="family">Ní Dheoráin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruth</namePart>
<namePart type="family">Holmes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nicholas</namePart>
<namePart type="family">Evans</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Laurent</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anthony</namePart>
<namePart type="family">Ventresque</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ellen</namePart>
<namePart type="family">Rushe</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 Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis (SPEAKABLE) @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nina</namePart>
<namePart type="family">Hosseini-Kivanani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alessio</namePart>
<namePart type="family">Brutti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Matassoni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sandipana</namePart>
<namePart type="family">Dowerah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Davide</namePart>
<namePart type="family">Liga</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christoph</namePart>
<namePart type="family">Schommer</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In recent years, there has been an upsurge in research on automatic speech recognition (ASR) for low-resource languages. Particularly, transfer learning using multi-lingual models has become a popular remedy for the lack of available datasets for target languages. However, given the complexities associated with each individual language, we argue it is unlikely that a single multi-lingual pre-training model will provide equal performance gains across all languages. We also recognise the important, and insufficiently studied influence that the specific pre-training dataset has on the performance of the model. In this paper, using the Irish language as a case study, we propose a more directed, incremental form of pre-training which we term multi-step fine-tuning. This method accounts for the complex relationships between the language and dataset features of the source pre-training and target datasets. We show multi-step fine-tuning improves performance over simple multi-lingual fine-tuning alone, and we investigate factors leading to certain pre-trained models achieving better results through linguistic and dataset similarity measures. This research also investigates the uniformity of the performance gains across different demographics. We show that the optimal pre-training strategy can differ between demographics suggesting that more careful pre-training dataset selection is necessary to ensure equitable outcomes in practice.</abstract>
<identifier type="citekey">ni-dheorain-etal-2026-less</identifier>
<identifier type="doi">10.63317/43cqo4vswrry</identifier>
<location>
<url>https://aclanthology.org/2026.speakable-1.18/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>162</start>
<end>173</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Doing More with Less: Determining Optimal Pre-training Model for Irish Automatic Speech Recognition through Multi-step Fine-tuning
%A Ní Dheoráin, Caoilfhionn
%A Holmes, Ruth
%A Evans, Nicholas
%A Laurent, Thomas
%A Ventresque, Anthony
%A Rushe, Ellen
%Y Hosseini-Kivanani, Nina
%Y Brutti, Alessio
%Y Matassoni, Marco
%Y Dowerah, Sandipana
%Y Liga, Davide
%Y Schommer, Christoph
%S Proceedings of Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis (SPEAKABLE) @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F ni-dheorain-etal-2026-less
%X In recent years, there has been an upsurge in research on automatic speech recognition (ASR) for low-resource languages. Particularly, transfer learning using multi-lingual models has become a popular remedy for the lack of available datasets for target languages. However, given the complexities associated with each individual language, we argue it is unlikely that a single multi-lingual pre-training model will provide equal performance gains across all languages. We also recognise the important, and insufficiently studied influence that the specific pre-training dataset has on the performance of the model. In this paper, using the Irish language as a case study, we propose a more directed, incremental form of pre-training which we term multi-step fine-tuning. This method accounts for the complex relationships between the language and dataset features of the source pre-training and target datasets. We show multi-step fine-tuning improves performance over simple multi-lingual fine-tuning alone, and we investigate factors leading to certain pre-trained models achieving better results through linguistic and dataset similarity measures. This research also investigates the uniformity of the performance gains across different demographics. We show that the optimal pre-training strategy can differ between demographics suggesting that more careful pre-training dataset selection is necessary to ensure equitable outcomes in practice.
%R 10.63317/43cqo4vswrry
%U https://aclanthology.org/2026.speakable-1.18/
%U https://doi.org/10.63317/43cqo4vswrry
%P 162-173
Markdown (Informal)
[Doing More with Less: Determining Optimal Pre-training Model for Irish Automatic Speech Recognition through Multi-step Fine-tuning](https://aclanthology.org/2026.speakable-1.18/) (Ní Dheoráin et al., SPEAKABLE 2026)
ACL
- Caoilfhionn Ní Dheoráin, Ruth Holmes, Nicholas Evans, Thomas Laurent, Anthony Ventresque, and Ellen Rushe. 2026. Doing More with Less: Determining Optimal Pre-training Model for Irish Automatic Speech Recognition through Multi-step Fine-tuning. In Proceedings of Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis (SPEAKABLE) @ LREC 2026, pages 162–173, Palma, Mallorca (Spain). ELRA Language Resources Association (ELRA).