@inproceedings{concina-etal-2026-scalable,
title = "Scalable Expansion of Multilingual Speech {LLM}s for {ASR}: A Continual Learning Approach",
author = "Concina, Lorenzo and
Matassoni, Marco and
Brutti, Alessio",
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-1.7/",
doi = "10.63317/5b86xj9f28dp",
pages = "59--65",
abstract = {Speech Large Language Models have recently enabled the processing of spoken language by coupling powerful language models (LLMs) with pre-trained speech encoders. However, their multilingual scalability remains limited, particularly for low - resource and unseen languages, while na{\"i}ve fine- tuning often triggers catastrophic forgetting of previously learned languages. This work investigates how Continual Learning (CL) can be used to sustainably expand multilingual Speech LLMs. We first demonstrate that multilingual projectors can be efficiently bootstrapped to new languages , even with extremely small datasets, but at the cost of severe degradation on the original supported languages. To address this, we adopt rehearsal-based CL strategies and show that interleaving even small amounts of replay data effectively stabilizes multilingual performance. Through extensive ablations, we quantify the minimum rehearsal budget required to prevent forgetting and identify fragile languages that require more targeted reinforcement. We further evaluate sequential acquisition of four linguistically diverse languages (Ukrainian, Japanese, Thai, and Vietnamese), revealing the trade -offs between buffer size and long- term stability. Finally, based on these empirical observations, we propose a Fragility-Based Sampling heuristic as a pathway to allocate rehearsal data more efficiently by tiering languages according to their stability thresholds. Our findings provide a practical roadmap for scalable, resource-efficient multilingual expansion of Speech LLMs, enabling inclusive ASR systems that can grow over time without sacrificing prior knowledge.}
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="concina-etal-2026-scalable">
<titleInfo>
<title>Scalable Expansion of Multilingual Speech LLMs for ASR: A Continual Learning Approach</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lorenzo</namePart>
<namePart type="family">Concina</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Matassoni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alessio</namePart>
<namePart type="family">Brutti</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>Speech Large Language Models have recently enabled the processing of spoken language by coupling powerful language models (LLMs) with pre-trained speech encoders. However, their multilingual scalability remains limited, particularly for low - resource and unseen languages, while naïve fine- tuning often triggers catastrophic forgetting of previously learned languages. This work investigates how Continual Learning (CL) can be used to sustainably expand multilingual Speech LLMs. We first demonstrate that multilingual projectors can be efficiently bootstrapped to new languages , even with extremely small datasets, but at the cost of severe degradation on the original supported languages. To address this, we adopt rehearsal-based CL strategies and show that interleaving even small amounts of replay data effectively stabilizes multilingual performance. Through extensive ablations, we quantify the minimum rehearsal budget required to prevent forgetting and identify fragile languages that require more targeted reinforcement. We further evaluate sequential acquisition of four linguistically diverse languages (Ukrainian, Japanese, Thai, and Vietnamese), revealing the trade -offs between buffer size and long- term stability. Finally, based on these empirical observations, we propose a Fragility-Based Sampling heuristic as a pathway to allocate rehearsal data more efficiently by tiering languages according to their stability thresholds. Our findings provide a practical roadmap for scalable, resource-efficient multilingual expansion of Speech LLMs, enabling inclusive ASR systems that can grow over time without sacrificing prior knowledge.</abstract>
<identifier type="citekey">concina-etal-2026-scalable</identifier>
<identifier type="doi">10.63317/5b86xj9f28dp</identifier>
<location>
<url>https://aclanthology.org/2026.speakable-1.7/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>59</start>
<end>65</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Scalable Expansion of Multilingual Speech LLMs for ASR: A Continual Learning Approach
%A Concina, Lorenzo
%A Matassoni, Marco
%A Brutti, Alessio
%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 concina-etal-2026-scalable
%X Speech Large Language Models have recently enabled the processing of spoken language by coupling powerful language models (LLMs) with pre-trained speech encoders. However, their multilingual scalability remains limited, particularly for low - resource and unseen languages, while naïve fine- tuning often triggers catastrophic forgetting of previously learned languages. This work investigates how Continual Learning (CL) can be used to sustainably expand multilingual Speech LLMs. We first demonstrate that multilingual projectors can be efficiently bootstrapped to new languages , even with extremely small datasets, but at the cost of severe degradation on the original supported languages. To address this, we adopt rehearsal-based CL strategies and show that interleaving even small amounts of replay data effectively stabilizes multilingual performance. Through extensive ablations, we quantify the minimum rehearsal budget required to prevent forgetting and identify fragile languages that require more targeted reinforcement. We further evaluate sequential acquisition of four linguistically diverse languages (Ukrainian, Japanese, Thai, and Vietnamese), revealing the trade -offs between buffer size and long- term stability. Finally, based on these empirical observations, we propose a Fragility-Based Sampling heuristic as a pathway to allocate rehearsal data more efficiently by tiering languages according to their stability thresholds. Our findings provide a practical roadmap for scalable, resource-efficient multilingual expansion of Speech LLMs, enabling inclusive ASR systems that can grow over time without sacrificing prior knowledge.
%R 10.63317/5b86xj9f28dp
%U https://aclanthology.org/2026.speakable-1.7/
%U https://doi.org/10.63317/5b86xj9f28dp
%P 59-65
Markdown (Informal)
[Scalable Expansion of Multilingual Speech LLMs for ASR: A Continual Learning Approach](https://aclanthology.org/2026.speakable-1.7/) (Concina et al., SPEAKABLE 2026)
ACL