@inproceedings{hassan-bhatti-etal-2026-multi,
title = "Multi-Task Instruction Tuning via Data Scheduling for Low-Resource {A}rabic {S}peech{LLM}s",
author = "Hassan Bhatti, Hunzalah and
Alam, Firoj and
Chowdhury, Shammur Absar",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.157/",
pages = "3473--3490",
ISBN = "979-8-89176-439-2",
abstract = "Audio large language models (LLMs) enable unified speech understanding and generation, but adapting them to linguistically complex and dialect-rich settings such as Arabic-English remains challenging. We present a controlled study of multi-task instruction tuning for an Arabic-centric audio LLM across generative tasks including ASR and speech and text summarization, and discriminative tasks including dialect and emotion recognition, in a resource-constrained setting. To support end-to-end Arabic speech summarization, we introduce AraMega-SSum, a first speech summarization resource for training and benchmarking Arabic-centric Audio-LLMs. We compare four training strategies (i) Uniform Mixing (UM), (ii) Task-Progressive Curriculum (TPC), (iii) Aligner-Based Diverse Sampling (ADS) for training-time batch construction, and (iv) A two-stage TPC{\textrightarrow}ADS strategy. Our results show a clear task-dependent trade-off: TPC and uniform mixing favor generative tasks, while ADS is strongest for SER. The two-stage TPC{\textrightarrow}ADS strategy achieves the best DID performance across both backbones while preserving competitive generative quality. The adapted models also outperform zero-shot proprietary baselines, including Gemini-2.5-Pro, on discriminative tasks. We made AraMega-SSum and all experimental resources available to the community (https://huggingface.co/datasets/QCRI/AraMega-SSum)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hassan-bhatti-etal-2026-multi">
<titleInfo>
<title>Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic SpeechLLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hunzalah</namePart>
<namePart type="family">Hassan Bhatti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Firoj</namePart>
<namePart type="family">Alam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shammur</namePart>
<namePart type="given">Absar</namePart>
<namePart type="family">Chowdhury</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Audio large language models (LLMs) enable unified speech understanding and generation, but adapting them to linguistically complex and dialect-rich settings such as Arabic-English remains challenging. We present a controlled study of multi-task instruction tuning for an Arabic-centric audio LLM across generative tasks including ASR and speech and text summarization, and discriminative tasks including dialect and emotion recognition, in a resource-constrained setting. To support end-to-end Arabic speech summarization, we introduce AraMega-SSum, a first speech summarization resource for training and benchmarking Arabic-centric Audio-LLMs. We compare four training strategies (i) Uniform Mixing (UM), (ii) Task-Progressive Curriculum (TPC), (iii) Aligner-Based Diverse Sampling (ADS) for training-time batch construction, and (iv) A two-stage TPC→ADS strategy. Our results show a clear task-dependent trade-off: TPC and uniform mixing favor generative tasks, while ADS is strongest for SER. The two-stage TPC→ADS strategy achieves the best DID performance across both backbones while preserving competitive generative quality. The adapted models also outperform zero-shot proprietary baselines, including Gemini-2.5-Pro, on discriminative tasks. We made AraMega-SSum and all experimental resources available to the community (https://huggingface.co/datasets/QCRI/AraMega-SSum).</abstract>
<identifier type="citekey">hassan-bhatti-etal-2026-multi</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.157/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3473</start>
<end>3490</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic SpeechLLMs
%A Hassan Bhatti, Hunzalah
%A Alam, Firoj
%A Chowdhury, Shammur Absar
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F hassan-bhatti-etal-2026-multi
%X Audio large language models (LLMs) enable unified speech understanding and generation, but adapting them to linguistically complex and dialect-rich settings such as Arabic-English remains challenging. We present a controlled study of multi-task instruction tuning for an Arabic-centric audio LLM across generative tasks including ASR and speech and text summarization, and discriminative tasks including dialect and emotion recognition, in a resource-constrained setting. To support end-to-end Arabic speech summarization, we introduce AraMega-SSum, a first speech summarization resource for training and benchmarking Arabic-centric Audio-LLMs. We compare four training strategies (i) Uniform Mixing (UM), (ii) Task-Progressive Curriculum (TPC), (iii) Aligner-Based Diverse Sampling (ADS) for training-time batch construction, and (iv) A two-stage TPC→ADS strategy. Our results show a clear task-dependent trade-off: TPC and uniform mixing favor generative tasks, while ADS is strongest for SER. The two-stage TPC→ADS strategy achieves the best DID performance across both backbones while preserving competitive generative quality. The adapted models also outperform zero-shot proprietary baselines, including Gemini-2.5-Pro, on discriminative tasks. We made AraMega-SSum and all experimental resources available to the community (https://huggingface.co/datasets/QCRI/AraMega-SSum).
%U https://aclanthology.org/2026.aacl-long.157/
%P 3473-3490
Markdown (Informal)
[Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic SpeechLLMs](https://aclanthology.org/2026.aacl-long.157/) (Hassan Bhatti et al., AACL-IJCNLP 2026)
ACL
- Hunzalah Hassan Bhatti, Firoj Alam, and Shammur Absar Chowdhury. 2026. Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic SpeechLLMs. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3473–3490, Hengqin, China. Association for Computational Linguistics.