@inproceedings{hisham-etal-2026-arabic,
title = "{A}rabic-Centric Preference Optimization for Efficient Dialect-Sensitive Machine Translation: A {S}eq2{S}eq Approach",
author = "Hisham, Abdelrahman and
Fathallah, Malak and
Khalafallah, Ayman",
editor = "Abdul-Mageed, Muhammad and
Elsayed, Tamer and
Bouamor, Houda and
El-Beltagy, Samhaa and
Al-Khalifa, Hend and
Abdelali, Ahmed and
Elmadany, AbdelRahim and
El Mekki, Abdellah and
Darwish, Kareem and
Alkhereyf, Sakhar and
Luqman, Hamzah and
Khalifa, Salam and
Magdy, Samar and
Alyafeai, Zaid and
Hamed, Injy and
Zaghouani, Wajdi and
Zitouni, Imed",
booktitle = "Proceedings of The Fourth {A}rabic Natural Language Processing Conference",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.arabicnlp-main.25/",
pages = "356--372",
ISBN = "979-8-89176-473-6",
abstract = "Arabic machine translation remains challenging because Modern Standard Arabic and dialectal varieties such as Egyptian Colloquial Arabic differ substantially in morphology, lexical choice, register, and everyday usage. We introduce a compact Arabic-centric encoder{--}decoder machine translation model designed to strengthen translation involving Modern Standard Arabic and Egyptian Arabic while preserving multilingual translation ability. We pretrain from scratch with multilingual Seq2Seq training and contrastive cross-lingual alignment, using an Arabic-centered data mixture covering Arabic, English, Egyptian Arabic, and 111 additional languages. We further apply Direct Preference Optimization (DPO) for our Seq2Seq model using grounded preference pairs targeting Arabic-specific translation errors, including literal translation, register mismatch, lexical ambiguity, domain terminology, and dialectal naturalness. We evaluate our model on FLORES, WMT24++, and NTREX using spBLEU. Results show that our model remains competitive with substantially larger LLM baselines and achieves its clearest gains on Arabic-centered directions, especially Egyptian Arabic. DPO improves FLORES translation into English by approximately +2 spBLEU for Egyptian Arabic and +2 spBLEU for Modern Standard Arabic. Our findings support targeted Arabic-centric Seq2Seq modeling as an efficient path for dialect-sensitive machine translation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hisham-etal-2026-arabic">
<titleInfo>
<title>Arabic-Centric Preference Optimization for Efficient Dialect-Sensitive Machine Translation: A Seq2Seq Approach</title>
</titleInfo>
<name type="personal">
<namePart type="given">Abdelrahman</namePart>
<namePart type="family">Hisham</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Malak</namePart>
<namePart type="family">Fathallah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayman</namePart>
<namePart type="family">Khalafallah</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 Fourth Arabic Natural Language Processing Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Muhammad</namePart>
<namePart type="family">Abdul-Mageed</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tamer</namePart>
<namePart type="family">Elsayed</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Houda</namePart>
<namePart type="family">Bouamor</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samhaa</namePart>
<namePart type="family">El-Beltagy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hend</namePart>
<namePart type="family">Al-Khalifa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ahmed</namePart>
<namePart type="family">Abdelali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">AbdelRahim</namePart>
<namePart type="family">Elmadany</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abdellah</namePart>
<namePart type="family">El Mekki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kareem</namePart>
<namePart type="family">Darwish</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sakhar</namePart>
<namePart type="family">Alkhereyf</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hamzah</namePart>
<namePart type="family">Luqman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Salam</namePart>
<namePart type="family">Khalifa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samar</namePart>
<namePart type="family">Magdy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zaid</namePart>
<namePart type="family">Alyafeai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Injy</namePart>
<namePart type="family">Hamed</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wajdi</namePart>
<namePart type="family">Zaghouani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Imed</namePart>
<namePart type="family">Zitouni</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">979-8-89176-473-6</identifier>
</relatedItem>
<abstract>Arabic machine translation remains challenging because Modern Standard Arabic and dialectal varieties such as Egyptian Colloquial Arabic differ substantially in morphology, lexical choice, register, and everyday usage. We introduce a compact Arabic-centric encoder–decoder machine translation model designed to strengthen translation involving Modern Standard Arabic and Egyptian Arabic while preserving multilingual translation ability. We pretrain from scratch with multilingual Seq2Seq training and contrastive cross-lingual alignment, using an Arabic-centered data mixture covering Arabic, English, Egyptian Arabic, and 111 additional languages. We further apply Direct Preference Optimization (DPO) for our Seq2Seq model using grounded preference pairs targeting Arabic-specific translation errors, including literal translation, register mismatch, lexical ambiguity, domain terminology, and dialectal naturalness. We evaluate our model on FLORES, WMT24++, and NTREX using spBLEU. Results show that our model remains competitive with substantially larger LLM baselines and achieves its clearest gains on Arabic-centered directions, especially Egyptian Arabic. DPO improves FLORES translation into English by approximately +2 spBLEU for Egyptian Arabic and +2 spBLEU for Modern Standard Arabic. Our findings support targeted Arabic-centric Seq2Seq modeling as an efficient path for dialect-sensitive machine translation.</abstract>
<identifier type="citekey">hisham-etal-2026-arabic</identifier>
<location>
<url>https://aclanthology.org/2026.arabicnlp-main.25/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>356</start>
<end>372</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Arabic-Centric Preference Optimization for Efficient Dialect-Sensitive Machine Translation: A Seq2Seq Approach
%A Hisham, Abdelrahman
%A Fathallah, Malak
%A Khalafallah, Ayman
%Y Abdul-Mageed, Muhammad
%Y Elsayed, Tamer
%Y Bouamor, Houda
%Y El-Beltagy, Samhaa
%Y Al-Khalifa, Hend
%Y Abdelali, Ahmed
%Y Elmadany, AbdelRahim
%Y El Mekki, Abdellah
%Y Darwish, Kareem
%Y Alkhereyf, Sakhar
%Y Luqman, Hamzah
%Y Khalifa, Salam
%Y Magdy, Samar
%Y Alyafeai, Zaid
%Y Hamed, Injy
%Y Zaghouani, Wajdi
%Y Zitouni, Imed
%S Proceedings of The Fourth Arabic Natural Language Processing Conference
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-473-6
%F hisham-etal-2026-arabic
%X Arabic machine translation remains challenging because Modern Standard Arabic and dialectal varieties such as Egyptian Colloquial Arabic differ substantially in morphology, lexical choice, register, and everyday usage. We introduce a compact Arabic-centric encoder–decoder machine translation model designed to strengthen translation involving Modern Standard Arabic and Egyptian Arabic while preserving multilingual translation ability. We pretrain from scratch with multilingual Seq2Seq training and contrastive cross-lingual alignment, using an Arabic-centered data mixture covering Arabic, English, Egyptian Arabic, and 111 additional languages. We further apply Direct Preference Optimization (DPO) for our Seq2Seq model using grounded preference pairs targeting Arabic-specific translation errors, including literal translation, register mismatch, lexical ambiguity, domain terminology, and dialectal naturalness. We evaluate our model on FLORES, WMT24++, and NTREX using spBLEU. Results show that our model remains competitive with substantially larger LLM baselines and achieves its clearest gains on Arabic-centered directions, especially Egyptian Arabic. DPO improves FLORES translation into English by approximately +2 spBLEU for Egyptian Arabic and +2 spBLEU for Modern Standard Arabic. Our findings support targeted Arabic-centric Seq2Seq modeling as an efficient path for dialect-sensitive machine translation.
%U https://aclanthology.org/2026.arabicnlp-main.25/
%P 356-372
Markdown (Informal)
[Arabic-Centric Preference Optimization for Efficient Dialect-Sensitive Machine Translation: A Seq2Seq Approach](https://aclanthology.org/2026.arabicnlp-main.25/) (Hisham et al., ArabicNLP 2026)
ACL