@inproceedings{al-shaibani-ahmad-2026-lossy,
title = "Lossy Orthography and Linguistic Inference: Computational Approaches to {A}rabic Rasm Restoration",
author = "Al-shaibani, Maged S. and
Ahmad, Irfan",
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.2258/",
pages = "51351--51371",
ISBN = "979-8-89176-459-0",
abstract = "Restoring dots to dotless Arabic text, historically known as {\_}rasm{\_}, is a foundational yet underexplored problem in Arabic NLP. While Arabic diacritization focuses on recovering omitted vowels, rasm restoration addresses a deeper level of orthographic ambiguity in which consonantal distinctions themselves are absent from the written text. Beyond its linguistic and historical significance, the task has applications in representation learning, tokenization, OCR, manuscript digitization, information retrieval, and content moderation. In this paper, we present a large-scale comparative study of Arabic rasm restoration across statistical, neural, pretrained, and prompt-based paradigms. We evaluate n-gram language models, BiLSTMs and Transformers trained from scratch, pretrained character-level CANINE models, and large language models (LLMs) using few-shot prompting. Our experiments were conducted on a diverse corpus exceeding 431M words spanning social media, news, religious texts, poetry, and academic writing. Results show that Transformer models achieve the best overall performance while remaining computationally efficient. BiLSTMs perform competitively, and classical n-gram models remain surprisingly strong despite their simplicity. In contrast, modern LLMs perform substantially below specialized models, with flagship systems only matching the performance of a 4-gram baseline. These findings position rasm restoration as a challenging benchmark for contextual orthographic inference and a natural extension of Arabic diacritization research."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="al-shaibani-ahmad-2026-lossy">
<titleInfo>
<title>Lossy Orthography and Linguistic Inference: Computational Approaches to Arabic Rasm Restoration</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maged</namePart>
<namePart type="given">S</namePart>
<namePart type="family">Al-shaibani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Irfan</namePart>
<namePart type="family">Ahmad</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>Restoring dots to dotless Arabic text, historically known as _rasm_, is a foundational yet underexplored problem in Arabic NLP. While Arabic diacritization focuses on recovering omitted vowels, rasm restoration addresses a deeper level of orthographic ambiguity in which consonantal distinctions themselves are absent from the written text. Beyond its linguistic and historical significance, the task has applications in representation learning, tokenization, OCR, manuscript digitization, information retrieval, and content moderation. In this paper, we present a large-scale comparative study of Arabic rasm restoration across statistical, neural, pretrained, and prompt-based paradigms. We evaluate n-gram language models, BiLSTMs and Transformers trained from scratch, pretrained character-level CANINE models, and large language models (LLMs) using few-shot prompting. Our experiments were conducted on a diverse corpus exceeding 431M words spanning social media, news, religious texts, poetry, and academic writing. Results show that Transformer models achieve the best overall performance while remaining computationally efficient. BiLSTMs perform competitively, and classical n-gram models remain surprisingly strong despite their simplicity. In contrast, modern LLMs perform substantially below specialized models, with flagship systems only matching the performance of a 4-gram baseline. These findings position rasm restoration as a challenging benchmark for contextual orthographic inference and a natural extension of Arabic diacritization research.</abstract>
<identifier type="citekey">al-shaibani-ahmad-2026-lossy</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2258/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>51351</start>
<end>51371</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Lossy Orthography and Linguistic Inference: Computational Approaches to Arabic Rasm Restoration
%A Al-shaibani, Maged S.
%A Ahmad, Irfan
%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 al-shaibani-ahmad-2026-lossy
%X Restoring dots to dotless Arabic text, historically known as _rasm_, is a foundational yet underexplored problem in Arabic NLP. While Arabic diacritization focuses on recovering omitted vowels, rasm restoration addresses a deeper level of orthographic ambiguity in which consonantal distinctions themselves are absent from the written text. Beyond its linguistic and historical significance, the task has applications in representation learning, tokenization, OCR, manuscript digitization, information retrieval, and content moderation. In this paper, we present a large-scale comparative study of Arabic rasm restoration across statistical, neural, pretrained, and prompt-based paradigms. We evaluate n-gram language models, BiLSTMs and Transformers trained from scratch, pretrained character-level CANINE models, and large language models (LLMs) using few-shot prompting. Our experiments were conducted on a diverse corpus exceeding 431M words spanning social media, news, religious texts, poetry, and academic writing. Results show that Transformer models achieve the best overall performance while remaining computationally efficient. BiLSTMs perform competitively, and classical n-gram models remain surprisingly strong despite their simplicity. In contrast, modern LLMs perform substantially below specialized models, with flagship systems only matching the performance of a 4-gram baseline. These findings position rasm restoration as a challenging benchmark for contextual orthographic inference and a natural extension of Arabic diacritization research.
%U https://aclanthology.org/2026.emnlp-main.2258/
%P 51351-51371
Markdown (Informal)
[Lossy Orthography and Linguistic Inference: Computational Approaches to Arabic Rasm Restoration](https://aclanthology.org/2026.emnlp-main.2258/) (Al-shaibani & Ahmad, EMNLP 2026)
ACL