Language identification
Predict the language of an input text from a defined set of languages written in Arabic or Arabic-derived scripts. The label set and treatment of mixed-language text are to be confirmed.
AbjadNLP 2027 · Shared task
AbjadLanguageID 2027
Identifying languages written in Arabic-derived scripts
A proposed shared task on automatic identification of languages and potentially language clusters represented in Arabic and Arabic-derived scripts. Final language coverage, datasets and evaluation tracks will be confirmed by the organising team.
Proposed language coverage
Arabic · Persian · Urdu · Kurdishالعربية · فارسی · اردو · کوردی
Ajami and other Arabic-derived scriptsCoverage to be confirmed
Script examples are illustrative. The final inventory of languages, varieties and writing conventions is TBC.
About the shared task
Which language is this text written in?
Many languages use Arabic or Arabic-derived writing systems, sometimes with overlapping character inventories and similar orthographic conventions. AbjadLanguageID explores whether systems can distinguish these languages reliably, particularly for short, noisy or under-resourced texts.
Language-cluster identification
Identify broader groupings of languages or varieties where suitable data are available. Regional groupings such as Magharibi, African, Gulf and Indian have been suggested, but their definitions and inclusion are not final.
Multiple Abjad languages
Proposed coverage includes Arabic, Persian, Urdu, Kurdish and languages using Ajami writing traditions, with additional languages and varieties subject to confirmation.
Related scripts, varied data
Possible challenges include shared characters across languages, script variation, transliteration, code-switching and limited labelled data. The official task will specify the permitted inputs and outputs.
Data and evaluation
Training, development and hidden test data, language-label definitions, licensing, scoring metrics and submission format will be announced by the organisers. No dataset sizes or benchmark results are being claimed at this stage.
Taking part
Participation details
Registration, dataset access and evaluation instructions will be published once the shared-task arrangements are confirmed.
This page describes a proposed task. Please refer to the workshop website for updates and applicable policies.
Visit the AbjadNLP 2027 workshop page →- Languages
- Arabic, Persian, Urdu, Kurdish, Ajami and others (TBC)
- Tracks and labels
- To be confirmed
- Data access and licensing
- To be announced
- Evaluation metrics
- To be confirmed
- Registration and submissions
- To be announced
- Shared-task dates
- To be announced
The team
Shared-task organisers
The previous edition's organising team is shown below; the 2027 team is subject to confirmation.
- Saad EzziniKing Fahd University of Petroleum & Minerals
- Irfan AhmedKing Fahd University of Petroleum & Minerals
- Salmane ChafikMohammed VI Polytechnic University
- Shadi AbudalfaKing Fahd University of Petroleum & Minerals
- Mo El-HajVinUniversity
- Ahmed AbdelaliHUMAIN
- Mustafa JarrarHamad Bin Khalifa University / Birzeit University
- Nadir DurraniHamad Bin Khalifa University
- Hassan SajjadDalhousie University
- Farah AdeebaUniversity of Massachusetts Amherst