Grade handwritten Arabic poetry answers in seconds.
ShiirGrader is an automated system that reads and evaluates students' handwritten answers on Arabic poetry — cursive script and diacritics included — so teachers spend less time marking and more time teaching.
والسيفُ والرمحُ والقرطاسُ والقمر
8 / 10
expected القلمُ
Arabic handwriting is one of the hardest scripts for a machine to read.
Grading handwritten poetry answers takes teachers hours of careful reading — and the features that make Arabic script beautiful are exactly what make it hard to automate.
Cursive by nature
Arabic letters join together within words, so a recognition system can't simply separate and read one character at a time.
Shape-shifting letters
Most letters change shape depending on whether they appear alone or at the start, middle or end of a word.
Diacritics carry meaning
Short-vowel marks above and below letters matter in poetry, where they shape both meaning and metre.
people use Arabic every day, and it is one of the six official languages of the United Nations. Its classroom tools haven't kept pace.
Built for the Arabic classroom.
For teachers
An automated marker that reads stacks of handwritten answers and highlights what needs attention.
- Faster marking of poetry exams and homework
- Consistent evaluation across every paper
- Flagged words show exactly where a student went wrong
For students
Quicker, clearer feedback on their own handwriting — the way they naturally write in class.
- Answers stay handwritten; no typing needed
- Feedback that points to specific words
- Supports personalised tutoring and practice
From answer sheet to grade in four steps.
-
01
Capture
The student's handwritten answer is scanned or photographed.
ScanPhoto -
02
Clean
The image is normalised and binarised into crisp black-and-white text, removing background noise.
NormalizationBinarization -
03
Read
A deep learning model recognises the cursive Arabic handwriting, diacritics included.
DCNNBiLSTMCTC -
04
Evaluate
The recognised answer is compared with the expected verse, and the result is returned to the teacher.
Automated marking
A deep network designed for connected script.
ShiirGrader's recognition engine combines three deep learning components, each solving a different part of the Arabic handwriting problem.
- Convolutional layers learn the visual shapes of strokes and letters.
- A bidirectional LSTM reads the sequence both ways, using context from the whole word.
- CTC loss lets the model learn from whole lines without marking where each letter starts and ends.
High accuracy on Arabic handwriting.
The model was trained and tested on a public Kaggle dataset of Arabic handwritten characters.
Share of all handwritten samples recognised correctly.
When the model names a character, how often it is right.
How many of the true characters the model successfully finds.
Figures from Shuker, Ogla & Rahma, IET Conference Proceedings (2025). doi:10.1049/icp.2025.0157
الخيلُ والليلُ والبيداءُ تعرفني
والسيفُ والرمحُ والقرطاسُ والقلمُ
"The horses, the night and the desert know me — and the sword, the spear, the parchment and the pen."
— al-Mutanabbi, 10th centuryWhere Arabic literature meets computer science.
ShiirGrader was founded in 2025 by a research team that pairs deep knowledge of Arabic language and poetry with expertise in artificial intelligence.
Arabic Language
Defines what a correct answer looks like — verse accuracy, spelling and diacritics — and brings the classroom and teaching perspective.
Computer Science
Builds the handwriting recognition model, the image-processing pipeline and the assessment platform.
Technology that supports teachers, never replaces them.
Teachers have the final say
ShiirGrader proposes a grade and flags issues. The teacher reviews, adjusts and approves every result.
Respect for the language
Diacritics and correct verse wording are treated as part of the answer, not noise to be ignored.
Quality education
Faster feedback helps students learn — supporting UN Sustainable Development Goal 4, Quality Education.
Frequently asked questions
What does ShiirGrader do?
It reads students' handwritten answers to Arabic poetry questions and evaluates them automatically, giving teachers a proposed result with flagged errors.
Can it read cursive Arabic and diacritics?
Yes. The recognition model is designed specifically for cursive Arabic script and diacritical marks, using a deep convolutional network, a bidirectional LSTM and CTC loss.
How accurate is it?
The recognition model achieved 97.20% accuracy, 97.10% precision and 97.30% recall on a public Kaggle dataset of Arabic handwritten characters, as published in IET Conference Proceedings (2025).
Do students need to type their answers?
No. Students write by hand as usual; the answer sheet is scanned or photographed and processed by the system.
Does it replace the teacher?
No. It is a support tool. Teachers review and approve every grade, and keep full control over assessment.
Grounded in peer-reviewed research.
Read the study behind ShiirGrader's handwriting recognition and evaluation engine.
Shuker, H. S., Ogla, R., & Rahma, A. M. S. (2025). Evaluation of student's handwritten answers to Arabic poetry by an automated system. IET Conference Proceedings, 2024(34), 651–656.
https://doi.org/10.1049/icp.2025.0157