TAAG — Teacher-Authored Arabic Grammar

TAAG is a small benchmark of 171 hand-written multiple-choice questions in Modern Standard Arabic, built to measure how well a language model handles Arabic as a language: its syntax, its morphology, and its rhetoric.

Every question was written from scratch by a certified Arabic teacher and follows a school curriculum.

What is in it

Every item is a question with three or four answer options and exactly one correct answer. The questions cover:

• النحو (syntax) — case marking (الإعراب), sentence structure, function of particles, agreement.
• الصرف (morphology) — word patterns (الأوزان), derivation, root and pattern analysis, verb forms.
• البلاغة (rhetoric) — figures of speech, simile and metaphor, meaning of stylistic constructions.
• Alongside these, some items test lexical semantics — synonyms, antonyms, and odd-one-out.

Provenance

The questions were written from scratch by a certified Arabic-language teacher with 15 years of classroom experience. They were not scraped, not translated, and not generated by a model.

Their content and difficulty are inspired by the official Arabic-language textbooks used in Lebanese schools for the intermediate stage (المرحلة المتوسّطة) — grades 7, 8 and 9. The questions are original writing informed by that curriculum, not copies of textbook exercises.

Dataset structure

A single test split of 171 rows. There is no training split — the collection is meant for evaluation only.

field	type	description
id	string	Stable item identifier, e.g. TAAG_000. Zero-padded, and not contiguous — some numbers in the range are unused.
question	string	The question stem, in Arabic.
choices	list[string]	The answer options in order. Length 3 or 4.
answer	string	The correct option as a letter: A, B, C or D.
answer_index	int32	Zero-based index of the correct option into choices.
answer_text	string	The text of the correct option, for convenience.

Example:

{
  "id": "TAAG_000",
  "question": "مرادف كلمة ممتنّة هو",
  "choices": ["راضية", "شاكرة", "صارحة"],
  "answer": "B",
  "answer_index": 1,
  "answer_text": "شاكرة",
}

Composition

	count
Total questions	171
With 4 options	148
With 3 options	23

Correct-answer distribution: A 28, B 66, C 56, D 21. The options were not shuffled to balance this. If you are scoring by letter rather than by option content, shuffle the options yourself — otherwise a model that simply prefers B will look better than it is.

Usage

from datasets import load_dataset

ds = load_dataset("ali-issa/TAAG", split="test")

for item in ds:
    options = "\n".join(
        f"{chr(65 + i)}) {c}" for i, c in enumerate(item["choices"])
    )
    prompt = f"{item['question']}\n{options}\nالجواب:"
    ...  # score the model's continuation against item["answer"]

The text is undiacritized, which is how Arabic is normally written. If your setup needs diacritics (تشكيل), apply the diacritizer of your choice — the collection is left undiacritized on purpose so that this choice stays yours rather than being baked in.

Citation

If you use TAAG, please cite the dataset:

@misc{taag2026,
  title        = {{TAAG}: Teacher-Authored Arabic Grammar},
  author       = {{Author names withheld during review}},
  year         = {2026},
  publisher    = {Hugging Face},
  note         = {Author list and URL will be added once the accompanying paper is published}
}

TAAG was built for the evaluation of an Arabic tokenizer. That paper is currently under review, so there is no published reference for it yet. Until it appears:

• Cite the dataset entry above. That is the stable, citable reference right now.
• If you want to point at the accompanying work as well, cite it as unpublished:

@unpublished{tokenizer2026,
  title  = {<paper title withheld during review>},
  author = {{Author names withheld during review}},
  year   = {2026},
  note   = {Under review}
}

This dataset card will be updated with the full author list and paper reference once it is published. If you are citing TAAG in work of your own, check this page for a newer citation before submitting.

License

Released under CC BY 4.0. You may use, share and adapt the questions, including commercially, as long as you give appropriate credit.

Authors

Withheld during review. The author list will be added to this card once the accompanying paper is published.

Contact

Questions and corrections to individual questions are welcome — please open a discussion on this repository.