PoetryMTEB/ArabicPoetryMeterClassification

Arabic Poetry Meter Classification

Multi-class meter (ʿarūḍ / prosody) classification of Arabic poetry verses for PoetryMTEB embedding evaluation.

Derived from MetRec (Al-shaibani, Alyafeai & Ahmad, Data in Brief 2020), collected from Aldiwan. Method paper: Pattern Recognition Letters 2020. Code/data: ARBML/MetRec; HF mirror: Zaid/metrec.

Dataset Card

Item	Description
Dataset version (PoetryMTEB)	1.0.0
Source (GitHub)	ARBML/MetRec
Source (HF)	Zaid/metrec
Data paper	Al-shaibani et al., MetRec, Data in Brief 33:106497, 2020 (DOI)
Method paper	Al-shaibani et al., Pattern Recognition Letters, 2020 (DOI)
Languages	Arabic (ar)
Unit	Single verse / bait (poem; hemistichs in verse_part1 / verse_part2)
Labels	Single-label 14 classical meters (label / label_name; MetRec ids 0–13)
Size	train=47124; test=8316; total=55440
Splits	train / test only (no validation). Membership preserved from the previous PoetryMTEB release
License	CC BY 4.0 (Data in Brief open-access license)
Evaluation metrics	Embedding classification: accuracy, macro/micro F1 (see scripts/evaluate_arabic_meter.py)

Label taxonomy (14 meters)

id	label_name	name_en	name_ar	name_zh	train	test	total
0	saree	Saree (Sarīʿ)	السريع	速律	4269	731	5000
1	kamel	Kamel (Kāmil)	الكامل	完律	4240	760	5000
2	mutakareb	Mutakareb (Mutqārib)	المتقارب	近律	4242	758	5000
3	mutadarak	Mutadarak (Mutadārik)	المتدارك	续律	1084	178	1262
4	munsareh	Munsareh (Munsaṛiḥ)	المنسرح	奔律	3767	681	4448
5	madeed	Madeed (Madīd)	المديد	延律	1176	230	1406
6	mujtath	Mujtath (Mujtathth)	المجتث	截律	1819	303	2122
7	ramal	Ramal	الرمل	沙律	4281	719	5000
8	baseet	Baseet (Basīṭ)	البسيط	简律	4248	752	5000
9	khafeef	Khafeef (Khafīf)	الخفيف	轻律	4242	759	5001
10	taweel	Taweel (Ṭawīl)	الطويل	长律	4248	752	5000
11	wafer	Wafer (Wāfir)	الوافر	丰律	4231	769	5000
12	hazaj	Hazaj	الهزج	摇律	1033	168	1201
13	rajaz	Rajaz	الرجز	快板律	4244	756	5000

Codebook: label_taxonomy.json. Upstream Latin spellings (kamel, wafer, …) are kept as label_name.

Features

Field	Type	Description
id	string	Example id (apmc-{split}-{legacy_id})
poem	string	Classification input: full verse text (# separator already removed; formerly verse)
verse_part1	string	First hemistich (before # in upstream train.txt / test.txt)
verse_part2	string	Second hemistich (after #)
label	int64	Meter class id (0–13; formerly meter_id)
label_name	string	Meter name (formerly meter_name)

Schema aligns with other PoetryMTEB classification datasets (id, poem, label, label_name).

Upstream row format (reference)

1 يا ضيفَ طيفٍ ما هَداهُ لمَضجَعي # إلا لهيبٌ في الحشى يتوقّدُ

Meter id and verse separated by a space; two hemistichs separated by #.

Construction method

1. Load existing PoetryMTEB/ArabicPoetryMeterClassification parquet splits.
2. Normalize fields: verse→poem, meter_id→label, meter_name→label_name; keep hemistich columns; assign stable ids.
3. Preserve train/test membership and MetRec 0–13 label ids (no re-split, no validation).
4. Write parquet + dataset_infos.yaml + Dataset Card + supporting materials.

Supporting materials (this repo)

Path	Content
VERSION	PoetryMTEB packaging version
LICENSE	CC BY 4.0 notice
CITATION.cff / CITATION.bib	Citation metadata
metadata.json	Dataset metadata
label_taxonomy.json	Meter codebook + counts
splits/train_ids.txt / splits/test_ids.txt	Split legacy id lists
checksums.sha256	SHA256 of packaged files
prompts/annotation_prompt_template.md	Meter labeling prompt template
scripts/evaluate_arabic_meter.py	Baseline probe evaluation

How to load

from datasets import load_dataset

ds = load_dataset("PoetryMTEB/ArabicPoetryMeterClassification")
print(ds["train"][0]["poem"])
print(ds["train"][0]["label_name"], ds["train"][0]["label"])

Intended use

• PoetryMTEB / MTEB-style multi-class classification probing of Arabic verse embeddings (classical meter / ʿarūḍ recognition).
• Research on automatic Arabic poetry meter identification.

Citation

@article{ALSHAIBANI2020106497,
  author  = {Al-shaibani, Maged S. and Alyafeai, Zaid and Ahmad, Irfan},
  title   = {MetRec: A dataset for meter classification of arabic poetry},
  journal = {Data in Brief},
  volume  = {33},
  pages   = {106497},
  year    = {2020},
  doi     = {10.1016/j.dib.2020.106497},
  url     = {https://doi.org/10.1016/j.dib.2020.106497}
}

@article{al2020meter,
  author  = {Al-shaibani, Maged S. and Alyafeai, Zaid and Ahmad, Irfan},
  title   = {Meter Classification of Arabic Poems Using Deep Bidirectional Recurrent Neural Networks},
  journal = {Pattern Recognition Letters},
  year    = {2020},
  doi     = {10.1016/j.patrec.2020.05.028},
  url     = {https://doi.org/10.1016/j.patrec.2020.05.028}
}

Upstream: https://github.com/ARBML/MetRec · HF: Zaid/metrec
This Hub packaging: PoetryMTEB/ArabicPoetryMeterClassification (version 1.0.0)

License

Distributed under CC BY 4.0 (same as the Data in Brief MetRec release). Please cite the MetRec data article and the PRL method paper when using the data. Verse texts originate from public Aldiwan pages; see upstream authors' notice regarding scraping of publicly accessible pages.