ChandassuMeterClassification

Single-label Telugu poetic meter (chandassu / type) classification for PoetryMTEB, derived from the Chandassu corpus (Kaggle / Mendeley / Hugging Face; local Chandassu_Dataset.csv).

Dataset Card

Item	Description
Source	Kaggle; Mendeley Data (DOI: 10.17632/46wrt7v53c.1); Hugging Face; paper arXiv:2510.01233
Languages	Telugu (te)
Size	train=3721; test=930 (no validation)
Classes	8 meters (type field)
Splits	Stratified ≈ 80% / 19% by meter, with cross-dataset anti-leakage vs PoetryMTEB/Padyam2GadyamMeterClassification
License	MIT (same as upstream)
Evaluation metrics	Classification on embeddings: accuracy, macro/weighted F1

Cross-dataset split constraint

Relative to PoetryMTEB/Padyam2GadyamMeterClassification:

1. Chandassu train does not contain poems that appear in Padyam test
2. Chandassu test does not contain poems that appear in Padyam train

Matching uses whitespace-normalized poem text. Forced assignments:
Padyam-train overlaps → Chandassu train (20 poems);
Padyam-test overlaps → Chandassu test (6 poems).

Verified after split: chandassu_train ∩ pady_test = 0, chandassu_test ∩ pady_train = 0.

---

Features

Field	Type	Description
id	string	Example id
poem	string	Telugu poem (raw_padyam_text)
label	int64	Meter class index
label_name	string	Meter type (romanized), e.g. kandamu

Codebook: label_taxonomy.json.

Labels

id	label_name	gloss	count
0	aataveladi	Āṭaveladi / ఆటవెలది (戏舞体)	995
1	kandamu	Kandam / కందం (坎达体)	683
2	teytageethi	Tēṭagīti / తేటగీతి (澄歌体)	676
3	seesamu	Sīsamu / సీసము (席萨体)	672
4	mattebhamu	Mattebhamu / మత్తేభము (醉象律)	617
5	champakamaala	Champakamāla / చంపకమాల (瞻波迦鬘)	389
6	vutpalamaala	Utpalamāla / ఉత్పలమాల (优钵罗鬘)	329
7	saardulamu	Śārdūlamu / శార్దూలము (虎律)	290

---

Construction method

1. Load local Chandassu_Dataset.csv / upstream Chandassu.
2. Load PoetryMTEB/Padyam2GadyamMeterClassification train/test poems; force overlapping Chandassu rows into compatible splits (anti-leakage).
3. Stratified train/test split of remaining rows by type (seed=42).
4. Map type → contiguous label ids (frequency-descending).

---

How to load

from datasets import load_dataset

ds = load_dataset("PoetryMTEB/ChandassuMeterClassification")
print(ds["test"][0]["label_name"], ds["test"][0]["poem"][:60])

---

License

Distributed under the MIT License, consistent with the upstream Chandassu release.

---

Citation / provenance

@misc{pavan2025computationalsociallinguisticstelugu,
      title={Computational Social Linguistics for Telugu Cultural Preservation: Novel Algorithms for Chandassu Metrical Pattern Recognition}, 
      author={Boddu Sri Pavan and Boddu Swathi Sree},
      year={2025},
      eprint={2510.01233},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.01233}, 
}

• Kaggle: https://www.kaggle.com/datasets/boddusripavan111/chandassu
• Mendeley Data: https://data.mendeley.com/datasets/46wrt7v53c/1 (DOI: 10.17632/46wrt7v53c.1)
• Hugging Face: https://huggingface.co/datasets/BodduSriPavan111/chandassu
• Paper: https://arxiv.org/abs/2510.01233
• Related (anti-leakage partner): PoetryMTEB/Padyam2GadyamMeterClassification
• This Hub packaging: PoetryMTEB/ChandassuMeterClassification