README.md

metadata

license: cc-by-sa-4.0
task_categories:
  - text-classification
task_ids:
  - multi-class-classification
language:
  - en
multilinguality:
  - monolingual
size_categories:
  - 1K<n<10K
pretty_name: English Poetry Form Classification
tags:
  - poetry
  - english
  - form-classification
  - poetic-form
  - text-classification
  - mteb
  - poetrymteb
  - embedding-evaluation
annotations_creators:
  - found
source_datasets:
  - riazhsks/multilingual-annotated-poems
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
    default: true
dataset_info:
  - config_name: default
    features:
      - name: id
        dtype: string
      - name: poem
        dtype: string
      - name: title
        dtype: string
      - name: author
        dtype: string
      - name: label
        dtype: int64
      - name: label_name
        dtype: string
    splits:
      - name: train
        num_examples: 1080
      - name: test
        num_examples: 271

English Poetry Form Classification

Single-label English poetic form classification for PoetryMTEB embedding evaluation, derived from the English corpus in multilingual-annotated-poems (originally based on poetry-eval / Walsh et al., EMNLP Findings 2024).

Dataset Card

Item	Description
Source	https://github.com/riazhsks/multilingual-annotated-poems → corpora/english_poems_processed.jsonl
Languages	English (en)
Size	train=1080; test=271 (no validation; MTEB Classification: train→fit, test→score)
Classes	15 forms with frequency ≥ 10
Filtering	Drop empty/null form; poem length ∈ [40, 8000] chars; rare forms (count < 10)
Splits	Stratified by form ≈ 80% / 19%, seed=42
License	CC BY-SA 4.0 (upstream CC-BY-SA)
Dataset version	1.0.0
Evaluation metrics	Classification on embeddings: accuracy, macro/weighted F1

Features

Field	Type	Description
id	string	Example id
poem	string	Poem text (classification input)
title	string	Title (metadata)
author	string	Author (metadata)
label	int64	Form class index 0 … C−1
label_name	string	Form name

Codebook: label_taxonomy.json.

Label inventory (kept)

id	label_name	gloss	train	test	total*
0	sonnet	Sonnet (十四行)	468	117	585
1	couplet	Couplet (对句)	220	55	275
2	common measure	Common measure (通用格)	55	14	69
3	ballad	Ballad (民谣体)	50	13	63
4	elegy	Elegy (挽歌)	49	12	61
5	free verse	Free verse (自由诗)	39	10	49
6	blank verse	Blank verse (无韵诗)	34	9	43
7	ode	Ode (颂歌)	34	8	42
8	dramatic monologue	Dramatic monologue (戏剧独白)	33	8	41
9	prose poem	Prose poem (散文诗)	30	8	38
10	pastoral	Pastoral (牧歌)	29	7	36
11	haiku	Haiku (俳句)	13	3	16
12	quatrain	Quatrain (四行)	10	3	13
13	ekphrasis	Ekphrasis (艺格敷词)	8	2	10
14	ars poetica	Ars poetica (诗艺)	8	2	10

* totals are class counts after length filtering (same pool used for train/test).

Dropped forms (count < 10)

form	count
limerick	6
concrete or pattern poetry	5
aubade	3
sestina	3
ghazal	2
villanelle	1
pantoum	1

Construction method

1. Load English JSONL from multilingual-annotated-poems.
2. Keep rows with non-empty gold form and text.
3. Filter by character length; keep classes with frequency ≥ 10.
4. Map forms to contiguous ids (frequency-descending).
5. Stratified train/test split by form (no validation).

How to load

from datasets import load_dataset

ds = load_dataset("PoetryMTEB/EnglishPoetryFormClassification")
print(ds["test"][0]["label_name"], ds["test"][0]["poem"][:80])

Citation

@misc{riazhsks_multilingual_annotated_poems,
  title={Multilingual Annotated Poems},
  author={riazhsks},
  year={2025},
  url={https://github.com/riazhsks/multilingual-annotated-poems}
}

@inproceedings{walsh-etal-2024-sonnet,
  title={Sonnet or Not, Bot? Poetry Evaluation for Large Models and Datasets},
  author={Walsh, Melanie and Preus, Anna and Antoniak, Maria},
  booktitle={Findings of EMNLP 2024},
  year={2024},
  url={https://aclanthology.org/2024.findings-emnlp.914/}
}

Hub packaging: PoetryMTEB/EnglishPoetryFormClassification (v1.0.0).