README.md

metadata

configs:
  - config_name: jfleg
    data_files: jfleg/data.jsonl
    default: true
  - config_name: asset
    data_files: asset/data.jsonl
  - config_name: fruit
    data_files: fruit/data.jsonl
  - config_name: iterater
    data_files: iterater/data.jsonl
  - config_name: iterater_clarity
    data_files: iterater_clarity/data.jsonl
  - config_name: iterater_coherence
    data_files: iterater_coherence/data.jsonl
  - config_name: iterater_fluency
    data_files: iterater_fluency/data.jsonl
  - config_name: stsb_multi_mt
    data_files: stsb_multi_mt/data.jsonl
  - config_name: turk
    data_files: turk/data.jsonl
  - config_name: wafer_insert
    data_files: wafer_insert/data.jsonl
  - config_name: wnc
    data_files: wnc/data.jsonl
license: cc-by-nc-sa-4.0
task_categories:
  - text-generation
tags:
  - text-editing
  - benchmark

EditEval: The Instruction-Based Benchmark for Text Improvements

This dataset contains the EditEval benchmark data, converted to JSONL and organized by task/dataset.

Subsets

Config	Task	Examples
jfleg	Fluency	1,501
asset	Simplification	2,359
turk	Simplification	2,359
iterater	Mixed (all tasks)	621
iterater_fluency	Fluency	203
iterater_clarity	Clarity	342
iterater_coherence	Coherence	76
stsb_multi_mt	Paraphrasing	153
wnc	Neutralization	1,700
wafer_insert	Updating (insertion)	9,108
fruit	Updating	150

Usage

from datasets import load_dataset

# Load a specific subset
ds = load_dataset("bzz2/EditEval", "jfleg")

# Load with a limit
ds = load_dataset("bzz2/EditEval", "fruit", split="train[:10]")

Schema

Each record has the following fields:

• id — unique example identifier
• input — source text to be edited
• title — article/document title (if applicable)
• task_type — editing task (fluency, simplification, neutralization, etc.)
• retrieved_documents — supporting documents (used by updating tasks)
• meta — additional metadata (JSON string)

Citation

@inproceedings{dwivedi-edit-2022,
  doi = {10.48550/ARXIV.2209.13331},
  url = {https://arxiv.org/abs/2209.13331},
  author = {Dwivedi-Yu, Jane and Schick, Timo and Jiang, Zhengbao and Lomeli, Maria and Lewis, Patrick and Izacard, Gautier and Grave, Edouard and Riedel, Sebastian and Petroni, Fabio},
  title = {EditEval: An Instruction-Based Benchmark for Text Improvements},
  publisher = {arXiv},
  year = {2022},
}