README.md · textdetox/multilingual

metadata

language:
  - en
  - uk
  - ru
  - de
  - zh
  - am
  - ar
  - hi
  - es
license: openrail++
size_categories:
  - 1K<n<10K
task_categories:
  - text-generation
dataset_info:
  features:
    - name: toxic_sentence
      dtype: string
  splits:
    - name: en
      num_bytes: 24945
      num_examples: 400
    - name: ru
      num_bytes: 48249
      num_examples: 400
    - name: uk
      num_bytes: 40226
      num_examples: 400
    - name: de
      num_bytes: 44940
      num_examples: 400
    - name: es
      num_bytes: 30159
      num_examples: 400
    - name: am
      num_bytes: 72606
      num_examples: 400
    - name: zh
      num_bytes: 36219
      num_examples: 400
    - name: ar
      num_bytes: 44668
      num_examples: 400
    - name: hi
      num_bytes: 57291
      num_examples: 400
  download_size: 257508
  dataset_size: 399303
configs:
  - config_name: default
    data_files:
      - split: en
        path: data/en-*
      - split: ru
        path: data/ru-*
      - split: uk
        path: data/uk-*
      - split: de
        path: data/de-*
      - split: es
        path: data/es-*
      - split: am
        path: data/am-*
      - split: zh
        path: data/zh-*
      - split: ar
        path: data/ar-*
      - split: hi
        path: data/hi-*

MultiParaDetox

This is the multilingual parallel dataset for text detoxification prepared for CLEF TextDetox 2024 shared task. For each of 9 languages, we collected 1k pairs of toxic<->detoxified instances splitted into two parts: dev (400 pairs) and test (600 pairs).

Now, only dev set toxic sentences are released. Dev set references and test set toxic sentences will be released later with the test phase of the competition!

The list of the sources for the original toxic sentences:

English: Jigsaw, Unitary AI Toxicity Dataset
Russian: Russian Language Toxic Comments, Toxic Russian Comments
Ukrainian: Ukrainian Twitter texts
Spanish: Detecting and Monitoring Hate Speech in Twitter, Detoxis, RoBERTuito: a pre-trained language model for social media text in Spanish
German: GemEval 2018, 2021
Amhairc: Amharic Hate Speech
Arabic: OSACT4
Hindi: Hostility Detection Dataset in Hindi, Overview of the HASOC track at FIRE 2019: Hate Speech and Offensive Content Identification in Indo-European Languages

We also added toxic words from Toxicity-200 corpus from Facebook Research for all the languages.