Hugging Face's logo

Datasets:
community-datasets
/
tapaco

Tasks:
Text2Text Generation
Translation
Text Classification
Modalities:
Text
Formats:
parquet
Sub-tasks:
semantic-similarity-classification
Languages:
Afrikaans
Arabic
Azerbaijani
Size:
1M - 10M
Tags:
paraphrase-generation
Libraries:
Datasets
pandas
License:
Dataset card Viewer Files Community
5
tapaco / README.md
system's picture
system HF staff
Update files from the datasets library (from 1.18.0)
e92c33e
preview code
|
raw
history blame
10.2 kB
metadata
annotations_creators:
  - machine-generated
language_creators:
  - crowdsourced
languages:
  all_languages:
    - af
    - ar
    - az
    - be
    - ber
    - bg
    - bn
    - br
    - ca
    - cbk
    - cmn
    - cs
    - da
    - de
    - el
    - en
    - eo
    - es
    - et
    - eu
    - fi
    - fr
    - gl
    - gos
    - he
    - hi
    - hr
    - hu
    - hy
    - ia
    - id
    - ie
    - io
    - is
    - it
    - ja
    - jbo
    - kab
    - ko
    - kw
    - la
    - lfn
    - lt
    - mk
    - mr
    - nb
    - nds
    - nl
    - orv
    - ota
    - pes
    - pl
    - pt
    - rn
    - ro
    - ru
    - sl
    - sr
    - sv
    - tk
    - tl
    - tlh
    - toki
    - tr
    - tt
    - ug
    - uk
    - ur
    - vi
    - vo
    - war
    - wuu
    - yue
  af:
    - af
  ar:
    - ar
  az:
    - az
  be:
    - be
  ber:
    - ber
  bg:
    - bg
  bn:
    - bn
  br:
    - br
  ca:
    - ca
  cbk:
    - cbk
  cmn:
    - cmn
  cs:
    - cs
  da:
    - da
  de:
    - de
  el:
    - el
  en:
    - en
  eo:
    - eo
  es:
    - es
  et:
    - et
  eu:
    - eu
  fi:
    - fi
  fr:
    - fr
  gl:
    - gl
  gos:
    - gos
  he:
    - he
  hi:
    - hi
  hr:
    - hr
  hu:
    - hu
  hy:
    - hy
  ia:
    - ia
  id:
    - id
  ie:
    - ie
  io:
    - io
  is:
    - is
  it:
    - it
  ja:
    - ja
  jbo:
    - jbo
  kab:
    - kab
  ko:
    - ko
  kw:
    - kw
  la:
    - la
  lfn:
    - lfn
  lt:
    - lt
  mk:
    - mk
  mr:
    - mr
  nb:
    - nb
  nds:
    - nds
  nl:
    - nl
  orv:
    - orv
  ota:
    - ota
  pes:
    - pes
  pl:
    - pl
  pt:
    - pt
  rn:
    - rn
  ro:
    - ro
  ru:
    - ru
  sl:
    - sl
  sr:
    - sr
  sv:
    - sv
  tk:
    - tk
  tl:
    - tl
  tlh:
    - tlh
  toki:
    - toki
  tr:
    - tr
  tt:
    - tt
  ug:
    - ug
  uk:
    - uk
  ur:
    - ur
  vi:
    - vi
  vo:
    - vo
  war:
    - war
  wuu:
    - wuu
  yue:
    - yue
licenses:
  - cc-by-2-0
multilinguality:
  - multilingual
size_categories:
  af:
    - n<1K
  all_languages:
    - 1M<n<10M
  ar:
    - 1K<n<10K
  az:
    - n<1K
  be:
    - 1K<n<10K
  ber:
    - 10K<n<100K
  bg:
    - 1K<n<10K
  bn:
    - 1K<n<10K
  br:
    - 1K<n<10K
  ca:
    - n<1K
  cbk:
    - n<1K
  cmn:
    - 10K<n<100K
  cs:
    - 1K<n<10K
  da:
    - 10K<n<100K
  de:
    - 100K<n<1M
  el:
    - 10K<n<100K
  en:
    - 100K<n<1M
  eo:
    - 100K<n<1M
  es:
    - 10K<n<100K
  et:
    - n<1K
  eu:
    - n<1K
  fi:
    - 10K<n<100K
  fr:
    - 100K<n<1M
  gl:
    - n<1K
  gos:
    - n<1K
  he:
    - 10K<n<100K
  hi:
    - 1K<n<10K
  hr:
    - n<1K
  hu:
    - 10K<n<100K
  hy:
    - n<1K
  ia:
    - 1K<n<10K
  id:
    - 1K<n<10K
  ie:
    - n<1K
  io:
    - n<1K
  is:
    - 1K<n<10K
  it:
    - 100K<n<1M
  ja:
    - 10K<n<100K
  jbo:
    - 1K<n<10K
  kab:
    - 10K<n<100K
  ko:
    - n<1K
  kw:
    - 1K<n<10K
  la:
    - 1K<n<10K
  lfn:
    - 1K<n<10K
  lt:
    - 1K<n<10K
  mk:
    - 10K<n<100K
  mr:
    - 10K<n<100K
  nb:
    - 1K<n<10K
  nds:
    - 1K<n<10K
  nl:
    - 10K<n<100K
  orv:
    - n<1K
  ota:
    - n<1K
  pes:
    - 1K<n<10K
  pl:
    - 10K<n<100K
  pt:
    - 10K<n<100K
  rn:
    - n<1K
  ro:
    - 1K<n<10K
  ru:
    - 100K<n<1M
  sl:
    - n<1K
  sr:
    - 1K<n<10K
  sv:
    - 1K<n<10K
  tk:
    - 1K<n<10K
  tl:
    - 1K<n<10K
  tlh:
    - 1K<n<10K
  toki:
    - 1K<n<10K
  tr:
    - 100K<n<1M
  tt:
    - 1K<n<10K
  ug:
    - 1K<n<10K
  uk:
    - 10K<n<100K
  ur:
    - n<1K
  vi:
    - n<1K
  vo:
    - n<1K
  war:
    - n<1K
  wuu:
    - n<1K
  yue:
    - n<1K
source_datasets:
  - extended|other-tatoeba
task_categories:
  - conditional-text-generation
  - text-classification
task_ids:
  - >-
    conditional-text-generation-other-given-a-sentence-generate-a-paraphrase-either-in-same-language-or-another-language
  - machine-translation
  - semantic-similarity-classification
paperswithcode_id: tapaco
pretty_name: TaPaCo Corpus

Dataset Card for TaPaCo Corpus

Table of Contents

Dataset Description

Dataset Summary

A freely available paraphrase corpus for 73 languages extracted from the Tatoeba database. Tatoeba is a crowdsourcing project mainly geared towards language learners. Its aim is to provide example sentences and translations for particular linguistic constructions and words. The paraphrase corpus is created by populating a graph with Tatoeba sentences and equivalence links between sentences “meaning the same thing”. This graph is then traversed to extract sets of paraphrases. Several language-independent filters and pruning steps are applied to remove uninteresting sentences. A manual evaluation performed on three languages shows that between half and three quarters of inferred paraphrases are correct and that most remaining ones are either correct but trivial, or near-paraphrases that neutralize a morphological distinction. The corpus contains a total of 1.9 million sentences, with 200 – 250 000 sentences per language. It covers a range of languages for which, to our knowledge, no other paraphrase dataset exists.

Supported Tasks and Leaderboards

Paraphrase detection and generation have become popular tasks in NLP and are increasingly integrated into a wide variety of common downstream tasks such as machine translation , information retrieval, question answering, and semantic parsing. Most of the existing datasets cover only a single language – in most cases English – or a small number of languages. Furthermore, some paraphrase datasets focus on lexical and phrasal rather than sentential paraphrases, while others are created (semi -)automatically using machine translation.

The number of sentences per language ranges from 200 to 250 000, which makes the dataset more suitable for fine-tuning and evaluation purposes than for training. It is well-suited for multi-reference evaluation of paraphrase generation models, as there is generally not a single correct way of paraphrasing a given input sentence.

Languages

The dataset contains paraphrases in Afrikaans, Arabic, Azerbaijani, Belarusian, Berber languages, Bulgarian, Bengali , Breton, Catalan; Valencian, Chavacano, Mandarin, Czech, Danish, German, Greek, Modern (1453-), English, Esperanto , Spanish; Castilian, Estonian, Basque, Finnish, French, Galician, Gronings, Hebrew, Hindi, Croatian, Hungarian , Armenian, Interlingua (International Auxiliary Language Association), Indonesian, Interlingue; Occidental, Ido , Icelandic, Italian, Japanese, Lojban, Kabyle, Korean, Cornish, Latin, Lingua Franca Nova\t, Lithuanian, Macedonian , Marathi, Bokmål, Norwegian; Norwegian Bokmål, Low German; Low Saxon; German, Low; Saxon, Low, Dutch; Flemish, ]Old Russian, Turkish, Ottoman (1500-1928), Iranian Persian, Polish, Portuguese, Rundi, Romanian; Moldavian; Moldovan, Russian, Slovenian, Serbian, Swedish, Turkmen, Tagalog, Klingon; tlhIngan-Hol, Toki Pona, Turkish, Tatar, Uighur; Uyghur, Ukrainian, Urdu, Vietnamese, Volapük, Waray, Wu Chinese and Yue Chinese

Dataset Structure

Data Instances

Each data instance corresponds to a paraphrase, e.g.:

{ 
    'paraphrase_set_id': '1483',  
    'sentence_id': '5778896',
    'paraphrase': 'Ɣremt adlis-a.', 
    'lists': ['7546'], 
    'tags': [''],
    'language': 'ber'
}

Data Fields

Each dialogue instance has the following fields:

  • paraphrase_set_id: a running number that groups together all sentences that are considered paraphrases of each other
  • sentence_id: OPUS sentence id
  • paraphrase: Sentential paraphrase in a given language for a given paraphrase_set_id
  • lists: Contributors can add sentences to list in order to specify the original source of the data
  • tags: Indicates morphological or phonological properties of the sentence when available
  • language: Language identifier, one of the 73 languages that belong to this dataset.

Data Splits

The dataset is having a single train split, contains a total of 1.9 million sentences, with 200 – 250 000 sentences per language

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

[More Information Needed]

Licensing Information

Creative Commons Attribution 2.0 Generic

Citation Information 

@dataset{scherrer_yves_2020_3707949,
  author       = {Scherrer, Yves},
  title        = {{TaPaCo: A Corpus of Sentential Paraphrases for 73 Languages}},
  month        = mar,
  year         = 2020,
  publisher    = {Zenodo},
  version      = {1.0},
  doi          = {10.5281/zenodo.3707949},
  url          = {https://doi.org/10.5281/zenodo.3707949}
}

Contributions

Thanks to @pacman100 for adding this dataset.