Hugging Face's logo

Datasets:
projecte-aina
/
CA-EN_Parallel_Corpus

Tasks:
Translation
Languages:
Catalan
English
Multilinguality:
multilingual
Size Categories:
10M<n<100M
Tags:
DOI:
License:
Dataset card Viewer Files Community
CA-EN_Parallel_Corpus / README.md
fdelucaf's picture
fdelucaf
Update README.md
1b50e14 verified
preview code
|
raw history blame contribute delete
No virus
6.75 kB
metadata
language:
  - ca
  - en
multilinguality:
  - multilingual
pretty_name: CA-EN Parallel Corpus
size_categories:
  - 10M<n<100M
task_categories:
  - translation
task_ids: []
license: cc-by-4.0

Dataset Card for CA-EN Parallel Corpus

Dataset Description

Dataset Summary

The CA-EN Parallel Corpus is a Catalan-English dataset of 14.967.979 parallel sentences. The dataset was created to support Catalan in NLP tasks, specifically Machine Translation.

Supported Tasks and Leaderboards

The dataset can be used to train Bilingual Machine Translation models between English and Catalan in any direction, as well as Multilingual Machine Translation models.

Languages

The sentences included in the dataset are in Catalan (CA) and English (EN).

Dataset Structure

Data Instances

The dataset is a single tsv file where each row contains a parallel sentence pair, as well as the following information per sentence:

  • language probability score calculated with the language detector lingua.py,
  • alignment score calculated with LaBSE,
  • domain,
  • text type.

Data Fields

Each example contains the following 7 fields:

  • ca: Catalan sentence
  • en: English sentence
  • ca_prob: Language probability for the Catalan sentence calculated with the language detector lingua.py
  • en_prob: Language probability for the English sentence calculated with the language detector lingua.py
  • alignment: Sentence pair alignment score calculated with LaBSE
  • Domain: Domain (see list of domains)
  • Type: Text type (see list of text types)

Example: 

[
  {
    Pel que fa als motors de cerca, també es basen en l'estructura del seu contingut d'informació al lloc web per analitzar i indexar el seu lloc web.	As for the search engines, they also rely on the structure of your information content on the website to analyze and index your website.	0.9999799355804416	0.9993718600460302	0.91045034	MWM	SM

  },
  ...
]

List of domains (and number of sentences per domain):

AUT: Automotive, transport, traffic regulations (2.289.951)

LEG: legal, law, HR, certificates, degrees (498.676)

MWM: Marketing, web, merchandising, customer support and service, e-commerce , advertising, surveys (1.066.111)

LSM: Medicine, natural sciences, food/nutrition, biology, sexology, cosmetics, chemistry, genetics (457.647)

ENV: Environment, agriculture, forestry, fisheries, farming, zoology, ecology (681.813)

FIN: Finance, economics, business, entrepreneurship, business, competitions, labour, employment, accounting, insurance, insurance (292.865)

POL: Politics, international relations, European Union, international organisations, defence, military (451.569)

PRN: Porn, inappropriate content (597.926)

COM: Computers, IT, robotics, domotics, home automation, telecommunications (1.200.192)

ING: Pure engineering (mechanical, electrical, electronic, aerospace...), meteorology, mining, engineering, maritime, acoustics (581.722)

ARC: Architecture, civil engineering, construction, public engineering (663.985)

MAT: Mathematics, statistics, physics (216.635)

HRM: History, religion, mythology, folklore, philosophy, psychology, ethics, anthropology, tourism (1.362.302)

CUL: Art, poetry, literature, cinema, video games, theatre, theatre/film scripts, esotericism, astrology, sports, music, photography (2.774.420)

GEN: General - generic cathegory with topics such as clothing, textiles, gastronomy, etc. (1.832.164)

List of text types (and number of sentences per text type):

PAT: Patents (583.353)

SM: Social media, chats, forums, tweets (6.420.644)

CON: Oral language, transcription of conversations, subtitles (3.709.344)

EML: Emails (543.010)

MNL: Manuals, data sheets (1.379.021)

NEW: News, journalism (1.346.845)

GEN: Prose, generic type of text (985.761)

Data Splits

The dataset contains a single split: train.

Individual domain or style specific subsets can be extracted from the original dataset by filtering by the previously mentioned domains and text types.

Dataset Creation

Curation Rationale

This dataset is aimed at promoting the development of Machine Translation between Catalan and other languages, specifically English.

Source Data

Initial Data Collection and Normalization

The data is a collection of parallel sentences in Catalan and English, partially derived from web crawlings and belonging to a mix of different domains and styles. The source data is partially Catalan authentic text translated into English and partially authentic English text translated into Catalan.

The data was obtained through a combination of human translation and machine translation with human proofreading.

The obtained corpus consists of 14.967.979 parallel sentences.

Who are the source language producers?

The original data gathering was entrusted to an external company through a public tender process.

Annotations

Annotation process

The dataset does not contain any annotations.

Who are the annotators?

[N/A]

Personal and Sensitive Information

Given that this dataset is partly derived from pre-existing datasets that may contain crawled data, and that no specific anonymisation process has been applied, personal and sensitive information may be present in the data. This needs to be considered when using the data for training models.

Considerations for Using the Data

Social Impact of Dataset

By providing this resource, we intend to promote the use of Catalan across NLP tasks, thereby improving the accessibility and visibility of the Catalan language.

Discussion of Biases

No specific bias mitigation strategies were applied to this dataset. Inherent biases may exist within the data.

Other Known Limitations

The dataset contains data of several specific domains. The dataset can be used as a whole or extracting subsets per domain or text types. Applications of this dataset in domains other than the ones included in the domain list would be of limited use.

Additional Information

Dataset Curators

Language Technologies Unit at the Barcelona Supercomputing Center (langtech@bsc.es).

This work has been promoted and financed by the Generalitat de Catalunya through the Aina project.

Licensing Information

This work is licensed under a Creative Commons Attribution 4.0 International license.

Citation Information

[N/A]

Contributions

[N/A]