aiana94/polynews-parallel

Dataset Card for PolyNewsParallel

Dataset Summary

PolyNewsParallel is a multilingual paralllel dataset containing news titles for 833 language pairs. It covers 65 languages and 17 scripts.

Uses

This dataset can be used for domain adaptation of language models or machine translation.

Languages

There are 65 languages avaiable:

Code	Language	Script
amh_Ethi	Amharic	Ethiopic
arb_Arab	Modern Standard Arabic	Arabic
ayr_Latn	Central Aymara	Latin
bam_Latn	Bambara	Latin
bbj_Latn	Ghomálá’	Latin
ben_Beng	Bengali	Bengali
bul_Cyrl	Bulgarian	Cyrillic
cat_Latn	Catalan	Latin
ces_Latn	Czech	Latin
ckb_Arab	Central Kurdish	Arabic
dan_Latn	Danish	Latin
deu_Latn	German	Latin
ell_Grek	Greek	Greek
eng_Latn	English	Latin
est_Latn	Estonian	Latin
ewe_Latn	Éwé	Latin
fil_Latn	Filipino	Latin
fin_Latn	Finnish	Latin
fon_Latn	Fon	Latin
fra_Latn	French	Latin
guj_Gujr	Gujarati	Gujarati
hau_Latn	Hausa	Latin
heb_Hebr	Hebrew	Hebrew
hin_Deva	Hindi	Devanagari
hun_Latn	Hungarian	Latin
ibo_Latn	Igbo	Latin
ind_Latn	Indonesian	Latin
ita_Latn	Italian	Latin
jpn_Jpan	Japanese	Japanese
kaz_Cyrl	Kazakh	Cyrillic
khm_Khmr	Khmer	Khmer
kor_Hang	Korean	Hangul
lav_Latn	Latvian	Latin
lit_Latn	Lithuanian	Latin
lug_Latn	Ganda	Latin
luo_Latn	Luo	Latin
mkd_Cyrl	Macedonian	Cyrillic
mos_Latn	Mossi	Latin
mya_Mymr	Burmese	Myanmar
nld_Latn	Dutch	Latin
npi_Deva	Nepali	Devanagari
ory_Orya	Odia	Oriya
pan_Guru	Eastern Panjabi	Gurmukhi
pcm_Latn	Nigerian Pidgin	Latin
pes_Arab	Western Persian	Arabic
plt_Latn	Malagasy	Latin
pol_Latn	Polish	Latin
por_Latn	Portuguese	Latin
ron_Latn	Romanian	Latin
rus_Cyrl	Russian	Cyrillic
spa_Latn	Spanish	Latin
sqi_Latn	Albanian	Latin
srp_Latn	Serbian	Latin
swe_Latn	Swedish	Latin
swh_Latn	Swahili	Latin
tet_Latn	Tetun	Latin
tsn_Latn	Tswana	Latin
tur_Latn	Turkish	Latin
twi_Latn	Twi	Latin
urd_Arab	Urdu	Arabic
wol_Latn	Wolof	Latin
yor_Latn	Yorùbá	Latin
zho_Hans	Chinese	Han (Simplified)
zho_Hant	Chinese	Han (Traditional)
zul_Latn	Zulu	Latin

The heatmap shows the language pairs available, as well as the number of articles per language pair.

Dataset Structure

Data Instances

>>> from datasets import load_dataset
>>> data = load_dataset('aiana94/polynews-parallel', 'eng_Latn-ron_Latn')

# Please, specify the language code,

# A data point example is below:

{
"src": "They continue to support the view that this decision will have a lasting negative impact on the rule of law in the country. ",
"tgt": "Ei continuă să creadă că această decizie va avea efecte negative pe termen lung asupra statului de drept în țară. ",
"provenance": "globalvoices"
}

Data Fields

• src (string): source news text
• tgt (string): target news text
• provenance (string) : source dataset for the news example

Data Splits

For all languages, there is only the train split.

Dataset Creation

Curation Rationale

Multiple multilingual, human-translated, datasets containing news texts have been released in recent years. However, these datasets are stored in different formats and various websites, and many contain numerous near duplicates. With PolyNewsParallel, we aim to provide an easily-accessible, unified and deduplicated parallel dataset that combines these disparate data sources, and which can be used for domain adaptation of language models or machine translation in both high-resource and low-resource languages.

Source Data

The source data consists of five multilingual news datasets.

• GlobalVoices (v2018q4)
• WMT-News (v2019)
• MAFAND (train split)

Data Collection and Processing

We processed the data using a working script which covers the entire processing pipeline. It can be found here.

The data processing pipeline consists of:

1. Downloading the WMT-News and GlobalVoices News from OPUS.
2. Loading MAFAND datasets from Hugging Face Hub (only the train splits).
3. Concatenating, per language, all news texts from the source datasets.
4. Data cleaning (e.g., removal of exact duplicates, short texts, texts in other scripts)
5. MinHash near-deduplication per language.

Annotations

We augment the original samples with the provenance annotation which specifies the original data source from which a particular examples stems.

Personal and Sensitive Information

The data is sourced from newspaper sources and contains mentions of public figures and individuals.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

Users should keep in mind that the dataset contains short news texts (e.g., mostly titles), which might limit the applicability of the developed systems to other domains.

Additional Information

Licensing Information

The dataset is released under the CC BY-NC Attribution-NonCommercial 4.0 International license.

Citation Infomation

BibTeX:

[More Information Needed]