fiNERweb
Collection
A multilingual dataset for NER covering 91 langauges and 25 scripts
•
3 items
•
Updated
•
1
Datasets:
Dataset Viewer
The dataset viewer is not available for this subset.
The default config contains 91 while it should generally contain 3 splits maximum (train/validation/test). If the splits mal, fra, ukr, hin, xho... are not used to differentiate between training and evaluation, please consider defining configs of this dataset instead. You can find how to define configs instead of splits here: https://huggingface.co/docs/hub/datasets-data-files-configuration
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
fiNERweb
Dataset Description
fiNERweb is a multilingual named entity recognition dataset containing annotated text in multiple languages. Each example contains:
- Original text
- Tokenized text
- BIO tags
- Character spans for entities
- Token spans for entities
Languages
Currently supported languages:
- vi: Vietnamese
- ta: Tamil
- or: Odia (Oriya)
- sk: Slovak
- af: Afrikaans
- cs: Czech
- ga: Irish
- pt: Portuguese
- so: Somali
- sl: Slovenian
- cy: Welsh
- fy: Western Frisian
- uk: Ukrainian
- is: Icelandic
- la: Latin
- hy: Armenian
- bg: Bulgarian
- tr: Turkish
- uz: Uzbek
- nl: Dutch
- ps: Pashto
- be: Belarusian
- en: English
- xh: Xhosa
- jv: Javanese
- hi: Hindi
- my: Burmese
- br: Breton
- ur: Urdu
- sr: Serbian
- zh: Chinese (Mandarin)
- ka: Georgian
- hr: Croatian
- ml: Malayalam
- km: Khmer
- te: Telugu
- ru: Russian
- ar: Arabic
- de: German
- fr: French
- om: Oromo
- sw: Swahili
- az: Azerbaijani
- gl: Galician
- ko: Korean
- sd: Sindhi
- fi: Finnish
- lv: Latvian
- eo: Esperanto
- kk: Kazakh
- lt: Lithuanian
- mk: Macedonian
- eu: Basque
- am: Amharic
- he: Hebrew
- si: Sinhala
- ne: Nepali
- yi: Yiddish
- sq: Albanian
- it: Italian
- kn: Kannada
- mn: Mongolian
- ja: Japanese
- gu: Gujarati
- su: Sundanese
- ro: Romanian
- sa: Sanskrit
- ku: Kurdish
- ky: Kyrgyz
- ug: Uyghur
- gd: Scottish Gaelic
- es: Spanish
- et: Estonian
- th: Thai
- sv: Swedish
- hu: Hungarian
- bs: Bosnian
- bn: Bengali
- ca: Catalan
- mr: Marathi
- da: Danish
- pl: Polish
- el: Greek
- ms: Malay
- mg: Malagasy
- pa: Punjabi
- lo: Lao
- fa: Persian
- tl: Tagalog
- as: Assamese
- id: Indonesian
Dataset Structure
Each example contains:
{
"text": str, # Original text
"tokens": List[str], # Tokenized text
"bio_tags": List[str], # BIO tags for NER
"char_spans": List[Dict], # Character-level entity spans
"token_spans": List[Dict] # Token-level entity spans
}
Usage
from datasets import load_dataset
# Load a specific language
dataset = load_dataset("whoisjones/fiNERweb", "am") # For Amharic
# or
dataset = load_dataset("whoisjones/fiNERweb", "en") # For English
# Access the data
print(dataset["train"][0])
Citation
If you use this dataset, please cite:
@misc{fiNERweb,
author = {Jonas Golde},
title = {fiNERweb: Multilingual Named Entity Recognition Dataset},
year = {2024},
publisher = {HuggingFace},
journal = {HuggingFace Datasets},
howpublished = {\\url{https://huggingface.co/datasets/whoisjones/fiNERweb}}
}
Key changes:
- Updated all language codes to their ISO 639-1 equivalents
- Updated the language list in the metadata section
- Updated the language descriptions to use ISO codes
- Updated the usage examples to use ISO codes
This should resolve the language code validation errors in the metadata. The ISO 639-1 codes are the standard two-letter codes that Hugging Face expects in the metadata.
- Downloads last month
- 713
Size of downloaded dataset files:
271 MB
Size of the auto-converted Parquet files:
271 MB
Number of rows:
1,990,764