license	pretty_name	language	configs	tags	size_categories
apache-2.0	English Wiktionary Data in JSONL	en de la grc ko peo akk elx sa	config_name data_files Wiktionary split path German german-wiktextract-data.jsonl split path Latin latin-wiktextract-data.jsonl split path AncientGreek ancient-greek-wiktextract-data.jsonl split path Korean korean-wiktextract-data.jsonl split path OldPersian old-persian-wiktextract-data.jsonl split path Akkadian akkadian-wiktextract-data.jsonl split path Elamite elamite-wiktextract-data.jsonl split path Sanskrit sanskrit-wiktextract-data.jsonl config_name data_files Knowledge Graph split path AllLanguage word-definition-graph-data.jsonl	Natural Language Processing NLP Wiktionary Vocabulary German Latin Ancient Greek Korean Old Persian Akkadian Elamite Sanskrit Knowledge Graph	100M<n<1B

Wiktionary Data on Hugging Face Datasets

wiktionary-data is a sub-data extraction of the English Wiktionary that currently supports the following languages:

• Deutsch - German
• Latinum - Latin
• Ἑλληνική - Ancient Greek
• 한국어 - Korean
• 𐎠𐎼𐎹 - Old Persian
• 𒀝𒅗𒁺𒌑(𒌝) - Akkadian
• Elamite
• संस्कृतम् - Sanskrit, or Classical Sanskrit

wiktionary-data was originally a sub-module of wilhelm-graphdb. While the dataset it's getting bigger, I noticed a wave of more exciting potentials this dataset can bring about that stretches beyond the scope of the containing project. Therefore I decided to promote it to a dedicated module; and here comes this repo.

The Wiktionary language data is available on 🤗 Hugging Face Datasets.

from datasets import load_dataset
dataset = load_dataset("QubitPi/wiktionary-data")

There are two data subsets:

1. Languages subset that contains extraction of a subset of supported languages:

   dataset = load_dataset("QubitPi/wiktionary-data", "Wiktionary")

   The subset contains the following splits

   • German
   • Latin
   • AncientGreek
   • Korean
   • OldPersian
   • Akkadian
   • Elamite
   • Sanskrit

2. Graph subset that is useful for constructing knowledge graphs:

   dataset = load_dataset("QubitPi/wiktionary-data", "Knowledge Graph")

   The subset contains the following splits

   • AllLanguage: all the languages listed above in a giant graph

   The Graph data ontology is the following:

   

Tip

Two words are structurally similar if and only if the two shares the same stem

Development

Data Source

Although the original Wiktionary dump is available, parsing it from scratch involves rather complicated process. For example, acquiring the inflection data of most Indo-European languages on Wiktionary has already triggered some research-level efforts. We would probably do it in the future. At present, however, we would simply take the awesome works by tatuylonen which has already processed it and presented it in in JSONL format. wiktionary-data sources the data from raw Wiktextract data (JSONL, one object per line) option there.

Environment Setup

Get the source code:

git clone git@github.com:QubitPi/wiktionary-data.git
cd wiktionary-data

It is strongly recommended to work in an isolated environment. Install virtualenv and create an isolated Python environment by

python3 -m pip install --user -U virtualenv
python3 -m virtualenv .venv

To activate this environment:

source .venv/bin/activate

or, on Windows

./venv\Scripts\activate

Tip

To deactivate this environment, use

deactivate

Installing Dependencies

pip3 install -r requirements.txt

License

The use and distribution terms for wiktionary-data are covered by the Apache License, Version 2.0.