metadata

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

There are two data subsets:

Languages subset that contains the sub-data extraction of the following splits:
- German
- Latin
- AncientGreek
- Korean
- OldPersian
- Akkadian
- Elamite
- Sanskrit
Graph subset that is useful for constructing knowledge graphs:
- AllLanguage: all the languages in a giant graph
The Graph data ontology is the following:

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

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.