WikiPron is a command-line tool and Python API for mining multilingual pronunciation data from Wiktionary, as well as a database of pronunciation dictionaries mined using this tool.

• Command-line tool
• Python API
• Data
• Models
• Development

If you use WikiPron in your research, please cite the following:

Jackson L. Lee, Lucas F.E. Ashby, M. Elizabeth Garza, Yeonju Lee-Sikka, Sean Miller, Alan Wong, Arya D. McCarthy, and Kyle Gorman (2020). Massively multilingual pronunciation mining with WikiPron. In In Proceedings of the 12th Language Resources and Evaluation Conference, pages 4223-4228. [bibtex]

WikiPron requires Python 3.6+. It is available through pip:

pip install wikipron

After installation, the terminal command wikipron will be available. As a basic example, the following command scrapes G2P data for French:

wikipron fra

The language is indicated by a three-letter ISO 639-2 or ISO 639-3 language code, e.g., fra for French. For which languages can be scraped, here is the complete list of languages on Wiktionary that have pronunciation entries.

One can optionally specify dialects to target using the --dialect flag. The dialect name can be found together with the transcription on Wiktionary. For example, "(UK, US) IPA: /təˈmɑːtəʊ/". To restrict to the union of dialects use the pipe character '|': e.g., --dialect='General American | US'. Transcriptions which lack a dialect specification are selected regardless of the value of this flag.

By default, the segments library is used to segment the transcription into whitespace. The segmentation tends to place IPA diacritics and modifiers on the "parent" symbol. For instance, [kʰæt] is rendered kʰ æ t. This can be disabled using the --no-segment flag.

The scraped data is organized with each <word, pronunciation> pair on its own line, where the word and pronunciation are separated by a tab. Note that the pronunciation is in International Phonetic Alphabet (IPA), segmented by spaces that correctly handle the combining and modifier diacritics for modeling purposes, e.g., we have kʰ æ t with the aspirated k instead of k ʰ æ t.

For illustration, here is a snippet of French data scraped by WikiPron:

accrémentitielle    a k ʁ e m ɑ̃ t i t j ɛ l
accrescent  a k ʁ ɛ s ɑ̃
accrétion   a k ʁ e s j ɔ̃
accrétions  a k ʁ e s j ɔ̃

By default, the scraped data appears in the terminal. To save the data in a TSV file, please redirect the standard output to a filename of your choice:

wikipron fra > fra.tsv

The wikipron terminal command has an array of options to configure your scraping run. For a full list of the options, please run wikipron -h.

The underlying module can also be used from Python. A standard workflow looks like:

import wikipron

config = wikipron.Config(key="fra")  # French, with default options.
for word, pron in wikipron.scrape(config):
    ...

We also make available a database of 2.5 million word/pronunciation pairs mined using WikiPron.

We host grapheme-to-phoneme models and modeling software in a separate repository.

The source code of WikiPron is hosted on GitHub at https://github.com/kylebgorman/wikipron, where development also happens.

For the latest changes not yet released through pip or working on the codebase yourself, you may obtain the latest source code through GitHub and git:

1. Create a fork of the wikipron repo on your GitHub account.

2. Locally, make sure you are in some sort of a virtual environment (venv, virtualenv, conda, etc).

3. Download and install the library in the "editable" mode together with the core and dev dependencies within the virtual environment:

   git clone https://github.com/<your-github-username>/wikipron.git
   cd wikipron
   pip install --upgrade pip setuptools
   pip install -r requirements.txt
   pip install --no-deps -e .

We keep track of notable changes in CHANGELOG.md.

For questions, bug reports, and feature requests, please file an issue.

If you would like to contribute to the wikipron codebase, please see CONTRIBUTING.md.

WikiPron is released under an Apache 2.0 license. Please see LICENSE.txt for details.

Please note that Wiktionary data in the data/ directory has its own licensing terms.