Data augmentation for Japanese text, built on AugLy.
The current version requires Python 3.11. The latest PyPI release predates the
uv migration, so install the current version from the main branch until the
next release is published:
uv add "augly-jp @ git+https://github.com/chck/AugLy-jp.git@main"from augly_jp import text as text_augmentations
text = "あらゆる現実をすべて自分のほうへねじ曲げたのだ"
augmented = text_augmentations.replace_synonym_words(text)
print(augmented)| Function | Example output | Description |
|---|---|---|
replace_synonym_words |
あらゆる現実をすべて自身のほうへねじ曲げたのだ | Substitutes words using the Sudachi synonym dictionary |
replace_wordembs_words |
あらゆる現実をすべて関心のほうへねじ曲げたのだ | Substitutes words using word embeddings |
replace_fillmask_words |
つまり現実を、未来な未来まで変えたいんだ | Generates substitutions with a masked language model |
replace_backtranslation_sentences |
そして、ほかの人たちをそれぞれの道に安置しておられた | Augments text through back translation |
Development instructions are in CONTRIBUTING.md.
Licensed under the MIT License. Runtime models and datasets retain their own license terms.