nlpcda

Why we included this project

Chinese NLP projects that run short on training text can use this package to generate more of it quickly, without collecting new corpora. It installs with pip and packs nine augmentation techniques: random entity replacement, synonym and homophone swaps, character deletion, adjacent-character shuffling, and equivalence-character substitution, plus a dedicated mode for NER data in BIO format that augments sentences and their labels together, which most utilities won't do out of the box. The author tuned it to protect meaningful details, so dates, times, and digits survive intact and the generated samples keep most of their original meaning. You can also supply your own synonym and equivalence dictionaries, which makes it easy to steer the noise toward your domain. For teams training classifiers, retrieval models, or entity taggers on Chinese text and wanting more examples without more labeling, this is a simple, installable starting point.