GiNZA is an open-source Japanese natural language processing library from Megagon Labs, built on the spaCy framework and Universal Dependencies. Available on GitHub with about 870 stars under the MIT license and updated in 2026, it is the standard open toolkit for Japanese tokenization and syntactic analysis.
GiNZA’s parsing models come out of a joint research project between Recruit Co., Ltd. and the National Institute for Japanese Language and Linguistics. Because it uses spaCy as its framework, GiNZA plugs into familiar pipeline APIs while adding Japanese-specific components, including bunsetsu phrase extraction that improves syntactic structure analysis. From version 5.3 the library runs on Python 3.10 or later, and its dependency parsing can be significantly accelerated on a GPU.
Developers install GiNZA from PyPI and load it as a spaCy pipeline, after which Japanese text can be processed into tokens, parts of speech, named entities, and dependency trees using the same API patterns as other spaCy models. English documentation is provided alongside the Japanese documentation, and the project site collects release notes and background on the underlying research. The maintainers recommend recent Python versions for simpler dependency installation and better throughput, especially for larger document collections.
GiNZA is free and open source under the MIT license. There are no paid plans or hosted tiers; you install the library and models and run them in your own Python environment.
GiNZA is for NLP engineers, researchers, and product teams working with Japanese text who need reliable tokenization, part-of-speech tagging, and dependency parsing without building language-specific tooling from scratch.