Stanza is the Stanford NLP Group’s official Python library for linguistic analysis in many human languages. Built on PyTorch, it provides production-ready pipelines for tokenization, sentence segmentation, part-of-speech tagging, lemmatization, dependency parsing and named entity recognition. The project has about 7,900 stars on GitHub and was actively maintained in 2026.
Stanza downloads pretrained neural models trained on the Universal Dependencies treebanks and other corpora, covering more than 70 languages with a consistent API. A pipeline processes raw text end to end: it splits text into sentences and tokens, assigns grammatical annotations and builds full dependency trees, and it can recognize named entities in the languages that support it. The library is designed both for researchers who want state-of-the-art, paper-backed models and for engineers who need a dependable NLP toolkit that installs with pip and runs on CPU or GPU. It is the successor in spirit to the widely used Stanford CoreNLP toolkit, reimplemented natively in PyTorch.
Free. Stanza is an open-source research library from Stanford; the code and the pretrained models are available at no cost for research and development use.
NLP researchers, computational linguists and developers who need accurate multilingual text analysis โ from parsing and entity extraction to preprocessing pipelines for downstream machine learning.