stanza

It supports over 70 human languages with pre-trained models, making it one of the most linguistically comprehensive NLP libraries available in a single package.

The pipeline is designed for both research and production use each component is a trained neural model that produces competitive accuracy on standard NLP benchmarks, and the pipeline architecture processes documents efficiently in batch mode for throughput-sensitive applications.

Stanza's models are trained on Universal Dependencies treebanks and NER corpora specific to each language, reflecting the linguistic diversity of real-world text processing needs that monolingual NLP libraries cannot address.

The library integrates with spaCy via a bridge package for users who want Stanza's multilingual models within spaCy's component ecosystem.

Stanza is open-source under the Apache 2.0 license and developed by the Stanford NLP Group as both a research contribution and a practical tool for the NLP community.

It is used in academic research requiring accurate multilingual linguistic annotation, in clinical NLP pipelines where biomedical named entity models are required (Stanza includes a biomedical pipeline variant), and in information extraction systems that need dependency parse trees as structural features.

Models are downloaded automatically on first use from the Stanford NLP hub, covering major world languages and several low-resource languages.