Stanza

1.11.1 · active · verified Sun Apr 12

Stanza, by the Stanford NLP Group, is a Python NLP library supporting over 70 human languages. It offers a fully neural pipeline for various text analysis tasks, including tokenization, multi-word token expansion, lemmatization, part-of-speech and morphological feature tagging, dependency parsing, and named entity recognition. Stanza also provides a stable Python interface to the Java Stanford CoreNLP Toolkit. Actively maintained, it receives regular updates, with the current version being 1.11.1.

Warnings

breaking As of v1.11.1, Stanza's default model download location has changed from `~/stanza_resources` to system-specific cache directories via `platformdirs`. This may affect users who relied on the old default path or custom scripts expecting models in `~/stanza_resources`.
Fix: Update your code to check the new default location (e.g., `platformdirs.user_cache_dir('stanza')`) or explicitly specify the model directory using the `dir` parameter in `stanza.download()` and `stanza.Pipeline()`, or by setting the `STANZA_RESOURCES_DIR` environment variable.
gotcha Stanza's neural models require PyTorch. Users often encounter `ERROR: Could not find a version that satisfies the requirement torch` during `pip install stanza` if PyTorch is not pre-installed or if there are compatibility issues. Installing PyTorch separately first, especially via a system package manager (e.g., `conda install pytorch ...`), is frequently recommended for a smoother installation.
Fix: Ensure PyTorch is installed and compatible with your system and Python version *before* installing Stanza. Refer to the official PyTorch installation instructions for your environment.
gotcha Processing individual documents or sentences one by one in a loop can be significantly slower than processing them in batches. Stanza is optimized for batch processing.
Fix: Collect multiple texts into a list and pass the list to the `nlp()` pipeline for annotation to improve performance.
deprecated Prior to version 1.0.0, the library was named `stanfordnlp`. If you are looking for very old documentation or examples, you might encounter references to this legacy package name.
Fix: For new projects, always use `stanza`. If working with legacy code, be aware that `pip install stanfordnlp` would be required for older versions.
gotcha Stanza has a strict Python version requirement of >=3.9. Using older Python versions can lead to various runtime errors, including `OSError: [Errno 22] Invalid argument` during model loading on macOS with Python <=3.7.1.
Fix: Ensure your Python environment is running version 3.9 or newer. Upgrade Python if necessary.

Install

pip install stanza Install latest version

Imports

stanza
```
import stanza
```

Quickstart

This quickstart downloads the default English language model, initializes a Stanza pipeline, processes a sample text, and then prints out token-level annotations (UPOS, lemma, dependency relation) and named entities.

import stanza

# Download an English model (only needs to be run once)
# Stanza will auto-download if models are not found, but explicit download is good practice.
stanza.download('en')

# Initialize the English neural pipeline
nlp = stanza.Pipeline('en')

# Process some text
text = "Barack Obama was born in Hawaii. He was the 44th President of the United States."
doc = nlp(text)

# Access annotations
print(f"Processing: '{text}'")
for i, sent in enumerate(doc.sentences):
    print(f"\nSentence {i+1}:")
    for word in sent.words:
        print(f"  {word.text}\tUPOS: {word.upos}\tLemma: {word.lemma}\tDepRel: {word.deprel}\tHead: {doc.sentences[0].words[word.head-1].text if word.head > 0 else 'ROOT'}")

print("\nNamed Entities:")
for ent in doc.entities:
    print(f"  {ent.text}\tType: {ent.type}")

view raw JSON →