---
id: nagisa
version: "0.3.0"
license: MIT License
license_treatment: permissive
maintenance: active
---

# nagisa — A Japanese tokenizer based on recurrent neural networks

License: permissive · Maintenance: active · Downloads: 823.6K/mo

## What it is and what it does

Nagisa is a Python module that tokenizes and tags Japanese text using neural networks trained on character- and word-level features. It splits Japanese sentences into words and assigns part-of-speech tags (noun, verb, particle, etc.) in a single pass, with output normalized to Unicode NFKC form.

The package is designed for ease of use: import, call tagging() on a string, and receive a result object with .words and .postags attributes. It supports filtering/extracting words by POS tag, adding custom dictionaries, and includes a built-in Japanese stopwords list. Advanced users can train custom models on annotated datasets using the fit() method and load them with a custom Tagger instance.

Use it for:

- Extract nouns, verbs, or other POS categories from Japanese documents for downstream NLP tasks
- Tokenize Japanese text for search indexing, removing particles and auxiliary words via stopword filtering
- Train a domain-specific word segmentation and tagging model on annotated Japanese corpora
- Normalize and parse Japanese user input in chatbots or form processing pipelines
- Analyze Japanese social media or news text to identify named entities or key terms by POS tag

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Nagisa performs Japanese word segmentation and part-of-speech tagging using recurrent neural networks, outputting tokenized words with their grammatical tags.

Yes, if you need Japanese NLP. Nagisa is actively maintained, has no known vulnerabilities, carries a permissive MIT license, and is in the top 5000 PyPI packages by download volume. The medium install friction (neural network dependencies) is a trade-off for having pre-trained models; if you work with Japanese text regularly, the convenience outweighs the setup cost. Not suitable if you need only English or other languages.

## Install

pip install nagisa
uv add nagisa
poetry add nagisa

## Installing nagisa

Before you install: Medium install friction due to dependencies on numpy and DyNet38/DyNet (neural network libraries). Wheels are provided for Python 3.10–3.13 across Linux, macOS, and Windows. Maintenance is active with a recent release (39 days old) and ongoing repository updates.

License in practice: MIT License permits commercial and private use with minimal restrictions, requiring only license and copyright notice retention.

Quickstart:

pip install nagisa

import nagisa

text = 'Pythonで簡単に使えるツールです'
words = nagisa.tagging(text)
print(words.words)
print(words.postags)

Requires numpy and DyNet38/DyNet (compiled neural network libraries); installation may take time on first setup.

Verify before relying:

- Whether pre-trained models are bundled or require separate download on first use
- Memory and runtime performance characteristics for typical Japanese text volumes
- Compatibility and stability of DyNet38 vs DyNet dependency resolution

## Package facts

- License: MIT License (permissive)
- Python support: unspecified
- Install friction: medium
- Maintenance: active
- Downloads: 823.6K/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

japanese word segmentation, japanese tokenizer, pos tagging japanese, japanese nlp, japanese morphological analysis, japanese text processing, japanese language parsing, japanese-nlp, neural-networks, tokenization

[View on SkillFed](https://skillfed.io/packages/nagisa) · [View on PyPI](https://pypi.org/project/nagisa/)