stemlite 0.1.0

A tiny, zero-dependency collection of English stemmers (Porter, Snowball, Lancaster).

stemlite

A tiny, zero-dependency collection of English stemmers — standard library only, no nltk/numpy/anything. Pure-Python implementations of the classic Porter, Snowball (Porter2), and Lancaster stemming algorithms, plus a small registry so you can pick one by name.

It exists to be a single, self-contained, stable dependency shared across minsearch, zerosearch, and sqlitesearch — so those search libraries can normalize words the same way without each carrying its own copy or pulling in a heavyweight NLP stack. Designed to run anywhere Python runs, including constrained environments like Cloudflare Python Workers (Pyodide).

Note: these are pragmatic, simplified implementations tuned for search-time normalization, not bit-for-bit reference implementations of the published algorithms.

Install

pip install stemlite

Usage

from stemlite import get_stemmer, porter_stemmer, snowball_stemmer, lancaster_stemmer, STEMMERS

porter_stemmer("running")      # -> "run"
snowball_stemmer("running")    # -> "run"
lancaster_stemmer("running")   # -> "run"

# Pick a stemmer by name (case-insensitive).
stem = get_stemmer("porter")
stem("running")                # -> "run"

# None (or an unknown name) returns a no-op stemmer that only lowercases.
noop = get_stemmer(None)
noop("Running")                # -> "running"

get_stemmer(name)

def get_stemmer(name: Optional[str] = None) -> Callable[[str], str]: ...

Accepts "porter", "snowball", "lancaster", "none", or None, and returns a Callable[[str], str]. An unknown name (or None) falls back to the no-op stemmer, which just lowercases the input. Names are matched case-insensitively.

STEMMERS

The registry backing get_stemmer, a Dict[str, Callable[[str], str]] keyed by "porter", "snowball", "lancaster", and "none".

Choosing a stemmer

• Porter — the classic, conservative choice. Good default.
• Snowball (Porter2) — a refinement of Porter; slightly different handling of edge cases.
• Lancaster — the most aggressive; stems words down harder (higher recall, more collisions).

Development

make setup     # uv sync --dev
make test      # run the test suite
make coverage  # run tests with coverage

License

WTFPL.