Any Texts Statistics (anyTS)

anyTS is the language-independent core of the text statistics libraries ruTS (Russian) and esTS (Spanish). It holds the code that does not depend on a language, so that it is written, tested and documented once; a language library adds its tokenizer, lemmatizer, spaCy model and rules and re-exports the core under its own names.

Documentation · PyPI

Features

• Extraction - sentence, word and character N-gram extractors whose tokenizer, lemmatizer and number pattern are hooks a language library overrides
• Lexical diversity - 32 metrics: TTR and its variants, MATTR, MSTTR, MTLD, MA-MTLD, MTLD-W, HD-D, the indices of Simpson and Yule, entropy, the laws of Zipf and Heaps, over the whole text or over windows with confidence intervals
• Cohesion helpers - the overlap of sentences, adjacent and over all pairs, Dice, givenness and repetition
• Dependency tree helpers - dependency distances, tree depth, valency and coordination chains on the labels shared by Universal Dependencies and ClearNLP
• Corpus measures - keywords against a corpus or a frequency dictionary, collocations, the dispersion of a word, Burrows's Delta and its variants, Zeta, Kilgarriff's chi-square, the Mendenhall curve and the comparison of corpora with effect sizes, a bootstrap by texts and the Holm correction
• Exceptions - one hierarchy whose classes are also built-in exceptions

The reference of every function is a named section that the libraries include in their own documentation, so a formula is described in one place.

Installation

Requires Python 3.11 or newer.

pip install anyts

Or with uv:

uv add anyts

The dependencies are numpy, pandas, scipy and spaCy; no trained spaCy model is needed.

Quick start

>>> from anyts import DiversityStats, SentsExtractor, WordsExtractor

>>> text = "The cat sat on the mat. The dog sat on the log! The cat and the dog slept."

>>> SentsExtractor().extract(text)
('The cat sat on the mat.', 'The dog sat on the log!', 'The cat and the dog slept.')

>>> ds = DiversityStats(WordsExtractor(lowercase=True).extract(text))
>>> ds.ttr, round(ds.entropy, 3)
(0.5, 2.864)

A language library subclasses the extractors and overrides their hooks:

>>> class Words(WordsExtractor):
...     def lemmatize(self, word):
...         return {"sat": "sit", "slept": "sleep"}.get(word.lower(), word)

>>> Words(lowercase=True, use_lexemes=True).extract("The cat sat. The dog slept.")
('the', 'cat', 'sit', 'the', 'dog', 'sleep')

The corpus measures take lists of words:

>>> from anyts.corpus import keyness

>>> cats = WordsExtractor(lowercase=True).extract("The cat sat on the mat. The cat saw a bird. The cat slept.")
>>> dogs = WordsExtractor(lowercase=True).extract("The dog sat on the log. The dog saw a cat. The dog barked.")
>>> [(keyword.word, round(keyword.log_ratio, 2)) for keyword in keyness(cats, dogs, min_freq=2)]
[('cat', 1.58)]

Development

The project uses uv for dependencies and ruff for linting and formatting.

git clone https://github.com/SergeyShk/anyTS.git
cd anyTS

make deps                   # create the environment and install all dependencies
uv run pre-commit install   # hooks: linters on commit, tests on push
make lint                   # ruff check, ruff format --check, mypy
make test-cov               # pytest with doctests and full coverage
make docs-build             # mkdocs build --strict

The full list of commands is in make help. Contribution guidelines are in CONTRIBUTING.md.

License

MIT