scta-scikit

Texts, search, statistics and similarity for the Scholastic Commentaries and Texts Archive (SCTA): the medieval Latin commentaries, questions and treatises the SCTA publishes, as documents you can load at any level of their structure (whole work, book, question, paragraph, ...), search, compare and model.

Install

pip install scta-scikit

Python 3.12 or newer. (Maintainers who lemmatize the corpus: pip install "scta-scikit[build]".)

Get the data

The corpus and models are published separately from the package, on Hugging Face: scta/scta-corpus, scta/scta-doc2vec and scta/scta-para-tfidf. Download them once:

import scta_scikit as sc

sc.download()                                   # the latest corpus (all texts, with lemmatized versions)
sc.download("2026.10")                          # a given version
sc.download(texts=["augustinedetrinitate"])     # only some texts
sc.download(doc2vec=True)                       # with the Doc2Vec models and the corpus they were trained on
sc.download(para_tfidf=True)                    # with the paragraph TF-IDF (for `scta match`)

or from the shell: scta download, scta download 2026.10 --doc2vec, scta download --para-tfidf, scta download --texts augustinedetrinitate.

The download holds the texts and their structure; the package builds the document groupings, the question-title TF-IDF and the citation similarities from them on your machine (about a minute for everything, or on first use). Models come with the corpus version they were built from, unless you ask for another. The data goes to ~/scta-data unless you choose another folder:

sc.set_data_dir("/path/to/data", save=True)    # remembered for future sessions

(scta config --set-data-dir PATH, or the SCTA_DATA environment variable.)

Use

import scta_scikit as sc

# documents at a level of a text's structure (here: books of De trinitate)
docs, texts, idx2label, label2idx = sc.load_docs("augustinedetrinitate", level=2)

# one document's text by its SCTA resource id
sc.get_document_text("adt-l1")

# keyword in context
corpus = sc.load_docs_by_toplevel(structure="structureBlock", toplevel="augustinedetrinitate")
sc.display_kwic(sc.kwic_search(corpus, "trinitas"))

From the shell (scta --help):

scta get adt-l1                       # text of a resource
scta search trinitas --toplevel augustinedetrinitate
scta stats --toplevel augustinedetrinitate --level 2
scta sim adt-l1                       # most similar documents (Doc2Vec)
scta sim adt-l1 --citations           # by shared citations
scta match "omnis doctrina vel rerum est vel signorum"   # paragraphs quoting or borrowing a passage

See readme_cli.md for every command.

License

The code is MIT licensed. The SCTA texts, and the corpus and models published from them, are published under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

For maintainers

Building the corpus from the SCTA servers, lemmatizing it, training the models, running all of it on the cluster, and publishing a release are described in readme_cli.md (scta update, scta model build, scta publish) and CLAUDE.md.