ebible

eBible gathers Bible versions from eBible.org, strips everything but the verse text, and lays it out one verse per line, aligned to a canonical book, chapter, and verse list.

eBible is a curated corpus of parallel data derived from Bibles provided by eBible.org, aimed at natural language processing research on parallel Bible text. Copyright is handled up front: the Bibles were collected from eBible.org, and to the best of the maintainers' ability the resources are releasable in this form. That framing matters because religious text reuse is legally messy, and the repo is explicit about it.

Each Bible in the repository contains only verse text. Introductory paragraphs, comments, and footnotes are removed, and everything is laid out one verse per line, which keeps the corpus clean for alignment and language modeling across the parallel versions.

Verse order in each Bible is aligned to a canonical list of Biblical books, chapters, and verses. File names pair a language abbreviation with the Paratext project name, and verse references use the three letter book abbreviation plus the numeric chapter and verse. Blank lines mark verses missing from the source, and verse ranges put all the text on the first verse's line, so the alignment is explicit in the file format itself.

A regeneration script, tested on Python 3.10 with no third party dependencies, downloads a translations list and the zip files for each translation, unpacks them, and builds extracts with a bulk extract tool from the SIL NLP project. Zip files are cached with a UTC date suffix, defaulting to 14 days, with flags to override the age or force a fresh download.

• Source 1 · Project repository
• Source 2 · README