You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

NuBerea Research: Authorship Stylometry

Computational stylometric measurements over the Greek New Testament for two families of texts: the thirteen Pauline letters (with the composite-letter question for 2 Corinthians) and the Johannine Gospel and letters. Every table states what the data separate, relative to a named anchor at a named length. The tables are model-derived diagnostics of the available data; they are not a historical ruling on who wrote a book. Every result row carries evidence_class = model_result where the tool layer exposes it.

These are small findings tables for scholarly reference and reproducible discussion, not a text corpus: they ship numbers, labels and OSIS anchors only, never Greek text. The full technical specification (schemas, methodology and provenance) is in the gated SPEC that accompanies this repository.

Pauline letters

  • Anchor: the 7 undisputed letters (Romans, 1-2 Corinthians, Galatians, Philippians, 1 Thessalonians, Philemon), taken as received. They are the reference input class, not a finding.
  • Verdict vocabulary (pauline_authorship, and the authorship_verdict rows of authorship_stylometry): a measurement label relative to the calibrated leave-one-out range of the reference letters.
    • reference: one of the 7 input letters;
    • within_range: inside the reference letters' range;
    • just_outside_range: outside the range, meta-z at most 2;
    • well_outside_range: meta-z above 2.
  • Calibration limits are stated, not hidden: Philemon (334 tokens) is excluded from calibration for the length-sensitive metrics (function-word distance z, Yule's K) and kept for lexical distance; Romans carries a slight large-sample Yule's K bias.
  • 2 Corinthians: the Semler and Bornkamm partition theories and a single-letter null are compared on the book's own stylistic signal, with named scholarly cuts set against computed change points.

Johannine Gospel and letters

  • Anchor: the Gospel core, John 1:19-20:31 with the curated seam spans removed. Function-word and related feature families (Method 8 only) at 2,000-token windows, calibrated on New Testament control pairs: W (two windows of one text), G (same-author control pairs) and D (different-author control pairs). G and D label control pairs whose authorship is the received input.
  • 1 John is nearest the Gospel core in every feature family at 2,000 tokens; whether the method links the two under the G tolerance depends on the family.
  • No internal break is separated in the Gospel core.
  • "Hands" (johannine_hands) are stylistic groups the method separates; not persons. The feature families do not agree on one count; the stable pattern is the Gospel core with 1 John, separated from Revelation.
  • Revelation is outside the family, out of scope per owner ruling (2026-11-11). For the record: its function-word distance from the Gospel core is 0.286 [0.264, 0.310] at 2,000 tokens, at the 100th percentile of G control pairs and the 98th of D control pairs. The method cannot separate a change of author from a change of genre, and no control spans gospel and apocalypse.
  • Short units (the Prologue 252 tokens, John 21 549, 2 John 245, 3 John 219) are undetermined for counting stylistic groups. johannine_short_unit_consistency tests each at its own length against the Gospel core: all four are consistent with the Gospel core at their lengths and none is anomalous. The test's power is weak: it calls only 9-19% of D control windows anomalous, so a consistent result raises the odds of the G class only 1.09-1.23 fold.

License

Released under CC BY 4.0.

Attribution

Analytical outputs are original to the NuBerea project. Underlying data derive from the following upstream datasets:

Scholarly References

  • Beijen, E. & De Heide, R. (2025). "Authorship Verification of the Disputed Pauline Letters through Deep Learning." HIPHIL Novum 10(1).
  • Pracht, E. B. & McCauley, T. (2025). "Paul's Style and the Problem of the Pastoral Letters." Open Theology 11, 20240034.

NuBerea project. Licensed CC BY 4.0.

Downloads last month
2,600