Modern Brenton English Septuagint

A contemporary-English edition of the Septuagint (the ancient Greek Old Testament), produced by removing the remaining archaic second/third-person forms (thou/thee/thy/-est/-eth) from the public-domain Updated Brenton English Septuagint.

The result reads in plain modern English while keeping Brenton's wording — no paraphrase, no AI rewriting. Every change is a grammatical-form substitution only.

Deuteronomy 6:5
  source : And thou shalt love the Lord thy God with all thy mind, ...
  modern : And you shall love the Lord your God with all your mind, ...

Provenance (all public domain / CC0)

1. Sir Lancelot C. L. Brenton, The Septuagint Version of the Old Testament (1851) — public domain.

2. Updated Brenton English Septuagint — Brenton's text with vocabulary and accuracy updates by Adam Boyd (2020), released CC0. This is our input, taken from the USFM files at ebible.org/englxxup (bundled in data/source/usfm/). It modernized spelling and corrected the text against the Greek, but deliberately retained archaic pronouns.

   ⚠️ Do not use the BibleNLP/ebible corpus/eng-englxxup.txt for this text: that file is line-aligned to a fixed MT-shaped verse list, and the Septuagint's own versification doesn't fit it — LXX Psalm 9 loses 18 of its 39 verses, Psalm 151, Greek Daniel (DAG), Greek Esther (ESG), and Esdras B 11–23 (Nehemiah) are missing entirely (~2,200 verses silently dropped). src/parse_usfm.py extracts from the USFM, which preserves Brenton's native LXX structure, including Rahlfs lettered verses (1Kgdms supplements, ESG 4:17a–z).

3. This project — a deterministic "de-thee" pass over (2). Released CC0.

What the transform does — and doesn't

Deterministic and morphological only: it changes grammatical forms, never word choice. There is no LLM and no paraphrase — this is Scripture.

• thou/thee → you, thy → your, thine → your/yours, ye → you, mine → my
• 2nd-person verbs → modern: art→are, hast→have, shalt→shall, knowest→know, gavest→gave, walkedst→walked, madest→made, …
• 3rd-person verbs → modern: hath→has, doth→does, saith→says, walketh→walks, crieth→cries, teacheth→teaches, …

It does not touch archaic vocabulary (behold, unto, thither, verily) — those are word choices, not the "thee/thou" grammar, and changing them would alter Brenton's translation.

Known limitation — postposed-subject imperatives. Archaic imperatives put the subject after the verb, with ye or thou ("wonder ye and howl", "Turn thou, and fall", "Look thou upon"). Pronoun substitution alone yields "wonder you and howl" / "Turn you, and fall" — grammatical forms are modern, but the word order remains archaic. Truly modern English would drop the pronoun or front it (ESV prints "You turn and strike" at 1 Sam 22:18 — same words, reordered); dropping is a word deletion and fronting is a reorder, and this project stays strictly substitution-only, so the order is preserved. The same applies to inverted questions ("Know you that…?") — faithful, readable, but not reordered.

How it avoids mangling ordinary words

The hard part of any de-thee is not turning priest → pri, harvest → harve, teeth → tes, or the name Azmaveth → azmavs. Four guards, in order:

1. Curated table — pronouns + irregular/auxiliary verbs, exact.
2. Homograph gate — art→are and wilt→will only after thou (so the noun "art" and "wilt" = to droop are left alone).
3. Proper-noun detection — a token that is capitalized more often than not across the corpus and would modernize to a non-word (azmaveth→azmavs) is a name; it's skipped everywhere. Hyphenated name parts (Ish-bosheth) too.
4. Wordlist gate — an ordinary English word (priest, harvest) is left alone, unless it's a known archaic verb the wordlist happens to contain (giveth, cometh). The regular -eth/-est tail then falls to a rule whose e-restoration is decided by the wordlist (tak→take→takes).

Every substitution is logged (reports/mappings.tsv); every archaic-looking token left untouched is logged (reports/leftovers.tsv); auto-detected names are logged (reports/skipped_names.tsv). The corpus is closed, so the archaic vocabulary is finite and fully auditable.

Build

python src/fetch_source.py    # downloads the English wordlist + frequency list
python src/parse_usfm.py      # USFM -> data/source/verses_native.tsv
python src/modernize.py       # writes data/output/ + reports/
python src/verify.py          # quality gate + 28-verse golden suite

Output

• data/output/verses.tsv — the canonical output: BOOK C:V<TAB>text, one row per verse, in Brenton's native LXX versification (LXX Psalm numbering, Esdras B chapters 1–23 under EZR, Greek Daniel under DAG, Greek Esther under ESG with Rahlfs lettered verses like 4:17a).
• data/output/eng-modern-brenton.txt — the same verses as plain lines, parallel to verses.tsv row order (used by the audit scans).

Quality (current build)

metric	value
verses modernized	28,637
substitutions applied	25,786
distinct form-mappings	572
proper names auto-skipped	97 occ / 28 distinct
residual thee/thou/thy/thine/ye	0
residual hath/doth/saith	0
unhandled archaic leftovers	0
golden regression verses	46 (src/verify.py)

The pipeline went through nine rounds of adversarial review (an independent model attacking the complete mapping table, every thine/mine verse, every relative-clause verse, and random verse samples), each round's findings fixed and frozen into the golden suite: junk-wordlist arbitration, noun-plural homographs (hats/cars/DIMMs), string-collision pathologies (fling/fled, clothing/clothed), proper-name protection (Hereth, Azmaveth), substantival vs attributive thine/mine, and vocative-relative agreement ("O Lord, who judges" vs "you that judge" vs NRSV-style "who have made all things by your word").

Coverage: the full Septuagint canon including the deuterocanonical/anagignoskomena books (Tobit, Judith, Wisdom, Sirach, Baruch, 1–4 Maccabees, etc.).

License

CC0 1.0 Universal (public-domain dedication), matching the source. Brenton (1851) is public domain; the Updated Brenton (Adam Boyd, 2020) is CC0. Credit to Adam Boyd and Sir Lancelot Brenton is appreciated but not required.