New Testament

The Greek New Testament is a first-class, annotated corpus in pyaegean. One call, greek.load_nt(...), returns the Koine text of the NT (the Nestle 1904 edition) as a standard Corpus in which every token already carries gold annotations: a dictionary lemma, a Robinson-style morph parse, a strongs number, a reconciled Universal Dependencies upos, the normalized form, and a short Koine gloss. You never have to run a tagger to work with it: the analysis is already there, and it is gold (from the edition), not pyaegean's own prediction.

This page is the reference for that corpus and the aegean greek nt command family. For the wider "load real Greek texts" story (the classical literary works, the load_work / CTS-id path) see Greek Works and Books; for everything you can then run on the loaded text (syllabify, scan, tag, parse) see Greek NLP.

Quick start

from aegean import greek

corpus = greek.load_nt("John", ref="1.1-1.3")   # John 1:1–3, offline (bundled)
doc = corpus.documents[0]
doc.id                                # 'John 1'
doc.meta.name                         # 'John 1'

t = doc.tokens[0]
t.text                                # 'Ἐν'
t.annotations
# {'lemma': 'ἐν', 'morph': 'PREP', 'strongs': '1722', 'normalized': 'Ἐν',
#  'upos': 'ADP', 'ref': 'John.1.1', 'gloss': 'in, on, among'}

On the command line, the same read:

aegean greek nt John 1

John 1  (1 chapter, 828 tokens)
John 1
  1: Ἐν ἀρχῇ ἦν ὁ Λόγος, καὶ ὁ Λόγος ἦν πρὸς τὸν Θεόν, καὶ Θεὸς ἦν ὁ Λόγος.
  2: Οὗτος ἦν ἐν ἀρχῇ πρὸς τὸν Θεόν.
  3: πάντα δι’ αὐτοῦ ἐγένετο, καὶ χωρὶς αὐτοῦ ἐγένετο οὐδὲ ἕν ὃ γέγονεν.
  …

`load_nt(book=None, *, ref=None, force=False)`

The Koine counterpart to load_work. It returns a Corpus with one Document per chapter; each token's annotations live in Token.annotations, so to_dataframe(level="token") surfaces them as columns.

argument	meaning
`book`	a book name or abbreviation (`"John"`, `"Jn"`, `"1Cor"`, `"Rev"`). Omit it (or pass `None`) to load the whole 27-book NT.
`ref`	a chapter, verse, or range within the book (see below). A `ref` requires a `book`; you cannot address a verse across all 27 books.
`force`	re-fetch the full corpus asset even if it is already cached.

greek.load_nt("John")                  # the whole Gospel of John
greek.load_nt("John", ref="1.1-18")    # John 1:1–18
greek.load_nt("Rom", ref="8")          # Romans chapter 8
greek.load_nt("Rom", ref="8.28")       # a single verse, Romans 8:28
greek.load_nt()                        # the whole 27-book NT

Token text, lemmas, and normalized forms are NFC-normalized at load time (the source edition mixes precomposed oxia and tonos codepoints, and the rest of the library emits NFC), so the gold strings compare byte-for-byte with pyaegean's own output.

The CLI: `aegean greek nt BOOK [PASSAGE]`

aegean greek nt [BOOK] [PASSAGE] [--ref REF] [-o FILE] [--json]

PASSAGE is a positional chapter or range, so the common reads are short:

aegean greek nt John 1        # a whole chapter
aegean greek nt Matt 1-3      # a chapter range
aegean greek nt John 1.1-1.18 # a verse range
aegean greek nt Rom 8.28      # a single verse

Naming a book renders the text (chapter headers, verse-numbered lines):

aegean greek nt Matt 1-3

Matt 1-3  (3 chapters, 1220 tokens)
Matthew 1
  1: Βίβλος γενέσεως Ἰησοῦ Χριστοῦ υἱοῦ Δαυεὶδ υἱοῦ Ἀβραάμ.
  2: Ἀβραὰμ ἐγέννησεν τὸν Ἰσαάκ, Ἰσαὰκ δὲ ἐγέννησεν τὸν Ἰακώβ, …
  …

--ref is an alias for the positional PASSAGE (aegean greek nt John --ref 1.1-1.18 does the same as aegean greek nt John 1.1-1.18). Give the passage one way or the other, not both.

With no BOOK the command prints a summary of the whole NT rather than dumping 137,779 tokens to your terminal:

aegean greek nt

                                   Greek NT
┌──────────────┬──────────────────────────────────────────────────────┐
│ field        │ value                                                  │
├──────────────┼──────────────────────────────────────────────────────┤
│ scope        │ whole NT                                                │
│ ref          │                                                        │
│ documents    │ 260                                                    │
│ tokens       │ 137779                                                 │
│ first        │ Matt 1                                                 │
│ source       │ Nestle 1904 Greek NT — morphology/lemmas …            │
│ data_version │ nt-corpus-v1@713f28a3b7d4d66132f5aa809fa223fe79762e5d │
└──────────────┴──────────────────────────────────────────────────────┘
read a book:  aegean greek nt John 1   (a chapter or range, e.g. Matt 1-3)

Two more flags:

flag	meaning
`-o` / `--output`	save the selected corpus to a `.json` or `.db`/`.sqlite` file (chosen by extension) instead of printing.
`--json`	print a machine-readable summary (`scope`, `ref`, `documents`, `tokens`, `first`, `source`, `data_version`).

Because the per-token annotations ride in each token, exporting at token level spreads them into columns:

aegean greek nt John 1 -o john1.json
aegean export john1.json -f csv --level token -o john1.csv   # lemma, morph, strongs, upos, … as columns

Addressing a passage with `ref`

ref mirrors load_work, read as chapter.verse:

`ref`	selects
`"3"`	chapter 3
`"3.16"`	chapter 3, verse 16
`"3.16-3.18"`	verses 3:16–3:18
`"3.16-18"`	the same range, shorthand (the high end inherits the low end's chapter)
`"3-5"`	chapters 3 through 5

A malformed reference is rejected with a message that names the accepted shapes, never a raw int() traceback.

Book names and aliases

load_nt (and the CLI) accept any book by its canonical name or by a common abbreviation, case-insensitively, ignoring spaces and dots: "Matthew", "Matt", "mt", "1 Cor", "1cor", and "1Cor." all resolve. The canonical name is what appears in document ids (John 1, Phlm 1).

greek.nt_books() lists all 27 in canonical order (pure metadata, no download):

from aegean import greek
books = greek.nt_books()
len(books)         # 27
books[3]           # {'name': 'John', 'aliases': ['john', 'jn', 'jhn']}

From the shell:

aegean greek nt-books          # a table of the 27 books and the names load_nt accepts
aegean greek nt-books --json   # the same data as JSON

book	accepted names
Matt	matthew, matt, mt
Mark	mark, mk, mrk
Luke	luke, lk, luk
John	john, jn, jhn
Acts	acts, act
Rom	romans, rom, rm
1Cor	1corinthians, 1cor, 1co
2Cor	2corinthians, 2cor, 2co
Gal	galatians, gal, ga
Eph	ephesians, eph
Phil	philippians, phil, php
Col	colossians, col
1Thess	1thessalonians, 1thess, 1th
2Thess	2thessalonians, 2thess, 2th
1Tim	1timothy, 1tim, 1ti
2Tim	2timothy, 2tim, 2ti
Titus	titus, tit
Phlm	philemon, phlm, phm
Heb	hebrews, heb
Jas	james, jas, jms
1Pet	1peter, 1pet, 1pe
2Pet	2peter, 2pet, 2pe
1John	1john, 1jn, 1jhn
2John	2john, 2jn, 2jhn
3John	3john, 3jn, 3jhn
Jude	jude, jud
Rev	revelation, rev, rv, apocalypse

An unrecognised name gives a helpful suggestion:

aegean greek nt Genesis
# aegean: unknown NT book 'Genesis' — did you mean 'Eph'?
# (`aegean greek nt-books` lists all 27)

(Genesis is in the Hebrew Bible, not the Greek NT: load_nt covers the 27 NT books only.)

Per-token annotation fields

Every token in a loaded NT corpus carries these in Token.annotations:

field	what it is
`lemma`	dictionary headword (gold, from Nestle 1904)
`morph`	Robinson-style morphology tag, e.g. `N-NSM`, `V-PAI-3S`
`strongs`	Strong's number (e.g. `3972` = Paul)
`upos`	coarse Universal Dependencies part of speech, reconciled from `morph`
`normalized`	accent/diacritic-normalized form
`ref`	canonical address of the token, e.g. `John.1.1`
`gloss`	brief English gloss (bundled Dodson lexicon, when available)

Turn a chapter into a table where each field is its own column:

import pandas as pd
from aegean import greek

corpus = greek.load_nt("Philemon", ref="1.1")
rows = [{"text": t.text, **t.annotations} for t in corpus.documents[0].tokens]
pd.DataFrame(rows)[["text", "lemma", "morph", "upos", "strongs", "gloss"]].head()
#      text    lemma morph  upos strongs                             gloss
#    Παῦλος   Παῦλος N-NSM  NOUN    3972                              Paul
#   δέσμιος  δέσμιος N-NSM  NOUN    1198             one bound, a prisoner
#   Χριστοῦ  Χριστός N-GSM  NOUN    5547 anointed, the Messiah, the Christ
#     Ἰησοῦ   Ἰησοῦς N-GSM  NOUN    2424                             Jesus
#       καὶ      καί  CONJ CCONJ    2532           and, even, also, namely

The Robinson-to-UPOS mapping that fills upos is exposed if you want it directly:

from aegean.scripts.greek.nt import robinson_to_upos
robinson_to_upos("N-NSM")     # 'NOUN'
robinson_to_upos("V-PAI-3S")  # 'VERB'
robinson_to_upos("T-NSM")     # 'DET'
robinson_to_upos("CONJ")      # 'CCONJ'
robinson_to_upos("N-PRI")     # 'PROPN'  (proper noun)
robinson_to_upos("A-NUI")     # 'NUM'    (indeclinable numeral)

Source, licence, and provenance

The text is the Nestle 1904 edition (Novum Testamentum Graece, ed. Eberhard Nestle). The per-token morphology, lemmas, and Strong's numbers come from the biblicalhumanities/Nestle1904 digital edition and are dedicated to the public domain under CC0; the base Greek text is itself public domain. That CC0 status is what lets pyaegean both fetch the full corpus and bundle an offline sample without any attribution burden on you. The provenance travels with the corpus:

corpus = greek.load_nt("John", ref="1")
corpus.provenance.source
# 'Nestle 1904 Greek NT — morphology/lemmas (biblicalhumanities/Nestle1904)'
corpus.provenance.license
# "CC0-1.0 (morphology, lemmas, Strong's); base Greek text public domain"
corpus.provenance.data_version
# 'nt-corpus-v1@713f28a3b7d4d66132f5aa809fa223fe79762e5d'

The full corpus is a sha256-pinned release asset, so the same request gives you the same text tomorrow. Set PYAEGEAN_NT_CORPUS_URL to point load_nt at a local or alternate copy. See Data and Provenance for how the cache, pins, and licences work across the whole toolkit.

Offline sample vs. the full corpus

pyaegean bundles two fully annotated chapters inside the package as an offline sample: John 1 and Philemon 1. Those two reads work with no network at all:

greek.load_nt("John", ref="1")        # offline
greek.load_nt("Philemon", ref="1")    # offline  (Phlm 1)

Everything else (the rest of John, all the other books, or the whole NT with load_nt()) is fetched to your local cache on first use and then read offline thereafter. When the full asset cannot be fetched and you ask for a book outside the bundled sample, the error says so plainly rather than failing obscurely:

greek.load_nt("Romans")
# ValueError: 'Romans' is not in the bundled offline sample (only John, Phlm is
# available offline); the full 27-book corpus could not be fetched. Retry with
# network access, or set PYAEGEAN_NT_CORPUS_URL to a local copy.

Loaded corpora record which path was taken: the bundled sample notes in its provenance that it is the offline fallback.

Glossing without loading a book

To look up a single Koine word's gloss without loading any chapter, use the bundled Dodson lexicon (CC0, no download): greek.gloss_nt(...) / greek.lookup_nt(...) / greek.gloss_strongs(...), or the CLI aegean greek gloss-nt. This is the same lexicon the corpus self-glosses from, so a token's gloss annotation and a gloss-nt lookup agree. Full details are on Greek NLP.

aegean greek gloss-nt "ἀγάπη"            # love
aegean greek gloss-nt --strongs "3056"   # a word, speech, divine utterance, analogy

Measuring a tagger on the NT

Because the annotations are gold and independent of pyaegean's own predictions, the NT doubles as an out-of-domain benchmark. greek.evaluate_on_nt() (CLI aegean greek eval nt) scores the neural pipeline against the Nestle 1904 gold (lemma and reconciled UPOS), a Nestle-own-gold complement to the PROIEL check. Both are genuinely out-of-domain: the shipped models train on AGDT, Gorman, and Pedalion, never on the NT. The measured numbers and the honesty notes (lemma-convention differences, why finer features are not cross-comparable) are in Benchmarks.

Notes and limitations

Annotations are gold and fixed (Nestle 1904 morphology), not pyaegean's own prediction, which is exactly what makes the corpus useful as a benchmark.
One Document per chapter. A whole-book load is a corpus of chapters; a ref range trims the verses within the affected chapters.
A ref needs a book. You cannot address a verse across all 27 books at once.
Scope is the 27-book Greek NT only. Old Testament / Septuagint books are not covered here.

New Testament

New Testament

Quick start

load_nt(book=None, *, ref=None, force=False)

The CLI: aegean greek nt BOOK [PASSAGE]

Addressing a passage with ref

Book names and aliases

Per-token annotation fields

Source, licence, and provenance

Offline sample vs. the full corpus

Glossing without loading a book

Measuring a tagger on the NT

Notes and limitations

See also

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

pyaegean

Clone this wiki locally

`load_nt(book=None, *, ref=None, force=False)`

The CLI: `aegean greek nt BOOK [PASSAGE]`

Addressing a passage with `ref`