bcv-commons/word-parts

Word Parts: Greek New Testament and Hebrew Bible

Every word of the Greek New Testament cut into its parts: preverb, augment, reduplication, stem and ending. Every word of the Hebrew Bible in MACULA's morphemes: prefixed particles, stem, pronominal suffix. The parts of a word, joined in order, give its printed form exactly. Each part is keyed to the MACULA word id.

What it is for:

• aligning a translation to the source by parts, not only by whole words;
• drafting a word's form from its stem and the grammatical marks of its endings;
• transliterating a name from its stem;
• looking up words that share a stem.

This is route a: built from MACULA alone, so it is CC BY 4.0. Its sister dataset bcv-commons/word-parts-sa (route b) adds MorphGNT's preverb analyses and endings and Morpheus's noun stems, and is therefore share-alike.

Version	v0.1 (schema 1)
Built	2026-11-13 by shoresh/macula/build_word_parts.py --route a and release_word_parts.py (bcv-query df4cdd3)
Greek	137,779 words of the Nestle 1904 New Testament (MACULA Greek), 223,357 parts
Hebrew	305,517 written words of the Westminster Leningrad Codex (MACULA Hebrew), 469,476 parts
Licence	CC BY 4.0

Files

file	one row per	columns
data/src_part_grc.parquet	part of a Greek word	key (MACULA Greek word id, 11 digits: book, chapter, verse, word), n (order in the word, from 1), text, kind, features
data/src_part_hbo.parquet	Hebrew morpheme	key (the written word: the first 11 digits of the morpheme ids), n, token (MACULA Hebrew's 12-digit morpheme id), text, kind, features (class= MACULA's word class)
data/words_grc.parquet	Greek word	key, surface, lemma, morph (Robinson's parse code as MACULA gives it), group (lemma in plain letters / principal part), cut (the parts joined by |), flags
exceptions_grc.tsv	flag, lemma and principal part	the listed exceptions, with token counts and an example
build.json		sources (with hashes or commits), code, counts
SHA256SUMS		the sha256 of every file

Kinds of part:

kind	what	features
preverb	the prefix of a compound verb, as the form writes it (ἀπ-, ἀφ-, συμ-)	preverb= the preposition, in plain letters
augment	the syllabic augment ἐ- (ἠ- in ἤθελον)	augment=syllabic
reduplication	consonant + ε, ε, or ει	reduplication=consonant / ε / ει
stem	the stem	pp= the principal part (below); lengthened=temporal when a temporal augment or a vowel reduplication lengthened its first vowel. That vowel cannot be cut off.
ending	the ending	Robinson's parse code
particle	a Greek conjunction, preposition, particle or interjection, or a Hebrew prefixed particle
suffix	a Hebrew pronominal suffix
mark	punctuation inside a printed form, such as a dash

Principal parts (pp):

pp	what
P	present and imperfect
F	future
FP	future passive
A	aorist
AP	aorist passive
R	perfect and pluperfect active
RM	perfect and pluperfect middle and passive

How the Greek cut is made

• Stem and ending. The stem is the beginning that a lexeme's forms share within one principal part (for a noun or adjective, across all its forms); the ending is what varies.
  • The endings a parse allows are learned from the lexemes with several forms in the New Testament: an ending counts when at least three lexemes show it.
  • A group's stem is a beginning shared by at least 0.6 of its forms, chosen so that the most forms have an allowed ending after it. On a tie the commoner endings win (μισθ|ός, not μισθό|ς), then the longer stem.
• Suppletive lexemes take several stems: ἐγώ with ἐμ- and ἡμ-; πολύς with πολ- and πλει-; the root aorist of καταβαίνω with βη- and βα-. A form that shares none stays whole (μου, ἦν).
• Preverbs are found from the lemma: the remainder is itself a New Testament verb, or an augmented form shows the augment after the preverb (ἀπ-έ-κτειναν).
  • Each form's own variant is then cut, following elision and assimilation (ἀπ- before a vowel, συμ- before a labial, ἐγ- before a velar).
  • Route b takes MorphGNT's preverb analyses instead.
• Augment and reduplication are read against the principal part's own base, not only the lemma:
  • εἶπον is read against εἰπεῖν and εἰπών; ἔφαγον against φαγεῖν;
  • a μι-verb's present reduplication is set aside (τί-θημι, so ἔ-θηκεν).
• Formants stay in the stem. The tense formants σ, θη and κ stay in the stem of their principal part: πιστεύσ|ας, σωθήσ|ομαι.
• Third-declension stems. The stem is what all forms share: αἰῶν|ος, σαρκ|ός, πνευμα|τος. A nominative that does not share it is cut on its own (νὺ|ξ) or stays whole, and is flagged.
• Kept whole (one stem part): the article; the relative pronouns ὅς, ὅστις, ὅσπερ, which are declined like the article; indeclinable words (proper names, numerals, letters); crasis forms (κἀγώ); elided forms; adverbs.

The Hebrew file only reads MACULA's morphemes. The verb's own endings (the afformatives of קָטַלְתִּי) are not cut.

Checks

Greek words: 137,779. Rebuild failures (the parts joined against the surface): 0.

Augment against MACULA's tense (indicative aorist, imperfect, pluperfect; suppletive εἰμί and function words left out)

• syllabic augment cut: 4,418
• temporal (lengthened vowel, not cut): 2,037
• initial vowel unchanged (ει-, ευ-, ου-, ι-, υ-): 999
• none found: 195

Reduplication (perfect, pluperfect)

• consonant + ε cut: 744
• οἶδα (a perfect in form, no reduplication): 319
• ε cut: 218
• temporal (lengthened vowel, not cut): 150
• none found: 145
• initial vowel unchanged (ει-, ευ-, ου-, ι-, υ-): 65
• ει cut: 19

Stem consistency per lexeme and principal part

• groups (lexeme + principal part; nominals: lexeme): 6,568; every form with the group's stem: 6,408 (0.976)
• suppletive groups (no beginning shared by 0.6 of the forms; several stems taken): 71
• tokens, form has its group's stem: 66,613
• tokens, suppletive group: form has one of the group's stems: 7,515
• tokens, suppletive group: form alone (whole): 3,135
• tokens, form's stem differs: 1,716

Flags

• article: 19,786
• ending_unlisted: 9,143
• suppletive_stem: 7,515
• suppletive_alone: 3,135
• relative: 1,548
• indeclinable: 1,267
• stem_differs: 1,105
• no_cut: 611
• crasis: 117
• preverb_not_in_form: 42
• elided: 21

Read by hand: a sample of 200 words, seed 17, route a. Two cuts are wrong:

• ἀν|έ|β|η should be ἀν|έ|βη;
• συν|ε|τέ|θειντο, a minor case.

About twelve follow a convention that could be argued, among them πά|ντα, πατ|έρες, ζητημά|των, θεραπεῦσ|αι and οὐδε|ὶς. The rest are sound.

Flags in words_grc.tsv:

flag	what
stem_differs	the form does not carry its group's stem; it was cut by its ending alone
ending_unlisted	the ending was not among the learned ones for its parse
suppletive_stem	cut on one of a suppletive lexeme's several stems
suppletive_alone	a suppletive lexeme's form with no shared stem; kept whole
no_cut	the form shares no stem and has no allowed ending; kept whole
article, relative, indeclinable, crasis, elided	kept whole, for the reason named
preverb_not_in_form	the lemma's preverb is not at the head of the form (ἤνοιξεν of ἀνοίγω)

Known limits

• New Testament Greek only. The Septuagint is not cut.
• Not cut, and listed in exceptions_grc.tsv:
  • Attic reduplication (ἑώρακα, ἐλήλυθα, ἀπολώλεκα);
  • the double augment of ἀνοίγω;
  • some forms of ἀφίημι;
  • the pluperfect of οἶδα.
• Formants are not separated. A stem is per principal part: λυσ- for the aorist, λυ- for the present.
• No person has reviewed all rows. Corrections are welcome as issues on bcv-commons/bcv-query.

Sources and licence

source	licence	used for
MACULA Greek (Nestle 1904 lowfat)	CC BY 4.0	the Greek words, their lemmas and parsing
MACULA Hebrew (WLC lowfat)	CC BY 4.0	the Hebrew morphemes

This dataset is licensed CC BY 4.0. Please credit "Word Parts, bcv-commons (bcv-commons/word-parts v0.1)" together with the sources above.

Versions

• v0.1 (2026-11-13): first release.
• schema in build.json is the version of the file shape. It is raised only when a column is renamed, dropped or changes meaning.