bcv-commons/greek-nt-clause-layout

Greek New Testament Clause Layout

The Greek New Testament (Nestle 1904) one clause segment per line, indented by level: subordinate clauses step in under the clause they belong to, and speeches step in under the line that introduces them. It is built from the syntax trees of MACULA Greek and the quotation data of Clear Bible, by rules written down below. It is meant for translators and readers who want to see the structure of a sentence at a glance, and as a base for comparing the clause structure of translations.

Release	v0.5, built 2026-10-27 from bcv-query 00bd28e
Text	Nestle 1904, as MACULA Greek has it (137,779 words, 27 books)
Size	29,682 lines (clause segments) of 27,964 clauses; 11,697 lines inside a speech
Characters	389 people (speaker_id) over 1,721 speeches; 63 of them with a proper name (speaker_name, on 1,132 speeches)
Changes	v0.5: who the speaker is across speeches, as keys: speaker_id (one person, one id) and speaker_name / speaker_name_ref (that person's proper name in the text: Κλεοπᾶς where the inquit says εἷς, Luke 24:18); speaker unchanged. v0.4: a speech whose inquit names no one takes its speaker's Greek name from the same speaker's other speeches (speaker_basis); alt_eng keeps the speaker data's English label only where no Greek name is found. v0.3: speakers named in Greek from the text itself, no English labels. v0.2: the cross-check below, and its one fix (no line ends in a conjunction or question particle)
Companion	Hebrew Bible Clause Layout, the same idea for the Hebrew Bible

Matthew 5:1–4 in layout/MAT.txt:

5:1	    Ἰδὼν δὲ τοὺς ὄχλους
5:1	ἀνέβη εἰς τὸ ὄρος·
5:1	καὶ καθίσαντος αὐτοῦ
5:1	προσῆλθαν αὐτῷ οἱ μαθηταὶ αὐτοῦ·
5:2	    καὶ ἀνοίξας τὸ στόμα αὐτοῦ
5:2	ἐδίδασκεν αὐτοὺς λέγων
5:3	    Μακάριοι οἱ πτωχοὶ τῷ πνεύματι,
5:3	        ὅτι αὐτῶν ἐστιν ἡ βασιλεία τῶν οὐρανῶν.
5:4	    μακάριοι οἱ πενθοῦντες,
5:4	        ὅτι αὐτοὶ παρακληθήσονται.

Files

File
data/clauses.parquet	one row per line, in text order
layout/<BOOK>.txt	the indented text, one line per row: chapter:verse, a tab, four spaces per level, the text
build.json	the build record: counts, the SHA-256 of the MACULA trees, the MACULA Greek files read for the speakers, and the speaker data
crosscheck/crosscheck.json	the cross-check against a Hebrew translation (see Cross-check), overall and per book
crosscheck/crosscheck-disagreements.tsv	every pair of words where the two readings disagree, with the Hebrew segments and Greek lines side by side
SHA256SUMS	SHA-256 of every file

The rows

Column
book, chapter, verse	USFM book code; Nestle 1904's chapter and verse
seg	the line's number in the chapter (1-based)
text	the line's Greek text, with MACULA's punctuation
text_depth	the line's level: what the layout indents by (0 = the margin)
depth	the clause's depth in MACULA's sentence tree, wrappers and coordinations not counted (see How a line is made)
clause	the clause's id: the MACULA word id of its first word. A clause interrupted by another, or running over a verse break, has several lines
parent	the id of the clause it belongs to in the tree (empty at the top of a sentence)
kind	VC (with a verb of its own) or NC
rule	MACULA's clause rule, the order of its constituents (V-O, S-V-ADV, …)
role	the clause's role in its parent (adv, o, s, …), from the clause or the wrapper around it
via	the rules of the wrapper and coordination nodes between the clause and its parent (Conj-CL, sub-CL, CLaCL, …)
junction	MACULA's junction of the clause, where given
sentence	the MACULA sentence: the word id of its first word. A sentence can run over several verses
quotation	the innermost speech the line stands in: the word id where it starts in the speaker data; empty in narrative
speaker	who speaks it, named from the Greek text: the subject of the verb that introduces the speech (Ἰησοῦς, Πέτρος, Στέφανος, μαθητής, ἡγεμών, φωνή), as a lemma; several subjects separated by commas. Empty where the text names no one (a passive γέγραπται, an unnamed "they", a participle)
speaker_ref	the word ids of the words that name the speaker; a pronoun or an article used as one (ὁ δὲ ἔφη) is followed to the word it refers to
speaker_basis	where speaker comes from: inquit (the speech's own inquit), same speaker in book / same speaker in NT (the same speaker's usual name from their other speeches, see below; speaker_ref then points there); empty where there is no name
alt_eng	the speaker data's English label, only where speaker is empty (111 speeches: "scripture", "crowd", "passers by"), so that no name the speaker data gives is lost; empty everywhere else
speaker_id	who speaks it across speeches: a stable id (n0009) for one person or group, the same in every speech they make and in later releases (speaker names each speech by its own inquit, so one person can be Πέτρος in one and Σίμων or μαθητής in the next). A person with a proper name has one id across the New Testament; a group or person without one ("crowd", "someone") one per book. Empty for a speaker the speaker data names no one by
speaker_name	that person's proper name as the text gives it (MACULA lemma; θεός for God): Ἰησοῦς, Πέτρος, Κλεοπᾶς; empty where the text of that book gives none (the blind men of Matt 20:30)
speaker_name_ref	the word id of a word that gives that name: the speech's own inquit where it does, else the nearest speech's that does (in the same book first), else a word a hand table gives (Luke 24:18 ὀνόματι Κλεοπᾶς)
inline	clauses written on this line though they are clauses of their own (participles, see the rules)
split_at_quotation	the line was broken off its clause where a speech starts or ends (see the rules)
first, last, words	MACULA word ids (BBCCCVVVWWW, as n… ids in MACULA's files)

How a line is made

A line is a run of words, inside one verse, that belongs to one clause of MACULA's tree. A clause interrupted by an embedded clause gives two lines; a clause that runs over a verse break gives one line per verse.

Wrappers and coordinations. MACULA Greek wraps many clauses in a clause node that holds only the word introducing them (a conjunction: Conj-CL; a subordinator: sub-CL; ὅτι: that-VP; the article of an articular infinitive: DetCL; a particle: PtclCL; an adverb such as ὁμοίως: AdvpCL). Such a wrapper is not a level: its words go with the clause they introduce (a postpositive δέ, γάρ, οὖν stays after that clause's first word). A coordination of clauses (CLaCL, Conj3CL, …) is not a level either: its clauses are siblings; a short member that MACULA puts straight into the coordination (ἀλλὰ ὁ ἀνήρ) is a line of its own at their level. A clause of a vocative alone (Κύριε,) keeps its line and adds no level.

Rules. Each is anchored on a kind of expression, never on a place:

• A bare articular participle (the article and the participle, μή at most: οἱ πενθοῦντες) is a noun phrase in the line, not a line; so is a participle modifying a noun when the noun's line is nothing but that noun phrase (φωνὴ βοῶντος ἐν τῇ ἐρήμῳ). A participle with words of its own after a longer line keeps its line (τοῖς ἁγίοις / τοῖς οὖσιν ἐν Ἐφέσῳ). (610 participles.)
• A preposition, conjunction, negation or ascensive καί that ends a line goes with the clause below it that follows, which it governs or introduces (εἰς τὸ φαγεῖν). Such a word, and the question particles οὐχί and μήτι, never ends a line inside a verse (ἵνα ἀγαπῶμεν ἀλλήλους; Κύριε, / οὐχὶ καλὸν σπέρμα…).
• No line is function words alone (ἀλλὰ, οὐ γὰρ, καὶ ἰδοὺ, a relative pronoun): such words go with the line they introduce. A short question (τί γάρ;) stays a line.
• A line breaks where a speech starts or ends inside it (ὁ δὲ ἔφη / Ἄνδρες ἀδελφοί…, Acts 7:2), at the start of the phrase the break falls in (153 lines).

Levels (text_depth). Inside a sentence the levels are the tree's. Every speech of the speaker data is a level too: its lines stand at least one level in from the line that introduces it, and none goes further out while it lasts, also over the many sentences of a long speech (Matthew 5:3–7:27 stay in from 5:2 "λέγων"). A speech inside a speech is one further in; a speech that follows another directly (no introducing line) stands where that one stands. Two speeches of one speaker with only a few verbless words between them are one (45 joins; the speaker data leaves Luke 1:74 out of Zechariah's song). 1,610 of the 1,721 speeches have a speaker named from the text (1,539 by their own inquit). A speech whose inquit names no one (a quotation of scripture, a speech that goes on after a narrator's aside) takes the name of the same speaker from their other speeches: the speaker data's label decides only who it is, the name comes from the Greek there. The name is the speaker's usual one in the book (most of that speaker's named speeches there, at least two), else in the New Testament, and must be a proper name or θεός, and mostly this speaker's. The speaker data's own labels for who speaks are in English; they decide only who a speaker is and are not in the data (except alt_eng). Characters: each label gets a stable id, kept in a table in the repository (resources/speaker_quotations/nt_characters.tsv, not released: its labels are English); two labels of one person share an id (Jesus, Jesus (child)), and God, Moses, David and Solomon, who speak in quotations of scripture across many books, have one id each. The name (speaker_name) is the proper name or θεός that more than half of the person's speeches named by a proper name, and at least a third of those named at all, are named by in their own inquit, and that no other person is named by more than three times as often; where no inquit names a person the text names elsewhere in the book, the table gives the word that does (22 people: Κλεοπᾶς, Γαβριήλ, Στέφανος, θεός for God, whose inquits are mostly φωνή or γραφή). Narrators get no name. Last, a line steps in one level at a time: where the tree has a level with no line of its own, deeper lines move out to one in from the line above, keeping the order of the levels.

Lines per level: 0: 9,275, 1: 11,062, 2: 6,090, 3: 2,380, 4: 655, 5: 160, 6: 42, 7: 17, 8: 1.

What it does not do (yet)

• No text-level hierarchy between sentences beyond the speeches: a paragraph's sentences all start at the margin of their speech.
• Where MACULA's tree or the speaker data differs from a reading you'd expect, the layout follows them (the speaker data has no quotations inside a parable's speakers, so in Matt 13:27 the servants' second question stands at the parable's level): Rev 4:8 "ὁ ἦν" stands one level deeper than "καὶ ὁ ὢν καὶ ὁ ἐρχόμενος"; John 4:9 "οὐ γὰρ συνχρῶνται Ἰουδαῖοι Σαμαρείταις" is inside the woman's words; Acts 7:7 "ὁ Θεὸς εἶπεν" is inside Stephen's quotation of God. 9 lines have no clause in MACULA (titles such as Matt 1:1).
• Not checked by hand beyond reading passages from each kind of book; the cross-check below compares it with an independent reading.

Cross-check

The layout was compared with an independent reading of the same text: the clause segments of Salkinson–Ginsburg's Hebrew New Testament as given by a tagger trained on the Old Testament alone (bcv-commons/hebrew-nt-analysis v0.1), joined to the Greek words through the alignment bcv-commons/alignments-by-id@v1.0 (links of class kept, 25 of 25 right in its hand check). Neither reading uses the alignment, nor the other. The unit is a pair of neighbouring Hebrew words whose Greek words are known: are they in one clause segment in Hebrew, and on one line here? 58,189 pairs.

Pairs where both readings agree	87.0%
Hebrew segment boundaries that are also line breaks here	72.0%
Line breaks here that are also Hebrew boundaries	76.1%
Boundary F1	74.0%

This is agreement between two readings, not accuracy. Most of the 7,576 disagreeing pairs are the translation: Salkinson often makes a clause where the Greek has a phrase (Rev 9:11 adds "הוּא מַלְאַךְ"; Eph 2:8 renders διὰ πίστεως "עֵקֶב אֱמוּנַתְכֶם") or renders a participle as a finite clause, so the letters with the densest participle and noun style agree least. Reading the disagreements found one fault of the layout, fixed in v0.2: a conjunction or question particle ending a line (Matt 13:27 "Κύριε, οὐχὶ / καλὸν σπέρμα"). The full list is in crosscheck/.

Book	Pairs	Agree	Boundary F1
Matthew	8,489	88.3%	78.2%
Mark	4,592	86.5%	76.5%
Luke	8,327	86.7%	75.9%
John	7,028	88.9%	80.8%
Acts	7,277	86.2%	72.4%
Romans	2,905	86.8%	69.5%
1 Corinthians	2,749	86.5%	69.4%
2 Corinthians	1,718	84.0%	64.1%
Galatians	891	84.0%	64.3%
Ephesians	980	90.9%	70.2%
Philippians	616	83.1%	57.0%
Colossians	619	83.5%	53.2%
1 Thessalonians	568	84.7%	59.5%
2 Thessalonians	344	87.2%	65.1%
1 Timothy	639	83.1%	52.6%
2 Timothy	445	87.0%	67.8%
Titus	244	87.7%	63.4%
Philemon	141	87.2%	55.0%
Hebrews	1,999	85.4%	64.9%
James	744	86.4%	75.7%
1 Peter	676	86.5%	64.3%
2 Peter	386	78.5%	45.8%
1 John	989	87.7%	76.4%
2 John	92	84.8%	66.7%
3 John	96	91.7%	85.2%
Jude	187	80.8%	52.6%
Revelation	4,448	88.6%	74.7%

Sources and licence

Part	Source	Licence
Greek text, clauses, sentence trees	MACULA Greek (Clear Bible / Biblica), Nestle1904/lowfat, through trees-macula.db (SHA-256 efe5b5a1c5e8…)	CC BY 4.0
Where speeches run	MACULA Quotation and Speaker Data (Clear Bible), as resources/speaker_quotations in bcv-query (SHA-256 7231cb5e4fc2…); used for its word spans, and to tell whether two speeches next to each other have one speaker (to join them over a gap); its names are not used	CC BY 4.0
Who speaks	MACULA Greek's subject and referent links (Nestle1904/lowfat, SHA-256 a2c5be4e6367…)	CC BY 4.0
The layout	bcv-commons	CC BY 4.0
The code (macula/build_nt_clause_layout.py, bcv-query)	bcv-commons	MIT

Attribution when you use this dataset:

Greek New Testament Clause Layout (v0.5), bcv-commons, CC BY 4.0. Text and syntax: MACULA Greek, Clear Bible / Biblica (CC BY 4.0). Speeches: MACULA Quotation and Speaker Data, Clear Bible (CC BY 4.0).

Feedback

Issues and requests: bcv-commons/bcv-query. Please name the chapter and verse, the line, and what you expected.