bcv-commons/passage-facts

Passage Facts

Who is heard in a passage, what kind of speech it is, and how the text is set (song, poetic lines, prose), for the whole Bible. Each fact is tied to exact words of the Hebrew or Greek text (MACULA word ids) and names its source and licence, so that it can be carried through word alignments into any translation.

No fact is a model's reading of the text. A fact is a source's own value (b.source), read from the Hebrew or Greek grammar by a fixed rule (b.text), inferred by a published rule from other facts (b.rule), or set by a review row (b.review).

Built	2026-11-12 by macula/build_passage_facts.py (bcv-query 97c1728)
Size	66 books, 62,663 facts (19,355 of them CC BY-SA 4.0); CC BY subset: 43,308 facts
Schema	1 (see Versions)

Files

Path	What it holds
data/<BOOK>.jsonl	all facts of a book, one JSON object per line. Licence of the whole: CC BY-SA 4.0
data-cc-by/<BOOK>.jsonl	only the facts of sources that are not share-alike. Licence: CC BY 4.0
codes.tsv	the decoding table: every code in the data, with its English meaning, its source, and notes
sources.tsv	the sources: licence, share-alike or not, enabled or not, attribution wording, what each gives
build.json	sources used, input hashes and pins, counts per book, and checks: how well each source landed on the words
SHA256SUMS	the sha256 of every file

A fact

{"id": "EXO.00001", "kind": "quote", "value": "qt.normal", "first": "020010090041", "last": "020010100192", "ref": "EXO 1:9-10",
 "basis": "b.source", "precision": "p.word", "src": ["clear", "layer"], "lic": "CC BY 4.0",
 "level": 1, "quote": "q00503", "found": "qf.block", "speaker": "מֶלֶךְ", "speaker_ref": ["020010080021"], "speaker_by": "sb.inquit",
 "speaker_id": "c0073", "speaker_name": "פַּרְעֹה", "speaker_name_ref": ["020010220021"]}

Field	Meaning
kind	what the fact is about: voice, quote, frame, domain, delivery, force, addressee, section, form, note (and, once enabled, TBTA's listener, attitude, age_relation, tb_force, rhetorical)
value	a code from codes.tsv. For frame it is the verb's lemma in Hebrew or Greek
first, last	the anchor: the first and last word of the stretch, as MACULA word keys (Hebrew 12 digits BBCCCVVVWWWP, a morpheme; Greek 11 digits BBCCCVVVWWW)
pieces	where words of another voice interrupt the stretch: its parts as [first, last] pairs
lines	for a printed layout: each printed line as [first, last]
at	the words a grammar rule read (the imperative, the vocative, the second-person forms)
after	for section: the word after which the break stands
ref	the verses, for reading; the anchor is first..last
basis, precision	how the fact was found (b.*) and how exact its anchor is: p.word exact words, p.verse the source speaks of whole verses
src, lic	the sources (keys of sources.tsv) and the licence of the fact (CC BY-SA 4.0 if any of its sources is share-alike)
of	the fact (a quote) this one describes
level, parent	for quote: 1 = spoken in the narrative or text, 2 = a speech inside a speech, …; parent the enclosing quote
speaker, speaker_ref	for quote: who speaks, as the Hebrew or Greek lemmas of the words that name them, and those words' keys
speaker_by	how the speaker was named (sb.*)
speaker_id, speaker_name, speaker_name_ref, members	one person across speeches; that person's proper name in the text; for a group, its members' ids
divine	the source marks the speaker as God (Jesus, the Spirit)
fit	for a note: partial where the note quotes a text (UGNT, UHB) that differs a little from ours and at least 80% of its words were found in order; otherwise the match is exact
tn_id	for a note: the note's ID in unfoldingWord's Translation Notes, to look the note up there
markers	for the ULT's form: its paragraph marker per line (q1, q2: indent levels)
start, end	for a verse-precision form: whether the first and last verse are set that way in full (cv.all) or in part (cv.part)

The data holds no English. To show a fact in English, look its codes up in codes.tsv. Speakers are named by the Hebrew or Greek words of the text; a frame fact gives the verb's lemma.

What the facts are

Kind	Rule	Sources
voice	vc.narrator: the words outside every speech (Hebrew books and Gospels, Acts). vc.frame: a text that is one speaker's from its start (a psalm's psalmist), with speaker	speaker layer; Clear
quote	each speech with its speaker, nesting, quote type (qt.*) and how it was found (qf.*)	Clear (CC BY), speaker layer / Greek layout speaker columns
frame	the verb that introduces a speech: the nearest verb in the 15 words before it, not inside another speech of the same level, with no Greek sentence end between; a Hebrew לֵאמֹר gives way to the verb before it in its clause. About 2% of speeches have no introducing verb, and the nearest verb is then not one of speaking	MACULA, clause layout
domain	the meaning domain of that verb in its place (Speak, Sing, Shout; Louw-Nida Speak/Talk, Inform/Announce, Question/Answer)	UBS dictionaries (CC BY-SA)
delivery	the speaker data's note on how a speech is delivered (praying, rebuking, singing)	Glyssen, carried in Clear's data
force	per clause segment, by the grammar alone, inside speech and outside it (a psalm's plea, a letter's command; of names the quote where there is one): imperative; Hebrew jussive and cohortative; prohibition (Hebrew אַל with a jussive, cohortative or yiqtol; Greek μή with an imperative, or with a second-person aorist subjunctive in a main clause); Greek exhortation (a first-person plural subjunctive in a main clause); question (a Hebrew interrogative word; in Greek every clause from the stop before to the question mark); Greek vocative. A Greek main clause has no subordinator (ἵνα, ἐάν, ὅταν, ὡς, ὅς, τίς ...) in its tree node before the verb or in any node above it; μή ποτε / μή πως / μή που ("lest") and οὐ μή give no force. ἰδού, ἴδε and ἄγε are imperative in form but particles in use; they count neither here nor for addressee. Not given, as they need a reading: οὐ with a future as a command, לֹא with a yiqtol. The prohibition and exhortation rules and the whole-question extent were proposed by make-tanakh-research (2026-11-12)	MACULA, clause layout
addressee	ad.sg, ad.pl, ad.both: the number of the speech's own second-person forms. ad.named: the one spoken to in the inquit (Greek: the verb's A2 argument; Hebrew: the object of אֶל in the inquit's clause)	MACULA
section	petuchah and setumah, word-exact (a mark inside a verse stands after its word): the Leningrad Codex, and MAM's Aleppo Codex tradition, each as its own fact	OSHB; MAM (CC BY-SA)
form	fm.ms.song: the Masoretic song layout, line by line (Exod 15, Deut 32, Judg 5, 2 Sam 22, some lists). fm.ed.*: Nestle 1904 prints the words as poetic lines, a hymn block, or lines quoted from the Scriptures (from the page images; printed lines from the digital edition). fm.tr.poetry: a translation sets the words in indented lines: the ULT line by line through its word alignment, the BSB by verse	MAM (CC BY-SA); Nestle 1904; ULT (CC BY-SA); BSB
note	an unfoldingWord Translation Note of one of 27 articles on speech, address and poetry (rhetorical question, quotation in a quotation, apostrophe, parallelism, singular/plural 'you', inclusive/exclusive 'we', ...), on the Hebrew or Greek words it quotes. Only the article and the place are carried, not the note's English	unfoldingWord Translation Notes (CC BY-SA)

Sources and licence

Source	Licence	Share-alike	What it gives
MACULA Quotation and Speaker Data	CC BY 4.0	no	who speaks which words; quote type
Glyssen character data (SIL / Faith Comes By Hearing)	MIT	no	delivery notes, carried by Clear's data
Hebrew Bible Speaker Layer / Greek NT clause layout speaker columns (bcv-commons)	CC BY 4.0	no	nesting, speakers named from the Hebrew and Greek text, review rows
Hebrew Bible Clause Layout and Greek NT Clause Layout, segmentation (bcv-commons)	CC BY 4.0	no	clause segments; clause relations
MACULA Hebrew and MACULA Greek (Clear Bible / Biblica)	CC BY 4.0	no	words, lemmas, grammar, verb frames
Open Scriptures Hebrew Bible	CC BY 4.0	no	Leningrad Codex paragraph marks
Nestle, Η ΚΑΙΝΗ ΔΙΑΘΗΚΗ, British and Foreign Bible Society 1904: the printed layout	Public Domain	no	passages printed in poetic lines or as a narrower block; its preface says the arrangement chiefly follows Westcott-Hort 1881
Berean Standard Bible: poetry layout	Public Domain	no	a translation's layout: verses it sets as poetry
unfoldingWord Literal Text (ULT), release v91	CC BY-SA 4.0	yes	a translation's layout: the lines it sets as poetry, placed on the Hebrew and Greek words by its own word alignment
unfoldingWord Translation Notes (UTN), release v91	CC BY-SA 4.0	yes	the translation issue a note names (its support article), on the Hebrew or Greek words it quotes; the note's English text is not carried
Miqra according to the Masorah (MAM), MAM-parsed plus, MAM-basics commit cfdbb64e	CC BY-SA 4.0	yes	the Masoretic layout: the songs (Exod 15, Deut 32, Judg 5, 2 Sam 22, ...) and the poetic lines of Psalms, Proverbs and Job with their indent levels; its own petuchah and setumah (Aleppo Codex tradition)
UBS Dictionary of Biblical Hebrew (SDBH) and UBS Dictionary of the Greek New Testament (Louw-Nida), ubs-open-license commit 33dcc8c6	CC BY-SA 4.0	yes	the meaning domain of the verb that introduces a speech (speak, sing, pray, command, ...), as the dictionary gives it for that word in its place

Why two folders. Facts from share-alike sources (unfoldingWord, MAM, the UBS dictionaries) can only be shared under CC BY-SA 4.0, so data/, which holds all facts, is CC BY-SA 4.0 as a whole. Every other fact stays under its own licence (CC BY 4.0, MIT, public domain), and data-cc-by/ holds exactly those, under CC BY 4.0. Each fact also says its own licence in lic.

Attribution. Name this dataset ("Passage Facts, bcv-commons") and the sources of the facts you use, with the wording in the attribution column of sources.tsv. For Glyssen's delivery notes keep its MIT copyright notice.

Not included: The Bible Translator's Assistant (All the Word) and OpenText.org. Both are in sources.tsv but disabled, because neither states an open licence.

How well the sources landed

Check	Whole Bible
Speeches with the verb that introduces them	6,574 of 6,936
Speeches the speaker layer joins across a speaking formula, given as one span (no pieces)	88
Translation Notes placed on the words they quote (of them partly: the quoted text differs a little) / not found	8,762 (227) / 175
ULT alignment words placed / not found	366,620 / 3,015
Paragraph marks: Leningrad / MAM / at the same word / the same mark there	3,133 / 3,353 / 2,844 / 2,344
Nestle 1904 passages with their printed lines	105 of 122
Books not in unfoldingWord's release (no ULT lines or notes)	1CH, 2CH, ECC, EZK, ISA, JER, NUM

Per book in build.json (checks).

Versions

A new build gets a new revision; a revision is never changed. Pin the revision. schema in build.json is the version of the fact shape. It goes up only when a field is renamed, dropped or changes meaning; a new field, kind or code leaves it unchanged. Codes are never renumbered. Anchors (first, last, pieces, lines, at, after) are a first version and may follow a link-type specification later; that would be a schema change, listed here.

Feedback

Issues and requests: bcv-commons/bcv-query.