NuBerea/intertextuality

Cross-corpus intertextual relationship classifications for biblical and adjacent ancient literature: which passage cites, alludes to, or echoes which. The dataset covers relationships within the Bible (Old Testament inner-biblical reuse, New Testament use of the Old Testament, intra-New-Testament cross-references), between the Bible and Second Temple Jewish literature (Pseudepigrapha, Philo, Josephus), between those corpora and rabbinic literature (Mishnah, Tosefta, Bavli, Yerushalmi), and across Christian reception history (Church Fathers citing Scripture and each other; Reformation-era authors citing Scripture, the Fathers, and one another). It also includes alignment edges between English Bible verses and paragraph locators in the Christian Classics Ethereal Library, and a per-verse original-language syntax store supporting Greek/Hebrew comparison of linked passages.

For general within-Bible discovery, bible_intertextual_relations is the provenance-first front door over 9,211 independently supported assertions: Nestle-attested NT→OT rows, MiqraBERT-native OT→OT proposals, scholar-attested literary analogies, curated doublets, and curated gospel-synopsis parallels. It supports exact verse, chapter, or book membership on either endpoint, including verses contained inside ranges. OpenBible/TSK candidate spines, community-vote fields, and vote-derived ranking are excluded entirely.

Each table records source-to-target edges between passages with relationship classifications (quotation / allusion / echo) and supporting signals, serving scholarship in inner-biblical exegesis, Second Temple Judaism, patristics, and reception history. The full specification — per-config documentation, schemas, methodology, and provenance — is available in the gated SPEC accompanying this dataset.

Scripture-citation union (scripture_citations_all)

scripture_citations_all adds 349,888 additive rows in seven arms, one arm per published citation member (edge_kind). The member configs remain published and unchanged. The row count is the sum of the member rows; it is not a count of unique citations, because sch-gold-crossref repeats the 55,869 sch-scripture edges and both arms are kept. edge_id is unique within each arm except the two Scholastic arms, where 128 ids repeat exactly as published in the members.

osisRef keeps each arm's native reference (Hebrew numbering on talmud-ot and targum-ot) and osisRef_std the English standard reference. Where the member left a reference empty it stays empty: 916 talmud-ot rows and 137 targum-ot rows have NULL in both, because their candidate set was ambiguous and was never guessed. match_grain is the union's grain envelope (verse or segment); the native grain of every source row is kept in member_match_grain (verse, chapter or verse_citation), and 127,416 rows are chapter-level citations.

Each arm keeps the licence class of its source component in the per-row k column: computed on the Reformation, Talmud and Targum arms and public-domain on the Scholastic arms. The table ships citation identifiers, labels, scores and bibliographic locators only; no Talmud, Targum or other source text is shipped. README metadata is not a grant for any component.

Arm (edge_kind)	Source config	Rows	k	Source revision	Content
ref-ot	ref_to_ot	80,517	computed	5f8a3ce036a1	Reformers and Bellarmine citing Old Testament passages
ref-deut	ref_to_deuterocanon	2,103	computed	5f8a3ce036a1	Reformers and Bellarmine citing Deuterocanonical passages
ref-nt	ref_to_nt	132,513	computed	5f8a3ce036a1	Reformers and Bellarmine citing New Testament passages
talmud-ot	talm_ot_quotation_edge	22,854	computed	5f8a3ce036a1	rabbinic quotations of Old Testament verses (identifiers, labels and scores only)
targum-ot	targ_ot_quotation_edge	163	computed	5f8a3ce036a1	targum quotations outside the base text (identifiers, labels and scores only)
sch-scripture	scripture_citations	55,869	public-domain	43cbdd47b0de	Scholastic scripture citations
sch-gold-crossref	schol_apparatus_crossrefs	55,869	public-domain	43cbdd47b0de	gold-crossref overlay on the same Scholastic edges

License and attribution

Estate-computed edge and classification tables are released under CC BY 4.0. Upstream sources requiring attribution:

Source	Link	License / terms
OpenBible.info cross-references	https://www.openbible.info/labs/cross-references/	Creative Commons Attribution
Nestle 1904 Greek New Testament TEI (quotation references)	https://github.com/biblicalhumanities/Nestle1904	Public domain (CC BY 4.0 markup)
MACULA Greek (SBLGNT) and MACULA Hebrew, Clear Bible	https://github.com/Clear-Bible	CC BY 4.0
STEPBible data (TVTMS versification), Tyndale House	https://github.com/STEPBible/STEPBible-Data	CC BY 4.0
Christian Classics Ethereal Library (CCEL)	https://www.ccel.org	CC BY-SA — the ccel_* configs are facts-only alignment edges computed from the CCEL corpus; no CCEL prose is shipped or reconstructible. Consumers wanting the text itself join by key to NuBerea/patristics-ccel, which carries the BY-SA card.
Sefaria (rabbinic reference structure; facts only, no source text redistributed)	https://www.sefaria.org	Structure references

No copyrighted source prose is redistributed in any config: edges ship passage identifiers, labels, and numeric scores only.

Copyright boundaries — example_locus run (2026-08-29/30)

The example_locus build (857,341 edges from 8 grammar/lexicon sources; a Du Cange lane in progress) recorded these copyright-driven limits, at the owner's direction:

Source side — quarantined lexica not yet run as example sources. HALOT (© Brill), BDAG (© Univ. of Chicago), Lampe (© OUP), and TLL (CC BY-NC-ND, © BAdW) are held estate-side as copyright-isolated provinces and contributed no edges to this run. Project ruling (2026-08-30): their citation POINTERS — "HALOT s.v. X cites Gen 1:2" — are uncopyrightable facts, the same doctrine already applied by the live lexicon sidecars (greek_lampe_sidecar, latin_apparatus_crosswalk_sidecar), so pointer-edge lanes over all four are PERMITTED under the standing firewall (no quoted definition or example prose, ever). What blocks each today is readiness, not license: BDAG already has a structured verse→sense map (nearest-term lane); Lampe has OCR; TLL has only coverage parquets (citation lists still need extraction from the fascicle OCR); HALOT is a source PDF with no machine-readable table at all.

Target side — extraction fine, target corpus not linkable. Jastrow's 97,849 edges are overwhelmingly citation_only because the Talmud text plane is NC (Sefaria structure-facts only); Allen–Greenough and the classical share of Lewis-Short land citation_only because classical Latin prose (Cicero, Caesar, …) is held only in NC-quarantined treebanks never wired as targets; DSS targets are NC-gated (pointer edges only); catlib is copyright-isolated and excluded entirely.

Not copyright: Du Cange (PD, 1883–87 Favre edition) was deferred in the original plan on a cost estimate ("hardest extraction, lowest yield") that measurement disproved — 52% of its 79,453 entries carry regular citation shapes, on par with the Lewis-Short lane's 51%. The lane was ordered 2026-08-30 and is being built.

Every exclusion above is per-row honest in the data: unlinkable targets ship as anchor_kind='citation_only' / have_corpus=false edges — recorded, never dropped.

NuBerea project. Licensed CC BY 4.0.