Morphological, syntactic, and lexical-alignment tables for the 73-book Sixto-Clementine Vulgate. The word-level parses are generated with Stanza from the exact NuBerea/vulgate-clementina revision recorded in SOURCE.json.

Canon and provenance

tokens contains every parsed word and punctuation token in one table, together with a Lewis & Short dictionary-gloss candidate when the normalized lemma has a source match.
sentences contains every Stanza sentence boundary.
dependency_edges contains every child-to-parent dependency relation.
All three canonical relations carry testament with one of old_testament, deuterocanon, or new_testament.
latin_lemma_seed is regenerated from the source dataset's lemmatized config.
stanza_corrections and latin_form_lemma are explicitly curated auxiliary tables. Their source revision is also pinned in SOURCE.json.

The build validates complete 73-book verse coverage, token/sentence counts, dependency-edge referential integrity, and the 46-book OT/deuterocanon plus 27-book NT split before publication.

Main relationships

tokens.token_id joins dependency_edges.child_token_id and dependency_edges.parent_token_id.
tokens.osis_ref plus sentence_index joins sentences.

Factuality and derivation

Category	Treatment
Canonical references, book metadata, positions, and testament	Deterministic normalization from the pinned Clementine source.
Tokens, sentences, lemmas, POS, morphology, and dependencies	Stanza ITTB machine annotations labeled `machine_generated_unverified`.
`gloss`, lexicon headword/POS, and entry ID	Lewis & Short dictionary data joined by normalized lemma and labeled `candidate_unverified_sense`; this is a dictionary gloss, not a contextual translation.
Matched lexicon entry without a short definition	Entry provenance remains in `lexicon_*` fields, while `gloss_status=lexicon_entry_without_gloss` and gloss provenance/derivation remain null.
`sum` gloss override	The malformed Lewis & Short short-definition fragment is replaced with the cited definition phrase “to be, exist, live”; the entry ID and override method remain recorded.
Stanza corrections	Preserved manual curation labeled `manually_curated_unverified` unless independently reviewed.
Form-to-lemma table	Parser-output majority statistics labeled `inherited_audited_candidate`.
Latin-to-source-language alignment	Not published. No source-attested or independently verified word-level alignment is currently available. Verse co-occurrence is not treated as lexical evidence.

Future cross-language alignments must identify the Latin token and source token or lexical identifier, state whether the evidence is source-attested or independently verified, cite the evidence, and identify the responsible source or reviewer. Unresolved tokens remain unaligned.

The repeatable review process is documented in alignment_review/README.md. Its dedicated Copilot profile is .github/agents/vulgate-alignment-reviewer.agent.md. AI output is restricted to unresolved cases or candidates for human review; only the human-validation stage can emit independently verified alignment rows.

License and attribution

The dataset is provided under CC BY 4.0.

Clementine Vulgate source text: public domain; digitization and detailed source attribution are documented by NuBerea/vulgate-clementina.
Stanza, Stanford NLP Group: Apache License 2.0.
STEPBible Strong's data: CC BY 4.0.
Parse layer, corrections, and dataset packaging: NuBerea / Streams, LLC (2026).

Suggested attribution: “NuBerea Vulgate Latin Dependency Parse Primitives, Streams, LLC, CC BY 4.0,” together with attribution for the upstream source datasets identified in SOURCE.json.

Downloads last month: 210

Total file size:

33 MB