--- license: cc-by-sa-4.0 language: - hbo - arc - grc pretty_name: Entity mentions tags: - bible - hebrew - greek - named-entities - coreference --- # entity-mentions v0.1 Who or what each word of the Tanakh and the Greek NT refers to: a person, place, group, deity, object (realia), plant or animal. It covers names and nouns, pronouns and pronoun suffixes, and the implied subject a verb carries. Each row is a MACULA word key and an entity. The dataset also has the entities themselves and their kinship. **Two routes, kept side by side.** `acai`: ACAI's own word links. `chain`: MACULA's reference from the word (subject reference, participant reference, referent), followed up to 4 steps to a word ACAI links. Where both routes reach a word and name the same entity, it gets one row. Where they name different entities, it gets a row for each, both marked `differs`, and neither is preferred. **No English.** Entities have neutral codes (`e` + 10 hex digits: the same codes as bcv-commons/reference-status and addressees). ACAI's ids are English names, so they are not carried, and neither are its labels, descriptions, subtypes or roles. Lemmas are Hebrew, Aramaic and Greek. ## Tables (`data/`) `mentions.parquet` (138,428 rows): | column | | |---|---| | `key` | MACULA word key (Tanakh 12 digits, Greek NT Nestle 1904 11 digits) | | `lang` | `hbo` (Hebrew and Aramaic) or `grc` | | `entity` | the entity code | | `kind` | `explicit` (a name or noun), `pronominal` (a pronoun or suffix), `subject` (a verb's implied subject), `other` (a chain from another word class) | | `route` | `acai`, `chain`, or `acai+chain` (both, same entity) | | `agree` | `same`, `differs`, `acai_only`, `chain_only` | `entities.parquet` (5,243: person 3,130, place 1,215, realia 345, group 306, flora 104, fauna 97, deity 46): `entity`, `type`, `gender` (`m`, `f` or empty), `lemmas` (`he:` / `arc:` / `el:` + the lemma, from ACAI), `characters` (the speaker layers' character ids joined to it), `apocrypha_only`, `non_biblical`, `mentions` (rows here). `kinship.parquet` (10,925 rows): `entity`, `relation` (father, mother, partner, offspring, sibling, tribe, group_origin, birth_place, death_place, possibly_same_as: ACAI's relation codes), `other`. ## How the routes agree | | Hebrew and Aramaic | Greek NT | |---|---|---| | words with an entity | 107,176 | 27,343 | | both routes | 28,392 | 13,919 | | same entity | 26,306 (0.927) | 12,096 (0.869) | | different entities (two rows each) | 2,086 | 1,823 | | ACAI only | 53,735 | 10,720 | | chain only | 25,049 | 2,704 | The chain can only reach entities that ACAI links somewhere nearby. Where it reaches none, the word has no chain row. Pronouns that neither route resolves are not here. bcv-commons/reference-status lists them as `unresolved`. ## Sources and licence ACAI Biblical Entity Data, © Mission Mutual / BiblioNexus, CC BY-SA 4.0, github.com/BibleAquifer/ACAI tag v2026-11-07 (commit 5375d0d334bb). Its Greek links (SBLGNT) are carried to Nestle 1904 ids by aligning the two MACULA Greek editions on their lemmas. MACULA Hebrew and Greek (CC BY 4.0): word keys and the reference attributes. Nothing comes from BHSA or TBTA. This dataset: CC BY-SA 4.0. ## Provenance Code 341e013: `shoresh/macula/build_entities.py` (ACAI to MACULA keys), `shoresh/macula/release_entity_mentions.py`.