|
Download README.md from bcv-commons/reference-status: direct link, hf CLI and curl.
- Browser
- Download file 4.31 kB
-
https://huggingface.co/datasets/bcv-commons/reference-status/resolve/main/README.md
- Command line
-
hf download hf://datasets/bcv-commons/reference-status/README.md
-
curl -L -o README.md https://huggingface.co/datasets/bcv-commons/reference-status/resolve/main/README.md
4.31 kB
| license: cc-by-sa-4.0 | |
| language: [hbo, arc, grc] | |
| pretty_name: Reference status of the noun phrases of the Hebrew Bible and the Greek New Testament | |
| tags: [bible, hebrew, greek, reference, definiteness, participant-tracking, translation] | |
| configs: | |
| - config_name: hebrew | |
| data_files: data/hebrew.parquet | |
| - config_name: greek-nt | |
| data_files: data/greek-nt.parquet | |
| # Reference status (v0.2) | |
| Whether a translation writes "a", "the", "this", "that" or a pronoun depends on who a phrase refers to and whether the reader | |
| already knows them. This dataset gives every noun-phrase head of the Hebrew Bible (119,892) and the Greek New Testament | |
| (36,766) its referent, its status in the episode, how the source marks it, and its animacy. | |
| | | new | given | restaged | known | unresolved | | |
| |---|---|---|---|---|---| | |
| | Hebrew Bible | 47,900 | 59,799 | 10,039 | 492 | 1,662 | | |
| | Greek New Testament | 14,277 | 17,248 | 2,786 | 118 | 2,337 | | |
| ## Changes | |
| - v0.2: referents keep ACAI's full entity ids. v0.1 folded an id into its alias's main entity, so that, for one, Saul the king | |
| (ACAI Saul.2) took Paul's code (ACAI lists Saul as one of Paul's names). A reference chain ending on a word with no lemma gave | |
| an empty `lemma:`; those 454 Hebrew pronoun phrases are now `unresolved`. Against v0.1, referents change on 5,165 Hebrew and | |
| 3,701 Greek phrases, status on 214. | |
| ## Columns | |
| | column | description | | |
| |---|---| | |
| | `clause_key` | the clause in the bcv-commons clause layouts (as in bcv-commons/clause-columns) | | |
| | `head` | the phrase's head word (MACULA id: Hebrew 12 digits, Greek 11) | | |
| | `referent` | who or what it refers to: `e…` an entity code (ACAI's entity, coded), `lemma:<Strong's or lemma>` a chain on the word itself, `unresolved:<word>` a pronoun with no referent | | |
| | `basis` | acai (ACAI links the word to the entity), ref (MACULA's reference chain reaches it), lemma, unresolved | | |
| | `status` | new (first mention in the episode), given (mentioned within the last 20 clauses), restaged (mentioned earlier, more than 20 clauses back), known (God on first mention), unresolved | | |
| | `distance` | clauses since the last mention (null for new / known / unresolved) | | |
| | `definite` | how the source marks the phrase: article (Hebrew also inside בַּ / לַ, Aramaic determined state), proper, suffix, construct, demonstrative, pronoun, or empty | | |
| | `proximity` | near (זֶה / זֹאת / אֵלֶּה, Aramaic דְּנָה …, οὗτος) or remote (article + הוּא …, ἐκεῖνος) where a demonstrative stands in the phrase | | |
| | `animacy` | person, deity, animal, place, thing, or empty: ACAI's type, else the head's UBS domain (SDBH, Louw-Nida) | | |
| The episode is the chapter. Mentions are counted from every head and from every word ACAI links to an entity (verbs carry their | |
| implied subjects). Entity codes are stable across builds of this dataset; they are not ACAI's ids. | |
| ## How good it is | |
| Nothing was checked by hand. The source's own marking agrees in direction: new referents are unmarked far more often than given | |
| ones (Hebrew 0.30 against 0.11, Greek 0.42 against 0.13), and pronoun heads are mostly given. Greek "new" phrases with the article | |
| (0.41) are partly unique, generic or inferable referents ("the crowd", "the house"), which this version does not separate. | |
| ## Measured use | |
| A rule-based drafting engine, tested on the Tanakh in six rotating test groups, each drafted by an engine that had not seen it: | |
| | use | before | after | word F1 of the draft | | |
| |---|---|---|---| | |
| | Indonesian *itu* after a given or restaged phrase the source marks as definite | 0.060 | 0.671 right | +0.34 | | |
| | Swahili demonstrative of the noun's class where the source has one (near / remote) | 0.033 | 0.321 right | +0.03 | | |
| Swahili marks a given phrase far less often than Indonesian: adding a demonstrative at every restaged phrase was right 0.07 of the | |
| time and was dropped. | |
| ## Licence | |
| CC BY-SA 4.0 (ACAI's ShareAlike, UBS domains). Built from MACULA Hebrew and Greek (CC BY 4.0), ACAI Biblical Entity Data | |
| (CC BY-SA 4.0), the UBS dictionaries' domains (CC BY-SA 4.0) and the bcv-commons clause layouts (Hebrew CC BY-SA 4.0, Greek CC BY 4.0). Credit: MACULA | |
| (Clear Bible / Biblica), ACAI (BibleAquifer), United Bible Societies, make-tanakh-research (the measurements), and this dataset. | |