|
Download README.md from bcv-commons/argument-roles: direct link, hf CLI and curl.
- Browser
- Download file 5.64 kB
-
https://huggingface.co/datasets/bcv-commons/argument-roles/resolve/main/README.md
- Command line
-
hf download hf://datasets/bcv-commons/argument-roles/README.md
-
curl -L -o README.md https://huggingface.co/datasets/bcv-commons/argument-roles/resolve/main/README.md
5.64 kB
| license: cc-by-sa-4.0 | |
| language: [hbo, arc, grc] | |
| pretty_name: Argument roles for the Hebrew and Greek clause layouts | |
| tags: [bible, hebrew, greek, semantic-roles, rrg, prepositions, louw-nida, septuagint] | |
| configs: | |
| - config_name: hebrew | |
| data_files: data/hebrew.parquet | |
| - config_name: greek-nt | |
| data_files: data/greek-nt.parquet | |
| # Argument roles (v0.2) | |
| Every constituent of every clause of the two bcv-commons clause layouts, with its grammatical relation, the clause's voice, the verb | |
| frame's macrorole and a semantic role: actor, undergoer, recipient, addressee, goal, source, location, time, accompaniment, | |
| instrument, means, beneficiary, agent and others. These are what a translation into another language needs to choose case markers, | |
| prepositions and voice. | |
| | config | layout (pinned) | constituents | with a role | preposition phrases with a role | | |
| |---|---|---|---|---| | |
| | `hebrew` | [bcv-commons/hebrew-clause-layout](https://huggingface.co/datasets/bcv-commons/hebrew-clause-layout) v0.37.1 (`32f89b71`) | 146,733 | 75,903 | 28,604 of 55,735 | | |
| | `greek-nt` | [bcv-commons/greek-nt-clause-layout](https://huggingface.co/datasets/bcv-commons/greek-nt-clause-layout) v0.7 (`9168fac3`) | 48,186 | 25,662 | 7,837 of 9,632 | | |
| Join a row to its clause by `clause_key`: the Hebrew layout's `clause_key`, the Greek layout's `first`, as in | |
| [bcv-commons/clause-columns](https://huggingface.co/datasets/bcv-commons/clause-columns). | |
| ## Changes | |
| - v0.2: the Septuagint word is now looked up in its own verse through the verse map (`macula/lxx_verses.py`). GLAUx numbers the | |
| Septuagint's way (Psalms one off, Jeremiah 25-51 reordered, Daniel 3-4 with the Greek additions), so the lookup found its verse | |
| for 0.785 of Hebrew words before (Psalms 0.13, Jeremiah 0.45) and 0.867 now; the rest fall back to the form's commonest parse, | |
| as before. Daniel's words keep the Old Greek: MACULA aligns them to it. `lxx_case` changes on 348 Hebrew phrases (203 in the | |
| Psalms, 71 in Jeremiah), `role` on 88. | |
| ## Columns | |
| | column | in | description | | |
| |---|---|---| | |
| | `clause_key` | both | the layout clause | | |
| | `words` | both | the constituent's word ids inside the clause, in text order, space-joined | | |
| | `head` | both | its first noun, pronoun, adjective or number word; empty when none | | |
| | `relation` | both | the constituent's role in MACULA's tree: s, o, o2, io, p, adv, pp | | |
| | `verb`, `voice` | both | the clause's main verb and voice (clause-columns) | | |
| | `macrorole` | both | the verb frame's label for the constituent (MACULA frames: A0, A1, A2, AA); empty when the frame does not list it | | |
| | `case` | greek-nt | the head's case | | |
| | `prep`, `prep_lemma` | both | the preposition heading the constituent | | |
| | `prep_ln` | greek-nt | the Louw-Nida subdomain of that preposition's sense here (bcv-commons/word-senses) | | |
| | `prep_lxx` | hebrew | the Septuagint word MACULA aligns to the Hebrew preposition | | |
| | `lxx_case` | hebrew | the case of the Septuagint word aligned to the phrase's first noun (GLAUx) | | |
| | `role` | both | the semantic role; empty where the evidence does not decide | | |
| | `role_basis` | both | which rule gave it (below) | | |
| ## How the roles are assigned | |
| - **frame / frame_passive:** the verb frame's A0 is the actor, A1 the undergoer (frame_passive: the passive subject), A2 kept as A2. | |
| - **ln (Greek):** the Louw-Nida subdomain of the preposition's sense names the role. For example, 90.A agent, 90.B instrument, | |
| 84.A source, 84.B goal, 83 location, 67 time, 89.H purpose, 34 accompaniment. With a verb of speaking (Louw-Nida 33), an | |
| experiencer or goal is the addressee (`ln_speech`, `+speech`). | |
| - **Hebrew preposition phrases,** in this order: | |
| - `lxx_nt_domain`: the Septuagint's preposition, the case of its noun, and the Louw-Nida domain that noun's lemma has in the | |
| New Testament. These are read through the New Testament's own (preposition, case, domain) → role table, with at least 8 | |
| examples and 60% agreement. | |
| - `lxx_nt`: the same without the domain (at least 20 examples, 60%). | |
| - `lxx_dative`: לְ that the Septuagint gives as a bare dative is the recipient. | |
| - `hebrew_prep`: כְּ is a comparison (`standard`). | |
| - `hebrew_domain`: the Hebrew preposition with its noun's SDBH domain, learned from the rows above. | |
| - `+speech`: a goal with אֶל, לְ or עַל after a verb of speaking (אָמַר, דָּבַר, קָרָא, צָוָה, עָנָה ...) is the addressee. | |
| Nothing comes from BHSA or TBTA, and no English text is used as evidence. No row was checked by hand; a human review of samples is | |
| planned and will be reported here. | |
| ## Measured use | |
| A rule-based drafting engine was tested on the Tanakh in six rotating test groups, each drafted by an engine that had not seen it. | |
| The figure is the share of right choices where the role changed the engine's choice: | |
| | use | Indonesian | Swahili | Tagalog | | |
| |---|---|---|---| | |
| | prepositions chosen by Hebrew preposition and role | +7.4 | +16.2 | +5.4 | | |
| | Tagalog case markers (ang / ng) by actor and undergoer | | | +3.6 | | |
| Word-level F1 of the whole draft against the published translation rises by +0.46 (Indonesian) and +0.85 (Swahili) from the | |
| prepositions alone. The Tagalog figure (+0.20) also includes kay / sa chosen by role and by whether the noun refers to a person. | |
| The Greek side has not been measured in drafting yet. | |
| ## Licence | |
| CC BY-SA 4.0: the Hebrew layout's clause keys, the UBS senses and the GLAUx parses are CC BY-SA 4.0. Credit: MACULA (Clear Bible / | |
| Biblica), United Bible Societies (word senses), GLAUx (KU Leuven, Alek Keersmaekers), make-tanakh-research (the measurements), and | |
| this dataset. | |