Tyndale STEPBible Data development for machine analysis and computational linguistics

What it is about

A report on the data behind STEPBible.org (Tyndale House, Cambridge). Free Bible modules tag words inconsistently (Gen 2:20 tagged "Adam" in one, "man" in another), so interlinears fail to link. Tyndale extended Strong's numbers (Greek up to G9999 for LXX words; Hebrew split by BDB homonyms with letter suffixes, as OpenScriptures did: H1254a "create" vs H1254b "fatten") and refined the OSHB-tagged Westminster Leningrad Codex: corrected word divisions and ketiv pointing; tags for every prefix, suffix and particle (vav split into conjunctive and consecutive; he split into article, interrogative, directional, paragogic); a three-way jussive flag (formally distinct, formally an imperfect where a jussive would differ, or ambiguous); variants from the UBS Hebrew text project with their evidence; and fixed-width word ids that hold both Hebrew and English versification. An automatic aligner (Paratext + Berkeley Aligner) for a Swahili translation worked for most words but failed on idiomatic senses such as פָּנִים "face" in Gen 32:21. The new datasets were announced as CC BY-NC-ND.

Ideas

• I1. The jussive flag: for every yiqtol in the Torah, mark whether the jussive form would differ (hollow, III-he, hiphil short forms) and whether it does, giving a "formal jussive / formal indicative / undecidable" value. It feeds volitive detection in prag.
• I2. Split ה־ and ו־ prefixes into their functions in FEATS/MISC (article / interrogative / directional / paragogic; conjunctive / consecutive), checked against MACULA's own values.
• I3. A sense split test for the dictionary: lemmas whose English alignment in JPS scatters over unrelated glosses (פָּנִים, בָּרָא) are candidates for homonym or idiom entries.
• I4. Keep a mapping between Hebrew and English versification in the verse ids (Exodus, Numbers, Deuteronomy have offsets) so JPS alignment never drifts.

For Mova

I1 is a feature for tools/macula and a rule input for tools/macula/src/prag.rs (directives), described in The verb (volitives) and 07-weak-verbs.md. I2 belongs to 13-mova-tools.md §5 (FEATS and MISC). I3 is a check in macula dict (10-lexicon.md, polysemy). I4 is a check on verses.tsv and the JPS alignment. The STEPBible data is NC-ND: a reference to compare with, never copied.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.7146/hn.v5i2.142739

@inproceedings{instonebrewer2019tyndale,
  title = {{Tyndale STEPBible Data development for machine analysis and computational linguistics}},
  author = {Instone-Brewer, David},
  booktitle = {HIPHIL Novum 5},
  year = {2019},
  doi = {10.7146/hn.v5i2.142739},
  url = {https://doi.org/10.7146/hn.v5i2.142739}
}

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/hbo-2019-instone-brewer-tyndale-stepbible-data-development-machine/