Strong's Words — multilingual, provenance-marked

Given a Hebrew or Greek Strong's number, the actual words each language uses for it — anchored on the original languages, one language per file, and every word marked with how it was generated (so you keep only what you trust).

📦 Full dataset on Hugging Face → https://huggingface.co/datasets/bcv-data/strongs

from datasets import load_dataset
ds = load_dataset("bcv-data/strongs", "surfaces")   # or "glosses" / "attestations"

This GitHub repo is a taster. Small samples live in samples/; the complete data — all 12 languages, every tier (including the per-occurrence attestations tier) — is on Hugging Face.

What's inside (on HF)

tier	what
glosses	one canonical word per Strong's, per language
surfaces	attested renderings from real translations, with frequency
surfaces_by_method	the above split by method / source / edition
attestations	one row per aligned occurrence (verse + token ids) — opt-in, large

Every row carries provenance — method (manual / transfer / lexicon / llm / …), source, and review (human-verified vs machine).

Example (glosses, Spanish)

strong  lemma   gloss   methods       sources
G0025   ἀγαπάω  amar    lexicon;llm   inhouse-llm;ubs-dict
H2617   חֶסֶד   misericordia  lexicon  ubs-dict

Languages

12, as canonical ISO 639-3 / BCP 47 tags (eng, spa, cmn-Hant, …): arb, asm, ben, cmn-Hans, cmn-Hant, eng, fra, hau, hin, por, rus, spa

License

CC BY-SA 4.0 — see LICENSE. Derived from Clear-Bible/Alignments, UBS dictionaries (via BibleAquifer), and STEPBible-Data. Attribute the sources and keep share-alike derivatives compatible.

Produced by the bcv-query project.