bcv-commons/aligned-mwe

aligned_mwe — multi-word target expressions per lexeme

Where lexeme-alignments is one row per surface token, this is one row per lexeme rendered by a contiguous multi-word phrase (חֶסֶד → "kasih setia", בֵּית → "tempat pengirikan"). Mined from the aligner's per-verse t_idx positions: only spans whose target token positions are contiguous (max−min+1 == len) qualify — scattered tokens that merely all linked to a lexeme are dropped (and counted in the manifest as scattered_dropped, never silently). Rides on eflomal's grow-diag-final-and symmetrised alignment.

Schema (per row)

column	meaning
lexeme	MACULA lexeme anchor (hbo:2545 / grc:0026)
strong	rollup Strong's (H2545 / G0026)
phrase	the attested contiguous multi-word rendering (lowercased)
n_words	tokens in the span
source_corpus	which original-language text the span was aligned against (WLC, Nestle1904, ...)
base_text	the target edition this phrase was attested in (added 2026-09-30)
count	times this (lexeme → phrase) span was aligned in that edition
share	count / Σ count for that lexeme within the same base_text
contig	always true (contiguity is the inclusion criterion)

Every edition of a language is pooled into the one iso=<lang> partition — there is no primary edition. A phrase found in several editions appears once per edition, so a (lexeme, phrase) pair is not unique across rows: group by (lexeme, phrase) and sum count for a language-level frequency, or filter on base_text for one edition. share is a per-edition probability, so it does not sum to 1 across editions. The manifest lists each language's base_texts, rows per edition (by_base_text) and a per-edition sources licence pointer. Every partition now has base_text and covers every edition of its language; copies downloaded before October 2026 had no base_text column and covered only the first edition.

Licensing — CC0-1.0

Phrases + counts + public identifiers (Strong's / MACULA lexeme), no MACULA analytical columns — same basis as lexeme-alignments. Each language's source translation keeps its own terms; see the per-language source pointer in manifest.json.

Regenerate: python -m lexeme_aligner.export_mwe --iso <first-tag> --pool <other-tags> --method all (the chain does this). Bulk Parquet is git-ignored and published out-of-band; only manifest.json (per-language metadata + content_sha256) and this card are committed.

Verse mapping

Hebrew and English Bibles number some Old Testament verses differently (Psalm superscriptions, 1 Chronicles 6, Joel, Malachi, Daniel 4 and 6, ...). Source and target verses are paired with the verse-mapping tables of TVTMS (Translators Versification Traditions with Methodology for Standardisation), part of STEPBible Data by Tyndale House, Cambridge — CC BY 4.0, https://github.com/STEPBible/STEPBible-Data — with each edition's numbering detected from its own text (cross-checked against bcv-commons/bibles). Source verse references follow the Hebrew (WLC) numbering of the MACULA source text.