prior-pack / README.md
larsgson's picture
Add pos + translit + word_class columns (wishlist #1/#5/#6)
80cbe30 verified
|
Raw
History Blame Contribute Delete
1.72 kB

prior_pack/ — language-independent leverage for the aligner (CC-BY)

One row per original lexeme, bundling shoresh signals the aligner's gloss/neural runs consume as priors. Built once (language-independent) via shoresh/macula/build_prior_pack.py. Spec: internal-docs/prior-pack.md. CC-BY-4.0 (MACULA lexeme + lxx_bridge; label-free, no MARBLE).

column meaning
lexeme / strong / testament / is_content / lemma the lexeme + rollup
pos normalized POS {noun,verb,adj,adv,pron,prep,conj,det,num,name,particle} — dominant MACULA class; name = grammatical proper noun (Np) or TIPNR person/place (elohim/theos stay noun; YHWH/David/Jesus = name)
translit romanized form (da.vid, Iēsous) — for gap-name / cross-script matching
word_class content | function, derived from pos
keyness biblical-salience (function-word filter); null for non-content
lxx_greek / lxx_hebrew cross-testament bridge (OT→Greek / NT→Hebrew), freq-ordered
senses [{stem, sense, share}] — sense inventory / prior distribution (OT)
neighbors [{lexeme, score, relation, confidence}] — semantic field (OT)
xling_confidence # of published aligned-lex languages that align this lexeme with a hi_conf dominant (0–7); high=stable anchor, low=fragile

Consumed: gloss (keyness+lxx+senses extend/clean the mined dict); neural (neighbors tie-break + senses). Publish to bcv-commons/strongs as a priors config. neighbors/senses are OT-only for now.

xling_confidence is derived from the aligner's published aligned-lex (the loop-back); rebuild with --aligned-lex-dir <mirror> when partitions change.