bcv-commons/word-senses

Word senses (v0.1)

The sense and meaning domains of the UBS dictionaries on every word of the Hebrew Bible and the Greek New Testament that UBS links to one: 366,562 words (Hebrew and Aramaic 239,117, Greek 127,445), keyed by MACULA's word ids, as codes.

A translation choice often depends on the sense, not on the word: the same Hebrew or Greek lemma is translated differently when it means different things. This dataset gives each word occurrence its sense, so a lexicon or a rule can be keyed by lemma and sense.

Files

• data/words.parquet: one row per word and sense (a word UBS links to two senses has two rows; 370,105 rows).
• domains.tsv: the domain codes and their names in the UBS dictionaries (the only English in the dataset).
• checks.md: coverage per word class and the largest gaps.

Columns

column	description
key	MACULA word id: Hebrew 12 digits (WLC, MACULA Hebrew), Greek 11 digits (Nestle 1904, MACULA Greek)
lang	hbo (Hebrew and Aramaic) or grc
entry	the UBS dictionary entry (MainId)
sense	the UBS meaning id (LEXID): entry, base form, meaning
ln	Greek: the Louw-Nida number of the meaning (LEXEntryCode); Hebrew: empty
domains	the meaning's domain codes, ;-joined (see domains.tsv)
subdomains	its subdomain codes, ;-joined
basis	index: UBS's word index on the WLC (Hebrew); references: the dictionary's own references per meaning (Greek)
part	i/n when UBS links one sense to a phrase of n MACULA words (this word is the i-th), else empty

How it was made

• Hebrew and Aramaic: from UBS's own word index on the WLC. Each word's link names a lemma, a base form and a meaning in the Semantic Dictionary of Biblical Hebrew (SDBH, v0.9.3), resolved to the meaning's id and checked against the domain the index names.
• Greek: from the UBS Greek New Testament dictionary (v1.1). Each meaning lists the verses and word positions it covers.
• Source: ubsicap/ubs-open-license, CC BY-SA 4.0. MACULA's own copies of these UBS fields are not read. Nothing comes from BHSA or TBTA.

Code: shoresh/macula/build_word_senses_ubs.py and release_word_senses.py in bcv-query.

Coverage and gaps

	noun	verb	adjective	preposition
Hebrew	0.93	0.83	0.79	0.05
Greek	0.99	0.96	0.98	0.87

The open SDBH release has no entry, or no linked meaning, for 264 Hebrew content lemmas, among them some of the most frequent verbs (הָיָה, בּוֹא, נָתַן, שָׁמַע, לָקַח) and כֹּל. Hebrew particles, prepositions and conjunctions are mostly not covered. checks.md lists the largest gaps.

How far to trust it

The senses are UBS's own assignments, carried over by code, not judged here. No row was checked by hand. A human review of samples is planned and will be reported here.

Use in drafting has been measured. A rule-based drafting engine whose lexicon is keyed by Hebrew lemma and UBS sense, instead of lemma alone, was tested on the Tanakh in six rotating test groups. Each group was drafted by an engine that had not seen it. The measure is the share of words drafted correctly among those whose senses take different translations in training:

language	lemma only	lemma and sense	gain	chrF3 gain
Indonesian	0.337	0.387	+5.1	+1.4
Tagalog	0.244	0.272	+2.8	+1.1
Swahili	0.249	0.281	+3.2	+1.4

The Greek side has not been measured in drafting yet.

Licence

CC BY-SA 4.0, as the UBS dictionaries' open release. Credit: United Bible Societies (Semantic Dictionary of Biblical Hebrew, UBS Greek New Testament dictionary based on Louw and Nida), MACULA (Clear Bible / Biblica) for the word ids, and this dataset.