Dataset Viewer
Auto-converted to Parquet Duplicate
hebrew_strong
stringlengths
5
5
greek_strong
stringlengths
5
5
tier
stringclasses
1 value
count
int64
3
7.43k
kept
int64
0
1.65k
score
float64
0
1
quoted
int64
0
30
H0505
G0001
heb_scg
3
3
1
0
H0175
G0002
heb_scg
4
4
0.8
0
H0001
G0005
heb_scg
3
3
0.75
0
H1892
G0006
heb_scg
4
3
1
0
H0029
G0007
heb_scg
3
3
1
0
H0085
G0011
heb_scg
73
68
1
4
H3220
G0012
heb_scg
8
8
0.88889
0
H2896
G0015
heb_scg
3
3
0.42857
0
H2896
G0018
heb_scg
79
79
0.82292
1
H2898
G0018
heb_scg
4
4
0.04167
0
H3559
G0018
heb_scg
3
3
0.03125
0
H7307
G0018
heb_scg
3
2
0.03125
0
H2617
G0019
heb_scg
3
3
0.75
0
H1523
G0021
heb_scg
3
3
0.75
1
H0157
G0025
heb_scg
118
117
0.90076
5
H0160
G0026
heb_scg
87
82
0.84466
0
H0157
G0026
heb_scg
7
6
0.06796
0
H5030
G0026
heb_scg
4
4
0.03883
0
H3039
G0027
heb_scg
27
27
0.49091
0
H1730
G0027
heb_scg
6
6
0.10909
0
H0157
G0027
heb_scg
5
5
0.09091
0
H0034
G0027
heb_scg
3
3
0.05455
0
H3368
G0027
heb_scg
3
3
0.05455
0
H4397
G0032
heb_scg
163
158
0.94767
2
H6635
G0032
heb_scg
4
4
0.02326
0
H5739
G0034
heb_scg
5
3
0.71429
0
H6942
G0037
heb_scg
16
14
0.64
0
H6944
G0038
heb_scg
5
5
0.5
0
H6944
G0040
heb_scg
145
142
0.64159
1
H6918
G0040
heb_scg
75
72
0.33186
0
H6944
G0042
heb_scg
3
3
1
0
H3045
G0050
heb_scg
5
5
0.45455
0
H5927
G0050
heb_scg
3
3
0.27273
0
H1847
G0052
heb_scg
3
3
1
0
H7784
G0058
heb_scg
5
5
0.5
0
H7069
G0059
heb_scg
12
12
0.54545
0
H7704
G0068
heb_scg
31
28
0.91176
0
H0935
G0071
heb_scg
33
32
0.54098
0
H1980
G0071
heb_scg
10
10
0.16393
0
H2986
G0071
heb_scg
3
3
0.04918
1
H0120
G0076
heb_scg
9
8
1
0
H0269
G0079
heb_scg
19
19
0.82609
0
H0251
G0080
heb_scg
320
307
0.95522
3
H3117
G0080
heb_scg
3
1
0.00896
0
H7585
G0086
heb_scg
4
4
0.57143
0
H7586
G0086
heb_scg
3
3
0.42857
0
H6231
G0091
heb_scg
6
6
0.33333
0
H7843
G0091
heb_scg
4
4
0.22222
0
H5766
G0093
heb_scg
7
6
0.28
0
H0205
G0093
heb_scg
6
6
0.24
0
H5766
G0094
heb_scg
4
4
0.36364
0
H5765
G0094
heb_scg
3
3
0.27273
0
H6381
G0102
heb_scg
3
3
0.3
0
H7891
G0103
heb_scg
3
3
0.6
0
H8548
G0104
heb_scg
5
5
0.71429
0
H4682
G0106
heb_scg
8
8
0.88889
0
H8064
G0109
heb_scg
4
3
0.57143
0
H4714
G0125
heb_scg
25
24
1
4
H1818
G0129
heb_scg
88
80
0.97778
2
H5712
G0139
heb_scg
3
3
0.6
0
H5375
G0142
heb_scg
51
50
0.56044
2
H3947
G0142
heb_scg
21
20
0.23077
1
H0954
G0153
heb_scg
4
3
1
0
H7592
G0154
heb_scg
40
36
0.60606
0
H1245
G0154
heb_scg
10
10
0.15152
0
H4325
G0154
heb_scg
5
5
0.07576
0
H4480
G0154
heb_scg
3
3
0.04545
0
H1697
G0156
heb_scg
3
3
0.15789
0
H4100
G0156
heb_scg
3
3
0.15789
0
H5771
G0156
heb_scg
3
3
0.15789
0
H7628
G0161
heb_scg
3
3
1
2
H5769
G0165
heb_scg
91
84
0.76471
2
H5703
G0165
heb_scg
9
9
0.07563
1
H5704
G0165
heb_scg
5
5
0.04202
0
H5331
G0165
heb_scg
3
3
0.02521
0
H6924
G0165
heb_scg
3
2
0.02521
0
H5769
G0166
heb_scg
61
61
0.87143
0
H5704
G0166
heb_scg
5
4
0.07143
0
H8441
G0167
heb_scg
4
3
0.44444
0
H2932
G0169
heb_scg
17
17
0.60714
0
H2931
G0169
heb_scg
5
5
0.17857
0
H2930
G0169
heb_scg
4
4
0.14286
0
H6975
G0173
heb_scg
3
3
1
0
H8549
G0185
heb_scg
3
3
1
0
H0241
G0189
heb_scg
6
4
0.27273
0
H8052
G0189
heb_scg
5
5
0.22727
1
H8085
G0189
heb_scg
5
5
0.22727
2
H0310
G0190
heb_scg
71
71
0.48299
0
H1980
G0190
heb_scg
71
68
0.48299
0
H8085
G0191
heb_scg
373
367
0.90976
10
H7181
G0191
heb_scg
14
13
0.03415
0
H8088
G0191
heb_scg
4
4
0.00976
0
H8052
G0191
heb_scg
3
3
0.00732
0
H3190
G0199
heb_scg
7
7
0.77778
0
H8085
G0202
heb_scg
4
4
1
0
H6189
G0203
heb_scg
10
10
0.625
0
H7097
G0206
heb_scg
5
5
0.83333
0
H3273
G0207
heb_scg
4
4
0.8
0
H6746
G0211
heb_scg
3
3
0.75
0
H4417
G0217
heb_scg
6
5
1
0
End of preview. Expand in Data Studio

Hebrew NT Bridge

Which Hebrew word renders which Greek word of the New Testament, read from two Hebrew translations of it: Salkinson–Ginsburg (heb_scg) and Delitzsch (heb_mod). Every Greek word is a word of MACULA Greek (Nestle 1904) and every Hebrew word part carries the lexeme MACULA Hebrew uses for the Tanakh, so the two Testaments meet on the same keys.

It is the counterpart of a Septuagint bridge. The Septuagint goes from a Hebrew source to its Greek translation; this goes from the Greek source to a Hebrew translation, in the Greek of the New Testament itself.

Release v0.3, built 2026-11-19 from bcv-query 341e013
Editions heb_scg: Salkinson–Ginsburg (1885/86), pointed. heb_mod: Delitzsch, as eBible's "Modern Hebrew Bible" has it, without vowel points
Greek MACULA Greek, Nestle 1904: 137,779 words, of which 65,264 nouns, verbs and adjectives

This is machine-made and has not been checked by hand. The links are word alignments by lexeme-aligner's method (its statistical chain by word parts, then the Jev model of TypeSafe by verse or clause). The Hebrew lexemes come from a tagger trained on MACULA Hebrew. A model's review of every Greek noun, verb and adjective, for both translations, is a dataset of its own: hebrew-nt-review-model. See Quality before relying on a single link.

Files

File Rows
pairs/<edition>.tsv one per pair of a Hebrew and a Greek lexeme seen at least 3 times
links/<edition>.parquet one per Greek word and Hebrew word part it is linked to
quotes/<edition>.tsv one per verse of the translation and Tanakh verse it follows
articles/<edition>.parquet one per Greek article and the Hebrew part that renders it, by rule (from v0.3)
function_words/<edition>.parquet one per Greek preposition, pronoun and conjunction and the Hebrew part(s) that render it (from v0.3)
report.json the figures on this card, and more
SHA256SUMS SHA-256 of every file

The pairs

Column
hebrew_strong the Hebrew lexeme: Strong's number as MACULA Hebrew gives it, zero-padded, homonym letter dropped (H0559)
greek_strong the Greek lexeme: Strong's number of MACULA Greek (G3004)
tier the edition
count Greek words of this lexeme linked to a part of this Hebrew lexeme (links of class edited-wide are not counted)
kept of them, links of class kept, the safest class
score count divided by all counted links of the Greek lexeme: how often the translator chose this Hebrew word for it
quoted of count, links to a Hebrew word inside a stretch that follows a Tanakh verse (see Verses that follow the Tanakh). count minus quoted is the evidence that does not lean on the Tanakh's own wording

Rows are sorted by Greek lexeme, then by count. A Greek word rendered by two Hebrew words (ἀρχιερεύς by הַכֹּהֵן הַגָּדוֹל) counts for both.

heb_scg heb_mod
Links from Greek word ids, with classes (by-id) Greek word ids, with classes (by-id)
Greek nouns, verbs and adjectives linked to a Hebrew word 61,542 62,348
of them on a Hebrew lexeme of the Tanakh 51,196 53,255
Pairs (count ≥ 3) 3,209 3,202
Hebrew lexemes / Greek lexemes in them 1,154 / 1,557 1,161 / 1,677

The links

Column
edition heb_scg or heb_mod
greek_id the word's id in the MACULA Greek trees (n40001001001 = Matthew 1:1, word 1)
greek_strong, greek_lemma, greek_class MACULA Greek's lexeme, lemma and word class
book, chapter, verse the verse of the Hebrew edition
i, part the Hebrew word in the verse and the part in the word, as in hebrew-nt-analysis
hebrew_strong, hebrew_lemma, hebrew_class the part's lexeme (empty for a name or a word the Tanakh lacks), lemma and class
source where the part's analysis comes from: macula (the form occurs in the Tanakh with one analysis) or model (the tagger)
link the aligner's class: kept, inside, part, edited, edited-wide; compact where the links have no classes
form by-id (links by Greek word id to a word or part) or compact (links from the verse's content words to whole words)
tanakh the Tanakh verse whose wording the Hebrew word follows (ISA 9:1, Hebrew verse numbers), empty elsewhere

A link to a whole Hebrew word is resolved to parts by rule: a word of one part is that part; for a Greek noun, verb, adjective, adverb or numeral, the word's parts of those classes; for a Greek function word, the parts of the same family (conjunction, preposition, article, pronoun, particle). links has 123,728 rows for heb_scg and 128,516 for heb_mod.

Articles (from v0.3)

The Greek article (ὁ, ἡ, τό) is linked by a rule, not by the aligner: articles/<edition>.parquet, one row per article (an article with no Hebrew part has one row with i and part empty). For articles, read this file in place of links.

The rule: the article's head is the first content word after it in its Greek phrase; the head's Hebrew word is its reviewed link. The article is then rendered by that word's article part (status primary); else, where Hebrew marks definiteness another way, by the part that carries it (implied): the word's pronoun suffix (ὁ πατήρ μου = אָבִי, the suffix), or in a construct chain the next word's article or suffix (τὸ ὄνομα τοῦ κυρίου = שֵׁם הָאָדוֹן, the article of הָאָדוֹן); else by no part (not_rendered). Where the head has no Hebrew word the aligner's link stays, a whole word resolved to its article part (by aligner).

Salkinson-Ginsburg Delitzsch
articles 19,786 19,786
by rule 17,977 (primary 7,409, implied 6,179, not rendered 4,389) 18,468 (primary 8,490, implied 6,307, not rendered 3,671)
by the aligner 1,809 1,318

Checked on a sealed sample of 200 articles per edition, judged blind: the rule (else the aligner) is right in 0.805 (Salkinson) and 0.835 (Delitzsch) of them; the aligner's own article links in 0.440 and 0.455. The figure is optimistic: the rule's last change (link the part that carries the definiteness) was made after reading the misses on that same sample.

Prepositions, pronouns and conjunctions (from v0.3)

function_words/<edition>.parquet: one row per Greek preposition, pronoun or conjunction and Hebrew part (greek_class prep / pron / conj; a word with no Hebrew part has one row with i and part empty; part empty with i given = a whole Hebrew word the rule could not narrow). For these words, read this file in place of links.

  • Prepositions: the aligner's link narrowed to the preposition: a whole-word link to its prefixed preposition (or object marker) part, a compound kept whole (מֵ|אֵת, לִ|פְנֵי, עַל יְדֵי, עַל פִּי). With no link: the preposition prefixed to the Hebrew word of the Greek word it governs (by rule), else none.
  • Pronouns: the pronoun suffix of the Hebrew word of the Greek word the pronoun belongs to (ὁ πατὴρ αὐτοῦ = אָבִיו, the suffix; else its clause's verb) (by rule); else the aligner's link narrowed to its suffix or pronoun part.
  • Conjunctions: the conjunction part leading the Hebrew word of the next Greek content word (καὶ εἶπεν = וַ|יֹּאמֶר, the וַ), or the conjunction word just before it (by rule); else the aligner's link narrowed to its conjunction part.
Salkinson-Ginsburg Delitzsch
prepositions 10,914 (aligner, narrowed 9,546; rule 270; none 1,098) 10,914 (9,913; 153; 848)
pronouns 16,177 (rule 3,990; aligner 10,659; none 1,528) 16,177 (4,445; 10,627; 1,105)
conjunctions 18,227 (rule 9,701; aligner 6,184; none 2,342) 18,227 (9,725; 6,783; 1,719)

Checked on a sealed sample of about 200 words per class and edition, judged blind (exact match of the Hebrew parts; the rules were fixed before the sample was scored, and scored once):

Salkinson-Ginsburg Delitzsch the aligner's own links
prepositions 0.720 0.835 0.505 / 0.550
pronouns 0.693 0.704 0.588 / 0.588
conjunctions 0.709 0.725 0.497 / 0.450

Most of the gain is narrowing a whole-word link to the right part; for pronouns that is nearly all of it.

Verses that follow the Tanakh

The translators knew the Hebrew Bible. Where the New Testament quotes it, they could write the Hebrew verse instead of translating the Greek, and elsewhere they borrowed its phrases (Salkinson's אוּד מֻצָּל מֵאֵשׁ in 1 Corinthians 3:15 is Zechariah 3:2). In such a stretch the Hebrew word is the Tanakh's choice, not the translator's answer to the Greek word, so a link there does not count as a second witness beside the Septuagint. These stretches are marked.

They are found on the Hebrew side alone, with no list of quotations. The nouns, verbs, adjectives, adverbs and numerals of a verse that have a Tanakh lexeme are compared, as a sequence, with those of every Tanakh verse (MACULA Hebrew). What the two have in common in the same order is kept where it is a stretch of two lexemes or more. A stretch weighs the sum, over its lexemes, of ln(Tanakh verses / Tanakh verses with the lexeme): a lexeme found in one verse of 150 weighs 5. A pair of verses is listed in quotes from weight 16, and its stretches are marked in links and counted in quoted from weight 20.

quotes column
nt_ref the verse of the translation (MAT 4:16)
tanakh_ref the Tanakh verse, in MACULA Hebrew's (the Hebrew) verse numbers (ISA 9:1)
weight the weight of the shared stretches
lexemes, longest the lexemes in them, and in the longest one
heb_scg heb_mod
Pairs of verses listed (weight ≥ 16) 739 559
Verses with a marked stretch (weight ≥ 20) 226 179
Hebrew word parts in the marked stretches 1,450 1,181
Counted links to them 1,302 1,085
Pairs left with fewer than 3 links without them 68 43
Content links on a pair the Septuagint bridge has: inside / outside the marked stretches 80.1% / 65.8% 81.7% / 67.8%

156 verses are marked in both translations, 249 in at least one. The last row is the check the mark allows: inside a marked stretch the Hebrew is the Tanakh's own wording, which the Septuagint translated, and there the Septuagint's pairs fit more often.

How far to trust the weight, on heb_scg: a list of quotations made another way (from the Greek the New Testament shares with the Septuagint) has the same book and chapter for 107 of the 122 verses of weight 25 and more, for 53 of 104 between 20 and 25, and for 59 of 270 between 16 and 20. Read by sample (14 verses per band, by a model, not a person): from 25 nearly all are quotations; between 20 and 25 quotations and borrowed phrases; between 16 and 20 about half are phrases any Hebrew text shares (numbers, "eat and drink"). Matthew's genealogy is marked too: it follows Ruth 4 and 1 Chronicles 2 name by name.

Not found: short quotations in common words (Hebrews 1:5 from Psalm 2:7; John 19:37 from Zechariah 12:10), and a quotation the tagger read wrongly, which is more often the case in heb_mod. A quotation that runs over two Tanakh verses is weighed verse by verse.

Quality

The links. For heb_scg they are bcv-commons/alignments-by-id (v1.1), whose card gives a check per class: kept 25 of 25 right, inside 21 of 25, part 21 of 25, edited 25 of 39, edited-wide 17 of 36. For heb_mod (since v0.2) they are made the same way: lexeme-aligner's own alignments-by-id builder, run by make-tanakh-research on 2026-10-31 on its alignment of Delitzsch (the statistical chain by word parts, then Jev v4 by clause; USD 5.61) and handed over to bcv-query, its parts mapped onto hebrew-nt-analysis v0.2 (141 words split differently; 10 links moved). Function words are linked too. v0.1 had only the coarser form (bcv-commons/compact-alignments, 11583082: nouns, verbs and adjectives, whole words); on heb_mod, of 51,933 Greek content words that reach a Hebrew lexeme by both forms, 93.9% reach the same one (on heb_scg, 50,491 and 97.9%).

Function words (articles, prepositions, conjunctions, pronouns, particles) were checked blind on 2026-11-17: 40 Greek words per class and translation, one per verse, judged by a model (Claude Opus 5.5, not a person) from the two verses without seeing the link, then compared with it. Right = the same Hebrew parts; partial = an overlap:

Class heb_scg right / + partial heb_mod right / + partial
article 48% / 50% 40% / 48%
preposition 60% / 80% 57% / 88%
conjunction 62% / 75% 60% / 82%
pronoun 78% / 88% 75% / 85%
particle 62% / 80% 57% / 78%

Most misses are links to the whole Hebrew word or to the next word, and Greek articles left unlinked where the Hebrew has ה. Use function-word links with care; the content-word links are the ones the review dataset checks.

The Hebrew lexemes. The New Testament has no reference analysis. On Tanakh text the tagger was not trained on, with the same lookup of known forms: see the card of hebrew-nt-analysis (pointed text for heb_scg, unpointed for heb_mod, which is weaker).

The two translations against each other. Of 45,661 Greek content words that reach a Hebrew lexeme in both, 74.1% reach the same one, and 74.9% have a lexeme in common. They are two translators' choices, so a difference is not an error.

Against the Septuagint. Compared with the pairs of the Septuagint bridge (MACULA Hebrew's word links to the Septuagint, and verse co-occurrence with GLAUx):

heb_scg heb_mod
Pairs the Septuagint bridge also has 1,578 (49.2%) 1,539 (48.1%)
Share of the links that lie on those pairs 86.5% 85.4%
Pairs the Septuagint bridge lacks 1,631 1,663
Greek NT lexemes with no Septuagint pair that gain a Hebrew lexeme 448 (7,702 NT words) 500 (8,016 NT words)
Hebrew lexemes with no Septuagint pair that gain a Greek lexeme 118 130

Known limits.

  • The translators knew the Hebrew Bible and the tradition behind the New Testament's Greek. Agreement with the Septuagint bridge is therefore not independent confirmation, also outside the marked stretches.
  • Both translations follow the Textus Receptus; the Greek ids are Nestle 1904. Greek words the translation's base lacks have no link.
  • The heb_mod text has verses whose boundary is lost or moved (10 verses with no text, their words in the verse before). Links in such verses are wrong or missing. The text is used as the source gives it.
  • A pair with a low count can be a wrong link. Filter on count, score, and for heb_scg on kept.

Sources and licence

Part Source Licence
Greek words MACULA Greek, Nestle 1904 (Clear Bible / Biblica) CC BY 4.0
Tanakh lexemes, for the mark MACULA Hebrew (Clear Bible / Biblica) CC BY 4.0
Word links heb_scg: bcv-commons/alignments-by-id (v1.1), by lexeme-aligner. heb_mod: lexeme-aligner's method and builder, run by make-tanakh-research, handed over to bcv-query. Compared with bcv-commons/compact-alignments (11583082) CC0 1.0 (the aligner's); this dataset CC BY 4.0
Hebrew word parts and lexemes bcv-commons/hebrew-nt-analysis, trained on MACULA Hebrew (Clear Bible / Biblica) CC BY 4.0
Hebrew texts Salkinson–Ginsburg; Delitzsch (eBible "Modern Hebrew Bible"), via bible.helloao.org, by the route bcv-commons/bibles names public domain
This dataset bcv-commons CC BY 4.0
The code (bcv-query, macula/build_hebrew_nt_bridge.py) bcv-commons MIT

The files hold no verse text: ids, positions, references, lemmas, lexeme numbers and classes.

Attribution when you use this dataset:

Hebrew NT Bridge (v0.3), bcv-commons, CC BY 4.0. Greek: MACULA Greek, Clear Bible / Biblica (CC BY 4.0). Hebrew lexemes: analysis trained on MACULA Hebrew, Clear Bible / Biblica (CC BY 4.0). Word links: lexeme-aligner (CC0).

Feedback

Issues and requests: bcv-commons/bcv-query. Please name the edition, the verse and the word, and what you expected.

Versions

  • v0.1 (2026-11-16): both translations; heb_mod from the compact form.
  • v0.2: heb_mod by Greek word id, with classes, word parts and function words (handed over by make-tanakh-research); heb_scg read from alignments-by-id v1.1 (its files are the same as v1.0's); the function-word check.
  • v0.3: articles/: the Greek article linked by rule (the head's Hebrew word; the part that carries definiteness), with status primary / implied / not_rendered and by rule / aligner; function_words/: prepositions, pronouns and conjunctions linked by rule (by rule / aligner / none). Pairs, links and quotes are as in v0.2.
Downloads last month
89