Dataset Viewer
Auto-converted to Parquet Duplicate
clause_key
string
rank
int64
other
string
score
float32
46001001001
1
51001001001
0.9438
46001001001
2
49001001001
0.8516
46001001001
3
47001001001
0.835
46001001001
4
55001001001
0.8164
46001001001
5
45001001001
0.7935
46001001001
6
54001001001
0.7881
46001001001
7
56001001001
0.7515
46001001001
8
61001001001
0.7031
46001001001
9
50001001001
0.6782
46001001001
10
65001001001
0.6777
46001002001
1
50001001001
0.7183
46001002001
2
51001002001
0.7139
46001002001
3
45001007006
0.708
46001002001
4
49001001001
0.6704
46001002001
5
47001001001
0.667
46001002001
6
52001001001
0.6455
46001002001
7
53001001001
0.5825
46001002001
8
52002014010
0.5781
46001002001
9
59001001001
0.5747
46001002001
10
45001007002
0.5635
46001002017
1
46006011009
0.6978
46001002017
2
45001004013
0.6528
46001002017
3
46001009004
0.6401
46001002017
4
48006014006
0.6348
46001002017
5
53003018001
0.6147
46001002017
6
49005020001
0.6123
46001002017
7
52005023010
0.6084
46001002017
8
45016020015
0.6084
46001002017
9
52005028001
0.6084
46001002017
10
65001021006
0.6079
46001002029
1
45004021008
0.561
46001002029
2
45003029009
0.5552
46001002029
3
51001011016
0.542
46001002029
4
47001008025
0.5347
46001002029
5
42009002009
0.5322
46001002029
6
52005024005
0.5293
46001002029
7
52001001016
0.5283
46001002029
8
44026022028
0.5234
46001002029
9
66002008019
0.5161
46001002029
10
48006014021
0.5156
46001003001
1
47001002001
1
46001003001
2
57001003001
1
46001003001
3
48001003001
1
46001003001
4
50001002001
1
46001003001
5
49001002001
1
46001003001
6
53001002001
0.9863
46001003001
7
45001007010
0.9106
46001003001
8
49006023001
0.8511
46001003001
9
56001004007
0.814
46001003001
10
55001002004
0.7798
46001004001
1
53001003003
0.7622
46001004001
2
52001002001
0.7563
46001004001
3
57001004001
0.7295
46001004001
4
53002013004
0.7217
46001004001
5
47008016001
0.7134
46001004001
6
46014018001
0.6768
46001004001
7
51001003001
0.6309
46001004001
8
55001003001
0.625
46001004001
9
44027025004
0.6245
46001004001
10
52002013001
0.6162
46001004012
1
49001001014
0.7349
46001004012
2
55002001005
0.6665
46001004012
3
47002017020
0.666
46001004012
4
47003014023
0.6533
46001004012
5
55001013009
0.6274
46001004012
6
49002006003
0.6274
46001004012
7
60001013015
0.6177
46001004012
8
45015005011
0.6167
46001004012
9
46016024001
0.6133
46001004012
10
55001009019
0.6064
46001005001
1
53002017005
0.7109
46001005001
2
47009011001
0.6304
46001005001
3
47008007001
0.5923
46001005001
4
50004012007
0.5918
46001005001
5
49001008001
0.5889
46001005001
6
58013021001
0.582
46001005001
7
49006018012
0.5776
46001005001
8
49006018001
0.5654
46001005001
9
47011006011
0.5649
46001005001
10
47007011025
0.5635
46001006001
1
49003019007
0.5244
46001006001
2
47001005001
0.5142
46001006001
3
48005004001
0.5117
46001006001
4
47005014001
0.4958
46001006001
5
60001011007
0.4758
46001006001
6
47013003008
0.4727
46001006001
7
51003015001
0.4707
46001006001
8
45009003007
0.4697
46001006001
9
48004019006
0.4663
46001006001
10
50002030001
0.4658
46001007001
1
50001028001
0.6729
46001007001
2
47006003001
0.5889
46001007001
3
47007009015
0.5601
46001007001
4
45012003014
0.5581
46001007001
5
59001004012
0.5562
46001007001
6
49003013003
0.5425
46001007001
7
47009011001
0.5386
46001007001
8
54006017008
0.5381
46001007001
9
46005011005
0.5356
46001007001
10
52003003001
0.5322
End of preview. Expand in Data Studio

clauses-similar v0.1

For every clause of the Tanakh and of the Greek New Testament, the 10 most similar clauses of the same text outside its own verse, by contextual vectors: ids and scores only, no text. Computed (no model choice, no human check); the "-similar" ending marks a table computed from encoder vectors. Code 307099b (bcv-commons, shoresh/macula/build_clauses_similar.py).

File Rows Columns
data/hebrew.parquet 875,820 (87,582 clauses) clause_key, rank, other, score
data/greek-nt.parquet 296,820 (29,682 clauses) the same
checks.json the yardsticks below

Clause keys are those of hebrew-clause-layout (clause_key) and greek-nt-clause-layout (first), the keys of clause-columns. score is the cosine of the two clause vectors.

How

A clause vector is the mean of its content words' occurrence vectors (noun, verb, adjective, adverb; all its words when it has none), each text centred on its own mean. Hebrew: BEREL 3.0 on the unpointed Tanakh; Greek: GreBerta on the Greek NT; four layers. The Greek side is mapped onto the Hebrew space (orthogonal Procrustes on 175,608 Tanakh–Septuagint word pairs that MACULA Hebrew aligns), so both tables and tanakh-nt-clauses-similar share one space.

Yardsticks

Check Figure
Chronicles clauses whose best neighbour outside Chronicles is in Samuel–Kings 0.526 (those books hold 0.173 of the clauses: lift 3.04)
Matthew clauses whose best neighbour outside Matthew is in Mark or Luke 0.714 (share 0.279: lift 2.56)
Salkinson's Hebrew NT verse → its Greek NT verse among 7,943 (the cross-language space; no NT pair trains the map) recall@1 0.629, @10 0.877

Similar means close in the encoders' space: shared words and phrasing first, then theme. It is a retrieval aid, not a claim that two clauses mean the same.

Sources and licence

Part Source Licence
Words, clauses, Tanakh–Septuagint alignment MACULA Hebrew and Greek (Clear Bible) CC BY 4.0
Septuagint words (the map) GLAUx Septuagint (Alek Keersmaekers, KU Leuven) CC BY-SA 4.0
Vector models BEREL 3.0 (DICTA), GreBerta (Riemenschneider and Frank) Apache 2.0

CC BY-SA 4.0, because the map is trained on GLAUx.

Downloads last month
42