larsgson commited on
Commit
32beff0
·
verified ·
1 Parent(s): 4bf530d

clause-columns v1.2: Greek subj fixes (null argument id; own subject word fallback) from the S2 audit (code 0d293bf)

Browse files
Files changed (5) hide show
  1. README.md +11 -2
  2. SHA256SUMS +4 -4
  3. build.json +13 -13
  4. checks.md +20 -20
  5. data/greek-nt.parquet +2 -2
README.md CHANGED
@@ -10,7 +10,7 @@ configs:
10
  data_files: data/greek-nt.parquet
11
  ---
12
 
13
- # Clause columns (v1.1)
14
 
15
  Grammar and discourse facts for every clause of the two bcv-commons clause layouts, keyed by their clause keys: the main verb and its
16
  form, voice, stem and person; negation and its scope; force; quotation and speaker; the narrative line; actor and undergoer with the
@@ -26,6 +26,14 @@ layouts' own columns (rule, relation, parent, depth and so on) are not copied he
26
 
27
  ## What changed from the previous version
28
 
 
 
 
 
 
 
 
 
29
  **v1.1:** the Septuagint word is now looked up in its own verse through the verse map (`macula/lxx_verses.py`). GLAUx numbers the
30
  Septuagint's way (Psalms one off, Jeremiah 25-51 reordered, Daniel 3-4 with the Greek additions), so the lookup found its verse
31
  for 0.785 of Hebrew words before (Psalms 0.13, Jeremiah 0.45) and 0.867 now; the rest fall back to the form's commonest parse, as
@@ -119,6 +127,7 @@ row by row. A human review of samples is planned for later and will be reported
119
 
120
  ## Licence
121
 
122
- CC BY-SA 4.0: the Hebrew layout's clause keys are CC BY-SA 4.0, and so are GLAUx's parses. Credit: MACULA (Clear Bible / Biblica),
 
123
  GLAUx (KU Leuven, Alek Keersmaekers), bcv-commons (the layouts, the speaker layer), make-tanakh-research (the column design and
124
  the previous version), and this dataset.
 
10
  data_files: data/greek-nt.parquet
11
  ---
12
 
13
+ # Clause columns (v1.2)
14
 
15
  Grammar and discourse facts for every clause of the two bcv-commons clause layouts, keyed by their clause keys: the main verb and its
16
  form, voice, stem and person; negation and its scope; force; quotation and speaker; the narrative line; actor and undergoer with the
 
26
 
27
  ## What changed from the previous version
28
 
29
+ **v1.2:** two Greek `subj` defects, found by a blind audit of 100 clauses per language (2026-11-17), are fixed; nothing else
30
+ changes and Hebrew is identical. (1) MACULA Greek marks an unexpressed argument with the id `00000000000`; where a verb has no
31
+ subject reference, `subj` took it from the frame's A0 as if it were a word (1,223 clauses in v1.0-v1.1). It is skipped now. (2) A
32
+ clause's own subject word (tree role `s`: noun, pronoun, adjective, numeral) is the last fallback when MACULA gives no subject
33
+ reference or A0; Greek clauses with no subject go from 6,791 to 3,673. Greek `subj` changes on 5,522 clauses and `subj_switch`
34
+ on 6,659. On the audit sample, Greek `subj` agreement where the judge could decide went from 0.44 to 0.62 (shaped on that
35
+ sample, so optimistic); Hebrew 0.75. Main verb 0.97 / 1.00, negator 1.00, negation scope 0.98, subject switch 0.93.
36
+
37
  **v1.1:** the Septuagint word is now looked up in its own verse through the verse map (`macula/lxx_verses.py`). GLAUx numbers the
38
  Septuagint's way (Psalms one off, Jeremiah 25-51 reordered, Daniel 3-4 with the Greek additions), so the lookup found its verse
39
  for 0.785 of Hebrew words before (Psalms 0.13, Jeremiah 0.45) and 0.867 now; the rest fall back to the form's commonest parse, as
 
127
 
128
  ## Licence
129
 
130
+ CC BY-SA 4.0, because of GLAUx's parses (`lxx_tense`, `lxx_mood`). The Hebrew layout's clause keys are CC BY 4.0 (its
131
+ segmentation is; see the layout's card), and every other column comes from CC BY 4.0 sources. Credit: MACULA (Clear Bible / Biblica),
132
  GLAUx (KU Leuven, Alek Keersmaekers), bcv-commons (the layouts, the speaker layer), make-tanakh-research (the column design and
133
  the previous version), and this dataset.
SHA256SUMS CHANGED
@@ -1,5 +1,5 @@
1
- fbb9a5b614101c4b20e7e7d9732e2842a4fc14d0d11bf16e27af78f0b48643a8 README.md
2
- 83e8475481fbadb344168847d62b6f59fda5c367b83ecdd06aec8829de90f296 build.json
3
- 27d6fe882d9515919104e39aab298a25a69e7d170185bd7a832da014086b4bd1 checks.md
4
- 7ceccf98690b1a3938d63d7e5932df9a0e2d90dbadaafaf1d22c57f77656b3ab data/greek-nt.parquet
5
  bc2e8159bb1b27008e78b320f550958e32b42ef1565c5f9fd3214a72a3f1df3c data/hebrew.parquet
 
1
+ ca402253857c396e57fddf286aab73ffb14de25a91990151dcfa97b14cf8b3c1 README.md
2
+ b11cd23cafcb8e0c8aaa4282333c6dc2710c9526a58e902a8b94d33aad5e7ff9 build.json
3
+ 4fef03b9b3489544154c45b255242778195d9c1a148bc590a5ea27be77433430 checks.md
4
+ d9ec3ce0ef3ae84351e001b4e23762c7a1d6067008f32a60918d40a8523e00db data/greek-nt.parquet
5
  bc2e8159bb1b27008e78b320f550958e32b42ef1565c5f9fd3214a72a3f1df3c data/hebrew.parquet
build.json CHANGED
@@ -1,12 +1,12 @@
1
  {
2
  "schema": 1,
3
- "version": "v1.1",
4
- "built": "2026-11-14",
5
- "code": "4ee0442",
6
  "builder": {
7
  "file": "macula/build_clause_columns.py",
8
- "sha256": "593033fbfc3e50e7df00e4db219b09fb52b0539776f9202784b38bfee74fb37c",
9
- "last_commit": "5577bef",
10
  "contract_check": {
11
  "file": "macula/clause_columns_contract.py",
12
  "sha256": "e86d76ecad2ef42b0d6052e3001a791745f3d36239e6d9126b8a930899a1783b"
@@ -34,7 +34,7 @@
34
  "built": "2026-10-27",
35
  "clauses_sha256": "fad419bbdbcf97b4915560778034a308a07bb2e72dd2a2f6d1854b66c3e164a3",
36
  "release": "same clause keys as v0.37 (checked)",
37
- "licence": "CC BY-SA 4.0 (the release; its keys key this table)"
38
  },
39
  "greek-nt-clause-layout": {
40
  "path": "macula/data/nt-clause-layout",
@@ -259,7 +259,7 @@
259
  "undergoer": 29682,
260
  "undergoer_basis": 14717,
261
  "args": 23324,
262
- "subj": 22891,
263
  "subj_switch": 29682,
264
  "neg_scope": 29682,
265
  "subj_form": 29682,
@@ -301,9 +301,9 @@
301
  "implicit": 1479
302
  },
303
  "subj_switch": {
304
- "switch": 13943,
305
- "same": 8688,
306
- "none": 6791,
307
  "first": 260
308
  },
309
  "subj_form": {
@@ -349,9 +349,9 @@
349
  }
350
  },
351
  "files": {
352
- "README.md": "fbb9a5b614101c4b20e7e7d9732e2842a4fc14d0d11bf16e27af78f0b48643a8",
353
- "checks.md": "27d6fe882d9515919104e39aab298a25a69e7d170185bd7a832da014086b4bd1",
354
- "data/greek-nt.parquet": "7ceccf98690b1a3938d63d7e5932df9a0e2d90dbadaafaf1d22c57f77656b3ab",
355
  "data/hebrew.parquet": "bc2e8159bb1b27008e78b320f550958e32b42ef1565c5f9fd3214a72a3f1df3c"
356
  }
357
  }
 
1
  {
2
  "schema": 1,
3
+ "version": "v1.2",
4
+ "built": "2026-11-18",
5
+ "code": "0d293bf",
6
  "builder": {
7
  "file": "macula/build_clause_columns.py",
8
+ "sha256": "d3cef3a0a450733bca71d3645d5c04fc19f5685820502d70767e34b85f6b1a80",
9
+ "last_commit": "307099b",
10
  "contract_check": {
11
  "file": "macula/clause_columns_contract.py",
12
  "sha256": "e86d76ecad2ef42b0d6052e3001a791745f3d36239e6d9126b8a930899a1783b"
 
34
  "built": "2026-10-27",
35
  "clauses_sha256": "fad419bbdbcf97b4915560778034a308a07bb2e72dd2a2f6d1854b66c3e164a3",
36
  "release": "same clause keys as v0.37 (checked)",
37
+ "licence": "CC BY 4.0 (the clause keys: the segmentation is CC BY; the release as a whole is CC BY-SA)"
38
  },
39
  "greek-nt-clause-layout": {
40
  "path": "macula/data/nt-clause-layout",
 
259
  "undergoer": 29682,
260
  "undergoer_basis": 14717,
261
  "args": 23324,
262
+ "subj": 26009,
263
  "subj_switch": 29682,
264
  "neg_scope": 29682,
265
  "subj_form": 29682,
 
301
  "implicit": 1479
302
  },
303
  "subj_switch": {
304
+ "switch": 17099,
305
+ "same": 8650,
306
+ "none": 3673,
307
  "first": 260
308
  },
309
  "subj_form": {
 
349
  }
350
  },
351
  "files": {
352
+ "README.md": "ca402253857c396e57fddf286aab73ffb14de25a91990151dcfa97b14cf8b3c1",
353
+ "checks.md": "4fef03b9b3489544154c45b255242778195d9c1a148bc590a5ea27be77433430",
354
+ "data/greek-nt.parquet": "d9ec3ce0ef3ae84351e001b4e23762c7a1d6067008f32a60918d40a8523e00db",
355
  "data/hebrew.parquet": "bc2e8159bb1b27008e78b320f550958e32b42ef1565c5f9fd3214a72a3f1df3c"
356
  }
357
  }
checks.md CHANGED
@@ -28,8 +28,8 @@ Ours: `macula/build_clause_columns.py` (this folder's hebrew.parquet, greek-nt.p
28
  | hebrew | `subj` | 87,147 | 86,186 | 0.9890 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
29
  | hebrew | `subj_switch` | 87,147 | 85,879 | 0.9854 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
30
  | hebrew | `lxx_form` | 87,147 | 86,198 | 0.9891 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; 2 clauses: MACULA revision (our store has no Septuagint form there) |
31
- | hebrew | `lxx_tense` | 87,147 | 86,157 | 0.9886 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; plus differs by design: the parse of the form in its own GLAUx verse first (lxx_basis verse) |
32
- | hebrew | `lxx_mood` | 87,147 | 85,926 | 0.9860 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; plus differs by design: the parse of the form in its own GLAUx verse first (lxx_basis verse) |
33
  | hebrew | `neg_scope` | 87,147 | 87,061 | 0.9990 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
34
  | hebrew | `subj_form` | 87,147 | 86,406 | 0.9915 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
35
  | hebrew | `part` | 87,147 | 86,392 | 0.9913 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
@@ -84,7 +84,7 @@ Ours: `macula/build_clause_columns.py` (this folder's hebrew.parquet, greek-nt.p
84
  | greek-nt | `layout.parent` | 29,682 | 29,682 | 1.0000 | equal |
85
  | greek-nt | `layout.depth` | 29,682 | 29,682 | 1.0000 | equal |
86
 
87
- Columns of ours with no counterpart in research's table (not compared): Hebrew `speaker_id` (the layer's character id, as the Greek layout uses), `lxx_basis` (verse | form); Greek `args`, `subj`, `subj_switch` (from the Greek frames and references). Research columns not rebuilt here: copies of layout columns (`rule`, `relation`, `discourse`, `via`, `role`, `junction`, `parent`, `depth`: checked above as layout identity), morphology (`stem`, `person`, `number`, `gender`, `strong`, `causative`, `interrog`: in the spine), `force` / `legal_future` (the layouts' own `force` codes), `juncture` / `nexus` (wave 2, clause relation), `akt` / `akt_p` / `juss_p` (Jev model answers: model layer, not source facts), `heb_form` (research's private Delitzsch alignment and tagger).
88
 
89
  ## Disagreements
90
 
@@ -278,39 +278,39 @@ Columns of ours with no counterpart in research's table (not compared): Hebrew `
278
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,145 of 87,147 (1.0000)
279
  - 948 of 949 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
280
 
281
- ### hebrew `lxx_tense`: 990 disagreements
282
 
283
  | ours / research | clauses | examples |
284
  |---|---|---|
285
  | aorist / ∅ | 461 | 010010110031 010010210011 010020220011 |
286
  | future / ∅ | 156 | 010060070031 010080210101 010090020011 |
287
- | aorist / future | 86 | 010090050011 010090050061 010090050101 |
288
- | present / ∅ | 73 | 010010120091 010090120061 010300290031 |
289
- | future / aorist | 47 | 010200070151 010500050181 020100280121 |
290
- | imperfect / ∅ | 35 | 010090180011 010250270031 010360070071 |
291
- | future / present | 30 | 020230110071 020330030011 030060090031 |
292
- | perfect / ∅ | 23 | 010010290041 020090180081 040180210011 |
293
 
294
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,147 of 87,147 (1.0000)
295
- - 785 of 990 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
296
- - disagreements by our basis: verse 758 of 37278, form 221 of 13284, ∅ 11 of 36585
297
 
298
- ### hebrew `lxx_mood`: 1,221 disagreements
299
 
300
  | ours / research | clauses | examples |
301
  |---|---|---|
302
  | indicative / ∅ | 555 | 010010210011 010010290041 010020220011 |
303
- | imperative / indicative | 142 | 010120130011 010180050041 010200130151 |
304
- | subjunctive / indicative | 109 | 010090050011 010090050061 010090050101 |
305
  | imperative / ∅ | 101 | 010010110031 010130140121 020330010061 |
306
- | indicative / subjunctive | 47 | 010200070151 010500050181 020100280121 |
307
- | indicative / imperative | 39 | 020130040011 030100090011 030200230011 |
 
308
  | subjunctive / ∅ | 37 | 010320330011 020070150081 020100020011 |
309
- | optative / infinitive | 36 | 010160050161 010280030061 010480200091 |
310
 
311
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,147 of 87,147 (1.0000)
312
- - 791 of 1221 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
313
- - disagreements by our basis: verse 988 of 37278, form 222 of 13284, ∅ 11 of 36585
314
 
315
  ### hebrew `neg_scope`: 86 disagreements
316
 
 
28
  | hebrew | `subj` | 87,147 | 86,186 | 0.9890 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
29
  | hebrew | `subj_switch` | 87,147 | 85,879 | 0.9854 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
30
  | hebrew | `lxx_form` | 87,147 | 86,198 | 0.9891 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; 2 clauses: MACULA revision (our store has no Septuagint form there) |
31
+ | hebrew | `lxx_tense` | 87,147 | 86,133 | 0.9884 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; plus differs by design: the parse of the form in its own GLAUx verse first (lxx_basis verse) |
32
+ | hebrew | `lxx_mood` | 87,147 | 85,874 | 0.9854 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000; plus differs by design: the parse of the form in its own GLAUx verse first (lxx_basis verse) |
33
  | hebrew | `neg_scope` | 87,147 | 87,061 | 0.9990 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
34
  | hebrew | `subj_form` | 87,147 | 86,406 | 0.9915 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
35
  | hebrew | `part` | 87,147 | 86,392 | 0.9913 | differs by error (research's): a clause the layout splits into several segments (2,263 keys) is read from one segment (verb from the last, word order from the first); research's reading replicated from our stores agrees 1.0000 |
 
84
  | greek-nt | `layout.parent` | 29,682 | 29,682 | 1.0000 | equal |
85
  | greek-nt | `layout.depth` | 29,682 | 29,682 | 1.0000 | equal |
86
 
87
+ Columns of ours with no counterpart in research's table (not compared): Hebrew `speaker_id` (the layer's character id, as the Greek layout uses), `lxx_basis` (verse | form); Greek `args`, `subj`, `subj_switch` (from the Greek frames and references). Research columns not rebuilt here: copies of layout columns (`rule`, `relation`, `discourse`, `via`, `role`, `junction`, `parent`, `depth`: checked above as layout identity), `legal_future` (a reading, left out of the layouts' force on purpose), `juncture` / `nexus` (wave 2, clause relation), `akt` / `akt_p` / `juss_p` (Jev model answers: model layer, not source facts), `heb_form` (research's private Delitzsch alignment and tagger).
88
 
89
  ## Disagreements
90
 
 
278
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,145 of 87,147 (1.0000)
279
  - 948 of 949 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
280
 
281
+ ### hebrew `lxx_tense`: 1,014 disagreements
282
 
283
  | ours / research | clauses | examples |
284
  |---|---|---|
285
  | aorist / ∅ | 461 | 010010110031 010010210011 010020220011 |
286
  | future / ∅ | 156 | 010060070031 010080210101 010090020011 |
287
+ | aorist / future | 93 | 010090050011 010090050061 010090050101 |
288
+ | present / ∅ | 71 | 010010120091 010090120061 010300290031 |
289
+ | future / aorist | 52 | 010200070151 010500050181 020100280121 |
290
+ | imperfect / ∅ | 36 | 010090180011 010250270031 010360070071 |
291
+ | future / present | 31 | 020230110071 020330030011 030140280011 |
292
+ | perfect / ∅ | 24 | 010010290041 020090180081 040180210011 |
293
 
294
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,147 of 87,147 (1.0000)
295
+ - 787 of 1014 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
296
+ - disagreements by our basis: verse 853 of 42287, form 150 of 8275, ∅ 11 of 36585
297
 
298
+ ### hebrew `lxx_mood`: 1,273 disagreements
299
 
300
  | ours / research | clauses | examples |
301
  |---|---|---|
302
  | indicative / ∅ | 555 | 010010210011 010010290041 010020220011 |
303
+ | imperative / indicative | 148 | 010120130011 010180050041 010200130151 |
304
+ | subjunctive / indicative | 120 | 010090050011 010090050061 010090050101 |
305
  | imperative / ∅ | 101 | 010010110031 010130140121 020330010061 |
306
+ | indicative / subjunctive | 53 | 010200070151 010500050181 020100280121 |
307
+ | optative / infinitive | 44 | 010160050161 010280030061 010480200091 |
308
+ | indicative / imperative | 43 | 020130040011 030100090011 030200230011 |
309
  | subjunctive / ∅ | 37 | 010320330011 020070150081 020100020011 |
 
310
 
311
  - replica of research's reading (split clauses read as research reads them, layer's last row per key, Septuagint parse by form only), from our stores: agrees on 87,147 of 87,147 (1.0000)
312
+ - 791 of 1273 disagreements are on clause keys the layout splits into several segments (2263 of 87147 joined keys are such)
313
+ - disagreements by our basis: verse 1112 of 42287, form 150 of 8275, ∅ 11 of 36585
314
 
315
  ### hebrew `neg_scope`: 86 disagreements
316
 
data/greek-nt.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7ceccf98690b1a3938d63d7e5932df9a0e2d90dbadaafaf1d22c57f77656b3ab
3
- size 973582
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d9ec3ce0ef3ae84351e001b4e23762c7a1d6067008f32a60918d40a8523e00db
3
+ size 993500