gesenius-kautzsch-grammar

Gesenius' Hebrew Grammar, as edited and enlarged by E. Kautzsch, 28th German edition (1909), translated by A. E. Cowley — Second English Edition (Oxford: Clarendon Press, 1910; public domain), in the Wikisource collaborative transcription (https://en.wikisource.org/wiki/Gesenius%27_Hebrew_Grammar). This is a PROSE-witness t0 source repo, the standard reference grammar for Biblical Hebrew morphology and syntax cited throughout the NuBerea Hebrew corpus estate — the analogue of allen-greenough-grammar for Hebrew rather than Latin, and of jastrow-lexicon/hebrew-lexicon (BDB) for grammar rather than lexicon.

evidence_status=grammar-author-claim — every row is Gesenius/Kautzsch/ Cowley's own 1910 grammatical analysis, not an independently verified linguistic fact; downstream consumers should treat rules/claims accordingly.

License and Attribution

Content	License	Source
Underlying grammar text (1910, based on Gesenius 1813 / Kautzsch 1909)	Public Domain (US, pre-1929 publication; author + translator + editor all deceased)	Wilhelm Gesenius, ed. E. Kautzsch, tr. A. E. Cowley, Gesenius' Hebrew Grammar, 2nd English ed., Oxford: Clarendon Press, 1910
Wikisource transcription/markup (proofread text, section templates, cross-reference anchors)	CC BY-SA 4.0 (+ GFDL, unversioned, no invariant sections)	https://en.wikisource.org/wiki/Wikisource:Copyright_policy

Wikisource's Copyright policy, quoted verbatim (fetched 2026-08-22):

Unless otherwise noted, all user contributions to Wikisource are released under the [Creative Commons Attribution-ShareAlike 4.0 International] (CC BY-SA 4.0) and the [GNU Free Documentation License], unversioned, with no invariant sections, front-cover texts, or back-cover texts.

Because this repo ships the Wikisource encoding's own transcribed prose (section text, notes, Hebrew examples — the Adapted Material itself, not a fact-grade derivative of it), it is the ShareAlike license home for k: cc-by-sa-4.0 in this lineage: any consumer elsewhere in the estate must join to this table by key (section_id/row_id) rather than re-shipping these rows (ref-cc-by-sa-sharealike-scope, owner ruling 2026-07-03).

Citation

Wilhelm Gesenius, Gesenius' Hebrew Grammar, ed. E. Kautzsch, trans. A. E.
Cowley, 2nd English ed. Oxford: Clarendon Press, 1910. (Public Domain.)
Wikisource transcription: https://en.wikisource.org/wiki/Gesenius%27_Hebrew_Grammar
(CC BY-SA 4.0 / GFDL, contributors per page history).

Completeness

Wikisource's own Index page for this work (Index:Gesenius' Hebrew Grammar (1910 Kautzsch-Cowley edition).djvu) carries the banner "This book has not been completely proofread and the OCR is known not to be very precise with Biblical Hebrew" and a work-level {{incomplete}} template. We measured this two ways rather than trusting the banner alone:

1. Section-existence coverage: every one of the printed table of contents' 167 numbered sections (§1-§167) — including §84, which the book itself splits into §84a/§84b — has a corresponding Wikisource section page that transcludes real content. 167/167 (100%) present. (The task brief's assumed range of §1-§159 undercounts the work: the printed 1910 Contents runs to §167, "Aposiopesis, Anacoluthon, Involved Series of Sentences," followed by unnumbered Paradigms/Indexes; there is no separate lettered "Paradigms" appendix range distinct from §93/§95.)
2. Underlying page-proofreading quality (the real substance of the {{incomplete}} flag): across the 631 underlying djvu Page: namespace pages transcluded into the 168 section pages, ProofreadPage quality levels are: level 1 "Not proofread" 450 (71.3%), level 2 "Problematic" 5 (0.8%), level 3 "Proofread" 159 (25.2%), level 4 "Validated" 17 (2.7%). So while every section's placeholder/skeleton exists, roughly 7 in 10 underlying pages have never been checked against the scan — this is what Wikisource's incomplete flag is actually reporting, and it means prose_text for the majority of rows is uncorrected OCR (Wikisource's own caveat: "the OCR is known not to be very precise with Biblical Hebrew" — treat Hebrew examples especially with caution). See VERIFY.md for the full method and numbers.

Configs / Schema

sections (2,047 rows: one row per §, plus one row per lettered

subsection where Wikisource marks a subsection anchor)

Column	Type	Description
section_id	string	Bare § number, e.g. "116", "84a", "84b" (168 distinct values: §1-§167 plus the §84 split)
subsection	string (nullable)	Single/double lowercase letter (a.."z", occasionally doubled e.g. ee) Wikisource marks within the section body via id="GHGpar-N-x" anchors, or null for the section's own row / sections with no lettered subdivisions
row_id	string	section_id + subsection concatenated, e.g. "20a"; unique row key
part	string (nullable)	Top-level Part per the printed 1910 Contents: "Introduction", "First Part" (Orthography/Phonology), "Second Part" (Etymology/Parts of Speech), "Third Part" (Syntax)
chapter	string (nullable)	Printed Chapter heading within the Part, e.g. "Chapter II. The Verb"
heading	string	Section heading as rendered, e.g. "§20. The Strengthening (Sharpening) of Consonants."
printed_page	string (nullable)	1910 printed page number the section starts on (from the Contents; djvu image = printed page + 24)
prose_text	string	Clean section/subsection text (HTML/wiki markup stripped, Hebrew left as-is); never truncated
notes	string	Bibliographic footnote text belonging to this section (Wikisource's small-font sidenote bibliography paragraphs); pipe-joined when more than one
hebrew_examples	string (JSON list)	Distinct Hebrew example forms (class="he_GHG" / lang="hbo" spans) appearing in this row, order-preserving
scripture_refs	string (JSON list)	Raw regex-matched Biblical-reference-shaped tokens found in the prose (e.g. "Ex 15", "Gn 10"); NOT normalized to Book.ch.v (the grammar's own abbreviation scheme is non-standard and ambiguous without a lookup table — kept raw per the granularity rule rather than mis-normalized)
cross_refs	string (JSON list)	Other §-numbers (optionally with subsection letter, e.g. "§93ee") this row references, extracted from both anchor-linked and bare Hebrew_Grammar/N hrefs
wikisource_page	string	Full Wikisource section-page title fetched
revision_id	int64	MediaWiki revid of the section page at fetch time
fetched_sha256	string	SHA-256 of the raw cached action=parse JSON response for this section page

toc (168 rows, one per printed Contents entry — §1-§167, with §84 split into §84a/§84b)

Column	Type	Description
section_id	string	Full id incl. subletter, e.g. "84a"
base_section	string	Bare section number, e.g. "84"
subletter	string (nullable)	"a"/"b" for the §84 split, else null
title	string	Printed section title
printed_page	string	1910 printed page number
part	string (nullable)	Part heading
chapter	string (nullable)	Chapter heading
roman_division	string (nullable)	Roman-numeral sub-division where the Contents marks one (e.g. "II. The Weak Verb.") — best-effort, see VERIFY.md limitations
letter_division	string (nullable)	Lettered sub-division where the Contents marks one (e.g. "B. Verba Derivativa, or Derived Conjugations.") — best-effort

Parsed once, deterministically, from the printed CONTENTS pages themselves (djvu images 15-20 / printed pages xi-xvi), not reconstructed from memory.

coverage (167 rows, one per expected section number 1-167)

Column	Type	Description
section_number	string	1-167
expected	bool	Always true for this range (all appear in the printed Contents)
present	bool	Whether every id variant for this number (incl. the §84a/§84b split) has a transcluded Wikisource section page with fetched content
subsections_expected	string (JSON list)	id variant(s) expected, e.g. ["84a","84b"]
subsections_present	string (JSON list)	id variant(s) actually fetched
revision_id	string (nullable)	Comma-joined MediaWiki revid(s) for the present variant(s)

Build

build.py is fully deterministic from the locally cached MediaWiki API responses (no network access at build/verify time — the cache was populated once by fetch2.py, a scratch-side helper, at ~1 req/sec against en.wikisource.org/w/api.php, User-Agent identifying NuBerea research): python3 build.py writes all three parquet outputs; python3 build.py --verify rebuilds in-memory and diffs every row/column against the committed parquet (exits non-zero on any mismatch).

Known limitations

• OCR/proofreading quality is uneven (see Completeness above): ~71% of underlying pages are unproofread. Hebrew diacritics/vocalization in prose_text/hebrew_examples should be treated as unverified OCR unless the source page's quality level (recorded per-page in the cache) is 3 ("Proofread") or 4 ("Validated").
• scripture_refs is a raw pattern match, not a normalized reference list. Gesenius' own abbreviation scheme (e.g. "Ex", "Dt", "1 S") only partly overlaps standard OSIS/Book.ch.v abbreviations, and the OCR/markup makes disambiguating verse ranges from page/section citations unreliable without a dedicated abbreviation table; normalization is deferred rather than guessed.
• roman_division/letter_division in toc are a best-effort parse of the printed Contents' nested Roman-numeral/lettered subdivisions (used inside Second/Third Part); a few entries retain a stale division carried over from state-machine tracking across a page boundary. part and chapter (the columns actually required by the task spec) are accurate throughout; treat roman_division/letter_division as supplementary.