Skip to content

Use Unicode PASHTA for pashta's stress helper - #118

Open
bdenckla wants to merge 1 commit into
openscriptures:masterfrom
bdenckla:pashta-helpers
Open

bdenckla wants to merge 1 commit into
openscriptures:masterfrom
bdenckla:pashta-helpers

Conversation

@bdenckla

Copy link
Copy Markdown
Contributor

Why

A pashta stands on the last letter of its word. When the word is not stressed on its last syllable, a second pashta mark stands on the first letter of the stressed syllable to show where the stress falls: the pashta's stress helper. WLC's Unicode text, as tanach.us published it, writes the helper with the pashta's own code point, HEBREW ACCENT PASHTA (U+0599); in Genesis 1:2, for example, it has תֹ֙הוּ֙. So did wlc/, until 380f4219 (2014-02-04, "Corrected pashta in the middle of a word.") changed the helpers to HEBREW ACCENT QADMA (U+05A8), the code point of a conjunctive accent that looks like the pashta: תֹ֨הוּ֙. Nothing in this repository gives a reason for that departure from WLC. The likeliest is the helper's look, discussed below; but the mark is the pashta's, not a qadma.

Other digital texts write the helper with U+0599 too: UXLC, WLC's successor at tanach.us; Miqra according to the Masorah (MAM) on Hebrew Wikisource, and so this repository's own MAPM/MAPM.xml; and Miqraot Gedolot HaKeter at mgketer.org. The exception among the important editions is Mechon Mamre's, which writes it with U+05A8, as Helmut Richter's Hebrew Cantillation Marks And Their Encoding prescribes. That document is the technical reference that this project's Cantillation page cites, so its reasoning deserves an answer. Richter's reason is the helper's look: the pashta is repeated on the stressed syllable, "where it looks exactly the same as a Qadma". The look is probably the reason for 380f421 too, whose message speaks of the pashta "in the middle of a word". This repository's two web apps set the Hebrew in SBL Hebrew if it is installed, and SBL Hebrew draws U+0599 near the left edge of its letter wherever it stands, the right place for a pashta on its word's last letter but not for the helper, while it draws U+05A8 over the middle of its letter.

Choosing a code point for its look is the logic of the printed page, where only the look matters. A digital text is copied, pasted and processed elsewhere, usually in another font and often by programs that read code points rather than shapes, so its code points should say what each mark is. Where a mark is drawn is the font's business: a font can draw a pashta that is not on its word's last letter over the middle of that letter, by a contextual rule. A program that takes U+05A8 for what Unicode says it is finds a qadma and a pashta in 3,815 words; this repository's two accent catalogs, read/Script/AccentCatalog.js and structure/OshbVerse/Script/AccentCatalog.js, list \u05A8\u0599 as well as \u0599\u0599 as "Pashta", and \u05A4\u05A8\u0599 as well as \u05A4\u0599\u0599 as "Mahpakh with Pashta".

WLC's own history points the same way. The Michigan-Claremont transcription from which WLC descends first coded the helper 63, the code it gives both the qadma and the 'azla, on its policy of coding marks by their shape. WLC's editors gave the helper a code of its own, 33, which WLC's Unicode text writes as U+0599 (§2.6.1 of the supplement to the Michigan-Claremont code manual that comes with WLC). In effect, 380f421 brought back the coding that WLC's editors had given up.

Writing each helper with its accent's own code point is also the only regular rule. Segol, telisha gedolah and telisha qetannah have no lookalike, so their helpers can be written no other way. Elsewhere, only the zarqa's helper is written with a code point other than its accent's, U+0598 beside the zarqa's U+05AE, and only because of the confusion surrounding Unicode's names and descriptions of ZARQA and ZINOR; that is hardly the precedent to follow.

The Decalogues' וְיוֹם

One word in wlc/ has a genuine qadma and a pashta: וְי֨וֹם֙, the first word of Exodus 20:10 and of Deuteronomy 5:14. It is stressed on its last syllable, so its qadma is no helper: the qadma belongs to one of the Decalogues' two cantillations, the עליון, and the pashta to the other, the תחתון. At both places wlc/ has both marks in its main text, with no alternative note, and this pull request leaves them as they are. While the helper is written as a qadma, nothing in the code points tells these two <w> elements apart from the 3,815 helped pashtas; after this change, they are the only <w> elements in wlc/ with a qadma before a pashta. They are also the only words where wlc/ still differs from WLC 4.20 in this respect, since WLC 4.20 codes them 33 then 03, as a helped pashta. WLC 4.22 codes them 63 then 03, a qadma and a pashta, and UXLC 2.6 and MAM have the qadma too.

What changes

In 37 of the 39 books in wlc/ (all but Psalms and Proverbs, which have no pashta), every U+05A8 that stands before a word's one U+0599, on the word's last letter, becomes U+0599, except in the two words above: 3,823 code points.

  • 3,815 <w> elements: 3,794 in the main text and 21 qere readings.
  • In the Decalogues, the 6 <catchWord> elements that repeat such a word (Exodus 20:4, Deuteronomy 5:8 and Deuteronomy 5:15), so that each still equals the word it names, and the 2 alternative readings (<rdg type="x-accent">) of חֶסֶד with such a helper, in Exodus 20:6 and Deuteronomy 5:10.

Nothing else changes. Each file differs from master's only at those characters (in UTF-8, the bytes D6 A8 become D6 99), so no lemma, morph, id or n changes, and no word is added or removed. U+05A8 and U+0599 have the same canonical combining class, 230, so the order of the marks, and what normalization would do with it, are as before. No file outside wlc/ changes: the accent catalogs already give both forms the same names, and index.js has not been regenerated since 2020-06-07, though wlc/ has changed since then.

Known readers

Projects that use wlc/ keep their own copies, so each sees this change only when it updates its copy. A search of the code on GitHub finds a few that depend on the helper's being written as a qadma. Some read U+05A8 as U+0599 when they compare wlc/ with other texts; after this change they need not. One library finds a word's stress by ignoring every U+0599, as postpositive: today it finds these words' stress through the qadma, and after this change it will treat them as it already treats WLC's and UXLC's text.

The web site

hb.openscriptures.org serves the gh-pages branch, last merged from master on 2022-02-08, so the site shows this change only after master is merged into gh-pages again. Its two apps need no change: their accent catalogs, the same files as master's, already read both forms. What changes on the site is the helper's look. The apps set the Hebrew in the first of SBL Hebrew, Ezra SIL SR, Ezra SIL, Cardo, Times New Roman and David that the viewer has installed. In SBL Hebrew, Ezra SIL, Times New Roman and David, the helper written as U+0599 stands in most words near or past the left edge of its letter, as it does in WLC's and UXLC's own text; written as U+05A8, it stands over the middle.

How it was checked

  • Each changed file differs from master's only at its changed characters, and is well-formed XML.
  • On master, 3,817 <w> elements have a qadma before their one pashta and none has two pashtas. After the change, 3,815 have two pashtas and only the two וְיוֹם elements have a qadma before a pashta. The 37 files have 3,823 more U+0599 and 3,823 fewer U+05A8.
  • Compared word by word with WLC 4.20 over all 23,213 verses (WLC 4.20's Michigan-Claremont files, converted to Unicode): on master, 3,811 words differ from WLC 4.20 only in having the qadma for the helper. After the change, 3,809 of the changed words equal WLC 4.20's, and only the two וְיוֹם elements still differ in that way. The other 6 changed words, in Exodus 20:4 and Deuteronomy 5:8 and 5:15, differ from WLC 4.20, before and after, only by marks that wlc/ keeps in its alternative notes. Every other kind of difference from WLC 4.20 has the same count as before.
  • The data validator's change summary should list 3,815 words, each changed in its text alone.

🤖 Generated with Claude Code

Since 380f421 (2014-02-04), wlc/ has written the pashta's stress
helper (the second pashta mark, on the stressed syllable) as HEBREW
ACCENT QADMA (U+05A8). WLC, from which wlc/ is taken, writes it as
HEBREW ACCENT PASHTA (U+0599), and so did wlc/ before that commit.
This writes each such helper as U+0599 again: 3,823 code points
in 37 books, in 3,815 <w> elements, 6 catchWords and 2
alternative readings. Nothing else changes.

The Decalogues' וְיוֹם (Exodus 20:10, Deuteronomy 5:14) keeps its
qadma, which is no helper: there the qadma is one cantillation's and
the pashta the other's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@bdenckla bdenckla changed the title Write the pashta's stress helper as HEBREW ACCENT PASHTA, as WLC does Use Unicode PASHTA for pashta's stress helper Nov 15, 2026
bdenckla added a commit to bdenckla/MAM-basics that referenced this pull request Nov 16, 2026
Ben, 2026-11-16: "put this work "on ice": git-track it for deep
storage, perhaps never to return to it, but perhaps we will."

The pull request to openscriptures/morphhb that gives wlc/'s 24
dual-cantillation verses one scheme was prepared on 2026-11-15 and
not posted. In it the main text is CLC's תחתון, with labelled notes
for CLC's עליון and for MAM-informed versions where CLC's is not
legal. The 14 decisions on its decision page are open.

- doc/mam-morphhb-pull-requests/decalogue-strands-2026-11-15/:
  README.md, the record and how to resume; change.diff, the draft
  against openscriptures/morphhb#118's head 696e21ce, which gives
  the draft byte for byte under git -c core.autocrlf=false apply;
  work-folder.zip, the untracked work folder (scripts, reports,
  data, decision page, description, base files) less what can be
  regenerated, every entry checked against its source.
- doc/mam-morphhb-pull-requests.md: item 6 and the State line.
- .gitattributes: the archive's path is binary.

Checks: git diff --check; the nine tests that scan tracked
filenames, text and Python (test_tracked_filenames,
test_h_dot_below_nfc, test_prose_mark_order, test_cgj_between_marks,
test_receipt_update_links, test_no_machine_paths_in_artifacts,
test_worktree_retirement_citations, test_repo_visibility_declared,
test_stderr_error_handler): 41 passed. Not run: the mega and the
rest of the suite, since no generator and no other test reads these
files, and the .gitattributes rule names only this archive's path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant