Zarka is not U+0598

If you are writing software that reads Hebrew cantillation and you picked the zarka accent by its Unicode name, it is wrong. This is the trap, how often it bites, and how to check that you have fixed it.

The short version

• In the twenty-one prose books — which includes the whole Torah — zarka is U+05AE, the character Unicode names HEBREW ACCENT ZINOR.
• U+0598, the character Unicode names HEBREW ACCENT ZARQA, is a different mark. In the Torah it is the same zarka written a second time, on a stressed letter that is not the last one.
• Resolve both to zarka. Match on the codepoint, never on the Unicode name.

Unicode says so itself

This is not a reading anyone has to argue for. The Unicode Consortium publishes a list of character names it considers wrong — Technical Note #27, Known Anomalies in Unicode Character Names — and both of these are on it.

Character names in Unicode are immutable — once assigned, a name is never changed, because too much depends on it. So this will not be corrected in the standard. It can only be worked around.

U+05AE HEBREW ACCENT ZINOR — Should have been called Hebrew accent zarqa (= tsinor).

Unicode Technical Note #27, Known Anomalies in Unicode Character Names

U+0598 HEBREW ACCENT ZARQA — Perhaps should have been called Hebrew accent tsinnorit. May also be used for zarqa when shown on accented non-final letter.

Unicode Technical Note #27, Known Anomalies in Unicode Character Names

How often it bites

Counted across all 5,846 verses of the Torah, in Miqra according to the Masorah:

Two facts make the identification certain rather than probable. Every one of the 372 verses containing U+05AE also contains a segol (U+0592) — and zarka is defined as the mark that precedes segol. And every one of the 59 occurrences of U+0598 sits on a word that already carries U+05AE, never on a word without one.

So U+0598 is not a second accent competing with zarka. It is part of the same one.

Why the mark is written twice

Zarka is postpositive: it is written after the final letter of its word, no matter which syllable carries the stress. That is fine when the stress is final and misleading when it is not — the reader would put the accent in the wrong place.

So when the stress falls earlier, the mark is written a second time, on the stressed letter. That second mark is U+0598, exactly as Unicode's note anticipates. Two examples from Genesis, both stressed on the first syllable:

וַיֹּ֘אמֶר֮ — Genesis 2:23, vaYOmer. טֶ֘רֶם֮ — Genesis 19:4, TErem.

One mark, one accent, two characters. Software that treats them as two different te'amim will report a mark that is not there.

What breaks if you get it wrong

Two things, and only the first one is obvious.

Identification fails. A lookup keyed on U+0598 misses every zarka whose word is stressed on the last syllable — which is most of them. In our case 312 of 375 zarka words returned nothing at all.

Phrase boundaries move. This is the one that is easy to miss, because nothing errors and nothing looks broken. Zarka is a disjunctive: it ends a phrase. If it is not recognized as one, no boundary is placed there, and every segol group in the text gets folded into the phrase that follows it. Any feature built on phrase structure — coloring, segmentation, an alignment model, a teaching aid that shows where to breathe — is then quietly wrong on those verses, and it will pass every test you thought to write.

How to test it

Genesis 1:7 is the shortest useful check. Its second word carries a zarka and the segol follows on the word after it, so one verse exercises both the identification and the phrase boundary.

// Genesis 1:7, second word. Keyed on U+0598, this is unnamed.
const word = '\u05D0\u05B1\u05DC\u05B9\u05D4\u05B4\u05D9\u05DD\u05AE';
identify(word);            // must be Zarka

// Genesis 2:23. Both characters, ONE mark — not two.
const doubled = '\u05D5\u05B7\u05D9\u05BC\u05B9\u0598\u05D0\u05DE\u05B6\u05E8\u05AE';
identify(doubled);         // must also be Zarka, exactly once

// And the boundary: the phrase must END on the zarka word,
// not run on into the segol that follows.

If you have a corpus to hand, the stronger check is the co-occurrence one: count the verses containing U+05AE and the verses containing both U+05AE and U+0592. If those two numbers are not equal, something upstream is wrong before your code ever sees the text.

Two more traps in the same table

Both of these bit us in the same pass, and both come from the same root cause — building a Torah te'amim table from a list that covers the whole Bible, or from a list of characters rather than of marks.

• Munach legarmeh is not a character. It is a munach (U+05A3) on a word that is followed by a paseq (U+05C0), and it is a disjunctive — the opposite of the plain conjunctive munach the character alone reports. In this text paseq is always its own whitespace-separated token (387 of 387) and 325 of those follow a munach. A per-character scan calls all 325 of them a plain munach.
• Geresh muqdam (U+059D) does not occur in the Torah. Not once in 5,846 verses. It belongs to Psalms, Proverbs and Job. If it is in your Torah table, the table came from a list of the whole Bible — and the same is true of ole, iluy, dehi and atnah hafukh.

Method

Every figure here was counted directly over the Hebrew text this site ships: all 5,846 verses of the Torah in Miqra according to the Masorah, a scholarly Masoretic edition published via Sefaria and used under CC BY-SA. Counts are of characters in the pointed text, and verse-level co-occurrence is computed per verse, not per word.

Different editions point differently, so if you are working from another text, measure yours rather than trusting these numbers. The method is the transferable part: count the characters, check the co-occurrence, and do not trust the names.

This came out of fixing it in our own reader, where zarka had been keyed on U+0598 since the first version and nobody noticed, because nothing errors — the marks were simply never named and the phrases were quietly grouped wrong.

The trope marks  ·  Printable trope chart  ·  All Torah readings