ReadStep 1 of 4 in one reading layer
Add the cues Persian leaves unwritten.
Written Persian leaves out the short vowels you hear, and usually leaves the spoken Ezafe unwritten too. VowelMarks writes those cues back onto the same letters, so nothing is replaced and nothing is rewritten.
As written
گل زیبای باغ قدیمی در بهار شکفت.
With VowelMarks
گُلِ زیبایِ باغِ قَدیمی دَر بَهار شِکُفت.
What this does
- Short vowels return
- Zabar, zir, and pish appear as real combining marks on the original letters.
- Ezafe becomes visible
- The connecting e that links noun to adjective is written in, and one toggle hides it again.
- Context decides ambiguous spellings
- Where one spelling has more than one reading, the surrounding sentence chooses which one to show.
- Every mark can be questioned
- Ask why a mark is there, pick a different reading for one occurrence, or report a result that looks wrong.
Strongest on modern Persian prose. Poetry, difficult literature, and highly colloquial writing remain more challenging, and native-speaker quality review is not complete.
What has been measured
Zero base-text preservation failures in that test.
Measured on held-out gold data.
Measured on 491 usable held-out rows.