Extraction or recognition
A Sanskrit PDF saved from a Unicode document holds its text, so converting it is extraction and comes out close to exact. A scanned Sanskrit PDF holds pictures of pages, so it is recognised first and inherits recognition’s mistakes, described on the Sanskrit OCR page. A scanned page is read with the Hindi or Marathi model unless you convert it with the OCR tool and choose Sanskrit there.
Sanskrit in legacy Devanagari fonts
A great deal of Sanskrit was typeset in the same legacy Devanagari fonts as Hindi: Kruti Dev, DevLys, Chanakya and Shree-Lipi, which store keys rather than letters. Their tables read Sanskrit as they read Hindi. On a 511-page Gita commentary set in Chanakya, 79.5% of the long words converted are in our Devanagari word lists, which are lists of modern Hindi and miss much Sanskrit vocabulary, so the true share is higher.
Verse numbers
A Sanskrit verse ends with its number between double dandas, ।।१।। or ॥१॥. When PDF to Word rebuilds a page’s text, from a scan or from a text layer it has to put back in order, a spacing rule meant for prose used to break that apart into । ।१। ।. It keeps the dandas and the number together now.
What is kept
Sanskrit text, reading order and paragraphs come through, and so do the fonts a digital PDF names, with their sizes, bold, italic and colour; its tables, as Word tables; its pictures; and its running heads, in Word’s header. A PDF whose every page is scanned is rebuilt a page at a time, at the page’s size, with its lines where the scan sets them. Text wrapped round a picture, or a box laid across columns, comes out simpler than the page.