Check that the Hindi is Unicode
Hindi government and newspaper files are very often set in Kruti Dev, DevLys or Chanakya. They look right and copy as text like “Hkkjr”. IndicPDF refuses such a document instead of translating the symbols. Convert it first with the Kruti Dev converter; the Unicode Word file it writes can then be translated.
The Word file is edited in place
The English replaces the Hindi inside the original file, so paragraph styles, alignment, indents, tables, numbering, header and footer stay the file’s own. We compared five translated documents element by element with their sources and found no difference in paragraph properties, table grids, page geometry or the look of the first run; the page counts were equal, at 1, 1, 1, 1 and 3.
Numbers and dates survive the trip
Dates, section numbers (धारा 24(2)), amounts, file numbers, e-mail addresses and web addresses are protected during translation and checked afterwards. In the Hindi-to-English run of our test notice all 6 of 6 came back exactly.
How close the round trip is
Translated into Hindi and back, our test notice recovered about nine in ten of its English words, with “deadline” for “last date” and “assistance” for “help”. These are the small differences of any machine translation, and readable. The Hindi reader accepted the one Hindi-to-English sentence in the first round; that is a single sentence, not a measure of quality.
What it does not take
A scan with no text layer is not accepted yet: read it with Hindi OCR first. Footnotes, endnotes and comments are not translated yet, and a sentence that cannot be translated stays in Hindi, highlighted yellow.