Your layout is the starting point
IndicPDF does not retype your document. It edits the Word file where it stands, so heading styles, alignment, indents, numbering, table borders and widths, tab stops, the header and the footer are the source’s own and only the words change. In the check we ran, five documents, from a structured notice to a three-page file, showed no difference in paragraph properties, table grids, page geometry or the look of the first run, and the page counts were equal.
A PDF is accepted if it has a text layer. It is converted to Word first, and the Word file is translated.
Amounts, dates and section numbers
Hindi government and legal text is dense with section references, dates, amounts and file numbers, and a translation that alters one is worse than none. Each is replaced with a marker before translation and put back after, and an answer in which a marker is missing is refused. In the Hindi run of our test notice all six protected items were back exactly.
Hindi writes some of them its own way: the amount “Rs 250” came back as “250 रुपये”. That is the engine writing the same amount in Hindi, not a change to the figure.
One edit a Hindi reader asked for
A Hindi reader read a first round of sentences and accepted 5 of 6. The sixth was a table cell reading “12.5% रुपये की छूट”, in which the engine had added “रुपये” (rupees) after a percentage. The right Hindi is “12.5% की छूट”. Nothing in the pipeline catches this, so check the percentages and amounts in a translated table.
Both languages in one file
Official Hindi documents often carry the English beside the Hindi. “Both languages” puts each paragraph’s original before its translation, in the same file, so the two can be checked line against line. Tables are included; headers and footers are translated only.
What it does not take
A scanned PDF with no text layer is not accepted yet. A Hindi document set in a legacy font such as Kruti Dev is refused, not translated from garbled text. Footnotes, endnotes and comments are not translated yet.