Very often PDF files in non-Latin scripts (especially RTL scripts such as Arabic, Hebrew and Farsi) are generated by software which sort of LTR-ifies the text at the word or sentence-fragment level, or just somehow gets the right glyphs to display but you get gibberish for the 'logical' text. In these cases there is very little to be done except write a custom back-converter which is effectively not an option.
However, if you can figure out how the file is created - which is often indicated in the meta-data accessible using common PDF readers - there might be an option to open the file in the application which generated it, or at least you could make your question more specific.