The job just can't be done automatically: DVI, PostScript and PDF are "final" formats, supposedly not susceptible to further editing - information about where things came from has been discarded. So if you've lost your (La)TeX source (or never had the source of a document you need to work on) you've a serious job on your hands. In many circumstances, the best strategy is to retype the whole document, but this strategy is to be tempered by consideration of the size of the document and the potential typists' skills.
If automatic assistance is necessary, it's unlikely that any more than text retrieval is going to be possible; the (La)TeX markup that creates the typographic effects of the document needs to be recreated by editing.
If the file you have is in DVI format, many of the techniques for converting (La)TeX to ASCII are applicable. Consider dvi2tty, crudetype and catdvi. Remember that there are likely to be problems finding included material (such as included PostScript figures, that don't appear in the DVI file itself), and mathematics is unlikely to convert easily.
To retrieve text from PostScript files, the ps2ascii tool (part of the ghostscript distribution) is available. One could try applying this tool to PostScript derived from an PDF file using pdf2ps (also from the ghostscript distribution), or Acrobat Reader itself; an alternative is pdftotext, which is distributed with xpdf.
Another avenue available to those with a PDF file they want to process is offered by Adobe Acrobat (version 5 or later): you can tag the PDF file into an estructured document, output thence to well-formed XHTML, and import the results into Microsoft Word (2000 or later). From there, one may convert to (La)TeX using one of the techniques discussed in "converting to and from (La)TeX".
The result will typically (at best) be poorly marked-up. Problems may also arise from the oddity of typical TeX font encodings (notably those of the maths fonts), which Acrobat doesn't know how to map to its standard Unicode representation.