Is this trivial, hard, impossible?

It is hard (probably impossible) problem but it is probably well studied , and many (universities published) open source solutions exist , which are state of the art and produce good-enough results

Off the top of my head, you might start with Lingua::Wordnet or starts with Lingua::TreeTagger, identify the parts, then look-up each noun in a dictionary of homophones ( say common english homophones for dutch speakers), maybe with Text::Soundex

You'll notice WordNet description doesn't exactly mention homonyms, but that is where a non-linguist like me would start.

See also Perl and Linguistics, Text Analysis Tools to compare Slinker and Stinker?, Natural Language Software Registry, Perl and Linguistics