While the package makes it's way on to the CPAN, you can also grap a copy over here.Keywords are flagged as being any word, or words, between double quotes which are then looked up in the glossary. If no match is found, the text is left unaltered.
If a match is located, the result is then parsed with Robert Cameron's REX shallow parsing regular expressions. Chunks of balanced markup are then re-inserted into the SAX stream via XML::Filter::Merger. Anything else, including markup not deemed well-formed, is added as character data.