Well, this ugly setup (below) does what I want.

If anyone would like to help me clean it, I'd appreciate learning how!

Word1: /^[A-Z0-9\/:"'.-]+$/i {$item[1]}
Word2: /[A-Z0-9\/:"'.-]+/i   {$item[1]}
Term: Word1 | (<skip: ''> Word2 ('_' Word2)(s))
[download]

(Why am I going to all this hassle, you may ask? Words are nested into terms, terms into superterms, and so on, each with a unique delimiter. So the simple regexp 1 token doesn't quite work. rkg