Lexical analysis and parallel prefix sums anyone?

Ive read a few times in articles about cuda and parallel prefix sum operations.

One thing I’ve always read about parallel prefix sum operations…is that they can be used as building blocks to implement lexical analyzers. I was wondering if anyone might have some idea of how one might do this (theoretically or actually), that is, to implement a lexical analyzer using parallel prefix sum operations…with the ultimate goal (for me at least) is of course to implement this for cuda.

Any pointers you might have…or links to interesting articles would be welcome.