I'd probaby use a hand-written front-end to identify everything inside a comment and pass the resulting strings to ANTLR. I'd then do the rest of the lexing and parsing using ANTLR (i.e. the front end doesn't need to tokenize the input). ...Richard