How do lexical tie-ins work?

[Zachary Carter]

I'm curious on how Bison's algorithm works wrt lexical tie-ins[1]. The
situation appears to me as a bit of a catch-22. The parser checks the
lookahead token to decide when to reduce (right?) but the semantic action
may alter the scanner so that the next token shifted is actually different
than what the lookahead token had been. It seems counter intuitive, but is
that really how it works?

[1]: http://www.gnu.org/software/bison/manual/bison.html#Lexical-Tie_002dins
-- 
Zach Carter