Teach the lexer to match and return specific Fortran keywords as tokens.
Update the parser to use these instead of always using a WORD token and
then checking the text.  This avoids extra string comparisons and will
allow more grammar productions to be unambiguously added later for
additional Fortran statements.