To summarize the answer, the parser rule ('NOP' | 'XCHG' 'AX,' 'AX') was trying to match 'AX' which was already being handled by the lexer under the REG rule.