What if we compiled a similar token frequency table for non-spam? Could that be “subtracted” from this table for an even more useful table?