It would make more sense to just use IRT for grading the responses than trying to add more complexity to the answers themselves.