A Formal Basis for Performance Evaluation of Natural Language Understanding Systems