THE IMPACT OF SEMANTIC AMBIGUITY ON BAYESIAN WEIGHTS

Generally, human users provide evidence assessments for intelligent sy stems. Users see or hear data D and determine if it is evidence of the type E or not E. Unfortunately since expert systems often model compl ex and ambiguous processes, there is a probability distribution that D will be categorized as E or not E (E'). This evidence categorization problem is referred to in this paper as semantic ambiguity. The purpos e of this paper is to model the impact of semantic ambiguity in the co ntext of a well-known set of weights. In particular, this paper uses t he Bayesian AL/X weights as the basis of that model. The resulting mod el shows that semantic ambiguity can have a substantial impact on the resulting probabilities. The same approach can be extended to other fo rms of weights on rules or other similar structures.