A MUTUAL-INFORMATION-BASED APPROACH TO RULE QUANTIFICATION

The rule quantification problem is the key to a hybrid language model which integrates grammar rules into a statistical language model. This paper discusses a new type of hybrid language model, from which the concept of rule combination degree is derived, and then presents a mutual\|information\|based approach to rule quantification. The experiments show that this approach is a valuable guide for both the selection of system rule base and the combination mechanism of rule and statistic.