Towards an Improved Performance Measure for Language Models

In this paper a first attempt at deriving an improved performance measure for language models, the probability ratio measure (PRM) is described. In a proof of concept experiment, it is shown that PRM correlates better with recognition accuracy and can lead to better recognition results when used as the optimisation criterion of a clustering algorithm. Inspite of the approximations and limitations of this preliminary work, the results are very encouraging and should justify more work along the same lines.