Maximum a posterJori optimization of parameters and the Laplace approximation for the marginal likelihood are both basis-dependent methods. This note compares two choices of basis for models parameterized by probabilities, showing that it is possible to improve on the traditional choice, the probability simplex, by transforming to the softmax' basis.

Many of the tasks associated with natural language processing (NLP) can be viewed as classification problems. Examples are the computer grading of student writing samples and speech recognition systems. If we accept this view, then the objective of learning classifications from sample text is to classify and predict successfully on new text. While success in the marketplace can be said to be the ultimate test of validation for NLP systems, this success is not likely to be achieved unless appropriate techniques are used to validate the prototype. This paper discusses useful validation techniques for classification-based NLP systems and how these techniques may be used to estimate the true performance of the system.