“…The pronunciation evaluation problem can then be de®ned as an estimation problem, where we try to estimate, or predict, the value of the ideal human grade h by using a set of predictors. These predictors are the machine scores that we have presented in our previous work (Neumeyer et al, 1996(Neumeyer et al, , 1998Rypa, 1996;Bernstein et al, 1990;Bernstein, 1992;Digalakis, 1992;Digalakis and Murveit, 1994), some of which were summarized in the previous section. We investigated the use of linear and nonlinear regression methods to predict one random variable, the human grade, from a set of others, the machine scores.…”