The paper illustrates a linguistic knowledge acquisition model making use of data types, innite memory, and an inferential mechanism for inducing new information from known data. The model is compared with standard stochastic methods applied to data tokens, and tested on a task of lexico{semantic classi cation.