Memory-based Reasoning Applied to English Pronunciation

Memory-based Reasoning is a paradigm for AI in which best-match recall from memory is the primary inference mechanism. In its simplest form, it is a method of solving the inductive inference (learning) problem. The primary topics of this paper are a simple memory-based reasoning algorithm, the problem of pronouncing english words, and MBRtalk, a program which uses memory-based reasoning to solve the pronunciation problem. Experimental results demonstrate the properties of the algorithm as training-set size is varied, as distracting information is added, and as noise is added to the data.