Logic for the New Ai

Research in connected speech recognition has shown that an incredible amount of computation is required to identify spoken words. This is because the contemporary approach begins by isolating and classifying phonemes by means of context-free features of the sound stream. Thus the stream must be reduced to acoustic atoms before classification can be accomplished. On the other hand, for people the context determines the phoneme, much as the melody determines the note.2 This is why a phoneme can vary so greatly in its absolute (i.e., context-free) features and still be recognized. People recognize a gestalt, such as a word or phrase, and identify the phonetic features later if there is some reason to do so (which there’s usually not).