ABSTRACT
Language is an essential faculty, equipped with human beings for
enabling smooth communication. Modeling the language faculty
of humans has become more intriguing, considering the recent
overwhelming advancements made in artificial intelligence, and is
expected to be inevitable for its further success. In the signal pro-
cessing context, “spoken language processing” embraces multiple
stages of processing, i.e., the feature extraction from raw speech
waveform data, followed by recognition, association, and other
higher-order processing of linguistically meaningful patterns. This
chapter deals with some of such processing, from the recognition of
individually spoken words to the formation of a composite network
representing syntactical structures, under a unified principle of
kernel memory. Kernel memory is a connectionist model, comprised
by a set of units and their connections, as in the ordinary artificial
neural network models, but is heavily based upon localistic repre-
sentation, unlike the well-known multi-layered perceptrons. Within
the context of kernel memory, while the model is represented by a
network structure, the notion of nodes, as well as their connections,
is extended in various aspects, and was originally proposed to
provide a basis for modeling various cognitive modalities. In this
chapter, with a review of some relevant works, a holistic model of
the spoken language processing is proposed based upon the kernel
memory concept.