| This research encompasses the design, development, and operation of a working artificial neural network model that provides for the acquisition, recall, and generation of temporally ordered information. These features are realized via a synergistic combination of subsystems that enable predictive learning, sequence interleaving, and new sequence creation. Together, these features provide a medium for demonstrating a variety of cognitive processes typical of intelligent behavior. Sequence learning is implemented as a Hebbian-based error-correcting paradigm that allows for rapid real-time training with minimum catastrophic interference and support for high-capacity sequence storage. Basic predictive learning is then used to illustrate common functions such as alphabet mastery, spelling, acquisition of mathematical facts, memorization of a script, basic motor skills, traditional associative memory, and an ability to form multiple associations with a single stimulus. Learned sequences can be interleaved to provide connectionist-based demonstrations of free association, transcription, route following, memory theatres, multiple trains of thought, complex motion, and a limited form of rehearsal. New sequences can then be crafted from previously learned sequences by using a generalized form of variable binding and can be used to help explicate such cognitive tasks as counting, solving mathematical expressions based on well-learned number facts, understanding simple pronoun referents in sentences, protolanguage reading comprehension, rule formation and the development of commonsense knowledge via inductive reasoning, the acquisition and deployment of external memory strategies, and a sophisticated nonstereotypical sequence-processing capability. By exhibiting such proficiency in a single temporally oriented network, this model takes a significant step toward the development of autonomous artificial systems capable of manifesting many of the characteristics that exemplify intelligence. In addition, the model embraces the search for common underlying cognitive principles and points to a number of promising areas for future research. |
