eCommons will be completely unavailable from 8:00am April 4 until 5:00pm April 5, 2018, for software upgrades. Thank you for your patience during this planned service interruption. Please contact us at firstname.lastname@example.org if you have questions or concerns.
A Model for Adaptive Problem Solving Applied to Natural Language Acquisition
Harris, Larry R.
Adaptive Problem Solving is the application of artificial intelligence learning techniques to practical problems. The approach taken in studying Adaptive Problem Solving is three-fold. First, to develop a model for Adaptive Problem Solving in order to specify the processes involved in computer learning, as well as the interaction between these processes. Second, theoretically well-founded, practical algorithms are developed for each of these learning processes. Third, as an application of this theory, the Natural Language Acquisition Problem is formulated in terms of the adaptive model. The specification of algorithms to perform the learning processes leads to the development of the Bandwidth Heuristic Search, an extension of the heuristic search, that includes many practical considerations without forfeiting any theoretical capabilities. A modification of this algorithm, the Bandwidth Heuristic Search for MIN/MAX trees is shown to be superior to the alpha-beta minimax process. The model is applied to the Natural Language Acquisition Problem in order to force an encounter with several critical problems involved with computer learning. The Natural Language Acquisition Problem is the problem of providing a robot the adaptive mechanisms sufficient to learn to converse with a human teacher using natural language. The robot first learns the lexicon of the language by correlating the teacher's description of the robot's actions with the robot's internal description. Then the robot infers a grammar that reflects the structure of the teacher's sentences. At this point the robot can begin conversing using a natural language. The linguistic capability of the robot includes the ability to disambiguate lexical and structural ambiguities, and the ability to formulate full sentence replies. After several learning sessions the robot converses in English using nested dependent clauses. This adaptive linguistic system successfully copes with many of the critical problems involved in computer learning and serves as an example of an adaptive program in which the learning, rather than yielding only minor improvements, provides the primary basis for successful performance.
computer science; technical report
Previously Published As