CISE Research Instrumentation: Parallel Processing for Machine Translation, Skill Learning, and Software Testing

点击翻译按钮获取中文摘要

A desktop parallel processing computer will be used to implement FORR, which is a general learning and problem solving architecture that supports learning about behavior in a broad domain of related problem classes. The architecture integrates diverse problem solving heuristic into a common framework and constructs decisions from their inconsistent advice. A parallel approach to machine translation based on the principles of blackboard architecture will also be implemented . Unlike conventional, sequential architectures for machine translation systems, the blackboard approach to MT offers a global workspace in which multiple knowledge sources, each expert in a particular aspect of translation, can work in parallel to solve difficult problems. For Japanese and English these include zero pronominals, number, definiteness, and language style. A third project involves an alternative architecture for experimental comparison of software test data adequacy criteria. This approach harnesses the parallelism inherent in such experiments by coordinating a pool of independent processors that implement the distinct criteria. The processors work together to decide which of the criteria they implement is better at exposing errors in the given program.