POWRE: Software Evaluation: The Application of Software Testing Techniques to Rule-Based Natural Language Processing Components

点击翻译按钮获取中文摘要

EIA-9973855Barr, ValerieColumbia University CISE/POWRE: Software Evaluation: The Application of Software Testing Techniques to Rule-Based Natural Language Processing ComponentsThis research will bring together the two fields of software testing and natural language processing (NLP) by applying software testing techniques for rule-based systems in the specific domain of NLP systems. The specific goal of this research is to adapt and extend an existing approach for coverage-based testing of rule-based systems in order to develop new methods for evaluation of rule-based NLP systems. In addition, the research will incorporate a study of how the accuracy of an NLP component affects the performance of a larger system, examining the robustness of systems which use some form of NLP and their ability to withstand inaccuracies in the analysis carried out by an NLP component. This research will be undertaken in the context of a number of research projects currently underway at Columbia University under the auspices of the Center for Research on Information Access and the Natural Language Processing group of the Department of Computer Science.