Automated Text Analysis of Unstructured Test in NASA's Aviation Safety Reporting System

Thomas A Ferryman, Battelle Pacific Northwest National Laboratory

Aviation Safety Reporting System (ASRS) has well over 100,000 reports submitted by pilots, air traffic controllers, and others in the aviation community.  The reports often are handwritten and resemble a stream-of-consciousness style, with creative spelling, grammar, use of jargon and other challenges.  Analysis of the reports has been done in pursuit of several different goals, including general survey of the corpus to characterize the nature of the reports, retrieval-by- example, multiple corpora analysis (with distinctly different written styles), and specific concept identification and have been approached using statistical and Natural Language Processing (NLP) techniques. We will report on the approaches taken, problems encountered, solutions developed and, very briefly, the results achieved.