In search of patterns in incident reports: A syntactic approach.

摘要

This research presents a novel methodology for automated analysis of text narratives. Current approaches hinge on statistical analysis of keywords and phrases and to a minor extent syntax. Due to their over-reliance on domain-specific knowledge and their lack of underlying behavioral models, these approaches have significant limitations particularly for problems dealing with human performance.; What is presented here is an approach based on text segmentation utilizing syntactic templates. This segmentation is inspired by linguistic theory and the idea of "mental event" types in narratives. The goal is to contribute with a complementary approach to the current methods, facilitating the detection of behavioral patterns in text. This work contains the method's conceptual formalization together with its computational implementation. Finally the data distribution is examined, raising the hypothesis about the method's potential usefulness. The results are applied to a specific problem in Aviation Safety dealing with Human Performance Errors.