| The Aviation Safety Reporting System collects voluntarily submitted reports on aviation safety incidents to facilitate research work aiming to reduce such incidents. To effectively reduce these incidents, it is vital to accurately identify why these incidents occurred. More precisely, given a set of causes, or shaping factors, this task of cause identification involves identifying the shaping factors responsible for the incidents described in a report. We have investigated three approaches to cause identification. The first two of these approaches exploit information provided by a semantic lexicon, which is automatically constructed via Thelen and Riloff's (2002) Basilisk framework augmented with our linguistic and algorithmic modifications. The first approach labels a report using a simple heuristic, which looks for the words and phrases acquired during the semantic lexicon learning process in the report. The second approach recasts cause identification as a text classification problem, employing supervised and transductive text classification algorithms to learn models from incident reports labeled with shaping factors and using the models to label unseen reports. The third approach utilizes annotation rationales, or text fragments identified by annotators as relevant to the labels assigned to the documents, to augment the training set by creating pseudo-instances. The pseudo-instances are created by selectively removing one or more rationales from a document. Then a supervised text classification algorithm is used to learn models from the augmented training set with additional constraints being placed on the misclassification of the pseudo-instances. Our experiments on the these approaches show that all of these approaches outperform their respective baseline systems significantly. |
