University of Hagen at QA@CLEF 2006: Interpretation and Normalization of Temporal Expressions

The German question answering (QA) system InSicht participated in QA@CLEF for the third time. InSicht realizes a deep QA approach: it builds on full sentence parses, inferences on semantic representations, and matching between semantic representations derived from questions and document sentences. InSicht was improved for QA@CLEF 2006 in the following main areas: temporal expressions are better normalized and temporal deictic expressions are resolved to explicit date representations; the coreference module was extended by a fallback strategy for increased robustness; equivalence rules can introduce negative information that should not occur in document sentences used for answering questions; answer candidates are clustered in order to avoid multiple occurrences of one real-world entity in the answers to a list question; and finally a shallow QA subsystem that produces a second answer stream was integrated into InSicht. The current system is evaluated on the German questions from QA@CLEF 2006. An ablation study indicates which changes had the most positive effects.