Overview of UI-CCG Systems for Event Argument Extraction , Entity Discovery and Linking , and Slot Filler Validation

In this paper, we describe the University of Illinois (UI CCG) submission to the 2013 TAC KBP Event Argument Extraction (EAE), English Entity Discovery and Linking (EDL), and Slot Filler Validation (SFV) tasks. We developed three separate systems. Our Event Argument Recognition system infers world knowledge from event argument overlaps to improve performance of a recognition/labeling pipeline. Our Entity Discovery and Linking system builds on the Illinois Wikifier, augmenting its candidate identification capabilities and using a clustering algorithm for cross-document coreference. Our Slot Filler Validation system follows an entailment formulation that evaluates each candidate answer based on the evidence present in the source document it refers to. 1 Illinois Event Argument Extraction System We adopted a two step approach for event extraction, and used a heuristic rule to induce world knowledge to improve performance. For event extraction, we first detected the triggers and then detected the arguments. For world knowledge induction, we computed statistics over ACE 2005 data to identify argument overlaps between event sub-types. 1.1 EAE System Description The UI CCG Event Argument Extraction system uses a two-stage approach. The first stage identifies event trigger words, and the second identifies event argument mentions. 1.1.1 Description of Training Data We used ACE2005 as our training data, as its event taxonomy is similar to that of the Event Argument Extraction task. We trained classifiers to recognize event triggers and event argument mentions; in the following sections, we describe the features we used. To detect the triggers, we train the classifiers based on the trigger features. To detect the arguments, we train the classifiers based on trigger features, mention features, and pair features. 1http://www.itl.nist.gov/iad/mig//tests/ace/ace05/