UVAVU: WordNet Similarity and Lexical Patterns for Semantic Relation Classification

We built a system for task 4 of SemEval 2007. The goal of the system is the verification of seven semantic relations: cause-effect, instrument-agency, product-producer, origin-entity, tool-theme, part-whole, and contentcontainer. The data-set for the evaluation of this task consisted of 140 training sentences and 70 test sentences per relation. An example sentence of the cause-effect relation is the following: “He derives great joy and happiness from cycling.” The two arguments in this sentence have been pre-annotated with WordNet 3.0 senses. The system should classify this example as positive. Our system consists of two parallel components, indicated by the blue and yellow boxes in the figure on the right. One component is based on WordNet similarity measures implemented in the Perl package WordNet::Similarity. The other uses lexical pattern methods on Google and the Waterloo Multi-Text System (WMTS) System Overview