On the categorization of Cause and Effect in WordNet

The task of detecting causal connections in text would benefit greatly from a comprehensive representation of Cause and Effect in WordNet, since previous studies show that semantic abstractions play an important role in the linguistic detection of semantic relations, in particular the cause-effect relation. Based on these studies on causality, and on our own general intuitions about causality, we propose a cover-set of different WordNet categories to represent the ontological classes of Cause and Effect. We also propose a corpus-based approach to the population of these categories, whereby candidate words and senses are identified in a large corpus (such as the Google N-gram corpus) using specific syntagmatic patterns. We describe experiments using the CauseEffect dataset from the 2007 SemEval workshop to evaluate the most effective combinations of WordNet categories and corpus data. Ultimately, we propose extending the WordNet category of Causal-Agent with the word-senses identified by this experimental exploration.