Automatic Extraction of Dutch Hypernym-Hyponym Pairs

In this study, we apply pattern-based methods to text for extracting lexical data, in particular the hypernymy relation. We automatically derive thousands of interesting lexical patterns like such NP as NP and evaluate the performance of these patterns by comparing the information they extract from a newspaper corpus with the information in the Dutch part of EuroWordNet. Additionally we investigate the usefulness of combining hypernymy relation evidence generated by different patterns and compare this approach with the application of fixed patterns to web data. We find that with larger quantities of data, individual fixed extraction patterns outperform the large combination of patterns applied to the corpus.