Linguistically-Motivated Automatic Morphological Analysis for Wordnet Enrichment

Abstract

Performance of NLP systems can only be as good as the lexical resources they employ. By modelling the evolved structure of language, there is scope for morpho-semantic enrichment of these resources. A set of linguistically-informed morphological rules is formulated from the CatVar database, implemented in a Java model of WordNet and tested on suffixation and desuffixation. Overgeneration and undergeneration are measured and an approach to improving these by using multilingual resources is proposed.