Introduction

In this section an overview is given of a variety of resources that contain lexical semantic information. There are many more resources available but here we only describe a selection illustrating the different types of resources, and focusing on the use of these resources for NLP. The selection covers: traditional machine readable dictionaries, wordnets, taxonomies, dictionaries with feature classifications, MT-lexicons, formalized lexical knowledge bases for NLP, resources for expert systems, traditional bilingual dictionaries. It is not easy to divide the lexical resources in more overall groups because each resource often contains different mixtures of data. For example, it seems obvious to group the Princeton WordNet1.5 with the EuroWordNet data as semantic networks but EuroWordNet also contains equivalence relations, making it a multilingual resource, and it incorporates a formalized ontology, making it partly a conceptual knowledge base. Similarly, we could make a division between traditional resources such as the Longman Dictionary of Contemporary English (LDOCE) and lexical resources developed for NLP purposes, but LDOCE includes a feature system (both syntactic and semantic) making it very useful as an NLP resource, as often has been shown. We therefore provide some grouping when it is obvious but also discuss resources individually, without however suggesting a difference in relevance or detail.

For each of these resources we give a description of the semantic content, both in terms of quantity and kind of data. Where possible we will describe the most frequent semantic values. This may either be the top-levels of hierarchies, definitions of features and concepts, or definitions of relations. Furthermore, we will provide references to the lexical semantic notions described in the previous section and the usage of the resources in NLP.