Real applications of natural language document processing are very often confronted with domain specific lexical gaps during the analysis of documents of a new domain. This paper describes an approach for the derivation of domain specific concepts for the extension of an existing ontology. As resources we need an initial ontology and a partially processed corpus of a domain. We exploit the specific characteristic of the sublanguage in the corpus. Our approach is based on syntactical structures (noun phrases) and compound analyses to extract information required for the extension of GermaNet's lexical resources.