On a pilot corpus of Dutch medical encyclopedia texts, we focus on the mechanism of taxonomic inference that involves the extraction of co-hyponym terms, and the taxonomic or domain-specific lexicosemantic relation in the form of a textual hypothesis, which coordinates these. An initial set of inference elements can be acquired by syntactic and semantic alignment, additionally driven by document structure. From the terms and the related hypothesis we can harvest lexical patterns, which are

more »

... d to annotated domain concepts. The aim of the process is to learn inference patterns and apply the system to short as well as unstructured documents where fewer or no discourse-level cues are available, in order to acquire new co-hyponyms linked to their coordinating term via a specified relation.