In this paper we present a language-independent and automatic approach to extend a wordnet by recycling different types of already existing language resources, such as machine-readable dictionaries, parallel corpora and Wikipedia. The approach, applied to Slovene, takes into account monosemous and polysemous words, general and specialized vocabulary as well as simple and multi-word lexemes. The extracted words are assigned one or several synset ids based on a classifier that relies on several features including distributional similarity. In the next step we also identify and remove highly dubious (literal, synset) pairs, based on simple distributional information extracted from a large corpus in an unsupervised way. Automatic and manual evaluation show that the proposed approach yields very promising results.