We present a fully automatic language-independent system for building domain-specific lexicons from a short list of terms defining the domain. Lexifield relies on a pre-trained word embedding model, a definition dictionary and a dictionary of synonyms. As compared to other word embedding based systems and a state-of-the-art sensorial lexicon, our system achieves better precision and recall on reference lists extracted from manually created resources such as Roget's Thesaurus.