GEMS (Geometrical models of natural language semantics) is the third instalment in an annual series of workshops on distributional models or semantic spaces. These models have become omnipresent in contemporary computational linguistics, in particular in lexical semantics, and neighboring fields, like cognitive science and corpus linguistics. The workshop will survey and summarize contemporary contributions to problems in meaning representation, acquisition and use, based on distributional models. We are interested in methodological innovations as well as tasks ranging from the induction of lexical, linguistic and world knowledge to practical and industrial NLP applications.