Distributional Composition using Higher-Order Dependency Vectors

This paper concerns how to apply compo-sitional methods to vectors based on gram-matical dependency relation vectors. We demonstrate the potential of a novel ap-proach which uses higher-order grammat-ical dependency relations as features. We apply the approach to adjective-noun com-pounds with promising results in the pre-diction of the vectors for (held-out) ob-served phrases.