# Vector Mixture Models Words are represented as vectors, which combine to produce new vectors. $ \mathbf{p}=f(\mathbf{u}, \mathbf{v}, R, K) $ Constraint: p lies in the same n-dimensional space as u and v. Assumption: all syntactic types are similar enough to have the same dimensionality. Additive models can add and multiplicative models can multiply the u and v vectors to get the phrase vectors. The additive and the multiplicative model are symmetric (commutative): they do not take word order or syntax into account. - John hit the ball = The ball hit John Correlate with human similarity judgments about adjective-noun, noun-noun, verb-noun and noun-verb pairs More suitable for modelling content words, would not apply well to function words: _some_ dogs, lice _and_ dogs, lice _on_ dogs --- ## References