Same same but different: type and typicality in a distributional model of complement coercion