Statistical spoken dialogue systems usually rely on a single- or multi-domain\ndialogue model that is restricted in its capabilities of modelling complex\ndialogue structures, e.g., relations. In this work, we propose a novel dialogue\nmodel that is centred around entities and is able to model relations as well as\nmultiple entities of the same type. We demonstrate in a prototype\nimplementation benefits of relation modelling on the dialogue level and show\nthat a trained policy using these relations outperforms the multi-domain\nbaseline. Furthermore, we show that by modelling the relations on the dialogue\nlevel, the system is capable of processing relations present in the user input\nand even learns to address them in the system response.\n