Conversational systems is a thriving research area with applications, such as call-centers, tourist information, car navigation, education, banking, health services, and games. Commercial applications exist as well, such Microsoft’s Cortana, Apple’s Siri, and Amazon’s Echo among others.
The PyDial case that we exploit here, is a Statistical Dialogue System (SDS) whose components are: natural language
understanding, belief state tracking, policy manager, and natural language generation. In PyDial, context is not taken into
account. This work: i) incorporates context information by taking into account past turns in a Toshiba patented featurisation
manner; and ii) investigates two different neural network architectures, namely a DNN and a CNN, and in doing so verifying
the importance of taking context into account