Contextual and Structural Language Understanding in Dialogues

Natural language understanding (NLU) is a core component of a dialogue system. Recently recurrent neural networks (RNN) obtained strong results on NLU due to their superior ability of preserving sequential information over time. Traditionally, the NLU module ignores the contexts of utterances and tags semantic slots for utterances considering their flat structures, as the underlying RNN structure is a linear chain. However, contexts and linguistic properties provide informative cues for better understanding. This paper introduces a novel model, a generalization of RNN to additionally incorporate 1) contextual utterances and 2) non-flat network topologies guided by prior knowledge. The model automatically figures out the salient and accurate contexts and substructures that are essential to predict the semantic tags of the given sentences, resulting in better understanding. The experiments showed the significant improvement on contextual and structural language understanding scenarios.