| dc.description.abstract | Task-oriented dialog systems hold numerous applications in assisting users to achieve various goals. They often comprise of a pipeline of individual components. In this work, our contribution is towards two such components, namely, dialog state tracker and natural language generator. A typical conversation comprises of multiple turns between participants where they go back-and-forth between different topics. At each user turn, dialogue state tracking (DST) aims to estimate user’s goal by processing the current utterance. However, in many turns, users implicitly refer to the previous goal, entailing the use of relevant dialogue history. Nonetheless, distinguishing relevant history is challenging and a popular method of using dialogue recency for that is inefficient. We, therefore, propose a novel framework for DST that identifies relevant historical context by referring to the past utterances where a particular slot-value changes and uses that together with weighted system utterance to identify the relevant context. Specifically, we use the current user utterance and the most recent system utterance to determine the relevance of a system utterance. Furthermore, we do empirical analyses to show that our method improves joint goal accuracy on WoZ 2.0 and MultiWoZ 2.0 restaurant domain datasets respectively over the previous state-of-the-art models. Secondly, we study a family of deep generative models for generating system response in a task-oriented dialog setting. The language generation tasks involve conditioning the output of the generative models on the current dialog state, system act and the previous user utterance. Finally, we do qualitative analysis and report the perplexity scores for a transformer encoder-decoder model and a conditional variational auto-encoder on schema guided dialog state tracking dataset. | en |
