From Documents to Dialogue: Context Matters in Commonsense-Enhanced Task-Based Dialogue Grounded in Documents

Abstract

Humans can engage in conversation to collaborate on multi-step tasks, and divert briefly to complete essential sub-tasks, such as asking for confirmation or clarification, before resuming the overall task. This communication is necessary as some knowledge in instructional documents can be implicit rather than grounded in the dialogue, meaning that people need to rely on their own and others' knowledge for problem-solving. We often attribute this capability to commonsense, i.e. the assumption that interlocutors perceive behaviours, temporality, context, space and object properties in a similar way. To facilitate these capabilities in a dialogue system, we frame flexible dialogue management as a hybrid setting of a document-grounded dialogue system (DGDS) and a knowledge-based grounded dialogue system. In a human evaluation, the hybrid system was more effective in capturing object knowledge (utility, appearance, storage, relationships, handling) and contextual knowledge (understanding of events and situations) compared to a rule-based baseline. In addition, we show that inferring context from different sources enhances the dialogue by allowing richer and more fluid interaction. To our knowledge, this is the first effort to model \textit{task-based} dialogue grounded in multiple documents

Keywords: natural language generation, Task-Based Dialogue, Common-sense Knowledge, Assistive chatbots