Creating a dialogue system that understands what people are really asking

My primary research project at ZAS aimed to create a real estate sales chatbot that could understand indirect questions. For example, if a potential renter or buyer asks of a property, "Is there a train station nearby?", the chatbot would be able to answer in useful ways about alternative transportation options rather than merely giving the distance to the nearest train station that might be far away.

The chatbot PRAGSales designed using linguistic theories on indirectness, implicature, and dialogue coherence from a computational perspective.  We tested theoretical claims by using game theory and decision theory to build models of human dialogue that were implemented within an automated dialogue system.

See the poster I presented at ACL 2015 in Beijing on this project. 

For a more in-depth explanation of the project, see the ACL proceedings paper, or the more detailed journal paper that followed.