Dialog Research Center (DialRC)

Projects

Research

DialPort: Spoken Dialog Research with Real Data

NSF-funded DialPort has 3 thrusts: Tools, Portal and Real Challenge. The Portal gives your connected dialog system a stream of user data. You can also request data that other systems have gathered. There is also a list of tools that are available to help you and your students build applications.

2244

Total Dialogs Collected

793

Dialog from Real Users

Let’s Go!: A Spoken Dialog System For The General Public

Let’s Go! is building a spoken dialog system that can be used by the general public. The system we are developing for Let’s Go! is designed to work with a much wider population, including groups that typically have trouble interacting with dialog systems, such as non-native English speakers and the elderly.

End-to-end Neural Dialog Systems

End-to-end neural dialog systems offer great promises in developing dialog agents that can handle complex conversations. Unlike traditional dialog frameworks, end-to-end systems are not constrained to hand-crafted semantic representations, and can keep improving on large training data. DialRC has been one of the first teams that focuses on the cutting-edge end-to-end dialog research and continues to lead and contribute to the advancement of future neural dialog systems.

Multi-domain Portal

Open domain conversation requires dialog agents to be capable of talking about many topics and master various conversation skills. The DialPort project at DialRC appraoches this problem by developing a dialog platform manager that can aggregate heterogenous dialog systems into a single powerful dialog agent. We tackle this problem by developing a shared dialog state representation across different dialog domains, and use a hierarhical dialog policy rooted from hierarhical reinforcement learning to learning to control all the dialog agents, which in turn creates a homogenous conversational experience to the end users.

Multi-modal Distraction Detection

Distracted Driving continues to be a cause of traffic accidents despite prevailing legislation. The goal of this project is to automatically determine when the driver is becoming distracted. This information can be sent to a warning system in a car or can be used to help shut down the distracting activity.

Crowdsourcing tool for Spoken Dialog Systems

Crowdsourcing has solved the issue of finding users, but it presents new challenges such as how to use a crowdsourcing platform and what type of test is appropriate. DialCrowd makes system assessment using crowdsourcing easier by providing tools, templates and analytics.