Dialog Systems

Advances in speech recognition, natural language understanding, and speech synthesis are enabling technologies for building useful conversational agents.  Users of technology are becoming habituated to talking to things. The consumer market is exploding with dialog systems, personal assistants and devices that allow voice interactions. The current slate of personal assistants can assist people by, for example, answering questions and helping to schedule appointments.  Moreover, there are task-based virtual agents that partially automate things like ordering food and making hotel reservations. However, these agents fall short of truly interactive, goal-based assistants that have been the focus of our research.

PARC’s technology for interactive and collaborative interactions have taken the form of virtual agents for customer service centers, conversational interfaces for multi-function printers, and enterprise intelligent assistants among others.  Our dialog platform has been used to implement textual chat agents as well as voice interactions handling textual, synchronous conversations with customers or users. Our goal is to build collaborative and cooperative agents that work together with people to accomplish their tasks providing more utility than shallow agents that lack the ability to maintain sustained interactions and context throughout a dialog.

Applications

• Intelligent Assistants
• Novel Conversational User Interfaces
• Accessibility Features
• Hands-free and eyes-free control
• Customer Service
• Training and novice user assistance

How the Technology Works

We model human-agent conversations according to four types of interactions: question/answering, transactional, procedural, and diagnostic. They are all goal-based to an extent, but the second two are better aligned with the notion of collaborating participants committed to working together to achieve an agreed-upon objective. In both procedural and diagnostic interactions, humans and agents must first establish a common goal, then work together to achieve it. Conversation Analysis (CA), a discipline specializing in the sociolinguistic study and understanding of talk-in-interaction, provides the theoretical underpinning of our technical approach. In particular, we examined the machinery that enables human social interactions across a number of activities. We believe that leveraging this machinery as much as possible in the design and implementation of intelligent agents is crucial to the success of human-agent interactions, especially as the realm of AI-enabled intelligent assistants continues to advance towards ever-increasing complexity and sophistication of situated, collaborative activities.