CAST 2019

Abstract: PolyAI is a London-based startup with a leading machine learning platform for conversational agents. The deployed AI agents understand users, hold conversations without getting confused and can easily scale to new use cases or other languages. This talk will present the machine learning techniques that underpin the PolyAI platform and results from its early deployments in contact centre environments.
[slides]

Leveraging Human Learning in Interactive Data Exploration
Sanad Saha (Oregon State University)*; Arash Termehchy (Oregon State University); Leilani Battle (University of Maryland)

DBPal: Weak Supervision for Learning a Natural Language Interface to Databases
Nathaniel Weir (Brown University); Andrew Crotty (Brown University)*; Alex Galakatos (Brown University); Amir Ilkhechi (Brown University); Shekar Ramaswamy (Brown University); Rohin Bhushan (Brown University); Ugur Cetintemel (Brown University); P. Ajie Utama (TU Darmstadt); Nadja Geisler (TU Darmstadt); Benjamin Hättasch (TU Darmstadt); Steffen Eger (TU Darmstadt); Carsten Binnig (TU Darmstadt)

Disambiguating Natural Language Queries with Tuples
Christopher Baik (University of Michigan)*; Zhongjun Jin (University of Michigan); Michael Cafarella (University of Michigan

Abstract: Conversational systems generally fall into two categories: task-oriented and social bots. Task-oriented systems aim to help users accomplish a specific task through multi-turn interactions, whereas socialbots focus on engaging and natural open-domain conversations. In natural interactions, even when conversation participants have a task or goal in mind, they can say things that are out of the boundaries of that task domain, and similarly they can switch to a task during a chitchat conversation. Hence, the ability to engage in knowledgeable social interactions and gracefully transition back and forth to the task is crucial for enabling natural conversations.

In this talk, I’ll summarize our recent work in both fronts, focusing on the convergence of approaches for the two categories of conversational systems. Starting with task-oriented interactions, I’ll present our approach for bootstrapping task-oriented dialogue systems from simulated seeker-provider conversations and dialogue state tracking using generate-and-copy mechanisms. This will be followed by a summary of learnings from previous Alexa Prize challenges and progression of our work as we approach the next challenge.

• - Dilek Hakkani-Tur, Senior Principal Scientist at Amazon Alexa AI
• - Nikola Mrkšić, Co-founder and CEO, PolyAI
• - Fatma Ozcan, Principal RSM, IBM Research