Paul A. Crook

Research Scientist

I’m a research scientist at Facebook in Seattle. My research interests are primarily focused on the application of machine learning to dialogue systems, such as accurately tracking the dialogue state, and optimizing the dialogue policy through the use of techniques like Reinforcement Learning (RL). The principal challenge in this work is the handling of uncertainty and ambiguity that exists in any natural language processing (NLP) task. Within the NLP world I focus on tasks like human-machine or human-human conversations through the medium of text, speech or multi-modal interaction. Thinking of dialogue as a navigation problem though a world of concepts provides the link to my earlier work on indoor mobile robotics navigation with noisy sensor information, and easily confusable locations.

Prior to Facebook I worked at Microsoft on the dialogue manager for Cortana, and was a research fellow and founding member of the Interaction Lab at Heriot-Watt University, Edinburgh (now famous as 3rd place runners up in the 2017 Amazon-Alexa Prize). I obtained my PhD from the University of Edinburgh, where I was a member of the Institute of Perception Action and Behaviour, in the School of Informatics, where I investigated the application of RL and active-perception in mobile robotic navigation. I also worked as a research assistant in the School of Informatics, applying partially observable (POMDP) models of RL to statistical spoken dialogue systems, and for a pre-spinout company project on applying machine vision for automatically labelling animal behaviours from video footage.


Application of machine learning, especially reinforcement learning, for AI systems learning to interact in uncertain environments, such as human-machine conversations

Related Links


Latest Publications

COLING - December 8, 2020

Situated and Interactive Multimodal Conversations

Seungwhan Moon, Satwik Kottur, Paul A. Crook, Ankita De, Shivani Poddar, Theodore Levin, David Whitney, Daniel Difranco, Ahmad Beirami, Eunjoon Cho, Rajen Subba, Alborz Geramifard

COLING - December 8, 2020

Resource Constrained Dialog Policy Learning via Differentiable Inductive Logic Programming

Zhenpeng Zhou, Ahmad Beirami, Paul A. Crook, Pararth Shah, Rajen Subba, Alborz Geramifard

EMNLP - November 5, 2019

Recommendation as a Communication Game: Self-Supervised Bot-Play for Goal-oriented Dialogue

Dongyeop Kang, Anusha Balakrishnan, Pararth Shah, Paul A. Crook, Y-Lan Boureau, Jason Weston