Language Understanding By Design


University of Waterloo Prof Pascal Poupart Talks Chatbots, Machine Learning and the Opportunities that Lie Between


Pascal Poupart confesses to receiving several marriage proposals a day. Well, not directly, mind you. Rather the UW professor, and a leading researcher in the fields of machine learning and reasoning, natural language understanding, and health informatics has received these admissions of love through Dan Bot™, an intelligent, self-learning chatbot he created with Waterloo messaging app creator Kik Interactive, allowing the company to more effectively engage and interact with its 120 million users.

Kik Interactive is not the only rising star in the start up world to beat a path to Poupart’s office door. The UW prof is also engaged in a research initiative with In the Chat, the creator of a second-generation social listening platform, to improve the harvesting and classification of social data to help large companies deliver real time customer support through social media channels.

The whole notion of having computers not only understand human language, with all of its nuances, but also intelligently communicate back and learn from that discussion feels like something out of 2001 ‐ A Space Odyssey.

But what was purely fictional speculation back in 1968 is now very real indeed. The pairing up of Poupart’s research in natural language understanding with the large text-based datasets generated by social networks is resulting in new advancements start ups like Kik Interactive and In the Chat find very exciting.

“Back in the days, telecom companies wanted to have a voice based system that could answer people’s questions,” says Poupart. “But the systems never really delivered on the promise. They were never good enough, and we never had the data sets to truly advance that research.” However, through collaboration with companies like Kik and In the Chat, we are now getting access to those large data sets that allow us to test the algorithms and advance natural language processing capabilities.

“I was blown away when I saw what In the Chat can do. Somebody can simply tweet a comment and without even a mention of a particular company, In the Chat analyzes the comment and within a matter of seconds that company can act to improve their relationship with the customer. People are expressing themselves on social media and most companies don't yet have a sense of how to approach this. In the Chat makes this easy for companies by routing social media posts like a traditional interactive voice response (IVR) routes phone calls, sending them to the right agents to address the customer's interests.”

Today, many solutions rely on keyword matching and manual sorting of social media posts, resulting in delays in customer support interactions. The new research partnership between Poupart and In the Chat seeks to find new ways to improve the automation of this process so that companies can focus on what is really important — taking care of their customers.

Pascal Poupart is reaping the benefit that comes through collaborating with startups. “Data comes in, the students get excited and everyone is happy.”

While Pascal Poupart’s research into natural language understanding and chatbots appeals to start ups seeking to mine the gold that exists in today’s social networks and better engage the youth demographic, his other research, focused on machine learning and health informatics addresses the extreme other end of the generational spectrum – our aging population.

Since his grad school days at U of T, Pascal Poupart has been collaborating with occupational therapists and kinesiologists on research initiatives aimed at creating approximate scalable algorithms for partially observable Markov decision processes, and their application to real world problems. Poupart’s research collaborations have resulted in the development of an automated system to guide and prompt individuals with Alzheimer’s through the hand washing process. Another collaboration with Eric Roy and Bill McIllroy, Department Chair of the University of Waterloo’s Department of Kinesiology resulted in a prototype ‘smart’ walker capable of gathering data and learning about a person’s movement in an effort to better predict and mitigate the risk of falls. And in a third initiative, Poupart worked with Laura Middleton (UW Kinesiology) and the Schlegel-UW Research Institute for Aging to develop an app trialed at the Village of Winston Park to help direct residents to organized group activities.

Pascal Poupart

“From a research perspective, I must say now that I have worked on a number of applications at both ends of the generational spectrum, there are more barriers that come with applications in the healthcare sector,“ confesses Pascal Poupart. “It is a highly regulated industry and there are many restrictions surrounding technologies. With the start-ups, you don’t face these concerns. You just go for it.”

That said, Poupart believes there may be an area of middle ground where his research can play in the start up realm and continue to aid those in need.

He muses, “For instance, is there the opportunity to create a healthcare app that leverages chatbots and natural language understanding to help coach people through exercise programs, or guide individuals suffering from chronic disease through daily routines? We’ll just have to see who shows up at my door next.”