Getting to Know Your Fellow Researchers: Dilek Hakkani-Tür

What sparked your interest in speech and language processing?

Early on, I was amazed by the ambiguity of natural language, that so many sentences could in fact be parsed and understood in different ways, and yet we can often times easily communicate with each other and interpret what we hear or read with the intended semantics. Then I found out in college that speech and natural language is actually quite an active area for computer science.

How do you think speech and language processing is changing the society for the new generation?

With the availability of several devices in the consumer space (such as smart phones and smart speakers), the new generation is growing up with easy access to the speech and language processing technology. It is especially interesting to observe kids interacting with conversational agents and these devices by talking to them. They figure out quickly what these agents can understand and how one can interact with them. They transfer what they learn about interacting with one agent to another.

What is your holy grail in speech and language processing? When will we achieve it?

Over the years, I’ve been getting busier and busier, so to me, the holy grail in speech and language processing is having an agent that I can talk to, that is present when I need help, that can assist me with the work that I need to do, that can learn from me how to do things in my way and that can get to know me and my preferences, just like the ones in movies and not a one size fits all kind of an agent.

Do you have any specific advice for students, junior faculty or others early in their careers? 

I suggest to focus on things that they are passionate about, as I believe success then follows that passion. Furthermore, instead of inventing problems that fit an idea of a solution, look for the real and common problems and focus on ideas that tackle them.

What development in the field has most surprised you? Was there a hard problem that turned out to be easy?  An easy problem that proved surprisingly difficult?

These days there is so much interest in dialogue systems from multiple research communities, there is certainly a lot of progress, so many good ideas that are being proposed, so many new papers and submissions, yet we are still so far away from understanding language at a level that kids can easily learn.

Oftentimes, the work that people get noticed for is not the same as the work which they find most exciting/rewarding/interesting. Which of your publications is your favorite? Why? 

My favorite publications are from my earlier work on active learning for speech and language processing. When I had my first job, I was asked to study semi-supervised learning methods, and although active learning was an earlier idea that was proposed in machine learning, it wasn’t a well-known area in speech and language processing. I thought it would be interesting to try this idea out and see if it would really help. My first experiments were promising and my mentors were very supportive, but I have seen so much resistance for the idea from the broader team, as, back then, the motto was “there is no data like more data”. Today, I am amazed how much industry picked up this line of work and how much savings in time and labor active learning resulted in.