- Feature Story
Department Welcomes New Associate Professor
Roger Levy studies the foundational architecture of human language processing and acquisition.
BCSN: Share a little bit about your research.
Levy: Human language is the most expressive communication system in the known universe. Hundreds of times a day, we say sentences that we’ve never said or heard before to express novel meanings, and we hear and read hundreds more sentences we’ve never encountered before and understand what they mean. We’re able to do this in real time, despite a number of competing factors – limitations on our memory and attention, a noisy environment, the rampant intrinsic ambiguity of language itself, and our incomplete knowledge of the people we speak and listen to and whose texts we read. Compare this, for example, with the state of the art in computer understanding of human language: even if you speak right into your phone and Siri is able to make out your words correctly, its understanding of what you say and consequently to translate your commands into actions remains incredibly limited and superficial. Cast in this light, our ability to communicate effectively with language is all the more extraordinary.
This ability immediately raises some very deep questions about how the human mind works. How is it that we are able to understand each other? How is our system for communication structured? How are we able to learn this system from the input that we get as children? We go far beyond the raw content that we get as children to be able to generate novel utterances and understand them. Our powers of acquisition are remarkable.
To study our uniquely human capability to acquire and use language, I combine computational modeling with psycholinguistic experimentation, ideas from linguistic theory, and analysis of naturalistic language data sets. I try to understand the statistical and structure properties of language as people actually use it on an everyday basis, and also how linguistic knowledge can be adaptively deployed in novel and unusual contexts to achieve a variety of communicative goals. Using all of these things together, I hope to understand more about how we process and acquire language.
BCSN: What drew you to MIT?
Levy: It was an offer that I couldn’t refuse. For the kind of work that I do, combining computational modeling, experimentation, large data set analysis, and rich theory development that I practice in my research, the graduate program here is the strongest in the world, and the overall community is unparalleled.
BCSN: What do you look for in grad students?
Levy: For me, the signature quality that makes a grad student stand out is the drive to build their own theory of the domain that they want to work in - having a very specific perspective that is theirs and that they want to develop. Their graduate training gives them practice at developing that perspective and bringing it to bear in studying problems by asking questions that are both important and tractable. The very hardest thing in science is figuring out what questions to ask – everything else follows from there.
At a more nuts-and-bolts level, I also I look for technical skills such as proficiency in mathematics, being able to work well with computers, coding, good communication and teamwork skills, enthusiasm for studying language, and a deep interest in the human mind.
BCSN: What excites you most about the next five years?
Levy: I’m very excited about extending the scope of my research program. Now that I’m in a department where brain and cognitive sciences are studied together, I’m particularly hopeful that we will discover new ways of interfacing my computational and behavioral work with any of a variety of techniques in neuroscience to deepen our understanding of the neural basis of cognition and language comprehension. This is the best possible environment for doing that, and I hope that organic collaborations will develop, allowing my skills and expertise to complement existing efforts to study the brain basis of language.
Learn more about Levy’s research at bcs.mit.edu/levy