Sometimes research projects come out of important annual conferences and sometimes they arise over a morning's breakfast. The former circumstance may be more common but the latter could lead to a new center on the University of Illinois at Urbana-Champaign campus for studying second language fluency.

Richard Sproat and Chilin Shih are Beckman Institute researchers, faculty members at the U. of I., and husband and wife. Shih is a member of Beckman's Cognitive Science group who focuses on linguistics and speech technologies, while Sproat is interested in linguistics, including computational linguistics, and speech processing as part of the Artificial Intelligence group. During a breakfast discussion about a possible proposal to the National Science Foundation's Human and Social Dynamics program, their research interests converged on an idea.

"We were trying to think of something we could do that was related to language," Sproat said of the conversation. "We don't work on social dynamics in the normal sense, but we were thinking about what kind of thing related to language seems to have some sort of social interaction component to it. We thought 'what about fluency' in the sense of getting people comfortable with interacting with others, in this case in another language."

Shih, a Professor of Linguistics, had worked on a project at the U of I's East Asian Languages and Cultures department that used the Toastmaster's program as a model for teaching fluency in a second language. Shih found that by teaching the art of extemporaneous speaking to Chinese students learning English as a second language, it improved the participants' overall English speaking skills.

"The idea was to get people talking, even if they made grammatical mistakes, even if they had problems with vocabulary and so on, to get them to practice talking to make them less afraid of talking," Sproat said. "One of the problems I've always had with a second language is I may know the grammar and the vocabulary very well but I'm afraid to open my mouth. This will get that. What happened is the students initially hated it, but they ended up loving it."

That concept was the starting point for the NSF Human and Social Dynamics proposal. The proposal included a three-year project for studying automatic methods for second-language assessment that would lay the groundwork for the Center for Second Language Learning Technology at the U of I. The proposed center would feature interdisciplinary research aimed at developing technological aids for second language learning. The proposal resulted in a critical research award that will get the project moving forward, with the ultimate goal of creating the second language learning center.

Sproat has joint faculty appointments in the departments of Linguistics and Electrical and Computer Engineering and has a wide range of research lines and projects. He is head of the Computational Linguistics Laboratory at Beckman and his research includes topics such as named entity detection and transliteration for multiple languages, prediction of prosody from text for affective speech synthesis, and language modeling for colloquial Arabic speech recognition.

The second language fluency assessment project is one of those projects currently at the top of his research list. It's a truly interdisciplinary endeavor with aspects of psychology, linguistics, speech processing, engineering, and teaching.

"What grew out of this was 'could we come up with a multidisciplinary approach to the question of fluency that would draw on second language teaching, methodology, psycholinguistics,'" Sproat said. "The idea would be that it would serve as a basis for interdisciplinary research between people who are interested in second language pedagogy, language testing, second language acquisition, in the psychology or linguistic sense."

Beckman researchers Kay Bock, Mark Hasegawa-Johnson, Dan Roth, and Brian Ross joined Sproat and Shih in making the proposal. They each bring a different research perspective to the project. Sproat and Hasegawa-Johnson are computational linguists who can contribute engineering expertise, while Shih, Bock, Ross, and Roth offer perspectives from areas of psychology and linguistics. Sproat said having input from researchers with various interests provides for a more comprehensive program.

"I think the main thing is all these very different insights from different backgrounds," he said. "We just come with different assumptions. I'm not a psychologist so I haven't thought about those issues. It often surprises me that when I'm talking to psychologists, they think about issues in some way that I never would have thought of. My background is in linguistics and to some extent, engineering. It does have a nice mix of people where you can really get very different insights on similar things."

The project's emphasis on speech fluency - broadly defined as the ability to formulate and execute successful speech plans - makes it distinct from programs that use automatic speech processing methods to aid in learning to speak a second language. The program's focus on spoken language is another difference with other programs.

Sproat said the emphasis on the spoken word is important because of the ultimate goal of teaching people to communicate more efficiently in their second language.

"It's not that learning to read and write is not useful but ultimately, for example, why do you want to learn Chinese? It's because China is becoming more important on the world stage and you want to have access to that culture and being able to speak to people is important," Sproat said. "So one goal would be to concentrate on people feeling comfortable speaking, rather than focus on making sure they get the grammatical forms right. I'd much rather be fluent in a language in the sense that I could go down the street and talk to somebody and get what I needed, even if I made a silly mistake. Unfortunately, a lot of language programs tend to emphasize the grammatical and pronunciation aspects rather than the fluency."

Developing technology that could quantify a person's fluency in a second language is an ultimate goal of the project.

"The third component is an engineering aspect, which is, however you define fluency, can one come up with a set of automatically derivable features that would allow you to determine whether this person is speaking fluently and this person isn't," Sproat said.

Sproat previously worked at Bell Labs and AT&T Labs, two places where he said the approach to research is similar to what he has found at Beckman.

"One of the things that was always very nice about Bell Labs - which was quite different from what I knew the case to be in academia - was that I could walk down the hall and talk to an engineer, or a computer scientist or psychologist, all on the same hallway," Sproat said. "You would never get that in a typical academic environment. So one of the things that was so appealing about Beckman is you do have people with all kinds of backgrounds and we really do work together."