Unraveling Language

Maxine Eskenazi knows how much can depend on the meaning of a single word. While she was living in France, her French mother-in-law sent one of her American aunts a gift of delicate chiffon fabric.

But the French have a different name for the sheer material--mousseline--while "chiffon" means "rag." Eskenazi's mother-in-law received, therefore, a well-intentioned note thanking her for the "nice rags."

Such incidents, along with Eskenazi's experiences teaching English in France, instilled in her a lifelong appreciation for the intricacies of language learning.
Learning a new language is hard work. Too many unknown words or a dull subject can make the effort seem unrewarding. Aware of these issues, Eskenazi, an associate teaching professor in the Language Technologies Institute and a member of the PSLC executive committee, and LTI professor Jamie Callan developed software to customize students' reading materials, helping them engage and learn more rapidly.

The software they created--Reader-Specific Lexical Practice for Improved Reading Comprehension, or REAP--combs the Internet for documents that fit the reading level and interests of the individual student. Finding real documents from the Web was crucial. "I wanted authentic texts for students, not things teachers make up that wouldn't reflect how language was really used," Eskenazi says.

REAP first gives a vocabulary test to find words that a student don't know. Because earlier studies found that students were more motivated to learn a language when they were reading about topics that interested them, students are also quizzed about their interests. The software then seeks out stories and articles with words the student needs to learn in the subject areas they enjoy--and with enough novelty to keep the budding reader challenged but not frustrated. REAP helps to teach vocabulary using context, rather than just definitions.

Teachers also were asked for their input on the technology. "It's extremely important when you're creating software that you have your ears wide open to what teachers need," Eskenazi says. During initial studies, instructors indicated they wanted to be more involved in the learning process, so researchers created an interface teachers can use to find documents for class discussion using specific topics and vocabulary.

REAP is presently in use at Pitt's English Language Institute, where teachers have noticed students--far from being bored--copying down URLs so they can go back and peruse the readings at their leisure.

"If we can motivate students to read more, that's a very good thing," Eskenazi says. REAP is also available in French, and the team recently received a grant from the CMU-Portugal Initiative to develop a Portuguese version.

Once a student has mastered written vocabulary and grammar, there's still the daunting prospect of pronunciation, where mistakes are inevitable and can be embarrassing, especially early in the language-learning process. To help with this problem, Eskenazi has created software called NativeAccent. A product of Carnegie Mellon spin-off Carnegie Speech, of which Eskenazi is chief technology officer, the interactive tool allows a student to practice speaking in front of a computer.

Pronunciation errors are corrected using speech-recognition technology. Through a cycle of speaking and receiving corrective feedback, the student gradually builds enough confidence to converse with other people. NativeAccent and other Carnegie Speech products have been used by speakers of more than 40 languages. In January, Carnegie Speech launched a new product specifically for international pilots and air traffic controllers, who must meet new standards for spoken English by 2011.

Eskenazi's future research will address the fact that students coming from different cultural backgrounds may present different learning challenges. For example, some Arabic-speaking students have an easier time learning to speak than read because they don't have a strong early exposure to written language. Students from East Asia can have the opposite tendency.

Eskenazi envisions an enhanced version of the REAP software that could gather video instead of text and use speech recognition to teach spoken language. She's also working on a tool to teach meta-linguistic cultural concepts.

Such technologies will help ensure that--when learning a new language--future students steer clear of too many avoidable faux pas.
For More Information: 

Jason Togyer | 412-268-8721 | jt3y@cs.cmu.edu