Found in Translation

Words and language have always fascinated Michael Goodman. Growing up in Florence, Oregon, he liked tracing the roots of words that most of us take for granted, and at Oregon State University, he has minored in Japanese. But it is his affinity for computers that is propelling the senior in the School of Electrical Engineering and Computer Science. Combining his interests, he has created software that overcomes a barrier in translation.

Along the way, Goodman lived in Tokyo for an academic year, collaborated with OSU faculty members and set the stage for graduate work in computational linguistics. Not bad for a young man who taught himself computer programming at home, as he explains it, “just by messing around.”

The problem he tackled for his senior project stems from a fundamental difference between Japanese and English. “The Japanese language is different from English in the way pronouns — words such as he, she or they — are used. They exist in the language, but their use is less common than in English,” says Goodman. Instead, subjects in a Japanese sentence usually refer to the last proper noun mentioned in a conversation. This practice can make it hard for people, whose primary language is English, to keep track of whom or what is being discussed. In order to address this problem, Goodman has created a software solution that he calls Co-reference Resolution. The goal is to point a translation system to the subject in scanned Japanese text, increasing translation accuracy.

Goodman had help in bridging two disciplines: computer science and linguistics. His adviser in the School of Electrical Engineering and Computer Science, Alan Fern, specializes in artificial intelligence and machine learning. Providing linguistics expertise was Setsuko Nakajima, a Japanese language specialist in the Department of Foreign Languages and Literatures.

Writing the software code might seem daunting enough, but the biggest challenge to Goodman’s research was getting access to annotated articles in Japanese. “To get all the articles I needed to be able to create the program and make sure it works,” he says, “would cost about $1,200 to $1,500.” Fortunately, a search for expertise on the science and development of translation systems led Goodman to a specialist in natural language processing at New York University who shared copies of scanned Japanese articles that his students had already analyzed to determine the subjects of each sentence. Using those articles, Goodman was able to focus on writing his software code.

“Doing this project has forced me to think long and hard about linguistic analysis and processing in a language that’s not my mother tongue, and has exposed me to the challenges and obstacles and ways to overcome them,” says Goodman.

Over the past year, he conducted his research and programming while taking a full load of courses. He credits his hardworking attitude to his parents who own a floor covering business in Florence. Watching them, Goodman learned that commitment and perseverance can lead to success.

Goodman plans to complete his project over the summer and enter a master’s degree program in Computational Linguistics at the University of Washington.