The computer as tutor

Imagine a language with a complex structure, where each word is represented by a unique symbol and where the meaning of a sentence is communicated not only with those words, but also with body movements and facial expressions.

Now, imagine an individual who thinks and communicates fluently in that language but must learn to translate it into standard written English—in order to send a business letter, perhaps, or to write a high school term paper or a graduate school thesis.

That’s the situation facing many deaf persons whose first language is American Sign Language. And, it’s a problem that Kathleen McCoy says she believes can be tackled with the help of a specialized computer program.

McCoy, professor of computer and information sciences, and various student research assistants have been working to develop such a program for several years. Known as ICICLE, for Interactive Computer Identification and Correction of Language Errors, the program is envisioned as a sophisticated grammar-checker and tutorial that
is specifically targeted to American Sign Language users.

“There are grammar-checking computer programs now available, and, while they often work well for native English speakers, they don’t catch the kinds of errors that American Sign Language users are likely to make,” McCoy says. “One common misconception about American Sign Language is that it’s standard English translated word by word into hand gestures. In reality, it’s a separate language with an entirely different structure that’s more similar to the Chinese or Navajo languages than to English.”

McCoy, who studies natural language processing, says it was about 10 years ago when she became aware of concerns in the deaf community about the difficulties in moving from fluency in American Sign Language to fluency in written English. At Gallaudet University, where much of the student body is deaf, students and faculty were reporting a need for improved ways to help sign language users perfect their English writing skills.

“I became fascinated by the question of what kind of computer program could be developed to tutor American Sign Language users as they learn to write English as their second language,” McCoy says. “I just kept thinking that there must be a way to do this.”

The result has been a decade of work by her and several graduate students who have helped develop and refine various aspects of ICICLE.

They began by collecting samples of written English from American Sign Language users and analyzing them to determine what types of errors were most common as the writers acquired standard English as their second language.

Today, the ICICLE program can help writers identify errors and correct their work, although McCoy says much work remains to enhance the program to the point that it can provide effective tutoring to a typical user.

Rashida Davis, a doctoral student who is working with the project, demonstrates the current capabilities of ICICLE. In the “input” box on her computer monitor, she types a short writing sample:

I went to school. Mr. Moore a teacher.

On her screen, a second box appears, with Davis’ second sentence underlined to indicate an error. Then, a third box opens:

There is a problem with this sentence. This sentence is missing a verb.

Davis says she’s working on ways in which the computer’s feedback can be more specific and instructive. She would like the program to be able to analyze a long writing sample and group similar types of errors together, showing the writer a pattern of problems that need to be addressed. Then, the program would generate examples the writer could use to practice the needed skills.

“My goal is to go beyond a simple, sentence-by-sentence analysis so it can be more helpful as a tutorial,” Davis says. “I’m interested in the whole field of how humans interact with computers and how to design them to facilitate that interaction.”

McCoy says she soon may be seeking volunteers to test ICICLE but that “it will still be many years before we can put this on a deaf person’s laptop for them to use as they write.” The process of developing the program is so labor-intensive, she says, partly because of the complexity of written English and also because the identification of errors in a writing sample depends on how advanced the writer’s English language skills are.

As an example, McCoy says, a person who writes, “She is teach piano on Tuesdays” might mean a number of different things. Someone learning basic English might be trying to write, “She teaches piano on Tuesdays,” while an intermediate English user might intend to write, “She is teaching piano on Tuesdays.” A more advanced student would not be likely to make those types of errors but might be trying to communicate, “She is taught piano on Tuesdays.”

The computer program eventually will have to distinguish different levels of English fluency and adjust for them, McCoy says. Throughout the project, she says, one key has been to recognize the importance of the language foundation that deaf people have.

“I think that if you learn American Sign Language first, that knowledge can only help you as you learn English—or another language—as your second language,” she says. “Having that first language is very positive, and that’s what we’re always trying to build on.” 

—Ann Manser, AS ’73, CHEP ’73