An interactive computer
program is helping children learn to read by listening to them as they read aloud
By Emily Bender
Computer software developed at IBM's Thomas J. Watson Research Center helps children learn to read. The technology takes a unique approach: it allows children between five and eight years old to practice reading and tells them how they are doing. The work involved two major technical challenges. First, the team had to develop a computer that would recognize children's speech - a particularly difficult problem because children don't pronounce words consistently. Then the researchers needed a child-friendly interface. They it achieved with an approach similar to that in video games.
Making use of speech recognition technology developed at IBM over the past 20 years, researchers at the Thomas J. Watson Research Center have produced computer software - called Watch Me Read - that helps children learn to read.This unique interactive program allows children to practice reading and lets them know how they are doing.
"Listening and responding to children read aloud has been the holy grail of technology-based literacy instruction," says Peter Fairweather, senior manager of the education solutions group at Watson. "Without speech recognition, technologists had to resort to off-center teaching strategies using picture matching or rebuses. No one could get close to what real reading is like."
It's not that people haven't tried. There have been attempts at IBM and elsewhere to develop a product that recognizes the speech patterns of children. What stopped most efforts was the difficulty of creating a separate model for children's speech.
Previous efforts had been hampered by the expense and size of the project, and a lack of computing power. But two years ago, Fairweather teamed up with Don Nix, a psychologist at Watson who has spent considerable time working in classrooms with kids, and Bill Adams, a veteran system designer. Together they decided they would take a serious run at the problem.
Watch Me Read was designed for use in the classroom, and aimed at children ages five to eight who are just learning to read. Nix emphasizes that the program is not intended to replace a teacher, but to offer children additional opportunities for feedback in crowded schoolrooms. And it can also be used at home.
"I don't think it's realistic to imagine you can teach reading with computers," says Nix. "A computer will never replace a teacher. And when it comes to comprehension, it's even less feasible. On the other hand, you can enhance skills and provide practice on a computer."
How Children Talk
There were two major technical challenges in developing the software. The first was to get the computer to recognize children's speech. The second was to develop an interface that would appeal to a child.
The voice-recognition problem was the diciest. Children have higher voices than adults and don't pronounce words consistently. When children's voices were tested with an adult speech-recognition system, the accuracy rate hovered at an unacceptable 75 percent. A new acoustic model had to be built.
Nix, with three graduate students from Columbia, began the process by collecting children's voice samples from all over the country. Subrata Das and Michael Picheny of Watson's human language technologies department analyzed the data and came up with a new speech model. Their colleague, Jean-Michel Le Roux, then integrated the new model into the ViaVoice tools used to build Watch Me Read.
Tests indicated the computer was 95 percent correct in recognizing the children's words. Says Nix, "We actually showed that the technology does work for kids, which was something that other people had not done." But it couldn't work just for children. "It was important," says Das, "that the model accommodate both children and adults, since that would allow a parent or teacher to help a child if needed."
A Kid-Friendly Interface
The next issue was making a program that would hold a child's attention in the midst of all the other distractions of a classroom. "The original idea," says Fairweather, "was to create a program that would mimic sitting on a couch with a child and a book, reading a little bit, and asking the kid to fill in a few words." And this is just what they have done. After loading Watch Me Read, one sees a replica of the pages of a book, say, Morris the Moose, with a charming addition: a little panda bear who walks, talks, wiggles and somersaults her way across the page, reading sentences aloud, then prompting the child to read a phrase alone.
"You are a funny looking moose," the panda reads, then repeats, "You are a funny looking ?" The child should then read "moose." If she says "mouse," the computer will tell her it doesn't recognize the word and ask her if she would like to try again.
As the child works through the book, the program does several things: it records the story as it is read by the panda and by the child, as well as any comments the child wants to make about the story. When the child has finished reading the book, the recording is then played back in the form of a performance, complete with the raising of a theatrical curtain and applause. Using a small, golf-ball-size video recorder, the child can make a video presentation to accompany the reading. As a more traditional study aid, the program also makes a list of the words with which the child had trouble.
Adams, who is manager of the K-12 tutoring group and an experienced developer of video games, programmed the user interface. "With video games," he says, "basically I'm in control and you're on a Disneyland ride. You make a direct response to get to the next scene, but I'm guiding the action. But with Watch Me Read, the goal was to give the kids more freedom of movement. On the other hand, we didn't want them to stall and not know what to do next." His job was to ensure that the system was "kid-friendly."
Adams accomplished this in various ways. For example, he wrote the code so that the program would automatically adjust itself to the child's reading level. If the child knows most of the words, the panda hangs back and lets the child read. If the child is hesitant, the panda reads most of the text, leaving only a few words for the child to sound out. The child also can set the difficulty level, choosing one of seven reading levels.
Children, of course, are not the only Americans who need help reading. One-third of adults are functionally illiterate, and Watch Me Read may help them as well. The product is being used in a pilot project in Philadelphia for an adult literacy class. And in the future, the technology may be adapted to teach foreign languages.
The software will be brought to market by IBM's K-12 Industry Solutions Unit after the final component is finished, in mid-1998. Known as an authoring tool, the software offers step-by-step directions for teachers or parents to scan in new books and record the text. The program automatically adds the new words to its vocabulary. Then, children will be able to head off into their imaginations with book in hand, and eyes glued to the computer monitor.
Emily Benedek is a writer based in New York City.