How the science of recall is finally helping us to learn other languages
A FEW YEARS AGO, Captain Emmanuel Joseph decided to learn Arabic before his deployment to Iraq. “At first it was easy,” he told me. At his base in the U.S., he explains, “we had native speakers teaching us basic things like greetings; imperatives like stop, go, walk; and some numbers and nouns. It was very much survival-level.” In Iraq, Joseph (not his real name) continued trying to learn Arabic with Al-Kitaab, the main textbook used by American universities and the military. But he struggled.
“I was forgetting more than I was learning,” he said. “With every chapter in the textbook came a hundred more vocabulary words. The language and the culture were accessible, but I also had a job to do. So I didn’t—and couldn’t—spend all my time studying.” Joseph cast about online for help and came across LinguaStep, an online Arabic-language program that quizzes a user in vocabulary and adapts to a user’s specific rate of learning.
LinguaStep was first developed in 2006 by Loren Siebert, an energetic computer-software entrepreneur with coppery hair crowning a triathlete’s build. Siebert has packed several lifetimes into his 40 years: computer coder at age 9, programmer for the Department of Defense at age 15, Marshall Scholar at age 21.
Siebert decided to learn Arabic on something of a lark: He took an aptitude test that told him he’d be good at languages. He thought Arabic was beautiful. So he signed up for a beginners’ class at the University of California, Berkeley. Like Joseph, Siebert struggled with the vocabulary. “Arabic is a language of memorization,” he said. “You just have to drill the words into your head, which unfortunately takes a lot of time.” He thought, “How can I maximize the number of words I learn in the minimum amount of time?”
Siebert started studying the science of memory and second-language acquisition and found two concepts that went hand in hand to make learning easier: selective learning and spaced repetition. With selective learning, you spend more time on the things you don’t know, rather than on the things you already do.
Siebert designed his software to use spaced repetition. If you get cup right, the program will make the interval between seeing the word cup longer and longer, but it will cycle cup back in just when you’re about to forget it. If you’ve forgotten cup entirely, the cycle starts again. This system moves the words from your brain’s short-term memory into long-term memory and maximizes the number of words you can learn effectively in a period. You don’t have to cram. In fact, if you do cram, you might learn a set of words and do well on a test, but you’ll forget it all a few weeks later.
Programs existed that followed the concept of spaced repetition, but for most of them, you had to input all the words you wanted to learn, which was tedious. And they didn’t have audio. “Memorizing the presidents is different from memorizing the first chapter of your Arabic textbook,” Seibert said. “You need the sound. You also want it to be tied to the textbook you’re using, not just a giant pile of flashcards.”
“There are subtle, key things that make all the difference in programs,” he added. “It’s like Google versus Lycos. Rosetta Stone is flashy and beautiful, but it didn’t work with my curriculum.”
Siebert’s program had audio and was keyed to the Al-Kitaab curriculum. By using spaced repetition, he programmed it to adapt to the user’s rate of memorization. He began to excel in class, and classmates asked him to share the software. So he put it online—and added a social component that allowed students to critique each other’s work. A year later, Siebert was asked to teach the class.
LinguaStep turned out to be an ideal tool for Captain Joseph. “It gave me a workout each day, and I did it.” He told me that his studies paid off within a year. In speaking with Iraqis, “I’d hear a word, and I could approximate the general topic of what they were talking about. I could quality-check the work that translators were doing for me. The benefit cannot be overstated.”
ARABIC IS ONE of the languages the U.S. Department of State dubs “extremely hard.” Chinese, Japanese, and Korean are the others. These languages’ structures are vastly different from that of English, and they are memorization-driven. But Arabic brings the importance of language learning into especially sharp relief. Even today, six years after LinguaStep (and nearly 12 years into the wars in the Middle East, not to mention fast-changing foreign-policy shifts with a number of Asian countries), there has been little new technology to help coordinate the myriad aspects of language learning.
“A comprehensive online solution is everyone’s holy grail,” Phil Hubbard told me. A senior lecturer in linguistics and director of the English for Foreign Students program at Stanford University, Hubbard has been studying computer-assisted language learning for close to three decades. “A lot of the people developing these programs have a good idea, but no particular experience in language teaching,” he said. “They leverage one part of it, but don’t do the other parts well.”
Joseph had tried Rosetta Stone, to learn French. “It’s not totally ineffective, but over time it was mind-numbing,” he said. “The problem is that the grammar rules are never actually explained. So you spend a lot of time trying to figure out what the point is.” When he started advanced Arabic classes in the next stage of his military training, he found himself once again slogging through vocabulary lists and grammar rules. After he had tried all the leading language software made available to him by the military, Joseph sought out LinguaStep’s Siebert anew.
He asked Siebert to customize LinguaStep specifically to his curriculum. Joseph shared the new version with his fellow students, who found it extremely useful. While Joseph went through his own chain of command to try to get the program into the curriculum, Siebert reached out to the Defense Language Institute, which is responsible for providing language instruction to the Department of Defense, and the Defense Advanced Research Projects Agency, which is responsible for developing new technologies for the military. Neither man had success.
“I said, ‘I have this program,’” Siebert recalled. “‘It’s done. Do you want to distribute it and get it into people’s hands, to give them another tool to use?’ They said, ‘We’re sorry, but it’s not something we can do.’ The Catch-22 is that if it’s just generic commercial software that’s off the shelf, like Rosetta Stone, they can buy it. But mine’s not generic, it’s tailored; it’s not shrink-wrapped, it’s interactive and online.” The military has an internal program to develop its own software. “They couldn’t use mine, because they could theoretically develop something like it themselves. But they hadn’t.”
After September 11, 2001, the U.S. government experienced a dramatic shift in priorities. To help meet its language-learning goals, in 2003 the Department of Defense established the University of Maryland Center for Advanced Study of Language. Since then, the government has invested millions of dollars in Rosetta Stone, with special programs for various departments. But in 2008, the Maryland center published a study that questioned the effectiveness of Rosetta Stone, stating that it failed to incorporate critical language-learning principles. Last year, Rosetta Stone lost one of its biggest contracts: the U.S. Army.
MICHAEL GEISLER, a vice president at Middlebury College, which runs the foremost language-immersion school in the country, was blunt: “The drill-and-kill approach we used 20 years ago doesn’t work.” He added, “The typical approach that most programs take these days—Rosetta Stone is one example—is scripted dialogue and picture association. You have a picture of the Eiffel Tower, and you have a sentence to go with it. But that’s not going to teach you the language.”
At Middlebury, for almost a century, students have been getting the equivalent of a year’s worth of language study in seven or eight weeks during summer immersion programs. The languages include Arabic, Chinese, French, German, and Spanish.
According to Geisler, you need four things to learn a language. First, you have to use it. Second, you have to use it for a purpose. Research shows that doing something while learning a language—preparing a cooking demonstration, creating an art project, putting on a play—stimulates an exchange of meaning that goes beyond using the language for the sake of learning it.
Third, you have to use the language in context. This is where Geisler says all programs have fallen short. “A lot of people think that learning with authentic materials”—audio or video in which native speakers are speaking naturally, without a script—“is just a gimmick. But what you will get out of it is all the nonlinguistic cues that you get in a real language-speaking situation. If you are in a doctor’s office, you know what they are saying due in large part to visual and audio clues, not linguistic clues.”
Fourth, you have to use language in interaction with others. In a 2009 study led by Andrew Meltzoff at the University of Washington, researchers found that young children easily learned a second language from live human interaction while playing and reading books. But audio and DVD approaches with the same material, without the live interaction, fostered no learning progress at all. Two people in conversation constantly give each other feedback that can be used to make changes in how they respond.
Over the last decade, the University of Maryland Center for Advanced Study of Language has conducted language research dedicated to national security and intelligence. The center has more than 150 research scientists and affiliates whose work focuses on second-language acquisition, technology, cognitive neuroscience, and less commonly taught languages. “You’re seeing a great movement to put language learning online or on a disk, without a teacher,” said Richard Brecht, the center’s executive director. “But our research shows that the ideal model is a blended one,” one that blends technology and a teacher. “Our latest research shows that with the proper use of technology and cognitive neuroscience, we can make language learning more efficient.”
The center’s scientists now have large-sample research data showing that our ability to learn a language can be predicted. Let’s say a woman with a strong ability to hold things in her memory and manipulate them makes a mistake in conjugating a verb. If her teacher simply repeats the word in its correct form, she can hold the material in mind and keep going. A man with a weak working memory would need his teacher to stop class and go over the correction.
“We can design environments that would respond to those strengths and weaknesses,” Brecht said. “Neither approach is better, but it’s important to tailor the approach to the student. We can cut the time to learn a language in half.”
A FEW YEARS AGO, Middlebury began developing its own online program, Middlebury Interactive Languages, a for-profit venture created in partnership with K12, a company specializing in technology-based online education. The challenge was how to translate Middlebury’s intensive teaching methods into usable software. Geisler said that LinguaStep, with its adaptability and social aspects, can approximate the much-desired “whole toolbox.” But if you’re aiming for the perfect toolbox, “it’s very complex to create an interactive, task-based environment online—and really expensive.” Right now, Middlebury’s reputation is at stake.
The school released its first two online programs, for French and Spanish, last year. The new courses use computer avatars for virtual collaboration; rich video of authentic, unscripted conversations with native speakers; and 3-D role-playing games in which students explore life in a city square, acting as servers and taking orders from customers in a café setting. The goal at the end of the day, as Geisler put it, is for you to “actually be able to interact with a native speaker in his own language and have him understand you, understand him, and, critically, negotiate when you don’t understand what he is saying.”
The program includes the usual vocabulary lists and lessons in how to conjugate verbs, but students are also consistently immersed in images, audio, and video of people from different countries speaking with different accents. Access to actual teachers is another critical component. It’s Geisler’s maxim—use it for a purpose, in context, and in interaction with others—in digital action.
Brecht told me his Maryland center, the one dedicated to research for national security and intelligence, now has the resources in place to conduct the empirical studies that every language scientist craves. “If language learning takes half the time it used to, the return for the government is hundreds of millions of dollars, if not billions,” he said. He hopes that by releasing the work into the public domain, the LinguaSteps and the Middleburys of the world will use it for further innovation.
Middlebury has just released its third online program. It’s for Chinese—one of the “hard languages.” I asked Geisler about the program’s promise for the military. “We haven’t signed any contracts with them,” he said, “but both sides have reached out to each other, and we’ve responded to a request for information from the Air Force.”
The various elements of language learning have long been disparate. Now intensive, integrated online programs like Middlebury’s may radically change the landscape—especially for people like Captain Joseph, for whom learning a language could mean the difference between life and death.