In a lab in Pittsburgh filled with sleek computers, doll houses and an assortment of colorful toys, two scientists are trying to find better ways to teach students who speak in non-mainstream dialects how to excel in school--and in life--by learning to communicate in mainstream English. Their surprising conclusion? By providing children technologies that converse in their own native dialects -- for example in Appalachian, Southern or African American Vernacular English -- schools can more effectively teach students standard English.
The technology tool these scientists have created manifests as a cartoon kid named Alex. A computer programmed "virtual peer," Alex brainstorms with students in their first dialect, helping to build their self-esteem while teaching mainstream English.
What follows is my lightly edited conversation with Professor Justine Cassell (left, above), Associate Dean for Technology Strategy and Impact at Carnegie Mellon University (CMU), and her advisee, Samantha Finkelstein, a doctoral student in Human Computer Interaction at CMU. Together, they have spearheaded a linguistic project that could be a game-changing tool for educators working to teach mainstream English -- and perhaps even foreign languages -- in our nation's neediest schools.
Virtual Peer "Alex" converses with an adult in "mainstream English."
Monica Gray: You've created Alex, who you refer to as a "virtual peer." What exactly is a virtual peer?
Justine Cassell: A virtual peer is a cartoon-like image of a child that moves, speaks and uses its body in the ways that children do. It can interact with children the way children interact with other children. It could just be a playmate. But the ones that we build have learning goals.
MG: How long have you been working on virtual peers? What are some of your learning goals for Alex?
JC: I've been working on virtual peers since 2000. Today, we are working on using virtual peers like Alex to teach children mainstream American English. In particular, we are evaluating the hypothesis that virtual peers can help children to understand that their native dialect is an aspect of their culture to preserve, but that at the same time they need to acquire the mainstream dialect, and use that in the classroom as way to succeed in school.
MG: Could you elaborate on what you mean when you say "mainstream dialect" and why you think teaching it should be a priority in the classroom?
Samantha Finkelstein: Children come into classrooms from all different cultural backgrounds. There is significant research that shows when you are able to bring that cultural background with you into the classroom, you learn better. We work with students who from the moment they come into the classroom are excited about learning, but whose excitement may be tempered, or even stamped out, when educators--very often with good intentions--say quite harsh things about the way their students speak (as in one case, for example, "if you speak like you're from the ghetto, you'll end up in the ghetto"). Perhaps it's out of a desire to help shield the children from prejudice that they are likely to face outside of the classroom, where mainstream English is often necessary for success and class mobility, and use of non-mainstream dialects is misinterpreted as a sign of ignorance or even stupidity, but it makes the children feel bad about themselves and their cultural heritage - a cultural heritage which includes the language of poetry and song. And this kind of harsh criticism and correction has been shown to be simply ineffective at helping the students to learn the mainstream dialect.
MG: Is there something to be said for simply allowing students to speak in their native dialect in order to respect their cultural identity or preserve their heritage?
SF: A dialect or language is as much a part of cultural identity as are certain foods or ways of dressing or a shared set of songs. Nevertheless, in today's America, we recognize that being successful is associated with speaking the mainstream dialect. For that reason, we want children to learn to speak what some educational researchers call "the code of power." At the same time we don't want them to lose the connection to their culture that can be represented by certain ways of speaking in certain contexts.
MG: How do you define 'dialect'?
JC: A dialect is a standardized and shared way of speaking, governed by rules of vocabulary, pronunciation and grammar, and spoken by members of a given community - geographical community, socioeconomic community, or ethnic community, among others. So I don't want anyone to go away thinking that dialect means slang, or poor English. The way I'm speaking right now is a dialect, the way Samantha's speaking is a dialect, the way you've been speaking is a dialect. As a shortcut we use the word "dialect" to refer to a "non-mainstream dialect." In truth, "mainstream dialect" or "standard dialect" is a fiction because we all speak slightly differently from one another, come from different places, and all of those influences change a little bit the way we speak.
MG: Why did you create a virtual student peer as opposed to a virtual teacher?
SF: We work with students who come from marginalized backgrounds, or backgrounds where school was not an option or was not welcoming. For them, learning in and of itself may be threatening. When those students work with virtual peers--computer characters who act like children and who are able to build rapport with them, who are perceived to be equals to them--those virtual peers may be able to call students out on misconceptions in a way that is not threatening.
MG: What are some of the benefits of collaborating with Alex as opposed to a real-life student peer?
SF: A really important thing about peer collaboration as well is that when it's going well, it's super effective. The problem is that it is really hard to know how to collaborate. Teachers rarely go over collaboration skills as part of their curriculum. A lot of classrooms will have times when teachers say things like, "Ok kids, now put your desk together and collaborate," and kids think that means put your desk close to each other, and continue working independently! So virtual peers are able to model good collaboration skills, and structure the environment for maximal learning.
JC: So in some sense virtual peers mix the benefits of peer learning with the benefits of learning from a teacher. The virtual peer can know how to evoke answers from a child in a way that another child might not know. They know how to collaborate because we've built those skills into them. They may know enough about the subject domain so that the real child actually learns about that subject domain. And yet at the same time they have the skills to build a personal bond that allows children to learn.
Virtual Peer "Alex" speaks in "Standard Dialect" with student.
Virtual Peer "Alex" speaks in "African American Vernacular English" with student.
MG: What dialects are you currently researching?
JC: Our current work has focused on African American Vernacular English, because that's a non-mainstream dialect that is spoken widely in the two cities where I've conducted this work. It's a dialect of English spoken primarily by African Americans, although not only by African Americans, and certainly not spoken by all African Americans. But many African Americans in the US speak some African-American Vernacular some of the time, and in those cases it serves as an index of identity. So we know, for example, that President Obama switches back and forth from the mainstream to the vernacular, probably as a way of indexing his affiliation with multiple cultural bases. There are other dialects of English that are just as important indices of cultural identity-there's Appalachian English, there's Southern English--and in all of those cases we've seen ourselves, and we've read about, cases where children speaking those dialects come to school and are denied learning opportunities because the teacher can become so focused on fixing their dialect as opposed to listening to their performance in the subject which they are studying.
MG: I've heard it referred to as "code-switching."
CS: Exactly. What we are doing with Alex is specifically evaluating two approaches to teaching children mainstream English. The a priori for this particular project is that children in the America of today, in order to be successful in the world, need to speak the mainstream dialect. We may not like that fact. We may believe that anybody should be able to speak however they want. We may believe that the way we speak is not an index of how smart we are, but we all recognize that policy-wise, something would need to change pretty radically for anybody to be able to speak any way they wanted and still be successful in the world. In today's world, children need to learn to speak the mainstream dialect. So, we're exploring the best way to teach children the mainstream.
MG: How did you test the code-switching virtual peer model in classrooms?
JC: The version that many teachers told us they wanted was to have a virtual peer that spoke only the mainstream, to serve as a model of how teachers would like students to speak without the hierarchy implicated in the teacher-student relationship. So that version of the virtual peer brainstormed with the child about the science task they were working on in mainstream English, then said, "OK now let's practice what we're going to present to the teacher," and practiced the presentation in mainstream English too. We contrasted that with the code-switching hypothesis, where the virtual peer brainstormed in the dialect that the child spoke, and when we say that, we actually mean that. We spent almost a year doing a grammar investigation in the schools where we worked. We worked with several linguists who made sure that syntactically, phonologically, we had correctly characterized the language that these kids spoke at home and at school. So, in that version of the virtual peer, the virtual peer brainstorms with the child in the non-mainstream dialect, and then says just like the other one: "Ok, now let's practice the presentation we're going to give to the teacher." And it adds one extra sentence: "I think they like it when we speak school English." Then it switches to the mainstream dialect, and practices the presentation in the mainstream dialect. That's the only difference between the two.
MG: How long did you run this experiment and how did you measure impact?
SF: This was a study over six weeks with an additional two weeks of pre- and post-testing. So, before and after the intervention took place, we gave kids a number of assessments and language assessments that would give us access to what the kids were doing in regards to dialect use and science reasoning. So, for example, how they were speaking when they were just talking to a peer, versus how they would speak when they were talking to a teacher. And in terms of science reasoning, whether these third-grade children were able to form hypotheses and describe evidence for their hypotheses. We also asked them a series of questions about their language ideologies too. For example, how much they agreed with several sentences. Like, "I think it's ok for kids to speak dialect when they're just hanging out with their friends." Or, "I think kids can sound smart when they speak dialect." We also looked at how engaged students were in the task, and we looked at the quality of their science reasoning as they were interacting with the virtual character over the six week period.
MG: How did the students who worked with the code-switching version of Alex perform?
JC: In order for this all to make sense I have to tell you a tiny bit about the experiment. What we did was ask kids to collaborate with Alex and brainstorm about a science task, and then to prepare a presentation for their teacher about what they found. It's a pretty normal science task for these third graders - they look at a picture of an animal in its habitat and they are supposed to come up with ideas about what the animal eats, where it sleeps, how it escapes from its enemies and things like that. Alex and the kid collaborate on coming up with ideas, which is what teachers in third grade generally do with tasks like these - the teachers say things like "break up into groups and work with your partner to come up with hypotheses about the animal and then find evidence for your hypotheses."
Now (and can you tell I'm building the suspense here before I answer your question and tell you the results?), let me say by way of caution that we have data currently on twelve students - six students working with the code-switching agent, and six students working with the mainstream English-only agent. Only twelve so far because an experiment like this, where students work with Alex one at a time once each week was a great way for us to see how the kids really interacted with the virtual peer, but it was time-consuming - to collect the data and then to analyze all of it. However, even though there were only 12 students total, the results are strong and, as well as statistically significant results, we also have a good detailed sense of what was going on.
So, in fact, we looked at three different ways that virtual peer's dialect pattern might have impacted students: first of all, their performance on the science task they were doing with Alex, next their own use of dialect, and third what is called their "language ideology" - that is, their beliefs about the relationship between non-mainstream dialect and intelligence, and about the merit of people who use non-mainstream dialects.
SF: Okay, to know the impact on children's performance in science, we measured that by calculating the amount of talk that would be considered good science reasoning. Coming up with hypotheses, like "I think the animal doesn't eat the bunnies," finding evidence for the hypotheses, like "I think he doesn't eat the bunnies because he doesn't have sharp claws or anything to eat them with, and he doesn't look fast enough to run after them," and asking question, like "I wonder if he sleeps on his belly so that the spikes on his back don't hurt the worms in the ground." And the results showed that after six weeks of working with the virtual peer, students who worked with the code-switching version of Alex demonstrated almost twice as much science talk during the preparation of a presentation to the teacher as the students who heard Alex speak only in mainstream English the whole time!
We also found that after six weeks, during their last session with the agent, the students who heard Alex code-switch were more likely to significantly increase their use of mainstream English for their teacher presentations, compared to the students who had an agent who spoke mainstream English the whole time. That's kind of wild, because what we found here was that kids who hear mainstream English only end up using less mainstream English than children who heard African American Vernacular English and mainstream English!
Finally, the third result that I think is most exciting in terms of the implications for building technologies in the future: using a questionnaire designed for children, we found that, compared to what they believed at the beginning of the six weeks, the kids who worked with the code-switching version of Alex were more likely at the end to agree with sentences like, "kids can sound smart when they speak dialect," and admit that they themselves spoke dialect sometimes. Conversely, kids in the mainstream English condition who were working with the technology - which, by the way, is the kind of mainstream English-speaking technology currently designed for classrooms were less likely to agree than they were at the beginning -- were less likely to agree that kids can sound smart when they speak dialect. Remember that all of these kids spoke African American English themselves, so these kids in the mainstream English condition, using technologies like they hear in their own classrooms, are basically ending up dissing the way they themselves speak.
JC: For each of these results, though, we need more work to understand exactly what's going on here. However, in the meantime, these results are quite striking the way they are. And to have found the results we found really tells us something about not just our project but about non-technological approaches to teaching mainstream English. We know that chastising children for speaking a non-mainstream dialect has negative effects on their self-image; here we've learned that modeling code-switching from non-mainstream to mainstream English improves their self-image and effectively reduces their use of the non-mainstream dialect in front of the teacher. In addition, this experiment demonstrates the positive effects on learning science of brainstorming in one's home dialect, without negative feedback from the teacher. These results are also a wake-up call to attend to the dialect spoken by existent classroom technology - systems that read books aloud, or computer tutors that give children feedback on their work. These everyday educational technologies - which often take on the form of a child (and, in my experience, quite often take on the form of an African American child) - are speaking mainstream English without anybody questioning the effectiveness of that choice. And yet here we saw that a technology that looks like an African American child and speaks only standard English results in lower learning of science and increased use of dialect!
MG: How much would Alex cost a U.S. teacher today?
JC: The program runs on a cheap laptop. It can be projected onto a beautiful 50-inch screen, which costs about $2,000. It can also be projected onto a wall, which costs nothing; many classrooms have projectors, even in under-resourced schools. And they can essentially have the same virtual peer that you would have on a fancy 50-inch screen for the price of the projector and a cheap laptop.
MG: Is the program available online now for teachers to download?
JC: No, we're not at that point. The reason we're not at that point is we are first and foremost scientists and not entrepreneurs. We haven't taken any shortcuts. When you take a product to market, inevitably you have to take some shortcuts in what you do. At the moment what we've done has been for scientific rigor, to determine which version of this works best. And to do that in a human centered design way - that is to build, and then assess effects on children and then rebuild until the technology improves learning gains. In addition, because speech recognition is very difficult with children, for technical reasons, and because speech recognition with children who speak a dialect other than the mainstream dialect is nonexistent, we can't have this yet be a stand-alone system.
MG: At what point would you consider this research a success?
JC: I would call it a success if maybe just some number of teachers and school systems recognized that dialect is not an index of intelligence. And it's not bad English. It is an index of identity and culture. As such, it needs to be respected and not dissed, it needs to be separated from the teaching of school subjects like Math and Science. That would be a success.
MG: If you had a blank check, how would you spend it?
JC: We are looking for that blank check! This has been a very hard project to fund because it is so politically charged in the America of today, particularly over the last year or so where the discourse around race has become so charged, and where African Americans have suffered from particularly bad treatment and, in response, many are trying to blend in as much as possible, including through diminishing their use of non-mainstream dialect as much as possible. Parents are worried about whether they are going to even see their teen boys at the end of the day, school principals are worried about how to keep the black children in their classes safe and so, quite understandably, they don't want to hear about the ways in which AAVE dialect is a positive index of culture and identity - they want to hear about ways in which they can help their children blend in, not just in the classroom but also on the street, and that includes through their use of language.
In this context, I have had tremendous difficulty in talking about our work. We found ourselves at odds with a school that no longer wanted us to do the research, and asked us to cease working with their students because they felt that the work we were doing would convey a negative stereotype of African American identity. There are many people who think like that. There are many African American colleagues of mine, and teachers I know, who have said, "In order to be a functioning successful member of society, you need to speak 'the mainstream dialect.'" I need to be very clear that I agree about that. The question we're asking is: What's the best way to teach the mainstream dialect? Because the way that schools are using now isn't working.
So, a blank check would allow me to work with more students in more schools in more cities, to work with children who speak other non-mainstream dialects (Appalachian English, Pittsburghese, and so forth) to make sure that our results generalize to a larger population, and to understand the ways in which different students may succeed in learning mainstream English in different ways. Work like this requires people to build the technology, go into schools, work with children - and also to work with teachers and school systems to understand the larger context in which we want children to be able to maintain their relationship with their home culture, and the language of poetry, song, and story of that culture, while still learning the mainstream English of school success.