The Center for Data Innovation spoke with Mait Müntel, CEO of Lingvist, a global startup that uses algorithms to improve the classroom language learning experience. Müntel discussed how some of Lingvist’s technologies originated from nuclear physics data science work.
Gillian Diebold: How can Lingvist improve the classroom language learning experience?
Mait Müntel: We built Lingvist as an extremely efficient self-study application; however, it turned out that when language learners use it in parallel with classroom learning they see an amazing boost in their results.
The secret is that Lingvist can teach some things, such as vocabulary, much more efficiently than a teacher can. When students use our platform even for 10 minutes per day, the jump in knowledge and ability to use the language increases drastically, and teachers see this. They can advance much faster in the classroom, as teachers can then focus on things they can teach the best, such as engaging in real-world conversations or pronunciation.
Sara, a language teacher from Desert Ridge High School in Arizona, USA, said that she has been teaching languages for 20 years but has never seen students learn languages this fast. At the beginning of each lesson, she has her students use Lingvist for 10 minutes. “We’re doing things in the first year that we have never even done with third-year students,” she said of her experience using Lingvist.
A perfect symbiotic relationship exists between technology and the classroom, which massively improves learning efficiency and makes the learning experience much more enjoyable.
Diebold: What technologies are behind the Lingvist platform?
Müntel: We use different technologies that, funnily enough, partially originate from nuclear physics. Who would have thought that nuclear physics could be linked to language learning? But this is what actually happened. The learning technology was developed while I was working at CERN (European Council for Nuclear Research), after our team discovered the famous Higgs boson—the so-called God particle. The first Lingvist prototype was actually built using CERN data science software tools.
Today, we use two key technologies. The first is for creating highly personalized learning content depending on the level and interest of the learner. When learning from scratch, everyone has to learn pretty much the same 1,000 most frequently used words to get started. We’ve selected those words based on a statistical analysis of how people actually use the language. Nevertheless, once a learner has moved beyond the complete beginner level, our true personalization expertise comes into play. Lingvist identifies what people know and want to learn and automatically generates the most relevant course for them in seconds. You can learn words from general topics, like cooking, traveling, or movies, to very specific topics, like molecular biology or even nuclear physics.
The second technology teaches this well-selected and relevant content in the most efficient way. We use spaced repetition algorithms that are individually adapted to each person. When people learn with Lingvist, they complete simple exercises—we call them flashcards. Sometimes they answer them correctly, and sometimes they make mistakes. These mistakes are crucial to learning because true learning actually occurs after making a mistake. Our spaced repetition algorithms measure how the learner’s memory works and they select the exercises so that every student can learn the most within the time they have. This hyper-personalized technology ensures Lingvist users learn extremely fast because they learn the way in which their brain is designed to.
Diebold: What datasets do you draw on to train your model, and how do you ensure its representativeness?
Müntel: The data we have collected about learning is truly unique in many ways. As our goal since the very beginning has been to improve learning, we have recorded every learner’s interaction with the application with precise timestamps. So we know how learners’ accumulation of knowledge evolves over time. Such longitudinal datasets about learners are truly unique in the world of science. It’s a huge advantage in building better learning technology.
We can take the existing learning data and predict what people will know in the future. We can predict whether they will answer any exercise correctly or incorrectly. When they actually learn and answer those exercises, we will know whether our predictions were correct. This enables us to train our models and improve them continuously.
We know that our data is very representative when we can take new learners for whom we have no previous data whatsoever and adapt to their learning patterns in a few minutes. When they’ve answered 50 words in about 10 minutes, we will already know which of the next thousands of words they know or don’t know with a high degree of certainty. This helps to select highly relevant content for them, making their learning very efficient, as they don’t have to spend time on learning things that they already know.
Diebold: How do you address linguistic slang or different dialects?
Müntel: We have different language courses for larger dialects, like UK and US English or Portuguese and Brazilian Portuguese. Nevertheless, we currently do not focus much on dialects. There are bigger focus areas that are important to us at the moment. We cover the largest European languages today: English, French, German, Spanish, Portuguese, and Russian. However, there are thousands of languages that are spoken in the world. Before deep-diving into dialects that are fully possible to cover with our technology, we consider it more important to expand our offerings to many more languages. We already have technology that enables us to bring many more languages to market, so watch out for your favorites!
Diebold: How do you hope the platform will expand in the future?
Müntel: Lingvist’s main differentiation from other methods of learning resides in its game-changing technology and vast amounts of analyzed learning data, which—when combined—increases learning speed and enhances memorization. The technology is transferable to any field of education, and it’s built with that purpose in mind.
In saying that, however, we are not in a hurry to expand immediately to all fields of education. It’s more important to provide an amazing learning experience in one field in order to deliver true value. Language learning is certainly not the easiest one, so we know that if we can make language learning radically more efficient, then we can transfer this learning technology to other fields as well. Nevertheless, one step at a time.
Today, we can say that we use science and the most intelligent technology to help you learn languages smarter than ever before. For the future, our goal is to help you reveal your true learning potential to learn whatever you need with our amazing technology.