Language learning app eyes up B2B and edge opportunities

Memrise discusses edge computing and corporate training possibilities for its immersive AI-enabled language learning app.
7 September 2023

App developers are exploring edge computing options to provide offline AI capabilities for device users.

Getting your Trinity Audio player ready...

What happens when a grand master of memory, a motorcycle-restoring expat, and a neuroscientist team up? In the case of Ed Cooke, Ben Whately, and Greg Detre, the answer is that the language-learning app Memrise gets built – a process that included driving a bus across Europe to film a video dictionary of native speakers in 10 countries.

To keep operations on the road and make sure that the language learning app hit its targets, the trio recruited Steve Toy – who’d gained COO experience working in the device software space – as CEO. Today, the application’s tech stack includes numerous AI integrations, such as the use of large language models (LLMs), as well as speech-to-text (STT) and text-to-speech (TTS) technologies.

Relatively inexpensive AI

Developers can sometimes have reservations about using generative AI based on the costs of GPU time, but Toy has a different view. He makes the point that it’s much more affordable to synthesize content rather than having to film everything from scratch, recalling the founders’ trip in the ‘Membus’ in the early days of the language learning app.

However, while the technology behind the user experience may have changed, the foundations can still be summarized in three words – learn, immerse, and communicate. “First, we teach you vocabulary that you’re actually going to use. Next, we show you that language in real-life contexts, through immersive videos. Finally, you use what you’ve learned to practice conversing confidently,” writes Memrise, describing its approach to language learning.

Digging a bit deeper into how the app works, users build up an ever-expanding dictionary of words, and experience how phrases are used and spoken by watching videos. And while this may just sound like a stock language learning app or online training experience, Toy emphasizes that Memrise does things differently. And he’s right.

Users have the chance to upload any video and turn that content into a lesson that’s customized for them. “Technology has kicked that door open,” Toy told TechHQ.

Thanks to AI tools working behind the scenes, users can select any scenario that appeals to them and understand how that plays out in a foreign language. Considering corporate training opportunities, the functionality could be a game-changer for firms that want to equip their staff with terminology in areas that matter to the business. And, in the future, Memrise could find itself supporting B2B services alongside its regular B2C revenue streams.

Toy explains that there’s more to the custom courses than just being a translation from one language to another. It’s important not to drop language learners straight in at the deep end, as users will grow frustrated if all of the words and phrases are completely alien to them.

Instead, the software compares the content with the user’s language learning app dictionary to find segments that are likely to be comprehensible, allowing proficiency to be built at steady intervals. And, again, you can see how this approach would work in a business environment, as well as for classic travel scenarios such as ordering food in a restaurant.

When it comes to practicing their skills, users have the option of interacting with an AI-powered chatbot. Conversational AI has made it possible to deliver realistic language training as never before. Chatbots can listen to spoken phrases, speak their replies, understand text, and offer words of help.

Language learning app Memrise CEO Steve Toy

Steve Toy, CEO of Memrise. Image credit: Memrise.

This year, Memrise partnered with Discord to bring its language learning app chatbot to the popular instant messaging social platform. Discord members can converse with the advanced chatbot, which is supported by GPT-3, much like they would chat with their friends on the server. And users have two options /learn_solo and /learn_together.

Reviews of the Discord app are positive and describe the experience as a fun way to learn a language. It gives users the chance to run through a scenario ahead of carrying out the same task in the real world. And it’s easy to see how businesses could benefit from such a tool to help train their staff.

In the past, building such a chatbot would have consumed a lot of programming time, with developers having to craft numerous rules. All of the user interactions would need to be hardcoded. However, that all changed with the introduction of GPT-3 and subsequent LLMs.

Conversational AI algorithms based on next-word predicting neural networks are happy to chat about a wide variety of topics, given the vast size of the training datasets used by OpenAI and other foundation model creators. Plus, those conversations can take place in multiple languages, which is remarkable.

Edge computing options for language learning apps

The future for language learning apps certainly seems like a bright one, but there are still some issues to solve. Currently, Toy and his team are looking at edge-computing options to bring AI features closer to the device. He wants users to be able to develop their language skills on a plane, for example, when internet connectivity isn’t available.

The next step is to identify which part of the LLM is relevant to the language learning app and cleave that portion off to user devices. Many software developers are in a similar boat, relying on API calls to the cloud to provide AI-enabled features. In the future, LLMs that have been constrained to support particular types of queries and with a much smaller footprint, could pave the way for serving features locally.

On TechHQ we’ve written about how it’s possible to dynamically trim the number of artificial neurons that participate in model inference. Companies such as ThirdAI have shown how the approach can lift the performance of algorithms running on CPUs to levels typically only seen using GPUs. And it’s something that could benefit app makers with AI features such as Memrise.

Finally, it’s worth covering off the competition between Memrise and other language learning apps such as Duolingo or OpenAI-backed Speak. If you’re asking what’s the best language learning app, Memrise or Duolingo – to highlight a popular search query – then you could be missing out on the opportunity to use multiple resources.

The more practice, the better, and the reward of using a combination of language learning apps is that you get to fire more neurons (real ones, not just AI versions) more often. It’s a strategy that’ll help vocabulary, key phrases, and sentence structures to stick in the brain for when you need them to seal your next business deal.