Duolingo Bears are Feasting on a Red Herring

The stock I am buying most heavily right now is Duolingo. It’s already a top 5 holding, and I wrote one previous article about the company, but it is time to revisit because I never realized just how misunderstood this company is. I’ll break down the proposed bear cases in a series of articles. I did this a long time ago for SoFi when I wrote for Seeking Alpha and it ended up being very valuable to me to be able to reference in future articles. The Duolingo bear cases basically fall into three categories: disruption from real-time translation, disruption from AI-generated competition, and valuation.

This article will focus on translation. I’ll cover the argument first, followed by why I think it’s flawed, and then we’ll look at the data.

Real-Time Translation as a Bear Case

The argument goes like this. Duolingo is primarily a language learning app. Therefore, its primary use case is to communicate with other people in a foreign language. If there is a suitable alternative to communicating with other people, it will eliminate demand for the product.

Here is Google’s recent demo where Google translates a conversation with Jimmy Fallon.

And here is a recent marketing material from Apple showing off live translation on their new Airpods (link goes to X)

https://x.com/i/status/1965462807253827725

And now Meta has joined in the fun with live translation and live captioning on their new smart glasses. Those are impressive tech, and the latency will only get better with time. Nobody is going to waste their time learning a new language when they can just put in an earpiece and AI will translate what is being said in real time. That sounds reasonable…

Real-time Translation is Actually a Red Herring

…if you’ve never learned or spoken a foreign language. I speak two foreign languages. I learned German and spent two years as a missionary in Germany. I learned Spanish because we visited my wife’s grandparents, who only spoke Spanish, multiple times a week for well over a decade. I’ve also taught others to speak a foreign language. After returning from Germany and while still an undergrad, I spent two-and-a-half training other prospective missionaries in German-speaking countries.

Here is what people who speak foreign languages understand that those who only speak one do not. Real-time translation is no substitute for learning or knowing a language. There is almost no overlap between the two. I know this sounds backwards to a lot of people, especially in the USA where only about 20% of the population is bilingual. However, the main use case for learning a language is not so that you can have a conversation with people when you go visit a country or you're at a business meeting.

Translation is Transactional Only

Go back and watch those demos again. There is still significant latency. Imagine how labored and difficult having a full conversation would be with those lags. “But Chris, that latency is only going to get better and at some point will be instantaneous”. That’s true, the processing will get so good that the time it will take to translate the words and sentences will be, for all intents and purposes, instantaneous. That will not solve the latency problem. I’ll prove it to you.

All languages are unique. The grammar, sentence structure, intonation, cadence, and wording are always unique. That means that even if the translation process is instantaneous, having an earbud doing the translating for you will still have lag. Here is the above paragraph written in English but with German grammar:

I speak two foreign languages. I have German learned so I two years as a missionary in Germany spend could. I have Spanish learned because we my wife’s grandparents multiple times a week for well over a decade visited have, who only Spanish spoken have. I’ve also others taught, a foreign language to speak. After home from Germany returning and while I still an undergrad was, have I two-and-a-half years prospective spent, in order other prospective missionaries for German-speaking countries to teach.

See what I mean? If you are translating from German to English or English to German, you can’t just translate in real time as the words are spoken. You have to wait until the end of every sentence or clause before you can start translating. Even with quantum computing, there is going to be latency in “real-time translation”. The use case for real-time translation is during transactional conversations including asking directions, understanding what that cool building is, or buying flowers at a market like in the Apple commercial. Even a business meeting can be done this way because the parties involved have a vested interest in exchanging information and will remain engaged for that reason. And honestly, Google Translate has been able to solve all those use cases for years already. Yes, the latency is going down, but this tech isn’t new, it’s just slightly faster, and it’s still only useful for transactional conversation.

Real-time translation will forever fail in casual conversation. Nobody is going to stick in an AirPod and have anything remotely resembling a real conversation with a friend. There is too much friction in translation. Even with the intonations, which were really good in the Google demo, there is still too much that is lost in translation. The syncing of gestures and eye movements, the timing of the jokes. The innuendo and subtleties of a normal conversation. Nobody learns a language so they can buy flowers or go to a business meeting. They may use Duolingo for a few days to get a feel for French before visiting the Eiffel Tower, but that individual has no value to Duolingo anyway. They are using the free version for a few days and then churning off anyway. I couldn’t care less about that as an investor.

It’s an Answer Looking for a Question

Why would I use Duolingo if AI can translate in real time is a question built on a flawed assumption.

Consider the following similar questions:

Why would I learn to play chess if a computer has been better than the greatest human grandmaster for over 20 years?

Why would I learn basic arithmetic since the pocket calculator has been around since 1970?

Why practice an instrument if a synthesizer can play the notes perfectly every time?

Even better, why learn anything about music at all if AI can just generate songs?

The assumption that people are making is that the main reason people learn a language is to be able to translate. That is incorrect. Very few people learn a language to be able to translate it. Just like you don’t learn math so that you can calculate the product of 468 and 3,792 in your head and you don’t learn an instrument so you can hear it played by a master. You learn a language to progress, to immerse yourself in another culture, to better yourself, to be able to connect with your own heritage, because you think it's fun or the foreign language is beautiful, to connect with a loved one, to maintain a language you already know, and a hundred other use cases that are rarely translation-related. For non-native English speakers, learning English is often a stepping stone to higher-paying jobs. That will continue to be true even as real-time translation gets better.

Which Opinion is True?

You can make a case for why translation disrupts Duolingo. I can make a case for why it doesn’t. I think my case is based on first principles and is correct, but is there a better way to determine who is right? Maybe we can look at the actual data? Does Duolingo keep track of why people learn a language? Yes, they do. So why do people start learning a new language?

English, Spanish, and French are the three top languages worldwide. It’s English first, by a wide margin, and then the other two. Above you find the chart provided by Duolingo for the main motivation of their new learners last year. If your reason for learning a language is for self-improvement, social purposes, or entertainment, then the ability to translate a language will not affect your demand for the app. The top three categories are absolutely and unequivocally not affected by translation.

Education won’t stop with better translation, we just established that translation does not support connection between people, and translation isn’t going to stop people who are using it just to have fun. “Spend time productively” is also pretty safe. The perception of “productive” may change slightly if people think that learning a language isn’t productive. However, I think it’s pretty clear that what is meant by productive is that they actually feel good about what they are doing. The bigger driver here is that people feel better using Duolingo instead of doom-scrolling TikTok, Instagram, or Facebook. And that will remain true. So the top four reasons for using the app are virtually unaffected by the advent of better real-time translation. Real-time translation also has no effect on people who start using the app to boost their career.

The only category truly affected is people preparing for travel. If you are going on a trip and you don’t have time or desire to become fluent, but you want to be able to function while traveling, or understand some key words and phrases, then these real-time translation items, whether earpieces or glasses, will almost fully fill that use case. As covered above, if these people are only using Duolingo to prepare briefly for travel ahead of a trip, they are almost certainly not high-value customers to lose. And those who are looking to travel to fully immerse themselves in the full cultural experience may still use the app anyway.

If real-time translation were a problem, it would have already eroded their business

I think the strongest argument against real-time translation undermining Duolingo’s growth is actually very simple. This type of translation has been available since long before Duolingo went public. Here is a demo of real-time translation from Google from 10 years ago on YouTube

Again, the latency and accuracy are better today than 10 years ago, but this functionality existed in 2015. That is the same year that Duolingo began to monetize its app. So literally every dollar they have ever made has been made in a market that included real-time translation as an option for their users. Duolingo went public in 2021, when Google Translate already had low latency and high accuracy. For anyone who thinks Duolingo will be disrupted by real-time translation, I have one simple question. Why is this time different? I have yet to see a single good answer to that question. I am convinced that real-time translation does not represent a credible risk to Duolingo’s business.

Subscriber Info

The DDI YouTube Channel is live. Work has been extremely rough this year and demanded a lot more of my time. I’m going to get back to a more regular cadence of articles and YouTube videos as things should be settling down some in the coming months. If you want to support my work, you can subscribe here on substack, or on X, or, now, on YouTube.

Paid subscribers get three perks.

1) Access to a private X chat.

2) I buy stocks every week, and every week I send out my weekly DCA weighting list to subscribers

3) I send out a portfolio snapshot at the beginning of each month which shows my total allocation to each of my positions.

If you are a paid subscriber and not in the X chat, please email me at datadinvesting@gmail.com and let me know and I’ll get you added. If you have any other ideas for things I can do to bring value to my paid subscribers, don’t hesitate to reach out.

You can also use my referral links for SoFi if you want free money when you sign up for SoFi, or discounts on Tesla or finchat.io:

SoFi Money Link - Get an extra $25 when opening an account
SoFi Invest Link - Get an extra $25 when opening an account
Tesla Referral Link - Get a discount on a new Tesla
finchat.io Link - Get 15% off

Disclosures: I have long positions in DUOL

The information contained in this article is for informational purposes only. You should not construe any such information as legal, tax, investment, financial, or other advice. None of the information in this article constitutes a solicitation, recommendation, endorsement, or offer by the author, its affiliates or any related third party provider to buy or sell any securities or other financial instruments in any jurisdiction in which such solicitation, recommendation, endorsement, or offer would be unlawful under the securities laws of such jurisdiction.