An Interview With Marko Jukic: Nobody Knows How To Learn A Language

In 2010, I founded what has now become vc.ru (sometimes called “Russian Techcrunch”) and spent the first 5 years actively contributing to the website by writing thousands of posts, ranging from from startup coverage to market analytics to — my favorite genre — interviews. Doing interviews is relatively both easy and edifying. You’re not confined to your own limited skull space and having conversations with smart people enriches everyone involved. The interviewer gets to grapple with new ideas firsthand, the interviewee gets to share their knowledge in real time, and the readers, of course, get to hear relevant insights and personal stories in a conversational tone that they otherwise never would.

Sometimes the outcome is a completely barren and insipid conversation that disappoints everyone involved. You feel less addressed than rhetorically bludgeoned by a corporate spokesperson that carefully hides behind the mask of a successful entrepreneur or researcher. Luckily, today that is not the case as we chat with Marko Jukic, an independent theoretical linguist researching language learning and in general, it looks like, a nice person.

I have always been interested in learning new languages. I spent my school years putting effort into dismal attempts to master German. The chances of your reading this text — which I craft mostly by myself — with English words and phrases combined into coherent sentences, are abysmally low. I still feel marrow-level dread in my bones staring at a blank page and in addition to the well-known phenomenon called “writer’s block”, I often feel the existential angst of not being able to force my neural pathways to form new ideas and thoughts in language that I have started to learn so recently. I hope that with the help of Marko, whom I discovered after reading his recent article ‘Nobody Knows How To Learn A Language’, we will solve most if not all of the problems concerning language mastery. And now, without further ado, let me take some time out of my busy schedule of trying to sound smart and give the stage to Marko and his answers to my many questions.

To start things off, can you please tell us a bit about your background? From what I can see online, you self-describe as a “theoretical linguist” who knows a bunch of languages and you currently live in San Francisco, where you’re working on a book that will (hopefully) teach us how to learn new languages. It would also be great if you could explain, in layman’s terms, what “theoretical linguist” means and what a theoretical linguist does.

I call myself a theoretical linguist for two reasons. The first reason is that it sounds impressive. One has to stand out on the Internet these days. The competition for attention is tough. The second reason is that I do in fact theorize about languages most of the day, every day. It’s basically my job to theorize about languages. I study languages and develop explanations, or models, for how they work. I test my models on myself and others that I teach, then improve them based on my experiences. I try to fit all of my models together into one complete explanation for how languages work and how one can learn them effectively — a paradigm, or theory. I am not an academic, in fact I dropped out of my very fancy college in America, but after a few years of entrepreneurship I have figured out a way to theorize about languages full-time. Without getting into too many details, cryptocurrency certainly helped.

My interest in languages dates back to my childhood. I grew up bilingual and, thanks to my father’s work, lived all around the world. Both of my parents studied and spoke, to various degrees of fluency, at least five languages each. The first words I spoke as a baby were, ironically, in Slovenian, which was not a native language for either of my parents. I formally studied French, German, and Azeri in school, then added Arabic, Russian, and Latin in college, among a number of others I studied in my free time. It was around then that I realized there was a critical failure in the mainstream language learning paradigm. What tipped me off was the realization that new language learners’ grammatical mistakes in a new language closely tracked the grammars of their native languages. That’s how I fell into the rabbit hole. The more I thought about it, the more it looked like the problem was theoretical. I began working on this problem seriously and, a few years later, I believe I am indeed getting close to solving it. At least, making significant progress in solving it that others can continue. I am now working on organizing all of my material into a book, while also publishing articles and excerpts online. Hopefully it will take no longer than one year from now. A tall order perhaps!

I really enjoyed your article ‘Nobody Knows How To Learn A Language’ where you make quite a few bold claims. Saying that institutions are a dead end as the means by which one attains fluency in a new language is understandable, as not many people graduate from high schools with any fluency in anything whatsoever. But you continue to disparage seemingly valid ways, such as commercial methods (do I need to cancel that Duolingo subscription?) and immersion, that seem to work for so many people. Can you elaborate: based on what did you come up with these conclusions? Research papers, data, books, experience, or something else? What method would you consider the lesser of those evils?

To be clear, I am not trying to disparage any methods or anybody’s efforts to learn languages. I hope my writing is not interpreted that way. I think that, within the theoretical paradigm for languages and language learning that we have, people are actually acting quite heroically in their efforts to learn new languages. For that they, both learners and providers of methods, deserve applause. I think the theoretical paradigm that underlies all of these methods is the problem. It is the reason that, despite the efforts of so many capable, intelligent, and hard-working people, a very small proportion end up learning new languages to fluency.

By theoretical paradigm, I mean the set of answers to questions such as “What is a language?”, “What does it mean to learn a language?”, and “What constitutes fluency in a language?”, that are assumed or stated outright by each method. Since every method purports to teach a language, they have explicit or, more commonly, implicit answers to these questions. This is maybe getting pretty philosophical, but I hope my meaning is clear. If our paradigm for what language is and what it means to learn a language is wrong, then that would be enough to explain the enormous disparity, broadly, between the inputs of effort and intelligence and the output of fluent speakers. I think the paradigm we have is largely implicit, largely shared by all existing methods, and largely comes from academia. I will describe it in detail in a future article, which is coming soon, so I won’t get too ahead of myself now.

I tend to think that commercial methods and institutional instruction are more or less tied in effectiveness, depending on which methods or institutions we are comparing. I think immersion is notably better, but, as I wrote in my article, very impractical. I reached these conclusions simply through careful observation, which is, of course, one of the basic principles of the scientific method.

The famous investor Peter Thiel has a principle that new technologies should be 10x better than existing ones. That is one order of magnitude better. So to give one example, I cannot remember ever having seen a relative comparison of commercial or institutional methods that argued a difference of effectiveness of orders of magnitude, only less than one order of magnitude. In my view, this is a notable signal of the distribution of effectiveness of methods. I generally try to avoid using statistics to make arguments or model reality — there is the famous quote about lies, damn lies, and statistics and all that. I tend to think statistics are misleading more often than not and I prefer to focus on the models that underlie them. The assumptions made are more interesting and more valuable to study.

I would not cancel my Duolingo subscription. I am a fan of Duolingo. Duolingo is one of the best and most convenient methods we have — it’s free, after all, and you can use it on your mobile phone — but I would be aware of the limitations, which are broadly common to all existing methods. The “least evil” method we have in the existing paradigm is immersion — if you can afford to practice it.

I am not a fan of commercial language practices either. It’s obvious that the business model of the majority of language-learning apps and services that promise to make you a fluent speaker is orthogonal to the interests of their users. Akin to fitness trainers, the goal of the learning apps is not to provide you with sound guidelines for self-practice or to make you a fluent speaker in a reasonable timeframe, but rather to lock you into the system, provide you with random dopamine rewards, and never ever let you go. What is your opinion of commercial services and who is the worst offender here?

As I said before, I do not intend to disparage any methods per se. Within the current paradigm, they are all valuable to some degree or another. Unfortunately, the paradigm sucks. So nothing really works. I don’t really expect people to develop a new paradigm on their own. Linguistics is a hard subject. Theorizing as an activity is difficult and skill at it is rare. The biggest problem with commercial methods as a whole is that they use the same theoretical paradigm as institutional instruction. In many ways, commercial methods are simply repackaging language learning methods from school or college. The way they repackage it is often preferable to school or college, but ultimately the core substance is the same.

Allow me to present an analogy. Imagine a tall mountain. Let’s call it Mount Fluency. The mountain is surrounded by rolling hills and forests. Hikers, of which there are many, want to climb to the top of the mountain. They know that navigating to the top of the mountain will require a lot of effort and intelligence. There will be impassable areas. There might be wild animals. They may run out of supplies. They have a map of the mountain that tells them the best paths to the summit. They look at the map and try to climb the mountain.

But something strange happens. Most hikers can’t get to the top. They get stuck in impassable areas. They get attacked by wild animals and retreat to the bottom. They run out of supplies and give up. A small number of hikers gifted with extraordinary brainpower, willpower, or resources manage to reach the peak. But they do it in spite of the map, not thanks to it. Most people who follow the map fail. But it gets stranger. Instead of blaming the map, the hikers blame themselves. They think that they just need to increase the amount of effort they put into hiking. No matter how many times they fail, they continue to blame themselves instead of the map. They blame themselves for being insufficiently heroic. That’s what I did for a long time. But that is an intellectual dead end. Not everyone can be a hero. Does that mean learning new languages is simply out of the question for the vast majority of people, even smart people? Possibly, but I began to doubt that. Eventually, I realized that we have to explore the possibility that the map is simply wrong.

All of my research has confirmed to me that the map is indeed wrong. To extend my analogy, the map is largely being produced by institutions. Commercial methods are simply reformatted, new editions of the map. The colors and labels on the map change, but the actual content of the map remains almost identical. There is a famous saying by the philosopher Alfred Korzybski that “the map is not the territory”. The map we have for scaling Mount Fluency is not the actual territory that we have to traverse. The formal study of language learning seems to focus on the attempts by people to navigate the mountain using the wrong map, rather than focusing on the territory that the map itself describes — language. To extend the analogy even further, it is unreasonable to expect map makers — producers of commercial language learning methods — to be land surveyors as well, let alone very skilled ones. The problem is not with the map makers. The map makers are very good at making maps. The problem is actually with the surveyors. The problem may very well be that there are not very many surveyors at all.

That said, I think the incentive problem you describe is real, although not the biggest issue. The problem is that since it is effectively impossible to learn new languages to fluency, commercial methods are competing on the dimension of which method is the least painful to use for studying. Users are not really able to evaluate the competing methods on their success at producing fluency, since it is so rare for any of them to achieve this.

I think it is a highly notable fact that Duolingo — of which I was a huge fan when it came out, and excitedly gave out beta access codes to all of my friends — originally had a business model of teaching users new languages and then using the users to translate texts for money at rates lower than professional translators. They abandoned this model in 2014. Their new business model is to provide language certification to users according to existing institutional standards of language proficiency. They are even building programs for use in schools i.e. institutional instruction.

Why did this change happen? Their official story is that business-to-business translation services did not accord with their values. I believe in reading between the lines: my suspicion is that they were unable to teach enough people new languages well enough that they could translate enough texts well enough to turn a profit. Since that wasn’t possible, they had to find a new way to make money. If they had truly produced a method for successfully learning new languages to fluency, they would have been able to have their users translate texts for money — a strong signal that they had cracked the code of language learning. Instead, they pivoted away from that to providing certifications according to a system of institutional instruction that we both agree fails to teach new languages to fluency.

The hidden signal is that they failed to figure out how to get people to learn new languages to fluency. I don’t fault them for this. Nobody else has figured it out either. I remain a fan of Duolingo.

The worst offender I have ever seen were these books in a Barnes & Noble bookstore in Los Angeles — I don’t remember the brand or company — that pledged to teach Korean and Hebrew to beginners. As you know, Korean and Hebrew both have their own scripts that are extremely different from Latin script. Nevertheless, both of these books were written entirely in Latin script, without a single character of Korean or Hebrew script to be found in over 20 chapters of “lessons”. When stuff like this is sold on bookshelves, it’s no wonder it’s so hard to learn new languages to fluency.

Regarding neuroplasticity and age — what is the best age to learn a new language and is it too late if you’re 40 or 50 year old?

I am not a neuroscientist or biologist so I cannot comment authoritatively on these matters. It seems broadly reasonable that it is easier to learn a new language when one is younger, as it is anecdotally easier to learn many things when one is younger. At the same time, there is plenty of countervailing evidence that this does not apply to institutional contexts, which are the most common when it comes to new language learning at a young age. I would predict, however, that there is no age at which learning a new language becomes metaphysically impossible. With a correct model, it is likely to become less than effectively impossible, which is how many correctly perceive it to be now.

Do you have any thoughts on the value of spaced repetition, brain games, and other similar things?

Much as with neuroplasticity and age, spaced repetition, brain games, etc. are all common topics of discussion when it comes to language learning. I think it is a very interesting and underrated fact that none of these things are actually exclusive to languages or linguistics. When you talk about neuroplasticity or whatever, you are not talking about languages or linguistics per se. You are talking about neuroscience, biology, or psychology. These things are obviously related to language, language learning, and linguistics in various ways, but they are not the core substance of the thing.

In fact, you are mainly talking about memorization. Memorization is essential to language learning insofar as the content of a language, the vocabulary, is arbitrary — therefore, it must be memorized. There are no shortcuts to this problem that I have found. I actually think it is logically provable that no shortcuts will ever be found, due to the arbitrariness of vocabulary. However, a language is not just a bunch of arbitrary vocabulary. A language has structure. It has patterns. It has rules. It has a grammar. It has syntax. These things are not simply memorized by rote, nor are they completely arbitrary. Moreover, they are common between many languages, even otherwise very different languages. They must be understood and incorporated at a deeper level.

While I do not want to prematurely reveal too much of my coming research, I notice that memorization is not the problem most people have with language learning. Memorization is generally a solved problem. Anybody can memorize something or other, whether it is linguistic or not. I think memorization is a red herring. The biggest problem seems to have something to do with internalizing the implicit structures of a language — the grammar and syntax — and using them correctly. I think this is actually a separate question from memorization entirely.

My weapon of choice in becoming fluent in English is the one-on-one Skype session. When I started my Skype lessons on italki, my teacher made me select a public speaker of my choice and try to repeat his speech. I chose Sam Harris and my poor attempts at emulating him make our sessions less gloomy because the difference is so stark, I cannot stop myself from giggling. What do you think of one-on-one sessions like this? Do they work? What’s the best way to use them?

Insofar as Skype sessions are conversations with native speakers, I think they are extremely helpful. You might consider them a form of immersion. One thing I notice about using conversation to improve language skills, however, is that many native speakers do not want to or do not feel comfortable correcting a learner’s mistakes in real time. Personally, I try to make an effort to correct and explain mistakes in English when I hear other learners make them. It can be a little awkward, but this will help a lot more than simple conversation without intervention. I think it’s great to practice speaking new languages in a way that is fun. It should be fun.

People often miss the fact that the skill of imitating the speaking style of another person, including in another language, is not a black box. The sounds produced by a human being can be meticulously and scientifically detailed using the International Phonetic Alphabet (IPA). We know which body parts produce them. The differences between these sounds, and the sounds one makes themselves, can be written down — made explicit — and explained and studied with the goal of changing the sounds one makes, such as, for example, to sound more like Sam Harris. Tone, cadence, volume, and so on can also be noted down and explicated. I would guess there is a lot of useful research to be done porting musical notation to production of speech.

This sort of phonetic study and practice is used both by theater students trying to adopt new accents for new roles as well as speech pathologists trying to fix the incorrect pronunciation of patients. Nobody seems to notice that the same techniques can be productively applied with new language learners. Phonetics are curiously absent from existing language learning methods. The question of how to get someone to correctly produce the sounds, as opposed to simply understand whether they are correct or incorrect, is a more difficult question, but I think there is low-hanging fruit here. I have personally used study of phonetics to improve the pronunciation of new language learners with great success. It is often enough to immediately fix a number of pronunciation mistakes. In my own learning, I have found it indispensable.

There is a popular software program called Grammarly that checks and offers corrections for English spelling, grammar, and writing mistakes in real-time. Do you think this kind of software is helpful or detrimental to language learning?

I’ve been aware of Grammarly, but haven’t used it extensively. I tend to think my English is pretty good already. I made an account and downloaded the browser extension and played around with it a bit. It looks like a great product. It would certainly help to improve one’s writing in English. The key feature that impresses me are the detailed explanations of grammatical concepts that you can access by clicking “See More” in areas that the program highlights. If one takes the time to note and understand those, their writing will improve in the long-term and they will not rely on the software as a crutch. The question of whether they will succeed in understanding the grammar is a separate one, but Grammarly looks to be accurately identifying mistakes and accurately diagnosing the type of mistake. This is obviously a very useful tool. One can still misuse it by automatically accepting all suggestions without bothering to understand why they are being made, but that can be said of any tool. It would be very cool to see this app developed for other languages as well!

Thanks for taking the time to answer my questions!

It was my pleasure!

You can follow Marko Jukic on Twitter, on Medium, or subscribe via e-mail on his website. You can read his inaugural article ‘Nobody Knows How To Learn A Language’ here.

Beautiful illustrations crafted by Oleg Kadka.