This Non-existent Book Supercharged My Language Learning

[00:00:00]I just leveled up my language learning by memorizing a book that doesn't exist.

[00:00:05]Let me explain. I've been an avid language learner for decades at this point. And after a PhD in linguistics, a few language learning successes, and even more failure, I spent the last year or two refining my approach to language learning. I did deep dives on the science of learning and memory, language acquisition, and on using cuttingedge technology in conjunction with more traditional methods for an empirically validated scientific approach to language learning. It's going to sound like I went off the deep end when I say I memorized the contents of an imaginary textbook that doesn't exist. You can't buy it. You can't read it. But ultimately, that's exactly what I did.

[00:00:41]How did I get here? To understand how any of this makes sense, I'm going to give away the basics of my new approach to language learning that has dramatically accelerated my learning.

[00:00:49]I'm launching a course, more on that in the description if you're interested, and I'll be discussing some of this at a conference in Bolognia later this year.

[00:00:57]So, let's get to it. This is Language Jones.

[00:01:00][music] So, we just made a pretty big purchase and a few years ago that would have sent me into a low-grade spiral for like a week. Here's how we actually handled it this time. This is Monarch. After the purchase landed, I just opened the app, saw it categorized automatically, and could see exactly where it sat against the rest of the month. All our accounts, cards, checking, investments, loans, already syncing in from our institutions, and they connect with over 13,000 in one clean dashboard. No logging into four portals to piece it all together. The thing that actually calms me down is fixed versus flexible spending. I can see what's locked in versus what we have room to move on.

[00:01:40]Monarch's AI insights flagged the change in plain English. Basically said, "Yeah, this month is higher. Here's why." And I can ask it follow-up questions about our money and get a real answer. So, instead of 20 alerts buzzing at me, I get one clear view and I move on with my day. A few things worth mentioning. It's private and adree. You're the customer, not the product. No data is being sold off in the background. It works on web and mobile, which matters because I check it on my phone, but do bigger reviews on my laptop. And it's not a bank. Monarch doesn't move money around.

[00:02:15]It just gives you a clear view of the accounts that you already have. Use my link in the description and promo code jones50 or scan the QR. If big purchases stress you out, like everybody else, this genuinely helps. Let's start with a little background. I have tried basically everything in the language learning space. That's part of what prompted me to make my various videos calling out fake polyglotss on YouTube.

[00:02:38]I know exactly how hard it is. I know how it's possible to grind away for years and still fail. Looking at you, modern standard Arabic. I've experienced actual immersion when I lived in Germany as a child and spent time in France with family that doesn't speak English. I've studied languages in high school and college. I worked for Rosetta Stone for a stent and fully drank the Kool-Aid.

[00:02:57]Maybe Joneses shouldn't make Kool-Aid jokes. Anyway, I've done the self-study thing with textbooks, podcasts, independent classes, you name it, I've tried it. I did an entire PhD in linguistics in part because a polyglot friend of mine told me it was going to help me see the matrix. It did, but that's a different story for a different day. And it didn't help me actually manipulate the matrix. And for every success, getting certified at B2 in French a decade ago or passing language exams in grad school or successfully chatting with people in French, Mandarin, Dutch, pretty sure I've lost that skill, and Italian. I've got a failure story. Arabic kicked my ass.

[00:03:32]Persian, a language I love, was a language I decided to dedicate a full 3 or 4 months to, honestly, on more than one occasion, and failed spectacularly last time. Like, could not get the words out of my mouth. In Hebrew, I studied for almost 5 years with next to nothing to show for it. So, I finally set out to do something new. Some of you have seen my videos on using LLMs for language learning, and this will draw on that a bit. It's actually the ideal use case for LLMs. They're not large truth models. They're not large psychiatry models. They're not even large friend models. They're large language models.

[00:04:07]Maybe the real friends were the words we learned along the way, but they're only part of the story. After over a year of deep dives on the existing empirical science on second language acquisition, on memory, and on learning, I realized that most of my tools were not great.

[00:04:20]There are practical implementations of each of those fields that could be implemented as best practices, and almost nothing actually integrates all of them. So, here's what I've been doing lately that led to memorizing an entire book that doesn't actually exist. I started by deciding I was going to try to memorize all of the pedagogical content of a real textbook that does exist. It's a finite goal and one that I could never really say I'd ever done, though I have. Now, if you want to memorize anything, the best practices for it are to be multi-ensory, for it to be comprehensible pieces, the minimum information principle, for it to fit with other things that you already know, and for it to be something that you actually have to generate, not just recognize, and for that generation to be repeated across time. There's actually an optimal way of doing all of this called spaced repetition. And there are all sorts of choices of software that will implement an algorithm to space out your repetitions optimally. I use Anki, which is free and open source, except for the iPhone app, which is well worth the $25 I spent on it. I actually started a while ago with Persian and Italian textbooks, just copying and pasting the Italian and writing the Persian by hand. This did not work. I did the same with the Routidge colloquial Hebrew, which is, I can say now with certainty, the weirdest textbook I have ever used. Ultimately, what I settled on was the smarter approach of using LLMs, in my case, Claude, to extract all of the material chapter by chapter and then make an Anki deck. I added CSS to each to help keep things separate. My Italian deck has an Italian flag theme and modernist font. I add images. I add pneumonics, one way of bootstrapping familiarity when there isn't any to start with. And I add audio with Hypert, a texttospech engine that allows me to use Azure 11 Labs and Gemini, which have really leveled up. It is much faster than trying to slice and dice the coursearse audio MP3s and add them to my Anki decks. Although I do that too for things where there isn't good service with Hypert. When I say all of the material, I mean all of the material, by the way, every vocab word, every sentence, every dialogue, every exercise, every answer key, everything.

[00:06:24]With this approach, I actually finished the Routidge colloquial Hebrew. And I internalized the grammatical lessons by really locking in memorizing examples.

[00:06:33]It was almost 4,000 cards. I did it in under a year. And now I know the contents of the book. It was incredible.

[00:06:40]I did the same extraction for Italian, which I love and which is relevant to me, and which not for nothing is so much easier that it serves as a reminder just how different Hebrew is. This is an aside, but do you know how to say an important question in the sense of an issue in Italian and Hebrew? In Italian, you just say it with an Italian accent in Hebrew.

[00:07:02]One of these is a lot easier to lock in than the other. So I did the same with the French textbook and with the Persian textbook that had bested me and I'm now slowly surely working through them with certainty that I am retaining and understanding everything and that I'm slowly developing the ability to generate all of it at speed. I have to remind myself when I'm doing my slow pace of 10 new words a day that actually taking a year to work through this book and actually learn all the contents is a lot better than doing 3 months of something that feels like intense learning and which leaves me with nothing to show for it. But I wanted another Hebrew textbook and there isn't one. By the way, if you want to know more about my methods and how to implement them yourself with any content you're interested in, not just textbooks, but books, podcasts, movies, TV, and even your language lessons with tutors and teachers, you might be interested in my upcoming digital course, the language learning accelerator. It covers everything from motivation, the basics of linguistics and memory to practical implementation and where technology fits in. You can sign up for updates and the weight list at www.languagejone.

[00:08:10]languagejones.com/acelerator.

[00:08:12]So, as somebody who's really trying to push the boundaries of how we can use technology like LLMs and text to speech and all that good stuff to accelerate language learning and cultural exchange rather than, you know, whatever else people are trying to accelerate, >> I thought to myself, what if there [snorts] were another textbook? What if there were a routidge colloquial Hebrew 2 textbook or Persian 2 or Italian 3?

[00:08:37]Now, you should know that I'd already started building out my own ankidex around various themes. I'm a linguist with a PhD in socioetics who specializes in variation and change. I'm a public educator with a YouTube channel. I'm a specialty coffee afficionado. What is it like? Spanish. And I lift weights and do combat sports, primarily Brazilian jiu-jitsu. How much of that is in your intro textbook or Dolingo or whatever?

[00:09:03]Precisely none of it. Not to mention using little conversational connectors like not to mention. So I was already using LLM to generate anex just so I could literally make small talk and it's going great so far. I can say something like and so on. But that was scattered and unsistatic. So I decided to give my entire ankoutlage colloquial Hebrew to Claude and just ask what's missing? What's over represented?

[00:09:32]if there were a second volume, what would it cover? And the results were stupidly good. That's in part because these LLMs, you don't have to use Claude, that's just what I find works best for me. They're trained on an absurd amount of data, including pretty much any textbook you could imagine. So, when I've given it examples in the past to format, it has correctly identified what they're from, even though I didn't say. And while LLMs excel at making plausible language, and that means they're incredible for coming up with example sentences and so on, their structural knowledge has gotten so good and so robust that they can also take an anode, infer the existing textbook and then outline the next volume based on an implicit comparison between the contents of the textbook and the contents of other textbooks at the same level and the kinds of things that are actually covered in advanced textbooks for other languages or for the same language from a different series and then generate that. Claude even asked if I just wanted to have the Anki deck and if I wanted it to bother with writing the chapters that would go around it all. So while I get around to getting some good explanations, most of it is immediately apparent to me. Naturally, I went with generate the Anki deck of content to memorize. So I'm memorizing vocabulary and sentences from chapter 2. What do you think? opinions, media, news, or from chapter five, on the phone and online, digital life, social media, and texting. Or for my favorite, chapter seven, at the ministry, bureaucracy, and waiting. And that's how I've leveled up my language learning by memorizing the book I wish existed, despite the annoying little fact that it doesn't exist. As a friend pointed out, Bourj Bourhees Borg Borgis, he said it in writing.

[00:11:15]Jorge, I want him to be French. Gorge Lou Bourj. I don't think that's right.

[00:11:19]Jorge Luis Bourhees.

[00:11:20]>> Anyway, that guy would be proud. I think Bodiard would too. It's like Tom Hardy says, >> "You mustn't be [music] afraid to dream a little bigger, darling.

[00:11:27]>> It's just that it's a good idea and a book like it can be inferred from massive amounts of language data." And of course, you don't have to, but if you want to learn how to take charge of your own language learning so you can talk about yourself and your interests or read the books that you've always wanted to read in the original or millions of other things cookie cutter courses don't cover, definitely be sure to sign up for the language learning accelerator. Spots are limited for the first cohort. Let me know what books I should imagine next.

[00:11:55]I'm thinking about doing a publish on demand LLM version of similacra and simulation 2 to simulacra 2 simulation by Jean Bodriard. If you like what I'm doing with the channel, you can support it on YouTube with super thanks or over on Patreon at patreon.com/languagejones.

[00:12:12]Please leave a comment. They are great for the algorithm. And subscribe if you haven't already. Until next time, happy learning.

[00:12:22]>> [music]