General
Learning multiple languages
Jun 1, 2025
Once you’ve learned another language to a good level, it’s very rewarding. Most people will then want to do it with another language — and many do not stop there!
But learning never really stops, and even learning two languages can be a lot. Many people will say that’s the max you can learn, and have all these regulations around which languages you can do simultaneously, specific level requirements, etc.
The truth is, the real hard part is managing the languages. The research in memory science is pretty clear: interleaved practice outperforms blocked practice in retention (i.e. how much you remember), discrimination (i.e. reducing cross-overs or inter language mixups), and application (i.e. the ability to use what you learn in conversation). Spaced repetition is incredibly effective for acquiring vocabulary and grammar.
Phrasing seeks to do the management for you. And, while we’re at it, optimize everything so you can learn more, and just enjoy the process.
You don’t have to take my word for it. Let’s go over the research together and you make up your own mind.
Direct evidence
Perhaps my favorite study is one titled Simultaneous Learning of Two Foreign Languages, English and French, by Adult Persian-Speaking Learners.
I like it not just because of the results, but because it is specifically studying the simultaneous learning of English and French by adult Persian-speaking learners, but it’s stratified by language level, has a decent sample size, and control groups to boot. The findings showed that simultaneous learners, particularly beginners in both languages, obtained higher averages in final exams and placement tests for French compared to controls, with a mean score difference of 15 points (67.33 vs. 52.33 for beginners). For intermediate learners, the simultaneous group also performed better in French (69.33 vs. 59.5).
To my engineering brain, this is like a behavior test. If you study two languages simulteously, will it help or hurt? Turns out, it helps. So let’s find out how, and lean into that.
Interleaving
Take two decks of cards. If you go through the first deck, than the second, that would be called blocked practice. On the other hand, shuffling the deck first, and going through the result would be called interleaved practice.
Interleaving can have dramatic effects on your retention.
2014 study by Rohrer et al showed an increase in recall from 42% to 74% after 30 days solely from the introduction of interleaving into the studies. By some measurements, this is more that double the performance (in the sense that it more than halves the errors made) — but even in straightforward “amount recalled” it is nearly double!
2024 study by Pan et al shows that the interleaving benefit is applicable to grammar acquisition when interleaving multiple languages.
2020 study by Suzuki et al “showed that interleaved practice led to more accurate performance on both immediate and delayed posttests than blocked practice,” and this effect actually was more universally beneficial regardless of the subject’s working memory.
2016 systematic review by Kang et al found interleaving improved retention and transfer when compared to one subject, even with equivalent total study time. This effect was strongest when subjects were related. It also showed increased ability to recall the information in novel contexts.
Rohrer & Taylor’s research on interleaving showed mixing practice across multiple topics leads to steadily increasing long-term learning benefits, up to about 3–5 topics per session. Beyond 5 topics, the improvement flattens out. You still learn better than blocked practice, but the extra gain diminishes. This effect was especially potent for tasks requiring discrimination or application
A 2016 study by Vlach et al. showed that interleaved learning of multiple categories (e.g., different types of animals) improved memory by strengthening category boundaries and reducing interference between similar items
Neural evidence from a 2019 fMRI study by Brunmair and Richter indicates that interleaving activates the prefrontal cortex more than blocked practice, supporting better categorization and memory consolidation.
What this tells me is that interleaving 3-5 languages will provide increasing benefits in your vocabulary gain, grammar understanding, ability to code-switch and distinguish between languages, and ability to apply what you learned in practice.
Batching and Block Practice
There’s plenty of research that indicates occasional benefits to blocked-practice though:
2016 study by Sorensen and Woltz which found that initial introduction to topics in a blocked manner can be beneficial to retention.
2013 study by Carpenter and Meuller found that blocked practice was beneficial for pronunciation in a foreign language. This makes extra sense, as pronunciation is as much a matter of fine motor control as it is recall.
For these reasons (and the fact that sometimes your just in an <insert language> kind of mood) you can quickly switch to focus on any individual language at any time.
There is also a study that I read years ago, but have failed to find at the time of writing this aritcle (if I find it I will update this section), that showed pure shuffling of ideas was suboptimal, versus spending a bit of time with each subject. It makes sense — there is likely an optimal speed to switch between languages (probably dependent on the individual, their languages, and their levels). Although, somewhat ironically, I can only find studies now indicating the opposite - that spending too much time with a subject will neutralize the benefits of interleaving.
Because of this research I originally implemented phrasing to have batching by default. You can configure say 10 cards per batch, and the deck would pick 10 cards in one language before switching, 10 cards in another language before switching, etc.
However, in testing this, I found the dopaminergic effect present without batching to be far more rewarding than batched practice. At the end of the day, Phrasing’s competition is not an annoying owl, but Instagram, TikTok, YouTube, X, etc — so by default we turned this off.
You can still turn on batching for maximum per card effect, but try it without batching. There is something innately rewarding about not knowing the language of the next card. It’s literally the difference between counting down the cards until you’re done, to constantly telling yourself “just one more card”.
What I’m interested in now is if there is a way I can automate the batching to be optimal. Perhaps something as simple as randomization would give you the best of both worlds, or perhaps there is a heuristic that could apply it to better effect. More development is needed!
Spacing
Continuing the analogy, spacing refers to splitting the decks into smaller portions and taking a break in between each portion.
Spacing is so ubiquitous that it has a Wikipedia article under its name: The Spacing Effect.
Zooming out, the idea that focused study in one language one year, with focused study the next year in another language, will underperform studying the two languages in parallel for two years, even when study time is controlled.
We take this to the extreme by removing any idea of sessions. We take a cue from our competitors (Instagram, TikTok, YouTube) and give you an infinite scroll deck of optimal cards to be studying. Do one review, do one hundred reviews — we will keep finding the optimal cards.
We want Phrasing to be just as easy to use in an elevator for two minutes as it is to use at your desk for an hour.
The more short sessions we can do throughout the day the better!
Note: Our algorithm is based on the next generation of Anki’s algorithm (at the time of writing this, FSRS v6 vs v5) — so if you just want to keep your habit of n cards in one session, once per day, it’ll work the same. The research is pretty split between once-per day sessions outperforming multi-day sessions.
Discrimination
Many of the studies in the interleaving section discuss its impact on interleaving, but I wanted to include a specific anecdote.
I studied French and Italian at the same time. I never had any issue with cross-over (although I did study to switch between them). I added Russian and German to my studies and still, no cross-over.
10 years later I started learning Dutch and experienced severe cross-over. Like I would switch mid sentence and not realize it until I got a weird look from whoever I was talking to.
5 years later I started learning Portuguese. To this day I cannot speak Italian without mixing up a bunch of Portuguese.
In my experience, with 100% correspondence, my ability to separate the languages was inversely correlated with the amount of time between the languages, with no significant relation to language relation.
Memory Schemas & Category Induction
These categories are more tangentially related to what might be an original hypothesis. I have to imagine this has been thought about and studied, but I can’t find anything studying precisely what I’m interested in.
When I find a language learning material, I try not to learn too many languages from the same content. For example, Greek and Turkish from Avatar: The Last Airbender; Cantonese and Mandarin from Kung Fu Panda; or French, Italian and Spanish from Archer. I translated a poem in French to study in French, Italian, Portuguese, and Maltese.
I do not, for example, create a dozen cards, with a dozen translations, and learn the exact same thing in each language.
I’m trying out different combinations and different numbers. My impressions so far are that for Turkish and Greek, it’s actually helped; Mandarin and Cantonese seem to have been negatively effected (although it seems to have improved after adjusting the ratio, so maybe there are optimal ratios per language pair?). The Romance languages also seem to be pretty neutral (but I have been studying them the least so it would be the hardest to tell). The song actually has helped a lot - perhaps verging on a little boring, but I think that has more to do with knowing they’re just machine translations.
I don’t think I can quite cite any of the research into memory schemas or category induction directly. They’ve both had positive results in the field of language learning, but their results feel very specific, and neither directly approach this issue.
However I think there is a benefit to studying the same subject with different languages, but I think studying from different materials will result in less cross-over between languages - or at least a certain ratio or volume is needed to properly distinguish the languages, perhaps related to language relation. Longer term progress is needed!
Language Acquisition Hypothesis
I don’t think this hypothesis needs any introduction. If you’ve made it this far through the article, you likely have a passing familiarity with the language acquisition hypothesis.
It’s important to realize that the time spent on Phrasing might result in 80% of your study time in a language, however extensive interaction with native material outside your comprehension will probably result in 80% of your language gains.
The recommended comprehension of a given material to passively acquire vocab is ~93%. That’s an incredibly high level of comprehension. For any interesting material, we’re talking a C1 level or higher. However, this treats every word as equal, whereas in reality there are a minority of words in any material that constitute a majority of the meaning (Pareto distribution).
Phrasing works by determine the optimal learning path from <your level> to <target content>, slowly introduce one piece of information at a time, inching you towards comprehension of <target content>. By teaching you the most relevant words, grammar, and patterns in a given Material, we hope to be able to reduce the required comprehension as much as possible. My own personal experience has been that when using this method, you only need to understand 40% (somewhere just under half) to start passively picking up vocab.
Encoding Variability
(Many of the previously links discuss encoding variability)
Phrasing understands words across expressions. If you add the word dog in the same grammatical context, we will intelligently alternate between uses of this word. If you add two forms of the same word (e.g. I am and you are) we we link these as well (although obviously not use them interchangeably).
Phrasing also shows you the text in line with the audio. It has consistent explanations for each word. It tests you on writing and spelling the word. We give people the tools to and strongly recommend that users practice speaking as they do their reviews as much as possible. We recommend people first do our Free Recall exercise by hand, then transfer it to the app.
We make sure to give you as many encoding clues as possible
Conclusion
Learning multiple language simultaneously is:
demonstrably beneficial
mechanistically beneficial
way more fun
However, it’s a lot to manage. With Phrasing though:
Learners who have not used spaced repetition should expect a 2x increase in recall. Some will argue that this doesn’t transfer to language progress - but even its staunchest critics would acquiesce when you combine that with binging the material.
Learners who have not studied via language acquisition should expect at least a 2x increase in comprehension alone (anecdotal)
Learners who have done both of these should be able to double their time studying (sentence mining and lookups take up about as much as flashcards reviews in my ~24 user interviews)
Learners who are learning multiple languages will likely improve at least 30%, although my hypothesis is we can get that to full 2x.
Now in reality, you can’t just multiply these numbers together. But I would venture to guess that we can increase the amount someone can learn by 4-10x eventually.
Are we there yet? No, I think our algorithm is a bit “leaky”. Currently I think we’re probably about 2-5x. And while I realize that I’m biased, but I find our algorithm so pleasant to use. I end up doom-scrolling (bliss-scrolling?) for way longer than I set out to study, easily for over an hour. Compare this to the stress of those last 10 Anki cards… well story for another article.
Extraordinary claims require extraordinary proof
I agree! Unfortunately, Phrasing is something new, and learning a language takes time. Anyone selling you fluency in a fixed amount of time is a carleton. We can increase the amount you can learn, but your brain has certain requirements for long term retention.
That’s why I’m documenting my progress learning 18+ languages with Phrasing. I’m taking it the most extreme possible, and documenting my progress in public.
But you can get started with Phrasing today