The best app for polyglots is built for more than one language: any pair of languages from one account, one subscription covering all of them, and a level and accent remembered per language. Mintza does exactly that, with fifteen languages in any of 210 directions. Most language apps assume the opposite: one language, learned from English.
That assumption is the quiet tax on every multilingual learner. This post covers the two problems it creates, what the science of language maintenance actually says, and why the right tool for someone juggling three or five languages looks structurally different from the right tool for someone learning their first.
The real polyglot problem is maintenance, not learning
Learning languages is the part polyglots have figured out. We covered their strategies, and the neuroscience behind them, in what polyglots do differently. The problem that never goes away is keeping the languages they already have.
The Italian polyglot Luca Lampariello is blunt about the limits of the polyglot life: “If you can speak thirty languages, you’ll need to practice all of them semi-regularly if you do not want to lose them.” And when you concentrate on one, he warns, the others suffer, and similar languages start interfering with each other.
Talk to anyone maintaining three or more languages and you hear the same two confessions. First, the languages they are not currently focused on are slipping, and they cannot find a routine that covers everything without consuming their life. Second, the rust always shows up in the same place: not in reading, not in listening, but in the mouth. You still understand everything. You just cannot say it anymore.
That second observation is not an anecdote. It is one of the most consistent findings in an entire field of research.
Speaking is the first skill you lose
The study of what happens to unused languages is called language attrition research, and its core findings map exactly onto polyglot experience. Monika Schmid and Teodora Mehotcheva summarize them in their review of foreign language attrition, and three are worth knowing.
Production decays before comprehension. In the researchers’ words, “productive skills are more vulnerable to attrition than receptive ones.” Studies found no signs of attrition in receptive vocabulary while recall, pulling the word out of your own head, degraded. The mechanism is the activation threshold: recognizing a word only needs a signal from outside, while producing it needs an impulse from within, which is harder. Every use lowers a word’s threshold, and disuse raises it. This is why a neglected language feels intact when you watch a film in it and broken when you try to order dinner in it. It is the same asymmetry that creates the gap we described in why you can understand a language but can’t speak it, now running in reverse.
Loss is fast, then flat. Attrition “seems to set in rapidly and then level off.” The studies found more loss in the first zero to three years of non-use than in the following five to twenty-five. Harry Bahrick tracked people tested up to fifty years after studying Spanish in school and found that after the initial drop, knowledge sat on a stable plateau for decades, content he called permastore.
Entrenched languages resist decay. The review notes that “once a language is sufficiently established/entrenched, it becomes impervious to deterioration.” Higher proficiency at the moment you stop predicts better retention. A language you drove to fluency is far more durable than one you abandoned at A2.
One honest caution: the same research found that casual rehearsal alone did not reliably prevent attrition in the populations studied, so nobody can promise that ten minutes a day makes a language immortal. What the evidence does support is narrower and more useful: frequency and recency of use keep words retrievable, and production is the side that needs the work. If you only have limited time per language, spend it speaking.
Dormant languages come back faster than they were learned
The most encouraging idea in attrition research is the savings paradigm. When a word drops below the level where you can even recognize it, it looks lost. It is not. Relearning studies, including work by Kees de Bot and colleagues, show that residual knowledge makes relearning a “forgotten” word measurably easier than learning a new one. The knowledge did not vanish. It became inaccessible, and access can be rebuilt.
For a polyglot, this changes the emotional math of rotation. Letting a language go quiet for a season is not throwing it away. The right mental model is languages in three states: active languages that get regular speaking practice, dormant languages resting on their plateau, and a reactivation process that is dramatically cheaper than the original acquisition. Polyglots who maintain many languages tend to work exactly this way, one focus language at a time while the rest cycle through maintenance, precisely because the alternative, keeping everything permanently active, does not fit inside a day.
The practical requirement that falls out of all this science is simple to state: a polyglot needs cheap, low-friction speaking practice in every language they care about, on a schedule they control. Which is exactly what the language app market fails to provide.
The app market assumes you are monolingual
Open the pricing and catalog pages of the major language apps and a pattern appears: they are architected for a person learning one foreign language, almost always from English.
Per-language subscriptions. Babbel’s own business FAQ draws the contrast directly, describing consumer Babbel subscriptions as ones “that typically focus on single-language access,” while access to all fourteen languages is the selling point of the corporate plan. Pimsleur’s official FAQ is similar: its Audio-Only and Premium subscriptions each cover a single language, and unlocking all fifty-one requires the separate All Access tier. For a person learning French, this is fine. For a person maintaining four languages, single-language pricing quietly multiplies by four.
English as the forced hub. Duolingo lets one free account study every course, but the catalog is lopsided. Its April 2025 launch of 148 new courses made the seven most popular non-English languages, Spanish, French, German, Italian, Japanese, Korean, and Mandarin, available from all 28 interface languages, and the press release notes that before this, speakers of many languages “only had access to learning English.” The rest of the catalog largely still requires English as your base. Speak, one of the strongest AI conversation apps, teaches English speakers six languages, while its localized versions around the world exist mostly so speakers of other languages can learn English. The market is a hub-and-spoke diagram with English at the center, and a polyglot’s life is a mesh.
Tutors scale linearly. Human tutors solve the speaking problem one language at a time. On italki, community tutors run 4 to 20 dollars per lesson and professional teachers 10 to 40, which is genuinely reasonable for one language. Maintain five, with one lesson each per week, and even at ten dollars a lesson you are at roughly 200 dollars a month, plus five standing appointments to schedule, reschedule, and defend. The cost is manageable for some. The calendar rarely is.
None of these products is bad. They are optimized for the most common customer, the single-language learner, and the polyglot pays the structural penalty.
Laddering: the technique the hub model locks out
There is a specific polyglot technique that English-centered catalogs make impossible: laddering, learning a new language through a language you already learned, rather than through your native one. A native English speaker studying German through Russian. A Spanish speaker learning Japanese through English.
The appeal is efficiency. One session moves two languages at once: the new language advances, and the support language, doing real work as the language of explanation, stays active instead of decaying on the shelf. It also keeps the support language from being a trophy. Attrition research says use is what keeps a language retrievable, and laddering builds that use into time you were already spending.
You will not find a coiner to credit; laddering is community folklore, passed between learners who discovered that a Russian-based German course was maintaining their Russian for free. But you also will not find it supported in most apps, because a catalog built around English as the base has no lane for it.
Mintza: one account, fifteen languages, any direction
Mintza is an AI voice teacher built by Paintingstack, and its architecture happens to be the one the polyglot case demands. Conversations are live voice, like a phone call: you talk about whatever interests you, the teacher adapts, corrects, and keeps the conversation moving at your level.
The structural decisions are what matter here:
Any pair, any direction. Mintza covers fifteen languages: English, Spanish, Portuguese, French, Italian, German, Greek, Chinese, Russian, Turkish, Swedish, Arabic, Japanese, Korean, and Hebrew. You pick the language you speak and the language you learn from the same list of fifteen, so all 210 directed combinations work. The language you “speak” is really the support language: when your sentence collapses, the teacher switches into it, rescues you, and brings you back. Choose Russian as your support language and German as your target, and you are laddering, with the rescue happening in Russian.
One pool of minutes, every language. There is no per-language subscription. Your minutes belong to your account, and a conversation in Japanese draws from the same balance as a conversation in Greek. Switching languages is a settings change, not a purchase. For maintenance rotation, this is the whole game: a twenty-minute week can be ten minutes of your focus language and two five-minute check-ins on languages you are keeping warm.
Your level and accent, remembered per language. You might be Advanced in Spanish, Intermediate in German, and a Starter in Japanese. Mintza stores level per language across its four levels (Starter, Beginner, Intermediate, Advanced), so switching back to a language restores you where you left it, and a language you have never studied starts fresh. Accents work the same way: ten of the fifteen languages offer regional accent options, thirty-seven in all, from Mexican Spanish to Kansai Japanese, each remembered per language. If you have not settled that question yet, our guide on which accent to learn walks through it language by language.
Honest scope. Mintza is conversation practice, not a full curriculum. It will not replace your grammar reference, your reading, or your flashcard system, and polyglots should keep the input tools they love. What it replaces is the hardest thing to arrange across multiple languages: regular spoken production with a patient partner, in every language you have, without five subscriptions and five calendars.
You can test the whole argument without spending anything: every account starts with 10 free minutes, no card required, and the free minutes never expire. Paid plans are monthly minute pools, Basic with 180 minutes, Plus with 360, and Pro with 600, cancel anytime, with current prices shown in the store.
The short version
The polyglot’s real enemy is not difficulty, it is decay, and decay attacks speaking first. The science says loss is fastest early, entrenched languages endure, dormant ones reactivate cheaply, and use is what keeps words within reach. The tools, mostly, say one language per subscription and English in the middle. An app for polyglots has to say the opposite: every language, every direction, one account. That is the architecture Mintza was built on, and for anyone maintaining more languages than the market planned for, architecture is the feature.
