Two separate costs get treated as one. Interference is a similarity problem: two languages close enough for their words and endings to compete produce constant cross-contamination, and Spanish and Portuguese, or Dutch and German, are the classic pairs. Time is an arithmetic problem, and it applies to every pair equally.
Choosing a distant pair solves the first entirely and does nothing whatever about the second, which is why the advice to pick unrelated languages is half a solution presented as a whole one.
Eight rules if you are going to do it anyway
- do not start two at once. Two beginner languages compete for the same slot in memory and both go slower than either would alone.
- one at B2 before starting another is the usual advice and it holds up: at B2 the first language needs maintenance rather than acquisition.
- maintaining is cheaper than learning. Twenty minutes a day holds a language that took a thousand hours to build.
- distant pairs interfere less. Spanish and Portuguese together is considerably harder than Spanish and Japanese.
- interference is real and temporary. Words from one language surfacing in the other is normal, annoying and does resolve.
- separate the times and the places. Morning for one and evening for the other, and preferably different rooms.
- different tools for each, so the two do not share a deck, an app or a notebook.
- expect the weaker one to stall. Attention is not divisible in half, and the one with a real use will win.
What interference actually is
It is competition at the moment of retrieval. Two items stored close together, meaning something similar, sounding similar, and both plausible in the slot you are trying to fill: the wrong one arrives, or neither does, and the pause is the cost.
That is why the effect is strongest exactly where the languages are most similar and why it is worst when both are being built at the same time. A word met last week in Portuguese and a word met last week in Spanish have no established route to distinguish them, so they collide, whereas a Portuguese word met last week and a Spanish word you have used for five years mostly do not.
Distance genuinely removes this. Nothing in Japanese competes with anything in Spanish, and a learner running that pair will not mix them up in a sentence. They will still find that seven hours divided between two languages is three and a half each.
Different stages rather than different families
The arrangement that actually works is one language being built and one being maintained, because those two are not the same size at all. Building takes hours; holding takes a fraction of them, and the fraction is small enough to fit beside a serious commitment to something else.
That also solves the interference problem for close pairs, and better than distance does. A language you have used for years is stable enough not to be pushed around by a new one, so somebody solid in Spanish can start Portuguese with far less mixing than two beginners in both would experience.
The maintenance half has to be the right kind of small, which means production rather than reading, since production is what fades.
Where it stops: two at the same stage
The stall is starting both together, which is the arrangement that looks most efficient and is the one that fails. Both are new, both need hours, and both are in the fragile state where everything is being confused with everything.
The rate halves, which is arithmetic and is fine, and the visible progress halves too, which is not fine, because the first three months of a language are the period that runs on visible progress. Halve that and both languages arrive at month four with half the momentum they needed.
What usually happens next is that one is quietly dropped, having consumed four months. The honest version of that decision is to make it at the beginning rather than in April.
Keeping the two apart in the one place they meet
A separate profile for each language in your review app is free and takes a minute. Two languages in one deck is where interference is manufactured deliberately: cards are shuffled together, the same prompt can belong to either, and every session trains the confusion the rest of this page is about avoiding.
It replaced one deck with two languages in it, which is what happens by default and which is why a great many people believe two languages cannot be run together.
Its limitation is scope. It separates the review and it does nothing for the rest of the week, so the other separations have to be deliberate: different times of day, different rooms if you have them, and never the same session.
What it looks like when it goes wrong
The recognisable failure is a sentence in one language with a word from the other in it, produced without noticing. It happens most with function words and small verbs, which are the items stored closest together.
The second is two languages both stuck at the same middling level for years, which is the outcome of splitting hours evenly and permanently. Two half-built languages are worth much less than one finished one, for the same reason described in difficulty, usefulness and the honest reasons people pick.
What a maintenance dose has to consist of, and why it has to be production rather than reading, is what fades first, and how little it takes to hold on, and where the hours are going to come from in the first place is turning a vague intention into hours on a calendar.
The open question is which of the two to build first, and the answer is not the easier one: it is the one with a person or a date attached, because that is the one that will still be running in month five: which languages an English speaker gets furthest in fastest.