You have the app, the podcast, a grammar book with a bookmark in it, and nobody to say anything to. The evening is free, the intention is there, and the only thing missing is the other person. So the question is not whether you can practise alone. It is which parts of speaking a language you can practise alone, and which part goes missing while you do.
Three things a person across the table supplies. The first is retrieval with nobody prompting you: you have to produce the sentence, not recognise it. The second is a signal when the sentence was wrong, arriving close enough to the sentence to still be about it. The third is continuity, somebody referring back to what you said last week, so the same words come up a second time without you scheduling them. Each of the five below is measured against those three, and each of them is described with what it cannot do on its own.
Daily contact, short
Ten minutes most days and an hour on Sunday are not the same hour, because the short version asks you to retrieve on more separate days, and retrieval is the mechanism the research covers. What you retrieve, you keep; what you leave alone gets harder to reach rather than disappearing, so the reaching takes longer while your vocabulary stays exactly where it was. Short and frequent keeps the paths short. It also survives a bad week, which an hour does not.
Its limit is what counts as contact. Reading and listening are contact with the language and they leave production untouched, so a month of podcasts can pass without your own sentences getting any easier to find. This is also the item where what we built is easiest to describe: a written conversation most days, short enough to survive a Tuesday evening.
Writing to someone who answers back
Producing a sentence nobody prompted is the expensive part, and the answer is the part self-talk cannot fake. A reply uses words you did not choose, in an order you did not plan, and it asks you to understand them before you can continue. That is where the second and third things on the list arrive: a correction can sit inside the reply rather than in a report you read afterwards, and whoever is answering can remember the last conversation and bring it up again.
Its limit is speed. Writing is slower than speech, so it does not train the timing of a real exchange. That is also exactly why it is easier to start: you get to think for four seconds, and the sentence still has to be yours.
A journal that gets read
A few lines a day is unhurried retrieval, which is a real thing rather than a soft version of the hard thing. You reach for words with no clock running, and what you wrote stays on the page, so it can be corrected later. Speech disappears as you produce it, and the mistake goes with it.
Its limit is in the second word of the heading. Unread, a journal preserves the error as faithfully as the sentence, and a month of it can lay down a wrong form very firmly. The errors that do this are the comfortable ones: the agreement in <em lang="it">molto</em> against <em lang="it">molti</em>, the tense that goes with <em lang="de">seit</em>, both of them mistakes that survive a year of solo study because nothing in the room contradicts them.
Saying things aloud before you need them
Rehearse the sentences the week is actually going to ask for. What you say at the till, on the phone to the landlord, in the first thirty seconds of meeting somebody. Said in advance, the sentence has been retrieved before the moment that needs it, so the first attempt is not also the first retrieval, and the pressure of the real exchange lands on your delivery rather than on your memory.
Its limit is that there is no signal. A wrong sentence rehearses just as smoothly as a right one, and rehearsal makes it faster to produce either way. It pays when the sentences in it have been checked by somebody, and it entrenches whatever is in them when they have not.
A weekly look at what went wrong
Read your own week back and a scatter of individual errors turns into one or two patterns. That is the form in which errors actually get fixed, because a pattern is small enough to hold in your head while you speak and a list of thirty corrections is not. It is also the only one of the five that tells you what to spend next week on.
Its limit is that it needs a record, so it depends on the second and third items existing. Nothing to read back, nothing to find. On our side this is one letter a week that reads your own week back to you, which is the same mechanism with the reading done for you.
That accounting is the honest one. None of the five is a substitute for the others, and only one of them, the second, reaches all three of the things a person across the table supplies, because it is the only one with an answer in it that you did not write. The other four still do their own work, and they do it better once you know which of the three each of them leaves out.
Sources: Roediger and Karpicke, Test-Enhanced Learning, Psychological Science 17(3), 2006. Bahrick, Semantic memory content in permastore, Journal of Experimental Psychology: General 113(1), 1984.
