Good news! Fossilised errors don't exist!
It has been claimed that errors that are made frequently enough "fossilise" -- that is to say that they "harden" and become difficult to fix. This, I feel, is overstating the case. It's certainly easier to learn something correct to begin with than to have to "unlearn" it later in order to learn the correct form, but even then, it may be less of a big deal than people think.
I learned French at high school from first year, and took direct entry into 3rd-year Italian as an additional language when I got the chance. I learned Italian quicker than French. Years later, I picked up Spanish, and learned it quicker than Italian. More recently still, I picked up Catalan... at lightning speed.
Well, that's to be expected, right? They are closely related languages after all.
But wait... if they are really similar, then one could describe one language in terms of another language and a set of differences. And if you think about it, an "error" is no different from a "difference" in cognitive terms -- the only distinction is that an error is a difference that isn't shared with a whole linguistic group.
So if fossilised errors really were such a big thing, wouldn't it have been harder for me to learn Spanish after Italian, rather than easier?
And this even holds as we get to languages increasingly distant from each other, because all we're doing is increasing the number of differences, which fossilised error theory claims would make things difficult.
Taken to its logical conclusion, fossilised error theory would make adult language learning an impossible task, because every language point we learn can be considered at attempt to overcome a fossilised language pattern -- ie the equivalent language point in the native language.
The paradox propounded by promoters of fossilised errors
The idea of a fossilised error is used as a warning against attempting to say things you don't know how to. This is pretty logical. But the idea of fossilised errors is particularly popular among the learn-by-listening crowd. Take the site Antimoon, and what they call "myths" about language learning:
Myth #2: "The best way to learn a foreign language is to speak it"
Myth #3: "It is OK to make mistakes"
Myth #4: "As a beginner, you're bound to make a lot of mistakes"
Basically, they reckon that if you make lots of mistakes, they will fossilise and you will learn badly. So you should listen to and read lots of examples so that you know it well before you attempt to say it.
But that's an inconsistent argument. How can it be that we learn mistakes by speaking and learn correct language by listening? Internally, the brain treats "correct language" and "errors" in the same way -- they're both just language forms.
So if it's the speaking that teaches us our errors, surely it's also the speaking that teaches us our correct language too? I mean if speaking does fossilise language, surely it's to our advantage to use that to fossilise good language?
It's certainly a good idea to learn things properly from the word go, because it's easier. Languages are internally more consistent, logical and systematic than many people give them credit for, and the real danger is that errors may distort the overall picture of the language, but they are never fatal and incurable.
It's better to avoid errors, but don't start distinguishing between "fresh" and "fossilised" errors and only attack the "fresh" ones. Do not give up, because... well, it's just lazy. If you start using "fossilisation" as an excuse, you'll find yourself making excuses for all your errors and you'll never really get as good at the language as you would like.