Learning primarily through listening and talking only works if you have an enormous amount of high quality data, ie, the situation you're in as a baby when you're surrounded by native speakers who are producing speech for hours every day.

If you are an adult and want to spend, say, a couple of hours each week learning a language from scratch, especially without constant access to a native speaker, your initial progress will be much faster if you study grammar and vocabulary in a traditional class from a text book, than if you just try to pick up patterns from listening to the TV or something.

I can't source my claim. I attended a public lecture years ago from a researcher about exactly this misconception.