You step into a forest, and a well-trodden path reveals itself in front of you. Even if you've never been there before, you have a pretty good idea of how to get from point a to point b.
Your brain is surprisingly similar. Memories you recall often are the strongest. Words you favor pop into your head most frequently.
Neurons like well-trodden paths. Your brain knows where to find oft-recalled things. Thus, it stands to reason, the best way to learn a language quickly is to trod some new paths into existence.
When you study with a traditional textbook, you're not really telling your brain where to find the information you need. You're connecting new ideas to ideas in your mother tongue. Which means when you want that information again, you have to go through your mother tongue, because that's where the neural connections are. Eventually, the contextual connections of the situations you're in will overpower the first-language connections, but that takes a lot of extra time.
Finding your own understanding of a new language pattern ensures that the first neural connections are with the concept matched to the new pattern, not your mother tongue. This is the first step in solidifying the right neural pathways. Forming at least three sentences connected to your life using the new pattern is step two. About a week later, stealing Pimsleur et all's research into the tipping point between short term and long term memory storage, playing a game designed to produce the pattern in natural conversation is phase three.
In early experiments, doing just steps one and two resulted in a 67% retention rate. But this is without question pseudoscience. There were no control groups or redundancies. The theory is built on a logical conclusion based on known information, plus observations of 20 years in classrooms plus some critical analysis of my own language acquisition habits, without a single scientific study to back it up. So don't take it as fact. Think about it for yourself. If it holds up under logical scrutiny, have a go and see how it works for you. If you're a neuroscientist and are curious, please feel free to start up an experiment or two.