I read an interesting article on Wired.com recently, relating to a paper in the Journal of Neuroscience. The study looked at a small set of musical skills. Common wisdom is that you can only learn music by performing it, which is analogous to learning language by production. Previous studies support this, and prove that you can't learn to play music just by listening, which is analogous to language learning by practicing receptive skills.
The study noted that previous research has focussed on an all-or-nothing approach -- all practice or no practice -- and instead looked at the effects on passive perception in addition to practice, and they claim that timed right, it's just as effective to spend 50% of your time listening to what you've just learned. They are, of course, talking about fairly early learners, as professional level musicians can show remarkable skills in learning and playing new music.
This makes a great deal of sense to me -- I have always said that I can only understand something that I can say (or, in the case of language I've forgotten, could have said previously). This doesn't mean that I ever would say it, but that it is possible in my internal model of the language. This, I feel, is analogous to the example of musicians.
A beginner in both fields has a small set of "devices" to employ (specific notes and practised combinations thereof vs specific words and practised combinations of them) and neither is going to be able to directly relate to any material that goes beyond that set. As the learner develops in both fields, the set of devices grows, and the learner will be able to generalise to all material made up of items within their arsenal of devices. Thus the musician who masters one style of music will be able to play any piece in that style by ear, because it is a recombination of musical devices he knows, but will not be able to do the same for a piece in a radically different style. The language learner with high fluency and proficiency in his new language will be able to understand lots of individual sentences that he has never heard before but are composed of language he knows, but if someone says an idiomatic phrase he has never learned, he obviously can't understand it -- are you hip to my jive, man?
If this research proves to hold true for language learning as it does for musical skills, then the language teaching industry is missing a trick by producing materials that students can't understand in their entirety. Perhaps instead students would be better listening to things that they have just been taught to say, even if they can only say them shakily. Some lessons in the Pimsleur method do this, repeating at the end of the lesson the dialogue that was used to teach that lesson. Assimil teaches you from a dialogue, and then asks you to listen to the dialogue again once you've learned it all, without reading any notes, so that you understand it in its entirety. This may be a lazy way to do it, because it doesn't really require true understanding, but many people are happy with it. It's curious then that neither Teach Yourself nor Colloquial, the two biggest book-shelf brands available in the UK, give the learner any instruction to relisten to each lesson's dialogues at the end of the lesson. It seems like a pretty quick win -- at present the dialogue is only listened to before the presentation of the associated language points, when the learner will most likely understand almost none of it.
Personally I would like to see the post-learning listening being a new and unheard piece, but built up of the elements used earlier in the lesson. This is easier said than done, though. If you start from a situational point of view (Chapter 1: Greetings, Chapter 2: Meet the family etc), then you are going to be recycling the same phrases and it will be an almost identical dialogue to what's been used in the lesson. If you go from a grammar-led course, on the other hand, it's going to be difficult to find anything interesting to say, and you're likely to end up with something that sounds quite contrived.
It's a difficult one.
(NB. The Journal of Neuroscience isn't available in my university library, so I've only read the abstract.)