Hoi, On Meta we wrote an article about the teaching of languages. http://meta.wikimedia.org/wiki/Language_training_in_Wikiversity Please have a read.
There are two things that are important to mention; the University of Bamberg is working on the implementation of what is written. The software will be a toolkit that can be used by people developing a curriculum or by people defining targeted exercises for specific conditions. As the University of Bamberg is in the business of teaching Italian to Germans, many if not most of the exercises will be created. When the curriculum is to be used for speakers of other languages, it will be a matter of defining the meta data about these languages and making sure that the translations for the key Italian words exist in the Ultimate Wiktionary
One other key thing in computer based training is how and where and under what conditions do you store information about the educational process. When the curriculum for the learning of a language gets to the stage where many of the exercises can be handled by the computer including evaluating and reporting to a possible teacher, it will free the teacher to teach the things where a computer is at a disadvantage; stimulating the active use of a language comes to mind.
There is always this argument that a good teacher cannot be replaced by material and I agree. What we can do is creating material where there is much more personalisation of the learning material to the student. When we have a broad array of texts, we can use these texts to find material that is of interest to a student. He/she still needs to learn all the basic words for a language the inflections and everything but this can be wrapped in topical texts that are of interest to the students. In Wikipedia we have LOADSA text. In Wikibooks we have even more texts and when we get a reputation, we may interest newspaper to use their current articles and in the process let students learn their product..
Thanks, GerardM