In human languages, only some types of morphological paradigms are attested, whereas others that would be equally logically possible seem to just not exist. For example, in English, there are triplets like “good - better - best”. These are called an “ABB” pattern: the first element of the triplet has one stem, called stem A, and the other two both have a different stem, stem B.
In other languages, “AAB” and “ABC” patterns exist. However, the one pattern that seems not to exist in any language is “ABA”. In English, this would look like “good - better - goodest”. There just seems to be no adjective in any language that shows this pattern.
One hypothesis about why this is the case is that the human language faculty, understood as a specific computational system genetically hard-wired in our brain, simply cannot generate ABA patterns. Our brains just can’t do that. Therefore, a child shouldn’t be able to learn patterns like that, even if that’s the input they get.
In an ongoing experiment at the Language Acquisition Lab at MIT we’re testing this hypothesis. We teach 6-7 years old children “an alien language”, consisting of triplets like these. Some kids will learn an ABB language, some kids will learn an AAB language, and another group of kids will learn an ABA language. If these can’t be learned, as our hypothesis says, these kids should perform less well than the other two groups.