Paper with a formal argument showing the possiblity of learning arbitrary languages in the limit from just positive examples when (1) the learner's hypothesis space includes all possible computations, and (2) the examples are drawn from a fixed distribution (i.e., not adversarially generated).