Several studies have demonstrated the benefits of self-explanation on learning well-defined domains like math, biology, and physics. However, these findings have yet to be replicated in probabilistic domains like second language acquisition. Working with adult English as a Second Language students (n=61) within the domain of the English article system (i.e. teaching students the difference between a dog vs. the dog) we conduct the first experimental study of the effects of prompting self-explanation on second language grammar acquisition. We compare two different modes of self-explanations (free-response and menu-based), each implemented in an intelligent tutoring system, to a control tutor with no explicit self-explanation prompts. Students in all conditions show significant learning gains but contrary to theoretical predictions, the self-explanation tutors did not lead to better learning over the no self-explanation condition. We discuss why and under what specific conditions target-specific practice without self-explanation may be a more effective instructional strategy.