This paper proposes a framework for modelling the process of teaching multiple concepts to a learner (e.g. a human student) with limited memory. The authors provide empirical justification for their approach and certain theoretical guarantees. Interestingly, there is the possibility that the insights from this paper could be applied to language learning in humans, which would I think represent an intriguing and uncommon domain of application for a Neurips paper. There is, of course, also potential application to human (or machine) teaching of machines. Significantly, the two reviewers expressed not only their approval of the work, but also the fact that they learned something of value from reading it. I share these sentiments. Another factor in my recommendation is how satisfied the reviewers were with the author response. Although scores did not change, the solid response of the authors seems to have been much appreciated. For these reasons, and the fact that the work tackles and interesting a non-typical learning situation, I think the paper should be accepted and will make a welcome addition to the conference programme.