Language system dynamics is studied at several distinct time scales (Lewis, 2014).

Weaknesses: Both general lines of work in this paper, on emergent language to learn speakers and listeners, and on explicit pragmatic reasoning to improve communicative success or efficiency, have been explored in prior work. Their integration here was limited to using speakers and listeners learned through emergent communication as the base models in pragmatic reasoning. I appreciate the approaches the paper uses, but I felt that its aims were currently ill-defined, and the contributions were too spread over several areas so that each of them is thin or unclear. For each of the contributions above, 1) Since, as the paper points out, the GameTable-sequential is essentially an upper bound on communicative accuracy, the results in 4.2 are largely useful for analyzing the relative merits of the methods, but details about the proposed game equilbria method, and the comparisons to past work, were unclear. Emergent communication gives a framework to investigate compositionality or efficiency of language, or concept formation, and the qualitative analysis in 4.4 is a partial step toward this, but it's unclear how representative these examples are.

All reviewers agree that this is an interesting, sound submission above acceptance threshold. I have read the reviews and author response and I would like to propose acceptance. Specifically, I agree with R1 that the idea of considering explicit equilibria methods in the context of multi-agent communication will inspire more research in the field. Moreover, the application on Starcraft domain is also a good contribution and overall this work provides good accuracy-based improvements with the proposed pragmatic reasoning method. However, I agree with R1 that a discussion beyond accuracy results (e.g., looking at the intrinsic properties of the learned communication) would have been really helpful. At the same time, the reviewers have raised a number of concerns, most of which appear have been clarified in the author response.

Emergentism and pragmatics are two research fields that study the dynamics of linguistic communication along quite different timescales and intelligence levels. From the perspective of multi-agent reinforcement learning, they correspond to stochastic games with reinforcement training and stage games with opponent awareness, respectively. Given that their combination has been explored in linguistics, in this work, we combine computational models of short-term mutual reasoning-based pragmatics with long-term language emergentism. We explore this for agent communication in two settings, referential games and Starcraft II, assessing the relative merits of different kinds of mutual reasoning pragmatics models both empirically and theoretically. Our results shed light on their importance for making inroads towards getting more natural, accurate, robust, fine-grained, and succinct utterances.

Emergentism and pragmatics are two research fields that study the dynamics of linguistic communication along substantially different timescales and intelligence levels. From the perspective of multi-agent reinforcement learning, they correspond to stochastic games with reinforcement training and stage games with opponent awareness. Given that their combination has been explored in linguistics, we propose computational models that combine short-term mutual reasoning-based pragmatics with long-term language emergentism. We explore this for agent communication referential games as well as in Starcraft II, assessing the relative merits of different kinds of mutual reasoning pragmatics models both empirically and theoretically. Our results shed light on their importance for making inroads towards getting more natural, accurate, robust, fine-grained, and succinct utterances.