The reviewers have different views on this paper - although the experimental results are not very strong the paper is well-written and introduces some interesting new ideas to the NeurIPS community. Overall I think the paper is worth presenting at NeurIPS. However, the final camera-ready paper MUST discuss the relation to heirarchical embedding schemes such as [1,2], as discussed by (R3) and also logic tensor networks [3], a related formalism for embedding logical expressions. [1] Poincaré Embeddings for Learning Hierarchical Representations. Maximilian Nickel, Douwe Kiela. NIPS, 2017. [2] Learning Continuous Hierarchies in the Lorentz Model of Hyperbolic Geometry. Maximilian Nickel, Douwe Kiela. ICML, 2018. [3] Serafini, Luciano, and Artur d'Avila Garcez. "Logic tensor networks: Deep learning and logical reasoning from data and knowledge." arXiv preprint arXiv:1606.04422 (2016).