Review for NeurIPS paper: Language-Conditioned Imitation Learning for Robot Manipulation Tasks

Additional Feedback: "objects vary in shape, size, color" could mention here that object category varies as well. Line 127 "30000 most used English words" no UNK token needed? Are all the words in the vocabulary already, even from human users for that evaluation? Language templates come from human "experts", but no detail is given about who these experts are, or what the agreement between them as annotators is. Table 2 is confusing to read.