Abstract
Procedurally generating cohesive and interesting game environments ischallenging and time-consuming. In order for the relationships between the gameelements to be natural, common-sense has to be encoded into arrangement of theelements. In this work, we investigate a machine learning approach for worldcreation using content from the multi-player text adventure game environmentLIGHT. We introduce neural network based models to compositionally arrangelocations, characters, and objects into a coherent whole. In addition tocreating worlds based on existing elements, our models can generate new gamecontent. Humans can also leverage our models to interactively aid inworldbuilding. We show that the game environments created with our approach arecohesive, diverse, and preferred by human evaluators compared to other machinelearning based world construction algorithms.