Digital sandbox games such as "Minecraft" can be used to assess and support creativity. Doing so, however, requires an understanding of what is deemed creative in this game context. One approach is to understand how Minecrafters describe creativity in their communities, and how much those descriptions overlap with the established definitions of creativity in the literature. Based on prior work, we used a modified rubric for assessing a creative product in this study. We investigated the evaluation of a total of 52 "Minecraft" builds (i.e., structures that are created by the Minecrafters) from five YouTube videos. In addition to qualitatively coding the video transcripts, we also used natural language processing (NLP) for automatic coding of our data at scale. Our qualitative analysis of the YouTube transcripts revealed evidence for all but one dimension (i.e., humor) of the creativity rubric we used in our dataset. In addition, two new dimensions of creativity emerged in our qualitative coding process (i.e., realism and effort). Finally, the NLP model that we developed achieved about 80% accuracy in distinguishing the sentences that showed evidence of creativity. This study provides insights for research and educational practices of creativity in "Minecraft" or similar game or simulation environments.