This entry investigates the nature of language used by businesses in disclosure documents (10Ks and 10Qs) and earnings calls during crises that had negative financial impacts. After applying three techniques from natural language processing, specifically document similarity, sentiment analysis, and complexity analysis, we are able to make important contributions to the literature. We discover statistically significant differences in communication before and after crises, characterized by additional discussion referring to the implications of the crisis on the business and increased negative sentiment. Furthermore, we contribute to the literature the finding that many traditional models correlating textual indicators with stock performance do not hold for crisis-constrained situations. We begin uncovering a possible trading strategy based upon sentiment surprise on a filtered set of tokens that excluded mentions related to the crises.