Scientific information expresses human understanding of nature. This knowledge is largelydisseminated in different forms of text, including scientific papers, news articles, and discourseamong people on social media. While important for accelerating our pursuit of knowledge, notall scientific text is faithful to the underlying science. As the volume of this text has burgeonedonline in recent years, it has become a problem of societal importance to be able to identifythe faithfulness of a given piece of scientific text automatically. This thesis is concernedwith the cultivation of datasets, methods, and tools for machine understanding of scientificlanguage, in order to analyze and understand science communication at scale. To arrive atthis, I present several contributions in three areas of natural language processing and machinelearning: automatic fact checking, learning with limited data, and scientific text processing. Thesecontributions include new methods and resources for identifying check-worthy claims, adversarialclaim generation, multi-source domain adaptation, learning from crowd-sourced labels, cite-worthiness detection, zero-shot scientific fact checking, detecting exaggerated scientific claims,and modeling degrees of information change in science communication. Critically, I demonstratehow the research outputs of this thesis are useful for effectively learning from limited amounts ofscientific text in order to identify misinformative scientific statements and generate new insightsinto the science communication process.