Measuring meaning is a longstanding methodological problem in social science – especially meaning in text data. It is also a theoretical problem: measurement requires us to specify (Merton 1948) the theoretically vague concept of meaning itself. Cultural sociologists have spent decades trying to clarify aspects of meaning, such as the extent to which meaning is structured and stable across contexts and the extent to which meaning is stored in our minds versus in external symbols (e.g., images, words, and even parts of words such as suffixes). Meanwhile, recent advances in computer science offer new measures and formal models of meaning in text data. For example, word embeddings quantitatively model the meaning of words in text data. This dissertation capitalizes on such recent advances in computer science to contribute to theoretical and methodological work on meaning in cultural sociology. The first paper theorizes the kinds of sociological meaning that word embeddings operationalize and describes how cultural sociologists can use word embeddings to empirically investigate meaning in text. The second paper uses word embeddings to empirically investigate the extent to which syntactically gendered language (e.g., “policeman”) conveys gendered semantic information. Finally, paper three develops a novel computational approach to measure latent thematic meaning in large-scale text data, integrating word embedding and topic modeling approaches to measuring meaning in text data. The third paper then applies this new approach to identify latent themes in a large, underutilized source of text data on violent death in the U.S.