This dissertation is a multidisciplinary approach that brings together computational methods in machine learning to aid quantitative methods in the social sciences towards the study of social conventions - in particular, linguistic meaning and ideologies. Recent advancements in machine learning have yet to be applied in the social sciences where they can help identify groups with distinct underlying properties in order to gain insight into their unique conventions. Here, social conventions can be thought of as regularities of behavior (eg. norms) and beliefs that are shared between members of a group, those that govern social interactions or ascribe meaning to actions which form through tacit agreement. We used universal features common to all groups as a means of identifying latent groups present in the data. This approach reveals extant patterns without relying on prior assumptions, cultural knowledge, or predefined subgroups to highlight endogenous features within and between groups. The four projects study various aspects of social conventions, identify salient concepts important to groups, and model mechanisms that drive group beliefs and behavior. Much of these studies are dedicated to developing and testing effective methodologies and findings from each study informs aspects of consequent ones. These studies consist of: 1. Universal schema of the World Color Survey (WCS), 2. Cultural Consensus of Ideological groups, 3. Group probabilistic ordering of moral concepts, and 4. Further evolution of natural categorization systems.