This paper contributes to an emerging literature that models votes and text in tandem to better understand polarization of expressed preferences. It introduces a new approach to estimate preference polarization in multidimensional settings, such as international relations, based on developments in the natural language processing and network science literatures -- namely word embeddings, which retain valuable syntactical qualities of human language, and community detection in multilayer networks, which locates densely connected actors across multiple, complex networks. We find that the employment of these tools in tandem helps to better estimate states' foreign policy preferences expressed in UN votes and speeches beyond that permitted by votes alone. The utility of these located affinity blocs is demonstrated through an application to conflict onset in International Relations, though these tools will be of interest to all scholars faced with the measurement of preferences and polarization in multidimensional settings.