The increased digitization of cultural heritage artifacts, such as books or manuscripts, as well as the prevalence of social media, create an ever-growing set of highly complex data which humanities researchers aim to analyze and understand. Recent advances in the accuracy of language analysis technologies, relying on generic language models trained on vast amounts of data, enable automated analysis of humanities data, but also pose a challenge: language models are essentially black boxes and it is unclear what exactly they learn and how. Our recent work focuses on visual analytics solutions for data exploration as well as topological analysis of high dimensional semantic spaces.