This thesis investigates how English noun compounds change in meaning across two centuries of historical text. It introduces a computational framework that extracts decade-specific sentence contexts and models them using contextualized embeddings. These embeddings are organized into semantic timelines that reveal both gradual shifts and abrupt changes in usage. This enables compound meanings to be examined at a finer temporal resolution. A central component of the framework is a visual analytics suite that supports interactive and exploratory analysis. It provides similarity heatmaps, temporal trajectories, variance patterns, and sentence-level scatter plots that show how individual contexts contribute to semantic behavior. The system also compares compounds with their constituent words, making changes in compositionality and modifier-head relationships more transparent. An additional feature identifies representative keywords within selected sentence groups, helping users interpret shifts in the semantics and what contextual cues drive them. A global clustering view further places each compound within broader patterns of similar temporal evolution across the dataset. Expert evaluation and focused case studies demonstrate that this integrated approach offers clear, interpretable, and scalable insights into the diachronic behavior of complex lexical expressions. The findings show that combining contextual embeddings, keyword-based interpretation, and interactive visualizations provides a powerful method for understanding how noun compounds evolve. Keywords: noun compounds, contextual embeddings, compositionality, clustering, temporal evolution, lexical expressions.