Historical tables, such as administrative registers, represent vast and valuable sources of information for researchers. However, despite large-scale digitization efforts, extracting and structuring their content remains challenging. The French 19th-century Land Registry is a notable example: rich in detailed land use information, yet highly heterogeneous, and still largely underexploited. Although recent deep learning methods have improved information extraction (IE) from digitised documents, they often lack semantic structuring. Conversely, Semantic Table Interpretation (STI) techniques, mostly applied to natively digital tables, offer structuring and linking capabilities but are rarely used on historical sources. In this work, we propose a pipeline that combines deep learning-based IE with STI, guided by a domain ontology. The approach produces a knowledge graph that enables querying and exploration of historical records. We evaluate the resulting knowledge graph using several metrics, demonstrating the potential of our method for semantic enrichment of historical data.