Social Network Extraction and Exploration of Historic Correspondences

Driven by the continuously increasing number of digitized and transcribed historic documents, natural language processing (NLP) and text analysis tasks are now frequently applied to historic texts to extract useful information and thus to enrich this cultural heritage. These tasks face several challenges, such as dealing with spelling variations, lack of orthography, and, oftentimes, missing reference language corpora. In this paper, we present our approach to information extraction from a large corpus of correspondences written in the first half of the 16th century. For around 9,700 letters written in Early New High German by the German reformer Philipp Melanchthon, we develop and extend approaches to named entity recognition to extract information about geographic places, persons and organizations, and time. The extracted information serves as the basis for creating a social network structure that includes information about senders and addressees of letters as well as people mentioned in letters. This correspondence network structure is to be exploited in terms of analyzing the evolution of the correspondences, key persons and communities, and eventually topics covered by the letters over time, thus resembling key analysis tasks applied to today’s typical social networks.