HistSearch - Implementation and Evaluation of a Web-based Tool for Automatic Information Extraction from Historical Text

Due to a lack of NLP tools adapted to the task of analysing historical text, historians and other researchers in humanities often need to manually search through large volumes of text in order to find certain pieces of information of interest to their research. In this paper, we present a web-based tool for automatic information extraction from historical text, with the aim of facilitating this time-consuming process. We describe 1) the underlying architecture of the system, based on spelling normalisation succeeded by tagging and parsing using tools available for the modern language, 2) a prototypical graphical user interface used by the historians, and 3) a thorough manual evaluation of the tool performed by the actual users, i.e. the historians, when applied to the specific task of extracting and presenting verb phrases describing work in Early Modern Swedish text. The main contribution is the manual evaluation, which takes both quantitative and qualitative aspects into account, and is compared to automatic evaluation results. We show that spelling normalisation is successful for the task of tagging and lemmatisation, meaning that the words analysed as verbs by the tool are mostly considered as verbs by the historians as well. We also point out the further work needed for improving parsing and ranking performance, in order to make the tool really useful in the extraction process.