Unsupervised Sumerian Personal Name Recognition.
Liang Luo, Yudong Liu, James Hearne, Clinton Burkhart · The Florida AI Research Society · 2015
This paper describes an unsupervised named-entity recognition (NER) system to identify personal names in Sumerian cuneiform documents from the Ur III period. We are motivated by the needs of social and economic historians of that period to identify specific persons of importance and such historically relevant facts as can be discerned by the surviving texts. The work was confronted by the challenges posed by the fact that Sumerian is not a well understood language and the texts come down to us in damaged condition. We based our recognizer on the Decision List CoTrain algorithm, subjecting it experimentally to modifications to accommodate the nature of the data and narrower task it was originally devised for. We achieved 92.5% recall and 56.0% precision, results that are usable by the economic and social historian. We described the results of our work and suggest further applications of the techniques we have devised, also in the analysis of ancient Sumerian texts.