Martina TROGNITZ     
(Heidelberg University, Heidelberg, Germany)

Abstract: EVA is an expert system for classification and dating of ceramics. It represents an application of natural language processing in archaeology. The program has a knowledge base that consists of a decision tree. This decison tree is walked through by taking the user’s answers to a set of predefined questions. The leaf nodes contain the output, a classification of the object. By answering one question at a time the user is encouraged to look closely at the object in question for relevant characteristics. In an alternative mode of the program, the user can input a description of the object in natural language. EVA then automatically extracts the answers to the predefined questions. This is achieved by information extraction, that mainly is done by searching the text for particular patterns. The knowledge base of this expert system is independent and fully exchangeable. Its syntax is simple and thus it can easily be expanded or changed by non-programmers. First tests were done with vases of the geometric style, showing that this is a practicable approach capable of good results if the knowledge base is good enough. Further work could include a component for image processing or the use of machine learning for an automatic acquisition of the decision tree or of the information extraction patterns.

Keywords: Computational Linguistics, Ceramic, Expert System