M. Járó - L. Költő szerk.: Archaeometrical research in Hungary (Budapest, 1988)

Analysis - SZILÁGYI Katalin: Computer testing used in the typology of pearls originating from the IX—XI centuries

SZILÁGYI Katalin* COMPUTER TESTING USED IN THE TYPOLOGY OF PEARLS ORIGINATING FROM THE IX-XI CENTURIES Abstract - In order to establish the typology of pearl findings a computer-aided analytical method was elaborated in which the author combined cluster analysis with distance analysis. This is discussed in detail in the paper. I have selected 84 pearls from the glass pearls found in the Halimba- Cseres cemetery in such a manner that they contain the maximum number of groups and sub-groups. I have also selected 16 other glass pearls originating from the same age but from seven different cemeteries; these 16 pearls only partially contained the features of the Halimba pearls. It was anticipated that they would show some kind of relationship with the Halimba pearls. This means that all together 100 pearls formed the basis of the computerized analysis [1]. I have used 121 features to describe the pearls. These include the quality of the material, the shape, the method of manufacture, the ornaments, the colour, the size. Some of these are qualitative, others are quantitative characteristics. All were coded in a form to be suitable to be worked with by a computer performing cluster analysis. I used the 2M cluster analysis program of the BMDP Biomedical Computer Programs package for processing the data. This program works with the so called average-link algorithm, and it is able to handle large amounts of data. The computer program characterizes the similarities between two pearls by the square-root of the number of dissimilarities between the pearls. The running of the program provides a detailed list on the merging of the different groups. For example the serial number of the groups to be combined, the average distance between them and the average of the features of the items belonging to this newly formed cluster. The computer produces from these combinations an easily understood graphical chart, the dendogram. The vertical axis shows the average combination distance, while the horizontal the serial number of the individual pearls (Fig 1). The dendogram can be split at any combinational distance thus producing separate groups. I have been able to find such a maximum combinational distance where the dendogram, when split, shows a grouping which coincides with that of the experiencal archeologist. This was 2,3, which means that if there are more than five diverging features, then we cannot talk any more about the same type. The program calculates the distances between two pearls for all the pearls, and the result is indicated both in numeric and graphic form. One element of the distance matrix is equivalent to one character on the graphicl diagram. The overlapping of this character is a good indication of the similarities of the two pearls. The groups of the dendogram appear here in the form of a rectangular triangle lying with its hypotenuse on the diagonal of the figure. With the aid of this figure the internal structure of the clusters can also be checked (Fig 2). The result of the comparative computer analysis is a control which is more objective than the opinion of the archeologist. This means that the glass pearls of the examined age can be better typified. As archeologists we would be inclined to take into * H-1026 Budapest, Pasaréti út 112/a.