EXPERIMENTING GENERALIZED PROCRUSTES ANALYSIS ON A CORPUS OF SCENES CARVED ON MESOPOTAMIAN CYLINDER SEALS AND ENCODED IN A TEXTUAL DATA SET

In a recently published study, a textual dataset encoding a group of scenes depicted on Mesopotamian cylinder seals from the third millennium was submitted to Procrustes Analysis, considering Correspondence Analysis variations due to progressive reductions of the forms involved. The results seemed to indicate that a slight reduction in the number of forms used to describe the raw data would improve the Correspondence Analysis results. Indeed, the actual impact of each reduction on the outcome of the analyses could not be adequately identified if the forms were not considered as elements forming segments, i.e., sub-images of the scenes. In this paper, the results of the same methodology are presented, this time applied to repeated segments within the same dataset, i.e., sequences of textual forms that may describe sub-images of the described scenes. The comparison between the progressive reductions of repeated segments showed relevant differences between small (around 10% of the rarest segments) and large reductions. Indeed, the latter may undermine a consistent interpretation of the different finding sites, yet well represented keeping 90% of segments. Moreover, a reduction performed by hand and not on the frequency of occurrences provided better results.