Classification automatique des manuscrits des discours de Cicéron fondée sur le choix et l'ordre des discours

The authors have followed a simplified procedure for examining the stemma of the 257 manuscript volumes surviving from the medieval period, limiting their purpose to the choice of orations given and the order followed. Their program computes the degree of similarity between two given manuscripts through alternative coefficients established for the purpose. The divergence of the latter is in practice rather small. They are considered large only when the volumes are similar as regards descent, barring the possibility of an abstract order which could have been followed by two scribes living in a different time and place, such as a chronological order. At a threshold controlled by a calculation of probabilities, the list of manuscript pairs by decreasing order of similarity is plotted on a graph. Kernels of highly similar manuscript classes then appear. The significance of these kernels is confirmed through a partial collation of the oration pro Balbo ; they occasionnally furnish more accurate information. To process similar data according to these procedures would require two or three days work, with maximum efficiency obtaining in the first stages of studies involving a large number of manuscripts, as a guide to further collation.