Tuesday 28 June 2011

Michael Penn - Computer Assisted Paleography of Syriac Manuscripts



I helped develop a software system bringing together recent advances in digital handwriting recognition with scholarship in Syriac studies. It will allow a scholar to scan a facsimile page from a manuscript of interest. The computer then compares the test document to those in its database and lists the degree of similarity between this document’s handwriting and that of other manuscripts indicating any close enough to suggest that they were written by the same scribe. It also compares the handwriting with documents of known dates, suggesting a year range during which the manuscript was originally written. 
The software for this project has been developed through a collaboration between myself, computer science Professor Nicholas Howe at Smith College, and a Smith engineering student. The software quantifies how closely the handwriting of a test document matches the handwriting of other documents in the computer’s database by measuring five different handwriting elements. It evaluates the overarching writing style in terms of contour hinge (roundedness) and run length (expansiveness). It then analyzes each of the twenty-two Syriac letters in terms of the whole letter, significant geometric portions of the letter, and blank space within parts of the letter. 
We recently tested the prototype on a database of 29 ancient codices. If the computer just analyzed overarching writing style it performed with 76% accuracy. When it compared corresponding letters the accuracy jumped to 98.7%. When it combined all five handwriting algorithms its accuracy was 100%. That is, every time the system would properly match a test page with the manuscript from which it came. The American Academy of Learned Societies recently awarded me a Digital Innovation fellowship for this work and the American Academy of Religions has provided me with a summer research grant to expand the database using manuscript data from the British Library.
I am proposing a paper to present the initial findings of this collaboration as well as obtain suggestions for how to best structure the finalized software package so that it would be of greatest use to scholars in the field.

No comments:

Post a Comment