Researchers use visual psychophysics and machine learning to try to read ancient texts

Researchers at the University of Notre Dame have developed an AI based on human perception to aid in the reading of old handwriting.

The hope is that this effort will allow for the preservation of millions of manuscripts. Some of these manuscripts have never been read. Although some texts can be accessed digitally, it is only a small fraction.

This project will make it possible to easily access handwritten manuscripts with automated transcription.

It will be easy to search the text for quick reference. Walter Scheirer, Associate Professor in the Department of Computer Science and Engineering, Notre Dame, says, “We are dealing with historical documents written using styles that have long been out of fashion, going back many hundreds of years, and in languages such as Latin, which rarely ever get used anymore.”

Scheirer states that the new system will use machine learning and visual psychology, which look at the relationships between the mental world and the physical world. Research focuses on the time it takes for a person to identify a letter, abbreviation, and handwriting.

Researchers have examined the transcriptions of Latin manuscripts in Latin that were digitized by readers. These manuscripts were written in the ninth century. To determine which words and passages were the easiest to comprehend, the researchers measured how long it took to read them.

It’s not a common strategy in machine learning. These psychophysical measures, which are directly derived from psychological studies of perception through behavioral measurements, will be used to label the data.

Scheirer states that we then inform the network about common problems in the perceptions of these characters and can make corrections based upon those measurements.” Scheirer adds.

The project is looking very promising. However, the Scheirer team continues to work on improvements to the system. The main problem is the accuracy of transcriptions. This is especially true for documents that are damaged or incomplete. Illustrations are not yet supported by the network.

Connon Wood, a historian for the education portal EDUCALINKAPP said that the network could prove to be especially beneficial for scholars in the humanities. He says that those who wish to gain a deeper understanding about specific historical events or ancient cultures should look to written material.

He continues, “So it is imperative that these manuscripts be preserved, especially when it comes to languages or cultures that are disappearing,”

There are many texts like these texts. The Acts of the Town Council of Santiago de los Caballeros, Antigua (Guatemala), which dates back to the 16th Century and records the beginning of self government in Guatemala, is one that springs to mind.