Archeology: Restoring ancient texts using artificial intelligence

 
A paper suggesting that up to 72% accuracy can be achieved when multiple historians use deep neural networks trained to restore ancient Greek inscriptions will be published in Nature this week.The findings of this study help restore newly discovered and obscure inscriptions and estimate their age and location by increasing speed and accuracy, helping to understand ancient history. It may be possible to proceed.

Historians are studying inscriptions written directly by people of the past on the materials (stones, pottery, metals, etc.) that remain to this day in order to elucidate the history of ancient civilization.However, many inscriptions have been damaged for centuries and are now unreadable and uncertain when they were created.Epigraphers who are experts in epigraphy can restore missing text, but their traditional methods are very complex and require a great deal of time.

Now, Yannis Assael and Thea Sommerschield are a type of artificial intelligence trained to restore ancient Greek inscriptions and estimate their age and location to overcome the limitations of current epigraphy methods. We verified a deep neural network (named "Isaka"). Assael et al. Revealed that 62% accuracy was achieved when restoring inscriptions damaged by Isaka alone, and 72% accuracy was achieved when multiple historians used Isaka.It also revealed that Isaka could help identify where and when the inscription was created.In the experiments in this study, Isaka estimated the location of the inscription with 71% accuracy, and the date of creation was within less than 30 years from the date of creation proposed by historians.

The findings of this study may open up the possibility of collaboration between artificial intelligence and historians, leading to further elucidation of human history.

doi: 10.1038 / s41586-022-04448-z
[Original English »]
 
"Highlights of Nature Magazines" is a translation of a release made by Nature's public relations department for the press.If you need more accurate and detailed information, be sure to read the original paper.

 
* This article is reprinted from "Nature Japan Featured Highlights".
Reprinted from: "Archeology: Restoration of inscriptions using artificial intelligence'
 

Nature Japan

Nature Japan, Inc. is part of Springer Nature, a world-leading publisher in research, education and expertise. Since its establishment in May 1987, Nature Japan Co., Ltd. has been a scientific journal.Nature We are engaged in all business related to publishing activities such as distribution of press releases related to Japanese printing and science, sales and marketing of academic journals and books.In addition, as a partner of universities, research institutes, government agencies and companies, we provide custom publishing and media production to highlight the characteristics of each institution, and advertising and sponsorship services to disseminate branding and research activities to the world. I am.As one of the major bases in the Asia-Pacific region, we are developing a wide range of business activities not only in Japan but also in Singapore, South Korea, Southeast Asia, Oceania, and India.