Old Tables of East Asia

About

This website began in September, 2026 as an experiment which asked the question: how far have large language models come in their ability to accurately extract both text and complex tables of information from historical sources? When it comes to decent scans of sources in English and East Asian texts the answer is, for some models, quite far (see some custom benchmarks, for example).

After beginning with just a small number of China and Japan year books, I expanded the focus and now include a somewhat arbitrary list of other interesting open access works published before around 1950 which have rich tables of information about East Asia and parts of the Japanese empire. Many of these tables are of very limited use to a historian, and some say more about the creators of these works than they do about the places they describe. Some of these tables, obscure though their information may be, will hopefully lead readers to ask interesting new historical questions, or reconsider old assumptions.

Another goal of this website is to introduce visitors to some of the wonderful array of open access materials available online. The types of sources here all beg for careful critical consideration, but many of them are put to effective use in the historiography of the region.

Accuracy

Only a sample of a few dozen tables have been manually checked by myself by comparing them to the original scans via the links provided. Many of the larger complex sample tables checked here here extracted numbers with 100% accuracy or only one or two errors - something that was certainly not the case in tests I carried out with large language model extraction of tables from sources in 2024 and 2025. Many of the mistakes found were very difficult to read (for a human as well) faded numbers (3 vs 8, for example). As you will notice, many tables mark cells as unreadable when the original scan was of poor quality. The highest error rates came from statistical tables with vertically stacked Chinese/Japanese numbers, especially the numbers one, two, and three (一、二、三) next to each other. These often required Claude Opus 5.5 or Gemini Argon 4 with higher effort rates, as well as cropping and zooming by the model to get very high accuracy rates. Only some books and tables were sent for extraction to these more expensive models. Mistakes do exist, however, and it was important to include the ability to chase tables back to the source scans via links here. In the future, I hope to make the human checking of sample tables more formal and post accuracy percentages here.

Further Reading

There is a rich historiography on the history of statistics and especially the role they play in state power and imperialist endeavors. Below is a reading list of works for students to explore on the topic.

East Asia