top of page

Insight into Taiwan: Inventory and Classification of Primary Sources in Taiwanese History

photo12178.jpeg
Su-Bing Chang

Emeritus Professor, Graduate Institute of Taiwan History, National Taiwan Normal University

7f761c33-7d10-414b-968d-0b70940daead.jpg
Kuan-Fei Chen

Assistant Professor, Department of History, National Taiwan University

Abstract

This session will feature hands-on demonstrations and showcases to compare the innovative achievements of Taihu AI in the intelligent application of Taiwanese historical documents, and explore its implications for future methodologies in digital humanities research.

​

"Taihu" is a knowledge-mining system developed from a digital humanities perspective, built upon primary historical materials and archival documents from Taiwan. The system's development team first surveyed a wide range of representative historical databases, and then categorized them based on the structure and characteristics of the materials. The databases currently included are Taiwan Provincial Assembly Gazette, Taiwanese Administrative Archives from the Ming and Qing Dynasties, the Chinese edition of Nichinichi Shinpo, and Taiwan Wenxian Congkan, among others. These databases have been constructed following rigorous academic standards.

​

So, how does the Taihu system compare with mainstream generative AI models in terms of the knowledge it produces or uncovers? We conducted three sets of comparative tests between Taihu and the widely used ChatGPT:

  1. Anti-government incidents in Taiwanese Administrative Archives from the Ming and Qing Dynasties

  2. Economic activities of individuals reported in the Chinese edition of Nichinichi Shinpo during the Japanese colonial period

  3. Resource development issues in the Taiwan Provincial Assembly Gazette

 

The test results demonstrate that Taihu has significant advantages in the following three areas:

  1. Accuracy of historical citations: Taihu provides complete source references, ensuring scholarly rigor in citations.

  2. Understanding of documentary context: The system accurately grasps the linguistic features and historical context specific to different periods in Taiwan’s history.

  3. Epistemic humility: Taihu refrains from fabricating answers when data is insufficient, avoiding confidently incorrect responses.

 

By leveraging these well-sourced historical document databases, Taihu allows users to conduct research based on solid materials with clear provenance. These strengths make Taihu a reliable tool for humanities research, enhancing both research efficiency and academic credibility. This is where the Taihu knowledge-mining system surpasses other generative AI models in scholarly research.

bottom of page