The Official Website of the People’s Government of Hunan Province Mobile 中文 Français 한국어 日本語

15 July 2015

Home>News>Coverage of Hunan from Global Sources

The World's First 'Jiandu' Dataset Enhances China's Traditional Lexicon

2025-04-09 Download Print


A tourist appreciates one of exhibits at the Gansu Jiandu Museum. Photo: VCG


Being in the artificial intelligence (AI) field for more than a decade, Zhang Qiang never thought that one day his field would intersect with jiandu - China's 3,000-year-old bamboo and wooden slips - until he and his team developed DeepJiandu, the world's first large-scale dataset dedicated to character recognition relating to Chinese relics.

Dubbed as China's earliest books made of wood or bamboo slips, jiandu has mediated the transition in Chinese writing forms from earlier oracle bone and bronze inscriptions to later paper-based manuscripts.

The recently debuted DeepJiandu is described by Zhang as a "latest experiment" to his team's long-run mission of building an online digital jiandu platform. The program's developing journey involves multiple difficult procedures, but its goal is clear - to make the thin bamboo slips embodied "ancient cultural-social beliefs, folk traditions and calligraphic arts" accessible to not only scholars, but also the general public. 

'A lexicon'

Unlike existing data detection tools designed for distinguishing legible characters and texts on ancient paper manuscripts, the current DeepJiandu was born to solve the conundrum of recognizing jiandu characters under complex scenarios.

These scenarios include specifying variably sized characters written in different dynasties, connected strokes in ancient handwritings and also characters ruined or blurred by natural weather, which is the "most common but inevitable dilemma" in jiandu conservation, Fang Beisong, a jiandu expert, told the Global Times. Fang is also the director of Jingzhou Cultural Relics Preservation Center located in Hubei Province.

"Prior to digital and AI tools, we almost 100 percent relied on manual works, but such works often took endless times, and have very narrow margin for errors," Fang emphasized.

Aimed at assisting researchers like Fang looking for jiandu research means, the DeepJiandu dataset provides 7,416 infrared images after scanning more than 40,000 physical jiandu specimens. The team has also examined on different individual samples to have created a nearly 100,000 annotations.

Such annotations, though they appeared to be minor, "greatly contributed to clarifying certain characters' meanings," allowing even amateurs to learn contents on jiandu.

"A common Chinese character like '大' (lit: meaning big in Chinese) can connote various meanings across different jiandu pieces. Our curated annotations form a lexicon that help scholars while engaging the public to learn the profound 'language' of jiandu," said Zhang.



Jiandu pieces showcased at the Gansu Jiandu Museum Photo: VCG 


'Relevant to modern life'

The DeepJiandu team's goal for providing an open-sourced "lexicon" for everyone actually started before the dataset was born. In June 2024, an even more detailed data sharing platform focuses on the jiandu relics was also created by the team with the support of Northwest Normal University. The platform contains categories such as "character glyph library" and "textual interpretation database." Checking into the entry featuring jiandu pieces unearthed from Gansu's Diwan ruins, one can see "filial virtue" and "morality" were praised by Han Dynasty (206BC-AD220) people.

At the site's character database, one can see common Chinese characters like '月' (lit: moon) and '印' (lit: seal or print) were written in styles such as the cursive and seal scripts, revealing the country's calligraphic aesthetics are rooted in ancient people's everyday lives.

"I'm not at an archaeology or history major, but browsing the site's digitalized jiandu pieces inspired me to create my own art prints," Yao Yao, a postgraduate at the Sichuan Fine Arts Institute, told the Global Times.

Similar to Yao Yao, gourmet and culinary researcher Zhong Yu'e told the Global Times that the food ingredients such as fish and fermented beans recorded in the Han Dynasty jiandu unearthed at Dunhuang's Xuanguanzhi site has inspired her to "recreate recipes emerged along the ancient Silk Road."

"Jiandu covers a vast range of information about ancient social and law regulations, literatures, the art of numerology, ethnic relationships and so forth," Zhang Yiwu, a professor of cultural studies at Peking University, told the Global Times.

"Many of such jiandu delivered information remains relevant to our modern lives," Zhang said. "Without digital technologies, the access to jiandu would never have reached such a wide range of audiences."

Just as what Zhang has suggested, the enlarged technological application on jiandu is now no longer unique to research labs, but has become integral to public sites such as museums. 

Housing over 100,000 jiandu pieces, the Changsha Jiandu Museum in Hunan Province is mostly known for its wooden and bamboo slips that belong to the Kingdom of Wu, which dates back to the Three Kingdoms period (220-280). To innovatively engage visitors with interesting stories behind jiandu slips, the museum has developed an AR picture book. By scanning the QR codes on book pages, visitors can grasp historical facts such as how household registration system was designed during the period.

As a fellow museum to the Changsha site, Zhu Jianjun, the director of Gansu Jiandu Museum, revealed to the Global Times that two AI robots, named "little Jian" and "little du," have already been used at the museum to provide interactive guiding services to visitors.

"The digital creativity on jiandu not only shows the innovation of the relics, but also stands as an example revealing that today's Chinese cultural heritage conservation is integrated with technological innovation," Zhu said.


Source: Global Times