Daejeon, July 27–31
DH2026 Text Submission Guidelines
This is the template for the international Digital Humanities conference 2026 of the Association of Digital Humanities Organizations (ADHO). Please read and follow the following information thoroughly in order to facilitate the conversion to TEI for the edition of the Conference Proceedings.
The template consists of sections with light grey background like the one you are reading right now that cannot be edited and will be processed automatically during the TEI-conversion. For the other sections please use only the paragraph styles provided by this templates named „DH-[Stylename]“:
In addition we support images, links, lists and tables and the following character level styles:
There are three main sections which are editable:
Please follow the additional indications for the typescripting:
Images: Images should embedded in the document and supplied at 300 dpi and uncompressed where possible. Screenshots will normally be the only exception, and these should be supplied at the highest possible resolution. Acceptable formats are TIFF, JPEG and PNG.
Image captions: All images should include below them a caption labeled “Figure (number) and a one-line description of what the image portrays.
Links: Make sure that embedded links are all active and functional.
Lists: Use the built-in list formatting tool in your document editor.
Notes should be used only for comments, not for simple bibliographic citations.
Program code and mathematical formulae should be included as images.
Quotations with a minimum of 5 lines of type should be in DH-Quotation format small type, set full left. For short quotations within the text, double quotation marks should be used. Single quotation marks should be used for quotes within quotes.
Bibliographic references should be cited in the text using the author's name, year of publication, and eventually pages, for example (Ellis 1987) or (Richardson / Braden-Harder 1988: 196). Works by more than two authors should be shortened using “et al.” (Baayen et al. 1996: 125).
Bibliography: The list of references should be headed References and placed at the end of the submission. It should be in alphabetical order and the paragraph style DH-Default should be used. Where an author has more than one publication, they should be arranged in chronological order, and if there is more than one publication within a year, they should be alphabetically ordered by title and labelled a, b, etc. (e.g. 1989a, 1989b). Single-author works precede co-authored works. If citing an electronic publication, please supply the full URL and the date when you last accessed it. Please follow the examples given below for bibliographic layout:
Baayen, Harald / Halteren, Hans van / Tweedie, Fiona (1996): “Outside the cave of shadows: using syntactic annotation to enhance authorship attribution”, in: Literary and Linguistic Computing 11, 3: 121–132.
Biber, Douglas (1988): Variation Across Speech and Writing. Cambridge: Cambridge University Press.
Ellis, David (1987): The Derivation of a Behavioural Model for Information Retrieval Design. Ph.D. thesis, University of Sheffield.
Garfinkel, Michele S. / Weiss, Steven. C. (1999): “In the court of history, Ehlers v. Bush”, in: Recent Science Newsletter 1, 1: 6-7 <http://web.archive.org/web/20030802233236/http://recentscience.gwu. edu/RSN/PDF.files/RSN1.1.pdf> [15.03.2021].
Hazadiah, Mohd Dahan (1993): “Topic as a Dynamic Element in Spoken Discourse”, in: Baker, Mona / Francis, Gill / Tognini-Bonelli, Elena (eds.): Text and Technology. In honour of John Sinclair. Amsterdam: John Benjamins 55-74.
Richardson, Stephan D. / Braden-Harder, Lisa C. (1988): “The Experience of Developing a Large-Scale Natural Language Text Processing System: CRITIQUE”, in: Association for Computational Linguistics (ed.): Second Conference on Applied Natural Language Processing, Austin, TX, February 1988: 195–202. DOI: 10.3115/974235.974271.
Note: The Guidelines were updated and corrected by Elisabeth Burr on 15 March 2021. The Harvard system described here diverges from the one used in DSH mainly with respect to person names. Person names are requested in full in order to be able to differentiate for example between Christoph, Christa, Christian, and in order to invalidate stereotypical concepts of “scholar”.
From China to Korea: Tracing a Divine Persona through Motif-Based Visual Analysis of the Yushu Baojing and Its Recompilations
Daoism is a native religious tradition of China organized around self-cultivation practices, ritual performances, and scriptural-register systems. Puhua tianzun is a major figure in Daoist Thunder Ritual traditions and serves in the capacities of presiding over thunder, exorcism, and moral transformation. The Yushu Baojing is a central text associated with the teachings of Puhua tianzun that conveys the cosmology and ritual logic of thunder rites.
During the 16th to 18th centuries, the Yushu Baojing circulated across in China and Korea through a process of textual re-composition and pictorial re-production that routinely redesigned the divine persona of Puhua tianzun. Illustrations across different editions create a representational thunder realm through visual components that are modular, separable, and replicable, and these same motifs occur in other media, including printed scriptures, ritual manuals and talismans. This study reconsiders the cross-edition transmission of Yushu Baojing imagery as a system of recombinable cultural units. Focusing on an eighteenth-century Korean re-edited version, it examines how Korean compilers inherited, transformed, and reconfigured visual vocabularies from earlier Chinese editions.
The dataset is derived from a sixteenth-century Chinese polychrome edition and an eighteenth-century Korean edition, comprising 27 illustrations. Using supervised learning and YOLO-based image detection, 198 visual motifs (225 image instances) were extracted to form a small, structured dataset. Methodologically and conceptually, this project aligns with the DH2026 Engagement theme by integrating computational image analysis with humanities-driven questions of transregional circulation, cultural transmission, and visual recombination.
Since the early 2000s, research on Puhua tianzun and the Yushu Baojing has focused on textual history, ritual institutions, and visual culture. Miura Kunio traces the scripture’s transmission across Chinese, Korean, and Japanese editions, locating its formation in the Southern Song period (1127-1279) (Miura 2002). Using Korean archival sources, Ku Chonghoe shows how the text, once excluded from official rituals, was popularized and locally adapted in Korea (Ku 2012). Li Yuanguo and Park Jongchun examine shifts in Puhua tianzun’s doctrinal and ritual roles, from institutional Daoist contexts to Korean local and shamanic practices (Li, 2013; Park, 2018). Visual studies by Wan and Liu further explore the ritual efficacy and regional reconfiguration of thunder-deity imagery (Wan 2015; Liu 2024).
In contrast, digital-humanities research is more focused on turning religious images into structured visual data. Ontology-based frameworks like ICON can encode and query patterns in different artworks (Sartini et al., 2023).There are also large annotated datasets from Dunhuang murals to Tibetan thangkas and Buddhist mudrā corpora, which show how religious images can be disassembled into small units that can be read by machines (Milani et al., 2020; Zhang et al. 2023).Research on CLIP-based models has also pointed out the potential and limitations of weakly supervised analysis in art history research (Conde et al. 2021; Baldrati et al. 2022; Toumarkine et al. 2025).Together, these work provides a methodological basis for analyzing Yushu Baojing's illustrations, allowing us to study how they can serve as reconfigurable visual units in cross-regional ritual communication networks.
Inspired by Richard Dawkins’s notion of the “meme,” originally proposed in evolutionary biology, this study approaches the Yushu Baojing tradition as a system of replicable cultural units. It examines how specific textual and visual elements were selectively inherited, modified, or discarded as the tradition moved from Chinese editions to Korean recompilations, reflecting divergent ritual and cultural contexts. Specifically, the study addresses three questions:
(1) Which visual motifs in Chinese editions were selectively adopted in the Korean re-edited versions?(1) Which visual motifs in Chinese editions were selectively adopted in the Korean re-edited versions?
(2) Which motifs were reorganized, expanded, or functionally transformed in the Korean context?(2) Which motifs were reorganized, expanded, or functionally transformed in the Korean context?
(3) How do full-page illustrations diverge from the textual narrative sequence, and what alternative visual logic structures the resulting mode of storytelling?(3) How do full-page illustrations diverge from the textual narrative sequence, and what alternative visual logic structures the resulting mode of storytelling?
To investigate the transregional transmission and recombination of visual motifs in the Yushu Baojing, this study integrates digital humanities methods with computer vision. Motif categories were first defined according to recurring iconographic and compositional units shared across the Chinese and Korean editions. Boundaries were drawn around the smallest coherent motif while excluding unrelated text, margins, or decoration. All annotations followed fixed criteria and were repeatedly reviewed by the author to enhance internal consistency and methodological transparency. Using these motif-level annotations as an interpretive guide, the study applies YOLO-based object detection to extract and localize motifs across both datasets. It then computes CLIP image embeddings and uses cosine similarity as a quantitative metric for cross-corpus image matching between the Chinese and Korean editions. Similarity scores, typically ranging from 0.72 to 0.86, are complemented by manual visual inspection to assess reliability and limitations. Activation-based visualizations using VGG16 further support the interpretation of motif reuse, spatial shift, and layout recontextualization across editions.
Figure 1. Research pipeline and methodological framework.
The YOLO11n model realized a motif detection and localization across both the Chinese and Korean datasets. In multiple test cases, motifs detected in the Korean edition correspond directly to motifs extracted from the Chinese edition, confirming the model’s ability to identify shared visual units across editions. Notably, bounding-box coordinates reveal systematic spatial shifts in motif placement within page layouts, indicating recomposition rather than simple duplication. These positional variations provide a quantitative basis for comparative layout analysis.
Figure 2. Motif detection results in the Chinese edition dataset.
Figure 3. Motif detection results in the Korean edition dataset.
CLIP-based similarity analysis reveals a strong degree of visual-semantic overlap between the two corpora. Most high-confidence matches fall within a cosine similarity range of 0.72–0.86. For example, the Chinese motif image C103-3 and the Korean motif image K89-3 achieve a similarity score of 0.8582, which is notably higher than that of other potential candidates. In several cases, a single Chinese motif corresponds to multiple Korean motifs with similarity scores above 0.83, suggesting repeated deployment of the same visual motif within the Korean edition.
Figure 4. Results of CLIP-based similarity analysis for selected Sino-Korean motifs.
VGG16 activation heatmaps indicate that similarity matching is driven primarily by structurally meaningful visual features. Model attention consistently concentrates on figures, seated postures, halos, and thunder-related attributes, rather than background or decorative elements. For example, comparison between one pair of full-page illustrations (C-103 and K-89) and their extracted motifs (C103-3 and K89-3) shows that motif-level matching yields more focused and interpretable activation patterns.
Figure 5. Results of VGG16-based visual matching for selected Sino-Korean motifs.
When reinserted into their original page contexts, matched motifs reveal non-linear patterns of transmission: some motifs retain their relative sequence across editions, while others are reordered, duplicated, or omitted. These patterns substantiate the interpretation of Yushu Baojing imagery as a system of recombinable cultural units operating across regional and temporal boundaries.
This study contributes to area studies and the study of cultural diffusion by modelling the transregional re-editing of the Yushu Baojing as a historically grounded and structurally verifiable process. Combining close reading, edition-based comparison, and computational image analysis, it translates Sino-Korean religious transmission from a descriptive narrative into an analyzable model of visual circulation. By treating Daoist imagery as modular units that can be selectively copied, repositioned, and recombined, the study demonstrates how religious images function as mobile cultural components within East Asian ritual networks rather than as fixed iconographic wholes. Methodologically, this study tries to fills a void in cross-edition visual comparison within Daoist and East Asian religious studies, offering a replicable framework for tracing cultural diffusion across regions, media, and time. More broadly, it tries to show how digital-humanities methods can enhance area-based scholarship by making processes of transregional adaptation, localization, and reuse visible at the level of visual structure rather than solely through textual evidence, which represents a significant practice of meme theory in Taoist text’s transregional transmission. More broadly, this study demonstrates how digital humanities methods can enrich area-based scholarship by making processes of transregional adaptation, localization, and reuse visible at the level of visual structure, rather than relying solely on textual evidence. In this sense, it offers a methodological application of meme theory to the study of Daoist texts’ transregional transmission.
As an exploratory and preliminary attempt, this abstract showcases that the Korean adaptations of the Yushu Baojing are not mere copies of the original Chinese texts; rather, they are thoughtful, structurally linked reworkings that build upon an inherited visual tradition. By employing a computational approach that contrasts the Chinese and Korean sources, the study constructs a methodological framework where historical knowledge, visual scrutiny, and machine learning converge. This foundation will be fortified before the official presentation through the augmentation of datasets, the enhancement of self-supervised annotation coherence. Expanding pretraining to include inter-religious visual collections will also place Daoist imagery within a larger East Asian visual context, allowing for a more robust comparative study even with limited data.
When comparing Chinese and Korean versions, four key insights come to light. First, motifs function as hypertextual nodes inherited from Chinese editions: labels, captions, and ritual instructions act as links that allow Korean readers and practitioners to recombine text, image, and ritual action across media. Second, the Korean editions intensify motif replication through clearer chaptering, explicit labels, and handbook-style layouts, effectively slicing complex Chinese ritual knowledge into callable modules that enhance engagement, usability, and transmission. Third, quantitative differences in motif distribution—marked by deletions, additions, and systematic re-chaptering relative to Chinese exemplars—indicate a purposeful adaptation to Korean local needs, shifting the emphasis from punitive thunder cosmology toward civic protection, healing, and popular–shamanic functions. Finally, the endurance of Puhua tianzun in Korea depends not on preserving a fixed Chinese text, but on the selective inheritance and continual recompilation of a Daoist visual–ritual toolkit. Through authoritative Chinese sources, stabilized motifs, and executable rites, the deity is repeatedly reassembled and reactivated, continually arriving in the Korean religious present through sustained forms of textual, visual, and ritual engagement.