Daejeon, July 27–31
Traditional Korean motifs acquire different symbolic meanings according to period, religion, social group, and context of use. A single motif may therefore carry multiple semantic layers. The lotus motif, for example, is associated with purity in Buddhist contexts and with fertility and abundance in vernacular auspicious contexts. This coexistence of meanings within the same formal element is closely related to multivocality in cultural heritage interpretation (Silberman 2008; Sitzia 2023). This study asks how such meanings can be represented not as undifferentiated descriptions, but as structured semantic units that can be traced, recombined, and examined.
Existing cultural heritage information models provide mechanisms for interpretive plurality. CIDOC CRM supports semantic interoperability (Doerr 2003; Bruseker et al. 2017), and E13 Attribute Assignment models interpretive acts or attributions as events (Bekiari et al. 2021). In iconographic modeling, the ICON ontology formalizes Panofsky’s levels of interpretation (Sartini et al. 2023), while IICONGRAPH extends iconographic and iconological information into a knowledge graph environment (Sartini 2024). These models are valuable, but they do not specifically target the decomposition of culturally specific motif meanings into atomic semantic units and their recombination across contextual conditions. The combinatory relationships among meanings therefore often require additional inference or reconstruction before they can be explicitly explored. This limitation is related to previous work on multiple interpretation and heterogeneous Korean cultural heritage data (van Ruymbeke et al. 2017; Kim et al. 2015).
This study proposes a two-layer ontology structure that distinguishes factual information from interpretive information. The Data Layer represents factual descriptions of artifacts, including Artifact, Pattern, Era, Donor, Material, and Technique. The Semantic Layer represents meanings, contexts, user groups, evidence types, and interpretive evidence. This separation reduces the conflation of factual and interpretive information and increases the reusability of interpretive structures (Doerr 2003).
Within the Semantic Layer, PatternMeaning is defined as an atomic semantic unit, and each distinct meaning is modeled as a discrete semantic entity. Compound interpretations are decomposed into smaller units so that each meaning can be associated with relevant cultural or historical contexts. Relationships among semantic units are represented through isCombinedWith, defined here as a symmetric relationship between PatternMeaning entities. It does not express hierarchy or causality; it records that two atomic meanings participate in the same interpretive configuration of a motif.
Interpretive evidence is modeled through EvidenceInstance, which links semantic units to textual sources, iconographic materials, historical records, and other evidence. Based on dcterms:BibliographicResource, EvidenceInstance records both the interpretive outcome and the evidential basis from which the interpretation is derived. This reflects paradata and supports transparency, traceability, and reproducibility in interpretation (Börjesson et al. 2020; Huvila et al. 2021).
Table 1. Core classes and object properties of the semantic layer
| Type | Name (URI) | Definition / Domain–Range |
| Class | PatternMeaning | Atomic semantic unit |
| Class | CulturalContext | Historical, religious, or social context |
| Class | UserGroup | Social or cultural group |
| Class | EvidenceInstance | Evidence source; subclass of dcterms:BibliographicResource |
| Class | EvidenceType | Type of evidence |
| Property | hasMeaning | Pattern → PatternMeaning |
| Property | isCombinedWith | PatternMeaning ↔ PatternMeaning |
| Property | belongsToContext | Pattern → CulturalContext |
| Property | belongsToUserGroup | Pattern → UserGroup |
| Property | supportedBy | PatternMeaning → EvidenceInstance |
The dataset comprises ten artifacts from the Donated Works Gallery of the National Museum of Korea, corresponding to the museum domain within GLAM. The selected objects include Goryeo celadon, buncheong ware, and white porcelain. Across these artifacts, 27 motif types were identified, from which 33 semantic units and 37 motif–meaning mappings were constructed. The semantic units were derived from literature and cultural heritage interpretation materials, and each unit is linked to its source through EvidenceInstance.
The proposed ontology was implemented in Neo4j and connected to an interactive D3.js prototype (Figure 1). When a motif is selected, related semantic units, contexts, combinatory relationships, and interpretive evidence are expanded hierarchically, allowing users to examine semantic branching and contextual differences.
Figure 1. Knowledge graph visualization showing motif meanings and contexts
For example, the lotus motif is decomposed into purity, innocence, fertility, and abundance. These meanings are not activated simultaneously in every instance. Instead, the active subset is determined by CulturalContext and by social or institutional groups, such as Buddhist temples, the royal court, and vernacular communities.
Table 2. Sample instance: atomic semantic decomposition of the lotus motif
| Atomic Meaning | Context | UserGroup | EvidenceInstance |
| Purity (PM_023) | Buddhist culture | Buddhist temple | Kumārajīva c. 406, Lotus Sūtra |
| Innocence (PM_011) | Buddhist culture | Buddhist temple | Kim 1992, Buddhist Thought |
| Fertility (PM_004) | Vernacular auspicious context | Royal court | Im 2004, Korean Patterns |
| Abundance (PM_025) | Vernacular auspicious context | Common people | Chung 2012, Minhwa |
| isCombinedWith | PM_023 ↔ PM_011 ↔ PM_004 ↔ PM_025 | — | Decomposition result |
This study contributes to cultural heritage data modeling in four ways. First, it represents Korean motif meanings as atomic semantic units. Second, it explicitly models combinatory relationships among semantic units through isCombinedWith. Third, it connects semantic units with interpretive evidence, supporting transparency and reproducibility. Fourth, it demonstrates a knowledge graph-based environment for examining motif meanings and contextual variation.
The approach may be applied to multilayered interpretation in Korean museum collections and may also help reduce the loss of interpretive plurality when Korean cultural heritage data are represented in overseas institutional collections (Twining 2017). Unlike attribute-centered representation, the proposed model structures relationships among meanings themselves, thereby extending the representational scope of multivocal semantic structures.
The model is not presented as a complete ontology for all Korean motifs, but as a testable structure for representing how motif meanings are separated, connected, and justified. This study remains an initial attempt. The dataset is limited to ten artifacts, and the validation scope is restricted. Future work should expand the dataset, refine the ontology through expert review, and test whether the model remains structurally valid across broader GLAM collections.
This research was supported by the Ministry of Culture, Sports and Tourism of the Republic of Korea under the Culture Service Diffusion Technology Development Program, “Development of AI Technology for Personalized Cultural Heritage Storytelling Based on Users’ Diverse Interests” (Project No. RS-2023-00220195).