Daejeon, July 27–31
The Tarim Basin, located in present-day Xinjiang (PR China), has long been a vital node in trans-Eurasian networks. As a region of encounter and transformation—rather than mere transmission—its importance surged during the heyday of the Silk Road. Among the numerous cultural centers of the Tarim region, the kingdom of Kucha stands out as a multilingual and multiethnic urban polity that actively adapted and transmitted Indian Buddhist culture into China from the second century CE onward. The two most significant remains of the Kingdom of Kucha are textual fragments and murals. Both continue to be the focus of scholarly research and are accessible through two major databases: CEToM(Comprehensive Edition of Tocharian Manuscripts) for the texts and KMIS(Kucha Murals Information System) for the murals.
This paper introduces a joint digital humanities project that combines two hitherto separate domains: the philological study of Tocharian texts, and the art historical study of Buddhist murals from Kucha. Through the development of digital infrastructures and semantic tools to integrate textual and visual corpora, the project showcases an interdisciplinary multimodal approach to fragmented cultural heritage. It promises methodological innovations and new insights not only for Central Asian studies but for digital humanities more generally.
From the second century CE onwards, Buddhism reached the Tarim Basin, carried by monks, merchants, and elite patrons. Kucha became one of the key centers in this cultural and religious transformation. The adoption of Buddhist culture was neither passive nor uniform. Local elites adapted and transformed Indian religious practices, literature, music, and visual art into a distinctive cultural idiom that radiated far beyond the region, influencing courts in China.
Until recently, the philological and visual corpora from Kucha have been studied in disciplinary isolation: philological research concentrated on Tocharian texts, while art historical studies focused on the Buddhist murals of cave complexes like Kizil and Kumtura. Yet, these corpora, are deeply interconnected, and our project seeks to explore the dynamic interplay between text and image. Recent work in digital humanities has increasingly explored multimodal corpora that combine textual and visual sources. Approaches in areas such as Egyptology (e.g. CaemmComhttps://www.archaeologie.hu-berlin.de/de/aknoa/forschung-und-projekte/projekte/caemmcom), and multimodal media research (Hiippala 2021) have demonstrated the analytical potential of linking text and image. However, these projects typically focus on corpora in which images are directly dependent on textual narratives or function primarily as illustrations.
In contrast, the materials from Kucha represent two largely independent yet historically entangled traditions. Our approach therefore addresses a different kind of multimodality, in which text and image must be semantically aligned across fragmentary and only indirectly related corpora.
The corpus—both textual and visual—is vast but highly fragmented. Textual materials include roughly 10,000 manuscript fragments in Tocharian the majority without clear provenance or dating. On the visual side, over 2,300 mural paintings from more than 300 caves have been catalogued, showing narrative and devotional scenes, as well as motifs from daily life.
Although both types of sources offer crucial insight into the economic, social, and religious practices of the period, they have rarely been analyzed in an integrated manner. As a result, substantial interpretative gaps persist: narrative imagery remains unanchored in textual sources; the socio-religious framing of the texts is underexplored; and visual portrayals of ritual, music, and cosmology lack systematic correlation with linguistic data.
To address this lacuna, we propose a multimodal and interdisciplinary research agenda centered on a new digital infrastructure that semantically links textual and visual data from Kucha. The project is structured around the following research questions:
Connecting textual and visual sources enables us not only to interpret each more fully, but also to illuminate the historical lived experience and cultural logic of Buddhism in the Tarim Basin.
Our philological foundation is the CEToM project (Vienna), which provides an edition of all known Tocharian manuscript fragments, enriched with lexical and grammatical data. Dating from the 4th to 10th centuries, the texts include Buddhist writings across various genres as well as secular documents related to monastic life. Complementing CEToM is the Tarim Brahmi project, which investigates the development of the local script and its adaptation to diverse linguistic varieties and literary forms. Both TEI-XML–encoded databases offer rich metadata on each fragment’s physical, textual, palaeographic, and linguistic features. The database currently comprises 11,000 manuscript fragments and 22,000 lexical entries supporting detailed analysis across multiple levels of annotation.
The art historical backbone is the Buddhist Murals of Kucha project (Leipzig), which catalogs wall paintings in a custom database enriched with a detailed taxonomy for iconography, pictorial elements, and ornamentation. This controlled vocabulary, called the Kucha Murals Thesaurus, comprises over 1,200 entries, many linked to precisely annotated polygonal regions within the images. To date, more than 2,300 murals have been cataloged and tagged with over 16,000 annotations, enabling highly granular visual queries.
While each corpus is robust in its own right, our aim is to integrate them through a shared semantic infrastructure.
Our strategy is grounded in linked open data principles and lightweight ontology design. We use SKOS (Simple Knowledge Organization System) to encode both the Kucha Murals Thesaurus and the textual lemma database (Miles & Bechhofer, 2009).
The Kucha Murals Thesaurus is hierarchically structured and comprises three interrelated thesauri: an Iconography Thesaurus (iconographic categories), a Thesaurus of Pictorial Elements (depicted figures, objects, and compositional units), and a Thesaurus of Ornamental Features. Crucially, entries in the iconography branch may themselves reference constitutive pictorial elements. As a result, iconographic concepts are not only situated within a taxonomic hierarchy but may also be analytically composed of multiple image components. This superposition of hierarchical (is-a) and part–whole (has-a) relations precludes a strictly tree-like structure and necessitates dedicated methodological treatment later in the paper.
The textual Lemma Database consists of a hierarchically structured dictionary as well as an TEI-tagged collection of the manuscripts.
The decision to merge the two resources into a single ontology was primarily driven by the requirement for interoperability with existing and future digital projects. In particular, our design closely follows the DiGA project (Digitization of Gandharan Artefacts) in Bochum, whose semantic modeling of thesauri we adopt in order to facilitate subsequent integration into linked open data networks (Autiero 2023). As the DiGA model itself is informed by the semantic representation of the Art & Architecture Thesaurus (AAT; J. Paul Getty Trust 2014), this approach ensures that the resulting data model remains interoperable with a wide range of external resources. Accordingly, the ontology is conceived chiefly as an integrative layer linking the textual and visual corpora and as a foundation for extended semantic analysis, rather than as a tool for direct image or text annotation, which is already implemented within the respective databases.
In addition, both resources are already hierarchically structured, so that their representation in SKOS does not require substantial restructuring. Our objective, however, is not to merge the two thesauri or to flatten their internal hierarchies. Rather, they are first modeled side by side in SKOS, with their respective taxonomic logics preserved through the use of skos:broader and skos:narrower.
For KMIS, we additionally employ the extended SKOS-Thesaurus relations (skos-thes:broaderGeneric and skos-thes:broaderPartitive) in order to distinguish generic hierarchical relations from part–whole relations among iconographic concepts and pictorial elements. In a second step, we introduce a separate layer of cross-references. Conceptual links between CEToM and KMIS are encoded using SKOS mapping properties, allowing for a differentiated treatment of semantic alignment.
Strict equivalence is expressed via skos:exactMatch or skos:closeMatch, for example when Tocharian B śarko “lute” is aligned with the KMIS category for depictions of pear-shaped lutes, or when Tocharian A ynāñmune “veneration” is mapped to the KMIS entry kneeling. Where only partial overlap exists, broader or narrower correspondences are modeled using skos:broadMatch or skos:narrowMatch, while looser associative relations are captured with skos:relatedMatch. This latter case applies, for instance, to the relationship between the two distinct Tocharian A and B terms for “sword” and the pictorial differentiation between long and short swords in the mural corpus.
While the present contribution focuses on linking the two thesauri, we also envisage integrating the underlying research data beyond the thesaurus structures through a dedicated ontology.
Aside from manual linking by our researchers, and simple word matching approaches, we also plan to apply Multiple Correspondence Analysis (MCA) and Cluster Analysis (CA) to map co-occurrence patterns across text and image features. MCA allows for joint projection of textual and visual features in a low-dimensional space, identifying thematic clusters and linking motifs with narrative types. CA enables us to group fragments—textual or visual—based on shared features, revealing tropes or iconographic traditions.
The presentation of the data linking approach will be an important part of the lecture and will be based on case studies (see next sections).
A particularly instructive example is provided by the depiction of the Supāraga narrative. A Kucha mural long interpreted as a scene of dragon-king rescue corresponds closely to a Central Asian recension attested in Old Uyghur and fragmentary Tocharian B sources, rather than to the brief Chinese accounts usually invoked in earlier interpretations. This recension foregrounds Supāraga’s self-sacrifice and subsequent rebirth as a colossal serpent—motifs absent from the Indian Jātakamālā yet rendered with striking clarity in the mural. The resulting textual–visual alignment thus points to a distinct local redactional trajectory and offers new evidence for reassessing the transmission and adaptation of this narrative in Kucha.
Another case is the Sutasoma-Jātaka, whose depictions in Kucha consistently show the man-eater Saudāsa with wings—an iconographic feature unknown in South Asia but recurring in Tocharian translations and regional folklore. Through our multimodal database, we can trace this motif across mural series and manuscript fragments, reconstructing a unique local adaptation of the story.
Kucha’s cultural identity was deeply musical. Philological data reveal native Tocharian terms for string instruments—like śarko (lute)—distinct from their Sanskrit counterparts. Paintings, meanwhile, depict several types of harps and lutes. By aligning lexical items with image annotations, we can reconstruct a rich musical lexicon and material culture and are able to provide a better understanding of the lexical meaning of the words. Our analysis reveals how Iron Age local traditions merged with imported Buddhist elements, producing a distinct artistic-musicological idiom.
Typically, multimodal editions are based on corpora in which images and texts are directly related—the images often depend on textual context for interpretation and rarely serve purposes beyond illustration. In contrast, our corpus is unique in that the visual and textual materials are independently meaningful yet historically and culturally interwoven. The murals function as an independent medium, capable of conveying narratives without direct reliance on textual sources. At the same time, their creation was shaped by textual traditions, just as the texts were received and interpreted within a visual culture shaped by the murals. This reciprocal dynamic is central to our project. By modeling their interaction, we aim to:
Create the first bidirectional, queryable corpus of linked Buddhist images and texts
Enable cross-modal identification (e.g., a scene clarifying a text, or a lemma revealing a narrative motif)
Our approach demonstrates how text-image relationships can yield insights into religious transmission, cultural translation, and identity formation on the Silk Road.
Our results are not only relevant for Central Asian philology and art history, but also for Buddhist Studies, musicology, historical linguistics, and digital heritage preservation. The modular design of our semantic infrastructure allows future integration of other media types, such as archaeological objects or epigraphy.
We plan to make all data FAIR (Findable, Accessible, Interoperable, Reusable) and available on the institutional servers of Vienna and Leipzig, ensuring long-term sustainability and reusability. Our metadata model and controlled vocabularies will be aligned with other DH projects in Buddhist Studies, such as DIGA (Digitization of Gandharan Artefacts), fostering broader data interoperability.
In a time where the humanities are increasingly shaped by digital methods and cross-disciplinary collaboration, this project stands as a model for how fragmentary cultural heritage can be reassembled through multimodal, semantic, and human-centered research. By bringing Buddhist images and texts of the Tarim Basin into conversation, we gain new perspectives not only on the past, but also on the future of the digital humanities.