The Facets of Intangible Heritage in Southern Chinese Martial Arts: Applying a Knowledge-driven Cultural Contact Detection Approach

China has a long history of martial arts traditions. Originated as methods of combat and self-defense, these practices have, since at least the Zhou dynasty (ca. 10th Century BC), acquired a humanistic dimension, thus becoming instruments for “whole-person education” [9]. Chinese martial arts nowadays are practiced in countless groups at various organizational levels, ranging from “families” to “schools” and “sects,” each with their own philosophies, concepts, techniques, and training systems [34]. With these groups as the hives of cultural transformation, Hong Kong has acted as a vibrant center for Southern Chinese martial arts throughout the 20th century due to its role first as a major port and trading center, then as a safe haven for refugees across China [9]. There is a tremendous wealth of martial arts resources and elite practitioners who represent the highest authority for their respective lineages or styles. However, these treasured cultural practices, made internationally famous by the movie industry, are being endangered by factors such as rapid urban development, population growth, cultural transformation, and the aging of the masters.

The HKMALA originated as a heritage project in 2012 to archive Hong Kong’s rich and diverse kung fu styles with traditions [9]. The HKMALA project has been a longitudinal research collaboration between the International Guoshu Association, the Laboratory for Experimental Museology (eM+) at EPFL, and the City University of Hong Kong. It encompasses a comprehensive analysis of digital strategy, including motion capture, motion-over-time analytics, 3D reconstruction, high-speed and panoramic video shooting, and photographic archiving of the masters involved. Its datasets represent one of the world’s largest motion archives for intangible cultural heritage, spanning over 130 sets of empty-hand and weapon sequences, or taolu, covering 19 styles with performances by 33 elite practitioners. The archive is accompanied by extensive contextual documentation, such as ritual descriptions and multimodal digital records of physical objects such as weapons and training tools.

The current, implicit, logical organization of resources in HKMALA reflects how they are assembled for exhibitions around a designated theme. Extracting open-access learning and study resources from the archive and exhibitions will enhance their role in the preservation, re-activation and revitalisation of traditional martial arts [34], as well as the basis for future educational programs [24]. However, realizing this requires a knowledge organization system that has both the flexibility to accommodate content in the diverse South Chinese martial arts spectrum, and awareness that said content is a carrier of ICH elements such as tradition, style, and influence. Part of this study was to survey the ontological landscape in search for what ontologies model, or could support a model of, martial arts in general, filling any gap over the course by ourselves.

In contrast to tangible material practices naturally embedding cultural identities in physical objects, ICH is defined through its reliance on tacit and embodied practices inextricably linked to people. These are usually bodily, performative, expressive, symbolic, rule-based, non-instrumental, and subject to a dynamic process [25, 47]. One of the critical channels where cultural expressions occur is kinesthesia—the movement. Through the perception of the body, physical qualities acquire significance in the presentation of culture. Relevant studies have been carried out in the field of dance, arguably the best-known performative discipline that values motion, perception, and interpretation, which either reveal the source of inspiration for a given technique or provide a means to facilitate its transmission. Similar considerations also hold in martial arts: Examples include techniques developed through the alliances with Japanese warriors to counter piracy incursions, or the osmosis between military and civilian martial arts, causing the latter to employ makeshift weapons or simulate them with hand techniques due to government prohibition to bear arms.

While there was no expectation that an existing formal data model would cover the nuances of Southern Chinese martial arts, still no validated ontological framework could be found to even only represent martial arts under general or limited aspects: The only online resource that was found is UMAO, an attempted ontology that does not seem backed up by experimentation or publications and, in the author’s stated intentions, merely constitutes a modeling exercise.²

A litmus test further attesting to such lack of representational coverage is the scrutiny of how martial arts concepts are represented in general-purpose knowledge bases: To richly authored English and Chinese Wikipedia articles on the Fujian White Crane style³\(^{,}\)⁴ corresponds a Wikidata representation that hardly covers anything more than the style’s geographical origin.⁵ Wikis are indeed among the key manifestations of interest in structured knowledge on martial arts, as corroborated by the Wiktenauer one on Western martial arts curated by the Historical European Martial Arts Alliance [46], whose information wealth is also not backed by a knowledge organization based on formal semantics.

We acknowledge, however, that the solution to part of the modeling problems in a transversal domain are to be found in how connected domains are modeled. In that sense, it is useful to look at models such as the Muninn Military Ontology [45] for what concerns its representations of weaponry or of regimental hierarchy, if anything, for alignment purposes. Likewise, foundational ontologies are able to provide the basic framework whereupon to lay descriptive models of the physical, social, and normative ramifications of the domain [40].

Following a similar reasoning, it makes sense to look at ontologies that model kinesthetic, movement-centered arts. One way of building these is by information-intensive knowledge mapping, in which performative movements are described by a dense sequence of micro-movements [31]. In recent years, there have been several attempts to capture and model movement expressions in performing arts. For example, Raheb et al. proposed an ontology that allows users to annotate classical ballet videos with a domain-specific vocabulary implementing the functionality of “motion concept search,” through which a user searches for specific movements performed within each video [15]. Kannan et al. focus on modeling dance video objects at multiple levels of detail by implementing the Dance Video Semantic Model (DVSM) [23]: This vocabulary attempts to combine and interconnect the underlying semantics between songs and movements. Chaudhry et al. propose an ontology to classify complex movements for folk Malaysian dances [10]. Finally, Lagrue et al. propose an ontology and web-based dance video annotation system for representing the semantics of dance videos at different granularity levels [27]. The above models are all very specific to the dance domain, providing only basic concepts for a martial arts ontology to align to and not always with a formalization that is openly available online. However, the use case of annotating performance media in a nonlinear, multilayered fashion has been retooled as part of the knowledge organization of HKMALA media, which falls beyond the remit of cultural contact detection and is described separately [35].

These few examples show a common problem being faced: the simultaneous capture, through an ontological model, of a kinesthetic aspect such as movement and a more static aspect such as intangible cultural tradition. On the one hand, the ontology must be able to express the modeling of kinesthetic processes. On the other hand, it should also make the cultural aspect visible, which such movement is intended to express [15]. On that, we argue that the cultural aspects of movement, style, technique, or influence do not need to be rendered explicitly in the ontology, rather, they can and should emerge from the interplay of the domain ontology with a cultural one, whereupon a computational method is applied. In other words, cultural aspects should be a connotation, rather than a denotation, of an ontology of martial arts. This allows us to still avail ourselves of cultural heritage models [39], such as the widespread CIDOC-CRM [13] and the emerging ArCo, the latter being the first to offer the basis for an intangible cultural framework [8]; only, the usage of these ontologies, and by extension the classification of domain features as cultural traits, is deferred to the application of the computational model, which in our case are the inference rules for detecting potential cultural contact. The knowledge representation device through which martial arts become evidence of cultural interaction is the CuCoO (Cultural Contact Ontology) ontology. CuCoO establishes a set of concepts and relationships that allows the description of contacts between cultures and their identification in the evidence. It enables the modeling of the cultural context itself, as well as the establishment of potential cross-cultural connections between the entities [30].

The question of cultural contact has been rooted in literature since the early attempts at defining it over the past three decades [11]. While disciplines such as sociology and anthropology offer the opportunity to explore the context and impact of this phenomenon in living communities, others like archaeology have to rely on the close examination of the material evidence to identify potential cultural traits that can be considered as evidence for cultural transmission. Although some may consider these as two different forms of exploring and understanding the potential outcomes of cultural contact—whether its effects are on people or on things—they actually rely on the same idea, which is to identify cultural traits that are exercised, perceived, and assimilated and can then be manifested in the material realm or not.

It should be borne in mind that cultural contact is not synonymous with influence: More often than not, this is the case and the evidence of it can be found in the characteristics of material manifestations of those cultures, or of their habits. Anything from the curvature of a weapon to the way an attack technique is named can bear such evidence. However, just as not every context of contact is peaceful and may have been enforced through, e.g., conquest, so does not every process cause the incorporation of traits from one culture into another. These latter ones are cases of impermeable cultural contact [41] and are characterized by the lack of evidence in cultural heritage, which is therefore to be sought in the historical discipline. Typical examples include the foundation of a colony or the resistance to modernizing trends, an example of which is found in the way martial arts in Zhangzhou developed, hardly ever amalgamating with more recent or international arts since under the Qing rule. In addition, permeable contacts may develop traits that appear in opposition to, rather than acceptance of, the culture that comes in contact with another, of which the development of unarmed martial arts, following the prohibition of weapons by conquering rulers, is an eminent example.

The previous attempt mentioned earlier aimed at identifying cultural traits that would have been exchanged and manifested in the material evidence of Southern Roman Spain. Using the same principle, this study explores the potential of an inference-based computational model to pinpoint evidence of interaction between cultures manifested in embodied knowledge and practices, rather than in material artifacts. In addition, we attempt to assess whether the factors that made sense in that study, e.g., (im)permeability and contexts of trade or conquest, suffice or should be expanded to accommodate the complexity of Southern Chinese martial arts history and, potentially, other historical disciplines.

Identifying which salient features of a setting are traits of a particular cultural identity or phenomenon is inherently a classification problem. As such, it naturally lends itself to supervised machine learning approaches that, however, are built upon models trained on a significant portion of curated gold-standard data [16] that, at the time of this article, are not openly available. Our work veers towards bootstrapping the construction of this ground truth as a byproduct.

Our stated intent when designing a martial arts ontology was not to constrain it to the use case of cultural contact detection, but to ensure its usefulness in multiple potential use cases, such as finding common and distinguishing traits between styles or publishing sports competition statistics as open data. However, ontology development methodologies that advocate modularization do allow for rationales that partition the landscape on the basis of the role that individual modules play in a driving use case [14]. By drawing inspiration from the NeOn Methodology, one of the most recently established scenario-based frameworks with reuse and modularity in mind [43], we based our knowledge engineering effort on a series of competency questions, of which the salient ones are reported below:

It is worth reminding how none of the above CQs are predicated upon explicit cultural indicators, though related considerations can be drawn from most if not all of them. This is a precise design choice: A user of this ontology must be given the tools to single out and highlight the entities with a potential to contribute to the use case at hand if needed.

Based on these design considerations, the martial arts ontology was developed as an ontology network that satisfies these requirements:

From a modularity perspective, the ability to single out relevant groups of terms is similar to the principle of ensuring that the subsumption hierarchy of a single class be as self-contained as possible, which is intended to be applied to the modularization of an existing ontology [12]. For instance, classes that subsume WeaponType, say, MeleeWeaponType, should belong in the same module, however, if parts of the human anatomy are considered akin to weapons in an offensive capacity (see, e.g., CQ3), then they should also belong together. Also, properties that connect classes (e.g., declarations of domain or range) should belong either in the same ontology as both classes or in ontologies that depend on one another. By following these principles, we obtain an organization of the Martial Arts Ontology into modules that happen to also play defined roles in the cultural categorization of an art by different facets, whether tangible or not.

Based on the above competency questions, requirements, and design principles, the martial arts ontology network was structured along three modules: the kinesthetic, stylistic, and social modules. These, in turn, reflect three possible dimensions and lenses for highlighting cultural phenomena.

To fulfill requirement #4 while taking #2 into account, we chose a foundational ontology as the base model to relate our key concepts to, namely, DOLCE UltraLite (DUL) [17]. This upper ontology provides fundamental notions such as agents, qualities, norms, and methods, which proved fitting for the organization of the martial arts terminology. We can therefore defer any mappings to cultural heritage models, such as CIDOC-CRM and ArCo, to the stage where the characterization of martial arts entities as cultural objects occurs.

The remainder of this article assumes the prefix mappings defined in Table 1.

The starting documentation for this modeling effort is the corpus of panel texts and media captions for the exhibitions that were established for HKMALA through the years, as well as multimedia subtitles of interviews to masters and their technique explanations. The base terminology was lifted from the corpus by project members and engineered as classes and properties. The resulting key concepts are synthetically shown in Figure 1, with nodes for classes and edges for properties relating them through either domain and range or class restrictions.⁷ The figure reflects the organization into modules denoting the principal cultural dimensions, with classes serving as contact points between them.

Each module of the Martial Arts Ontology was submitted to the OOPS! service for pitfall detection [37]. While no critical issues with the ontologies were detected on the outset, some that are deemed important on the OOPS! scale, such as the explicit declaration of disjoint siblings, were subsequently addressed. The only exception is on the explicit indication of domain and range definitions for each property, primarily due to some of them whose domain was deemed not entirely predictable, unless one commits to an explicit fallback to the owl:Thing top concept.

To quantitatively measure the Martial Arts Ontology, we rely upon the criteria offered by the OntoMetrics framework [28]. The salient statistics from the OntoMetrics report on information richness are summarized in Table 2.

From the contrast between the two richness figures, it can be observed that the model relies significantly more on the class hierarchy to convey its information than on property structures. This is primarily due to the fact that most properties exist for the primary purpose of providing one single connection between instances of two classes. A property hierarchy is currently only present to group properties of styles and techniques and is not foreseen for others in future developments, though it would boost relationship richness. Class inheritance richness, however, contributes to justifying why each class exists as an information carrier in its own right, which also explains their low tangledness score, i.e., there is little multi-class inheritance. Because we directly reuse foundational terms from DOLCE, and only use equivalence links to other third-party terms, the equivalence ratio is naturally low. As the ontologies are lightly axiomatized with universal or existential restrictions, as well as disjointness axioms for sibling classes, the axiom/class ratio is noticeably high. Last, the fact that subclassing is partly delegated to potential extensions of the ontology—we do not intend, for instance, to enumerate all the body part types that may serve as armament—explains why the dispersion measure, i.e., the fan-outness of the ontology, approximates one.

We have likewise verified that the Martial Arts Ontology and its closures with DOLCE UltraLite and with the HKMALA knowledge graph all return as OWL2-consistent, as of the time of writing, when run through the HermiT description logic reasoner [18].

According to ontology development methodologies based on competency questions, prototypical queries that encode these questions can be used as unit tests for the ontology itself [6]. In the following, we list prototypical SPARQL queries able to answer the competency questions illustrated in Section 3.

CQ1. What techniques does a martial art style use? This has a straight answer in SPARQL for all known styles.

CQ2. What type of weapons are employed in armed martial arts? Because the domain of employsArmament is dul:Method, therefore subjects can be techniques or styles, the solution to CQ2 concatenates the property to an optional inverse of belongsInSystem to ensure the propagation of weapons used only in some techniques.

CQ3. Which martial arts feature a hand technique that simulates wielding a weapon? CQ3 makes the reasonable assumption that a technique does not use a human body part to simulate another, therefore ?weapontype will be bound to types of actual weapons.

CQ4. Which styles, techniques, or forms are symbolically represented by a being of nature? CQ4 makes two assumptions: (a) natural beings are reused from, and described in, an external dataset like Wikidata; (b) that dataset contains the knowledge that the object is a natural being, so it needs to be queried in federation to answer the competency question. We also do not explicitly indicate the SymbolicObject type, which is intended to be inferred. Also recall that styles, techniques, and forms are all subclasses of dul:Method.

CQ5. In what group types (families, clans, schools, sports federations) is a certain martial art being taught?

CQ6. What aspects of a martial art style must be mastered to acquire a certain grade? Here, by “aspect” it is meant a form or a technique, which form the values to be bound to ?requirement.

CQ7. What belt color in a grading system corresponds to a grade in another system for the same martial art? A belt color is a degree. Here, we assume that equivalent degrees in different grading systems require mastery of the same syllabus.

CuCoO¹² is a monolithic (non-modular) ontology whose main goal is to identify entities that manifest cultural traits that have been related to two or more different cultural identities. For example, hand-to-hand combat techniques in Southern Fujian (an area primarily populated by Minnan and Hakka people) during the mid-Qing period testify to a reaction against the state-imposed proscription of arms by a rule of Manchu ethnicity. In another example, single-edged two-handed swords show features inherited from the alliances with Japanese ronin to counter piracy, whereas the staff-fighting techniques that emerged afterwards are the result of a contact with Japanese culture that is of an antagonistic nature, once Wokou (Japanese pirates) incursions intensified. For a better understanding of the case study implementation, we provide a brief description of the key concepts of the ontology.

cucoo:CulturalIdentity is an upper class that signifies a “self-conscious association of a group or person with a culturally-constructed group identity.”¹³ The Chinese Minnan or Hakka, or the Muromachi-period Japanese, are example instances of a primarily ethnic subclass of this class, but more subclasses, grounded on, e.g., religious or gender identity, are contemplated.¹⁴

Another upper class, cucoo:CulturalContactFeature, describes the phenomenology of a cultural interaction in three forms, each encapsulated by a subclass as follows:

cucoo:CulturalContactContext refers to “the context in which the situation of cultural contact takes place” (for example, in a context of migration). It is intended to be subsumed by a controlled vocabulary of possible phenomenological classes, of which the original study proposed those relevant for its remit, such as Conquest, Trade, War, or Migration.

cucoo:CulturalContactAspect refers to “the main feature that defines the cultural contact phenomenon,” thus assigning a connotation to it. Only the Linguistic and Iconographic aspects were contemplated by the study that the ontology initially aimed at, however, other taxonomies can be attached by subsuming the main class.

cucoo:CulturalContactTrait allows the identification of cultural traits in the evidence or in the actors of the cultural phenomenon. Once again, it should be regarded as an extension point through which the base model reaches into the cultural domain. Some of these traits, such as material, measure, carving, appearance, weight, iconography, or text, were hardcoded into the original ontology; however, another controlled vocabulary could be associated to it, whether the membership of classes is asserted or inferred.

Cultural contact traits are “characteristics of human action which are acquired and transmitted by any kind of communication in a situation of cultural contact.” The object property cucoo:isAssociatedWith relates a trait (e.g., a type of material or the grip over a weapon) with a cultural identity (e.g., Roman, Phoenician, Minnan, or Feudal Japanese) without restrictions, whereupon the former is classified as a cultural trait. For example, a Latin inscription in a sculpture can be associated with the Roman cultural identity because it is written in Latin, in the same way as it could be related to the Iberian cultural identity if it were to record an Iberian name in Latin script. In another example that will run through this section, the Shuang Shou Dao saber could be associated to either Feudal Japanese or Anglo-Saxon cultural identities, because two-handed swords belong in both (but not natively in South China), however, the fact that it is single-edged helps resolve the ambiguity in favor of the association to Feudal Japan.

By applying inference rules to CuCoO and to a domain dataset, it is possible to identify cultural contact traits of diverse provenance, each trait being one entity in the dataset. This is done by connecting an instance of cucoo:CulturalTrait to one of cucoo:CulturalIdentity (or of a subclass of it) through the property cucoo:isAssociatedWith. In this way, the objects that present cultural traits related to different cultural identities are considered in CuCoO as evidence for cultural contact.

One potential pitfall of using an ontology-based approach to detect culturally relevant aspects in data is that, when reusing arbitrary domain ontologies to describe the context, pinpointing these aspects in those ontologies would be both restrictive and unsustainable. While it is plausible that features identified in Roman coins, such as the clothing worn and weaponry wielded by those depicted on them, could be culturally relevant in the martial arts domain as well, making it explicit may give rise to knowledge representation issues, such as saturating the assignment of types to entities. CuCoO, the cultural contact model we adopted, is structured in such a way, therefore, we took it upon ourselves to generalize it so it would serve the needs of, among others, the Southern Chinese martial arts domain.

Because the present study required the association of contexts, aspects, and traits of cultural interaction beyond those originally defined, and in a way that cannot always be predicted in an ontology engineering phase, we aimed at improving the versatility of CuCoO while preserving its fundamental framework. Taking advantage of the existing modularization techniques mentioned earlier [12, 14], we created a separate, modular derivative of CuCoO. The resulting version used in this study is so partitioned:

Only the Core module of the reengineered CuCoO is necessary for the purpose of this study. Nevertheless, it is also useful to keep track of cultural traits to verify which ones correspond to features detected by the inference-based computational model described in the following. The alignments can be utilized if one wishes to, for example, materialize the inference as a cultural heritage dataset based upon CIDOC-CRM.

To detect potential characteristics of cultural contact found in the evidence of our domain dataset, we formalize inference rules, run them over the dataset, and materialize the new statements being generated by their execution.

Semantic Web technologies offer several means of expressing inference rules upon data in RDF format, including ontology rules in the SWRL language, SPIN rules, and CONSTRUCT queries in plain SPARQL language. To avoid language overheads for the comprehension of our computational model, we chose the last form, though all of them are for the most part interchangeable [5]. Our use case implementation, informed by the tools available for the construction of rule-based inferential models (cf. Section 2.2) and with generalizability in mind, consists of the following steps:

Writing the inference rules as indicated in step (1) allows them to be written with the sole knowledge of the domain model for the rule antecedent (i.e., the WHERE clause in the case of SPARQL CONSTRUCT) and of the cultural contact model for the rule consequent (i.e., the graph patterns in the actual CONSTRUCT clause). The data table in step (2) is as simple as a SPARQL VALUES clause that can be inserted into the rule to give it the desired scope.

We demonstrate the above through examples extracted from the six inference rules used for this study. Listing 8 shows a rule that generates CuCoO cultural associations out of martial art styles, possibly through techniques of them (line 11, ^:belongsInSystem?), that employ a weapon of some type that must be handled (line 13, ^:handles) in a way that has certain characteristics (the [ ?feature ?featureValue ] pair). At this stage, it is not yet known which features, such as grip or swing, will actually be considered: These features, as well as their values and the styles returned in output by the rule execution, will all be bound to entities from the HKMALA knowledge graph.

What the weapon types, cultural identities, or handling characteristics are should not be hardcoded into the rule, however, the available knowledge in that regard must be taken into account. This is the reason why, in Listing 8, there is (line 15) {%feature_table%} as a placeholder that can be substituted at runtime with a VALUES clause, which encodes the table of possible cultural associations. In our example, the following data table provides the scoping: It indicates that swords that are held with two hands are a trait of the Japanese Muromachi culture if single-edged (line 2) and of the Anglo-Saxon cultures if double-edged (line 3).¹⁵

At this stage, one does not yet know what such weapons and styles are or if they even exist. Applying the rule in Listing 8, scoped using the data table of Listing 9, attempts to detect cultural traits and to trace them back to the evidence that warrants them. At the time of writing, their execution over the HKMALA knowledge graph generates the following RDF statements:

The snippet in Listing 10 tells us that the martial arts of the Minnan people have two possible associations to the Muromachi-period Japanese culture (wd:Q334845, lines 9 and 15): one through the Shuang Shou Dao (a single-edged bladed weapon) and one through the typically Japanese custom of holding swords with both hands. Verification with HKMALA text documentation confirms these claims¹⁶ with the explanation that inhabitants of the Southern Chinese coast partnered with Japanese ronin warriors to counter regular pirate incursions. We also note that an object property, withGrip, is used in the data as the predicate for TwoHandedGrip, therefore the rule in Listing 8 automatically identifies that property (line 14) as the cultural contact trait to be associated to Muromachi Japan.

In another example, we create a rule that states that, if a martial arts style (e.g., Southern Praying Mantis or Fujian White Crane) or a single technique (e.g., Lock hand) are associated to a symbol, then this may be indicating cultural influence.

In Listing 11, this time, we fix (lines 9–12) the imagery-related properties we are interested in and leave to the data table only the responsibility of associating symbols to cultures. The {%feature_table%} is therefore substituted with the following, which has been simplified for example’s sake:

In Listing 12, Wikidata entity wd:Q25365 is the crane, and the corresponding cultural identity is a spatio-temporal framing tailored around the Zhangzhou province inhabitants during the Qing rule (modeled separately), arguably the first people known to have used the crane imagery in their styles, which has since trickled down to other martial arts.