Scientific methodology
Preface
The Rich Tradition of Chinese Martial Arts
The Chinese martial arts, often known collectively as "Kung Fu," embody a rich tradition comprising over 600 individual styles, extending far beyond their physical practice. These art forms, developed over centuries, reflect not only a variety of movement styles and techniques but also a profound cultural heritage manifested in both tangible and intangible cultural assets.
Kung Fu forms, referred to in Chinese as "Taolu" (套路), are structured sequences of movements. They are characteristic of different schools and styles within the martial arts traditions. These forms are transmitted gesturally from teacher to student and serve not only for physical training and self-defense but also as vehicles for philosophical concepts, aesthetic expression, and historical narratives. They are thus not merely athletic exercises but living repositories of cultural identity and collective memory. This became evident when the Kung Fu style "Tai Chi Quan" was inscribed as the first Chinese martial art on UNESCO's World Heritage list in 2020, alongside intensified efforts to preserve martial arts as intangible cultural heritage.
Evolution and Challenges in Modern Times
In contemporary times, Kung Fu forms are often practiced detached from their original practical applications. Historically, forms were used to enhance martial arts skills and prepare for real combat scenarios. However, today, they are frequently isolated as aesthetic performances or competitive athletic exercises. This decoupling of form and function has significant implications for the transmission and interpretation of cultural heritage.
For example, original technique names with rich references to Chinese mythology and history have been standardized in many styles over the last century into systems of pure posture description. Original names like "The Black Tiger Steals the Heart" or "The Jade Princess Works the shuttle" are now commonly replaced in modern documents with terms like "Stand wide-legged, extend arm sideways." This shift is partly due to the loss of many historical sources. The "Songshan" Shaolin Temple in Henan, which reportedly housed the largest library of martial arts manuscripts at the time, was burned down in 1928. The historical content was only partially reconstructed starting in 1981 through the Shaolin Encyclopedia, based on surviving fragments and oral traditions.
Synchronizing Multimodal Data Sources in Martial Arts Research
It is of critical importance to synchronize and connect various multimodal data sources for this purpose. These include, in particular, image and video material of Kung Fu forms (Taolu, 套路) as well as associated historical written records known as "fist lists" (QuanPu, 拳谱), supplementary structural data such as taxonomies and genealogies of the martial arts styles being studied. This approach not only enables the examination of changes in forms over time but also facilitates comparative analysis between different styles. Through this methodology, conscious and unconscious changes in the forms can be identified and contextually evaluated.
Challenges in Reconstructing Practical Applications
The original, practical applications are only rarely explicitly documented in the preserved images and written records. Therefore, even comprehensive archiving and analysis of historical materials cannot provide a complete picture. Most existing interpretations are of modern origin, and due to their authorship by individuals, the quality and transferability are often inconsistent. For instance, even current practitioners of Chinese martial arts within individual styles cannot agree on whether the contents were originally more focused on wrestling or boxing techniques, or if the forms had any concrete practical applications at all.
Hypothesis
As a basis for the analysis work, the following hypotheses were formulated:
- Kung Fu forms have a practical function (combat application).
- The practical function is encoded within the forms, specifically through the symbols of movements and postures, as well as the meanings of the historical technique names.
- The functions can be clearly assigned to the symbols in the sense of an isomorphism. That is, identical symbolism always corresponds to the same function, but each function can be represented through multiple symbols.
- Kung Fu forms share this symbolism across styles, even though the representation of individual symbols may vary stylistically.
- The movements within a Kung Fu form are interconnected. The practical application always arises from the situation, incorporating preceding and succeeding techniques.
- A Kung Fu form represents a "virtual" fight against a single opponent without context. Neither multiple attackers nor environmental conditions are utilized.
- Kung Fu forms originally served as a means to convey knowledge about combat tactics and methods. Forms thus always have an overarching "meaning" or "teaching content" that goes beyond the conveyance of the techniques they contain. Cultural elements are used to illustrate and facilitate the memory of complex movements in a non-written form.
Experiment
In order to validate the hypotheses, the following experimental and analytical workflows and tools were developed and established.
Database and Information Model
For the archiving, analysis, linking, and semantic annotation of source data, these are stored in a relational database model. The information model includes the following data points as source data:
- Image sequences (postures) of Kung Fu forms (Taolu)
- Listings of technique names, previously referred to as Quan Pu (Faust-Listen) for Kung Fu forms
- Genealogical information on masters and styles
- Form lists (curriculum) for each style
Multimodal Comparative Form Analysis
Different martial arts styles often share the same forms (Taolu). These can be identified by their names and the sequences of movements. For the comparative analysis of individual forms, an image database with an administrative interface was implemented, along with an information model for metadata, which allows the assignment of material to information about the form (Taolu), style lineage, master lineage, source, and other meta-information. The image sequences can then be synchronized across styles based on the shown (physical) postures. This makes it possible to identify deviations between different styles, both in the detailed execution of individual postures (symbolism) and in the overall sequence of the form (e.g., added or removed techniques).
The image sequences of the physical representation of forms can be synchronized with available (written) technique names. The written sources provide crucial clues for identifying similar/different techniques. Deviations between the physical execution of the form and the preserved written sources of the same style can also be detected.
Technique Catalog
Based on the analyzed image and written sources, a technique catalog was created. This has the advantage that, in further analysis, only one entity from the technique catalog needs to be assigned to the postures of a form. Content enrichments such as technique name, posture symbolism, and application interpretation can be automatically loaded. This provides the foundation for an automated validation process of the respective theses regarding technique interpretation.
Description System for the Physical Relationship of Combatants
The project hypotheses include the idea that techniques within a form are "connected," meaning that the interpretation of an application must always consider both preceding and subsequent techniques. Practically, this means that a form represents a (virtual) fight against a single opponent. For example, if the opponent is pushed away by a technique (thus creating distance), it is highly unlikely that a throw follows in the next technique, as this requires maintaining contact with the opponent.
To formally describe the situation within each technique (combat situation), a controlled vocabulary was developed to define the situation in terms of physical relation, distance, and binding. Such formalized vocabularies are already widely used in other martial arts, such as Brazilian Jiu-Jitsu and other ground fighting systems. In addition to the possibility of automatically validating form movements, such a vocabulary also allows communication of sections of a form using formalized language. This is essential, as describing the situation using natural language is often imprecise and therefore prone to errors. See Fighting positions for more details.
Evidence-Based Interpretation of Technique Applications
Each technique included in the catalog is described and interpreted within its own data space. This allows for the formulation of multiple hypotheses regarding the practical application of each technique. These hypotheses are evaluated using a standardized protocol. Currently, 26 questions must be answered for each hypothesis, with the responses automatically integrated into an evaluation system. This enables a qualitative comparison of different hypotheses regarding the practical application of techniques. The final scoring also represents an assessment of how well each hypothesis is supported by evidence and how well it aligns with the pre-formulated project hypotheses. A low score indicates a higher uncertainty of the results.
The formalized description of form sequences and their linkage to centrally stored theses on technique interpretation enables the automated validation of the project's entire dataset concerning the confirmation or refutation of the project hypotheses.
For each form, a color-coded representation indicates whether the assigned technique presents a plausible sequence in relation to preceding and subsequent techniques. When a thesis regarding a technique's application is modified, any potential conflicts in other forms resulting from the change become immediately apparent. This ensures a consistent theory for the practical interpretation of techniques.
Practical-Theoretical Translation of Technical Vocabulary
To utilize the available written sources for analysis, they must be translated. However, the challenge lies in the fact that the lists of technique names (QuanPu) consist of specialized vocabulary, where terms carry specific meanings within the knowledge domain of Chinese martial arts and culture. Processing these texts using traditional natural language translation methods often results in significant errors.
While helpful standard references exist for translating such terms (e.g., the *Falk Dictionary of Chinese Martial Arts*), these are broad in scope and typically do not cover all terms actually used within a specific style. Furthermore, these works lack a practical-theoretical approach, often reducing translations to descriptions of the physical posture associated with a term.
In the proof of concept, a specialized glossary for the translation of QuanPu texts was initiated. The glossary includes not only the translation and meaning of individual characters but also relevant character combinations and annotated synonym groups. Translation of a character or term is based on the following sources:
- Google Translate / DeepL – General, modern text translation.
- Falk Dictionary of Chinese Martial Arts – Domain-specific translation.
- Wiktionary.org – Character origin and historical semantics.
- Brennan Translation – Portal for translating martial arts texts and QuanPu.
- Practical interpretation of the associated technique in the context of all occurrences in forms.
The glossary currently contains approximately 1,300 terms, enabling the translation of new texts from the cultural context of the Praying Mantis style with 80–90% accuracy. Processing texts from other styles, such as Shaolin, Eagle Claw, or Black Tiger, has also shown promising results.
To facilitate the use of the glossary, a web-based translation tool was developed. This tool offers the following features:
- Translation with term length highlighting.
- Marking of term occurrences in the text.
- Display of background information.
- A language library to generate character variants (Mandarin/Cantonese, traditional/simplified).
- A text-to-speech function for Mandarin pronunciation playback.
This tool significantly enhances the ability to translate and interpret martial arts texts within their cultural and practical contexts.
AI supported image analyses
The creation of a practical-theoretical edition of Chinese martial arts requires significant resources. Beyond the research of visual and textual sources and their processing through the aforementioned workflows, the annotation of visual materials represents a substantial workload.
To make the handling of large datasets feasible, the workflows, software tools, and data formats used in the proof of concept were continuously optimized to minimize onboarding time and the effort required for annotations. For this purpose, a specialized tool for image management and annotation was developed.
In addition to manual annotation, initial tests were conducted to explore the use of artificial intelligence (AI) image processing methods. Results indicate that employing such methods can drastically accelerate annotation processes when they are used as decision-support systems (suggestion systems).
As part of the proof of concept, methods for posture recognition (marking joints in 2D images) were utilized to partially automate the recognition of hand and step symbolism. All image annotations are stored to ensure their availability for subsequent analyses.
This approach enhances the efficiency and scalability of annotation processes, enabling deeper and broader investigations into the martial arts forms under study.
Result Validation
Routine structure graph
The consistent use of data collection methods involving controlled vocabularies and formal linking enables the application of extensive and versatile analytical methods to the utilized data corpus. For example, structural analyses of forms can be conducted and visualized in interactive graphs.
The structured application of practical methods within forms can serve as an indicator of a correct interpretation (form harmony). Certain forms, for instance, are composed of multiple subsections, each starting with an initiation of contact with the opponent and concluding with a throw. If a technique is misinterpreted, it may result in a disruption of this structure, providing valuable feedback to maintain consistent theory formation.
These visualizations also make the structure of martial arts interactively accessible, offering a variety of didactic possibilities in teaching. This approach facilitates a deeper understanding of the forms and their practical application while enhancing the educational experience for students and practitioners alike.
Network graphs of techniques
The same metadata can be used to create network graphs of Kung Fu techniques. These graphs visualize which technique combinations are possible in combat. Interestingly, the graph also reveals combination possibilities that are not explicitly demonstrated in any of the analyzed forms but logically emerge through practical-theoretical interpretation.
The network graph thus represents the dimensional space of all possible Kung Fu forms within a single visualization. This serves not only to identify implausible interpretations and erroneous entries but also supports didactic communication and teaching. By providing a comprehensive view of potential combinations, it aids practitioners in exploring the full breadth of technique applications.
Practical Validation
A practical-theoretical interpretation cannot be created without considering and validating practical aspects. The validation process begins during the formulation of hypotheses for technique applications. Researchers require practical experience in technique training, sparring, and free fighting. This experience helps them assess whether a potential application is realistically feasible, including aspects such as biomechanical efficacy, timing, and tactics. Comprehensive training and experience in martial arts are indispensable for this process.
The formulated hypotheses are then discussed and critically evaluated by multiple researchers. Subsequently, the techniques are tested for their efficiency and effectiveness with a training partner. This testing is conducted both as standalone techniques and within the sequence of their respective forms, including preceding and succeeding techniques. Details of execution are refined, and any interpretative errors or misconceptions are identified.
Techniques that successfully complete this validation process are documented through studio recordings of application videos. The practical testing and video creation impose high demands on the physical fitness and martial arts expertise of the participants. A Kung Fu form typically includes around 50 techniques, many of which involve throws. For documenting a single form, multiple takes are required, as execution errors often become apparent only during review.
During the documentation of a form, a researcher may be thrown to the ground hundreds of times. This is feasible only with excellent breakfalling skills, as Kung Fu throws, unlike those in competitive sports such as Judo, are generally optimized not for "training together" but for "fighting against" an opponent. These throws are often designed tactically to make falls difficult to cushion.
Conclusion
All project hypotheses were confirmed within the scope of the proof of concept. A foundational theory of forms and a structural theory of Kung Fu forms were successfully developed. The general theory of forms provides clear rules for interpreting forms across different styles. The specifics of the interpretation are explicitly derived from the symbols used within the sequences of techniques (movements, postures, and written symbols).
Outlook
It is planned to extend the usage of AI technology to be able to compute larger amount of data in less time. The usage of 3D models would solve some problems with clear visualizations of postures in 2D images. Also the foundings of the proof of concept phase are planned to be published within a easy to understand practical format, the Warrior section.
We will not relent in our efforts until the historical legacy of Chinese martial arts is fully documented and understood.