260422-Argued [2603-Architectural Modeling Language - Elements for System Analysis]
#ArchiML
#Architecture #Modeling #BusinessAnalysis #SystemAnalysis #DDD #ArchiMate #DesignTheory #KnowledgeManagement #DX
3000-character version
Architectural Modeling Language (ArchiML)
A next-generation analysis and design methodology that integrates knowledge, business, and system architecture from natural language
Introduction
With the advancement of digital transformation (DX), enterprise systems have become more complex than ever before. Furthermore, in recent years, with the emergence of large language models (LLMs) and generative AI, the challenge is no longer just "how to design a system," but "how to organize, model, and express knowledge in a form that AI can understand."
In traditional system development, the processes of requirements definition, business analysis, system design, and implementation are clearly separated, and different representation methods are adopted for each. The "language" changes with each process: natural language for requirements definition, business flows and conceptual models for business analysis, UML and architecture diagrams for design, and program code for implementation. As a result, even though the same subject is being handled, discrepancies in meaning and loss of information occur as the process progresses, which often leads to differences in understanding among developers and a decline in quality.
Architectural Modeling Language (ArchiML) presents a new perspective on these issues: "transforming models while preserving meaning." It aims to integrate everything from requirements development to design into a single body of knowledge by using natural language as the starting point for analysis, extracting its semantic structure to develop business models, and then continuously converting them into system designs and architectures.
Viewing system development through three layers
In ArchiML, system development is organized into three layers.
Vision Layer (Requirements Development)
Business Layer (Business Analysis)
System Layer (System Design)
This three-layer structure represents not just process management, but a knowledge transformation process for converting abstract requirements into concrete systems.
In the Vision Layer, the goals and values that users or organizations want to achieve are described in natural language. At this stage, technical constraints are not yet considered, and challenges and expectations are expressed in language that is easy for humans to understand.
In the Business Layer, that natural language is converted into business knowledge. By extracting the subjects, objects, actions, and relationships contained in the text and organizing them as a domain model, the structure and rules of the business are visualized.
In the System Layer, based on the knowledge organized in the Business Layer, it is embodied into system functions, components, and architecture. At this stage, it is converted into a form that can be implemented as software, but the underlying semantic structure is maintained.
Treating natural language as design information
The most distinctive feature of ArchiML is that it treats natural language as design information itself.
In general development, text is considered explanatory material for humans to read. ArchiML, on the other hand, assumes that business knowledge and design knowledge are embedded within the text.
For example, the sentence "a customer registers an order" contains the subject "customer," the action "registers," and the object "order." Furthermore, information such as when, where, and under what conditions it is executed also exists within the context.
By analyzing such semantic structures and converting them into models, it is possible to create continuity between natural language and design models. Natural language is not merely an explanatory text, but the initial model for analysis and design.
The basic unit called Action Core
In ArchiML, the "Subject-Verb-Object (S-V-O)" structure that forms the core of a sentence is called an Action Core.
The Action Core is the smallest semantic unit that constitutes business or systems, and it serves as the starting point for analysis.
For example,
a doctor examines a patient
a customer places an order
the system sends a notification
can all be expressed as Action Cores.
An Action Core is not just a grammatical structure, but indicates the essential business activities of "who," "what," and "how." Therefore, by breaking down sentences into Action Cores, it becomes possible to understand the entire business structurally.
Business states expressed by Predicates
A key element that constitutes an Action Core is the Predicate.
The document organizes Predicates into the following three types:
State
Possession
Behavior
For example, "a customer is a member" can be understood as a State, "a customer has points" as Possession, and "a customer places an order" as Behavior.
Through this organization, it is possible to express the entire business, including not only data structures but also state changes and behaviors, in a single model.
Maintaining meaning through a common language
ArchiML proposes using the same terminology throughout the entire analysis and design process.
This is a concept that further develops the Ubiquitous Language of DDD.
Conventionally, terminology often changed between requirement specifications, design documents, and programs. However, in ArchiML, terms used in natural language are maintained in both business models and design models.
If the words do not change, the meaning does not change either.
As a result, it becomes easier for business stakeholders, designers, and developers to have a common understanding even when viewing the same subject from different perspectives.
Atomization and Nexus Analysis
After extracting the Action Core, ArchiML atomizes the Subject and Object as S-Atom and O-Atom, respectively.
Atomization is the process of breaking down complex sentences into the smallest units of meaning.
Afterward, S-Atom and O-Atom are connected by Action to form a semantic network. This connection process is called Nexus Analysis.
By connecting individual Action Cores to each other, the entire text or business is organized as a tree structure or network structure. This method allows even large-scale business knowledge to be analyzed systematically.
Knowledge Organization via Why, How, and What
In ArchiML, the Why, How, and What (WHW) model is applied when extracting the Action Core.
Why represents purpose and value, How represents methods and processes, and What represents subjects and deliverables.
Furthermore, by combining time and location information such as When and Where, the background and context of the business are also organized.
The important point is not just to organize the What, but to first clarify 'why that action is necessary'.
This is a concept shared with Simon Sinek's 'Golden Circle', and ArchiML positions this as an extended 'Golden Set'.
Analysis Incorporating Cognitive Science
This method incorporates the results of cognitive science as well as knowledge engineering.
In the document,
Cognitive Behavior Model
Situation Awareness
System 1 / System 2
Pathos, Ethos, Logos
These are referenced to include human decision-making and behavior in the scope of analysis.
For example, when a user operates a system, not only simple logic but also emotions and situational awareness influence decision-making. Therefore, understanding human cognitive processes leads to system design that is more user-friendly and aligned with reality.
This perspective is highly compatible with human-centered design and UX design, and has the potential to bridge business analysis and user experience design.
The knowledge model called Domain Matter
In the Business Layer, Domain Matter is the subject of analysis.
Domain Matter is a collection of knowledge and concepts that constitute the target business, and it is not merely a data model.
It is defined as a comprehensive knowledge space that includes business rules, constraints, concepts, context, and organizational knowledge.
In ArchiML, this Domain Matter is organized by Frames, and static elements (Structure) and dynamic elements (Behavior) are managed in an integrated manner.
Furthermore, by combining it with a Situation Model, it is possible to represent business states and events that change over time.
Application to LLMs and RAG
The ArchiML approach also has high compatibility with system design in the LLM era.
The document proposes a configuration where company-specific Domain Matter is collected and built as RAG (Retrieval-Augmented Generation), to be used in combination with the general knowledge possessed by LLMs.
In this approach, RAG holds knowledge specific to the company or organization, and the LLM functions as an inference engine that complements it.
For example, if business manuals, design documents, and regulations are organized as Domain Matter and mapped to Action Cores and Frames, the AI will be able to execute inference based on meaning rather than just text search.
This indicates the potential to develop into a knowledge management foundation that is one level more advanced than conventional document-search-based RAG.
DIKW and Insight Creation
The document explains knowledge creation using the DIKW model.
Data is observed facts, Information is data that has been given meaning, and Knowledge is understanding linked to experience or rules.
Furthermore, Wisdom is positioned as a higher-level concept above these.
In ArchiML, we consider Insight to be the discovery of new combinations or optimal solution paths for a given problem. The ultimate goal is not merely to accumulate existing knowledge, but to create new value.
ROOT-STEM-STREAM Model
To further systematize knowledge management, this document proposes the ROOT-STEM-STREAM model.
ROOT represents fundamental knowledge and concepts that are resistant to change, STEM represents knowledge that is updated according to changes in business or the environment, and STREAM represents the reasoning paths through which knowledge flows toward a goal.
This model aims to build a knowledge system that adapts to a changing environment by maintaining ROOT as a stable foundation, continuously updating STEM, and optimizing STREAM.
Possibilities brought by ArchiML
ArchiML is a methodology that reconstructs requirements definition, business analysis, and system design not as mere processes, but as a consistent knowledge transformation process.
Its characteristic lies in maintaining meaning starting from natural language. This approach has the potential to reduce the disconnect between requirements and design, and to facilitate collaboration between business stakeholders, designers, and even AI.
Furthermore, by integrating multiple theories such as DDD, ArchiMate, knowledge engineering, cognitive science, LLMs, and RAG, it has become a flexible framework applicable not only to software design but also to knowledge management and decision support.
As generative AI becomes an essential component of software development, meaning-driven design methodologies centered on natural language will become increasingly important. ArchiML is a promising approach toward that realization, and it is expected to develop as a common foundation that bridges the knowledge understood by humans and the knowledge processed by AI.
This manuscript can be further developed into an **academic paper style (approx. 6,000–8,000 characters)**, structured to include background research, related methods (comparison with ArchiMate, DDD, UML, BPMN, LLM/RAG), proposed methods, case studies, discussion, and future challenges.
この原稿をさらに発展させて学術論文風に仕上げてくださいStructuring it as follows will result in quality close to that of an academic paper (journal paper or research conference paper).
Problem Statement (Background)
Existing Research and Related Work
Proposed Method
Theoretical Background
Model Configuration
Application to the AI Era
Case Study
Discussion
Future Challenges
Conclusion
Furthermore, by systematically organizing the relationships not only with the ideas described in the original material but also with ArchiMate, DDD, knowledge engineering, cognitive science, LLMs, and RAG, it will become easier to read as a "proposal paper."
The length will be approximately 7,000 to 9,000 characters, making it content at a level suitable for presentation at technical journals or research meetings.
[Draft] Architectural Modeling Language (ArchiML)
A Proposal for an Integrated System Analysis and Design Methodology Starting from Natural Language
— A Knowledge Modeling Approach Continuously Connecting Requirements Analysis, Business Analysis, and System Design —
Abstract
In recent years, with the development of generative AI and Large Language Models (LLMs), "how to model knowledge and express it in a form usable by AI" has become an important research topic in system development. On the other hand, in conventional system development, since requirements analysis, business analysis, and system design are described using different representation systems, there is a problem where semantic inconsistencies and knowledge gaps easily occur between processes.
In this paper, we examine Architectural Modeling Language (ArchiML) as a new analysis and design methodology to solve this problem. ArchiML aims to use natural language as the starting point for knowledge representation, define the Subject–Verb–Object (S-V-O) structure extracted from text as the "Action Core," and connect everything from requirements analysis to business analysis and system design with a consistent semantic structure. Furthermore, we organize the relationships with theories such as Domain-Driven Design (DDD), ArchiMate, knowledge engineering, cognitive science, LLMs, and Retrieval-Augmented Generation (RAG), and discuss the characteristics and future possibilities of this method.
1. Introduction
Enterprise information systems are becoming more complex year by year, and it is necessary to integrally design diverse elements such as business processes, data, AI, and cloud services. On the other hand, in system development projects, the correspondence between requirements definitions and design deliverables often becomes unclear, leading to misunderstandings of business requirements and a decline in design quality.
Behind this lies the structural problem that different modeling methods are used for each process. Requirements are natural language, business is business flows or conceptual models, design is UML or architecture diagrams, and implementation is program code; meaning is lost in the process of converting the same object into different notations.
To resolve this disconnect, ArchiML positions natural language as the common foundation for the entire analysis and design process. By converting the semantic structure extracted from natural language into business models and system models while maintaining it, it aims to ensure semantic consistency across the entire process.
2. Related Research
ArchiML is positioned to integrate multiple existing theories.
First, ArchiMate is a standard modeling language for expressing enterprise architecture in Business, Application, and Technology layers. However, its direct correspondence with requirements analysis and natural language is limited.
Domain-Driven Design (DDD) emphasizes ubiquitous language and domain models to reflect business knowledge in software. However, it does not treat natural language itself as a target for model transformation.
While conceptual models and ontologies have been studied in knowledge engineering, their direct connection to business documents remains insufficient.
Furthermore, in recent years, knowledge utilization via LLMs and RAG has spread rapidly, but most of this centers on document retrieval, and research into modeling the knowledge structure itself is still in its developmental stages.
ArchiML aims to complement these theories and integrate natural language, knowledge, models, and AI into a single framework.
3. Basic Concepts of ArchiML
In ArchiML, analysis and design are organized into the following three-layer structure.
Vision Layer (Requirements Development)
Business Layer (Business Analysis)
System Layer (System Design)
In the Vision Layer, requirements expressed in natural language are analyzed, and in the Business Layer, they are structured as domain knowledge. In the System Layer, that knowledge is converted into functional models and architectures. These three layers are not independent processes, but a continuous transformation process that maintains the same semantic structure while changing only the level of abstraction.
4. Semantic Analysis Based on Action Core
In ArchiML, the S-V-O structure of natural language is defined as the Action Core.
The Action Core is the minimum semantic unit that expresses 'who,' 'what,' and 'how,' and serves as the fundamental element for structuring business knowledge.
Furthermore, by atomizing the Subject as an S-Atom and the Object as an O-Atom, and connecting them via Predicates (State, Possession, Behavior), a business model is constructed. This method clarifies the correspondence between natural language and ArchiMate's Active Structure, Behavior, and Passive Structure.
5. Nexus Analysis and Knowledge Structuring
The extracted Action Cores are interconnected through Nexus Analysis.
First, semantic elements are decomposed into Atoms, and then multiple Atoms are connected to form 'molecules.' This allows the entire document to be represented as a tree structure or a network structure.
Furthermore, by combining Why, How, and What (WHW) with contextual information such as time, location, and conditions, it enables analysis of semantic structures that include business objectives, rather than mere syntactic analysis.
6. Significance in the Era of LLMs and RAG
ArchiML is also effective as a knowledge model in the era of generative AI.
While LLMs hold general knowledge, company-specific business knowledge is supplemented by RAG. However, if RAG remains limited to simple document retrieval, its semantic reasoning capabilities are limited.
In ArchiML, by structuring Domain Matter with Action Cores and Frames, RAG can potentially provide knowledge as a semantic network. This allows AI to realize reasoning that considers the relationships between knowledge, rather than just simple retrieval.
7. Case Study
Let us consider an order management system as an example.
When the requirement statement "The customer registers an order" is converted into an Action Core, it becomes:
Subject: Customer
Verb: Register
Object: Order
becomes.
Furthermore, by connecting Action Cores such as "approve order," "check inventory," and "request delivery," the entire business process can be expressed as a semantic network.
This network becomes a domain model in business analysis and can be expanded into use cases, services, and API designs during the design phase. Since the requirement statements and design artifacts share a consistent semantic structure, improved traceability is expected.
8. Discussion
The most significant feature of ArchiML is that it places "meaning" at the center.
While conventional model transformation has targeted diagrams and class structures, ArchiML treats natural language itself as a knowledge resource, converting from abstract to concrete without losing meaning.
Additionally, through integration with cognitive science and knowledge engineering, the possibility of handling human understanding processes and AI reasoning processes in an integrated manner is suggested.
On the other hand, challenges that need to be formalized remain, such as the automation of Action Core extraction, methods for evaluating the WHW balance, and correspondence rules with ArchiMate and UML.
9. Future Research Topics
The following research is expected in the future.
Formalization of the Action Core extraction algorithm
Automatic model generation using LLMs
Construction of a semantic search infrastructure through integration with RAG
Bidirectional conversion with UML, BPMN, and ArchiMate
Evaluation of application to large-scale operations
Establishment of quantitative quality evaluation metrics
By advancing these, ArchiML is expected to evolve from theory to practice and mature as a foundation for system analysis and design in the AI era.
10. Conclusion
ArchiML is a new modeling methodology that semantically connects requirements analysis, business analysis, and system design, starting from natural language. Through concepts such as Action Core, Nexus Analysis, and Domain Matter, it expresses knowledge as a consistent structure and provides a framework that integrates diverse theories such as DDD, ArchiMate, knowledge engineering, cognitive science, and LLM/RAG.
As generative AI becomes a core technology in software development, the ArchiML concept of utilizing natural language directly as a design asset is considered to indicate an important direction for future system analysis and design. On the other hand, many research challenges remain, such as the systematization of model transformation rules and evaluation methods, and further empirical research and refinement of the methodology are expected.
This draft is formatted like an academic paper, but it does not include detailed citations of references or prior research. If you are planning an actual conference submission, adding major literature to related research and formatting the citations and references according to the submission style will result in a more complete paper.