TL;DR
This paper introduces a novel approach to maintaining knowledge representation systems by focusing on "structural enhancements" that modify the underlying domain model. It presents a language for specifying these changes and, crucially, a mechanism for automatically propagating their effects into existing system components (frames and operations), significantly easing the burden of system evolution.
What Problem Does This Paper Address?
The paper tackles the significant challenge of software maintenance, particularly for knowledge representation (KR) systems. While traditional software maintenance often focuses on bug fixing, the authors argue that "enhancement" – adding new functionality or modifying existing behavior – is the dominant and growing maintenance activity. This is especially true for KR systems, which are often developed using exploratory, "cut-and-try" programming techniques. Existing maintenance tools are insufficient for managing the complex evolution of these systems, leaving a critical gap in support for incremental development and system upkeep.
Methods
The core contribution lies in defining and automating "structural enhancements" to the domain model. The authors propose:
- A Domain Model Structure: Based on a conventional frame-based object model, where types have subtypes, and attributes are defined with specific ranges and cardinalities.
- A Language for Structural Enhancements: This language allows for explicit specification of changes to the domain model, categorized into two main types:
- Attribute Structure Changes: Modifying an attribute's domain (type of object it applies to) or range (type of value it can hold), or changing its cardinality (e.g., from optional to required, single-valued to multi-valued).
- Type Structure Changes: Modifying the membership of enumerated sets, such as refining a type into more specific subtypes or combining existing types into a new abstraction.
- Assumption Violation Detection: Each structural enhancement explicitly violates an assumption embedded in the previous domain model. The declarative nature of the model makes these assumptions explicit and detectable.
- Propagation Mechanism: The paper details how these violated assumptions can be used to automatically identify all sites within existing "frames" (data structures) and "operations" (code) that rely on the changed assumption. This allows for the systematic updating of these components to reflect the new domain model.
Datasets
Not discussed in the paper. The paper focuses on a conceptual framework and an example derived from the authors' work on an automated software development environment.
Key Insights and Ideas
- Enhancement is the Dominant Maintenance Activity: The paper strongly argues that system evolution through enhancement, rather than bug correction, is the primary driver of maintenance costs and effort, especially for exploratory development paradigms like those used for KR systems.
- Domain Model as the Key to Maintenance: By making the domain model explicit and declarative, it becomes possible to reason about and automate changes to the system. The structure of the domain model dictates the types of valid enhancements.
- Assumption-Driven Propagation: The insight that structural changes to the domain model violate explicit assumptions is crucial. This allows for a systematic, rule-based approach to updating dependent system components.
- Separation of Concerns for Maintenance: The paper effectively separates the task of modifying the domain model from the task of propagating those changes, leading to a more manageable and automatable process.
Key Results
The paper presents a theoretical framework and an implemented set of tools for performing structural enhancements. While specific quantitative results are not detailed, the key qualitative findings are:
- A complete categorization of structural enhancements for a subset of their domain modeling language.
- A detailed analysis of the propagation effects of these enhancements on both frames and operations.
- The development of tools that can modify the domain model and propagate some of these changes into existing frames. The example provided demonstrates how a type refinement and attribute specialization can be handled.
Limitations and Future Work
- Focus on Structural Enhancements: The paper explicitly states that it focuses on "structural" enhancements (modifying the domain model) and acknowledges that "functional" enhancements (modifying the procedural component, i.e., operations) are more difficult to support.
- Operations Propagation Not Fully Implemented: While the analysis for propagating changes into operations is presented, the tools for this phase were not yet fully implemented at the time of writing. This requires advanced static analysis and type checking capabilities.
- Need for More Declarative Specifications: For functional enhancements, the authors suggest that tools will need to understand the functional aspects of the system, which will likely require more declarative structural specifications for these systems.
- Future Work: The authors anticipate moderate difficulty in implementing the operation propagation tools due to the variability in code and the need for a robust transformation system. They also highlight the importance of a categorization of program modifications, composition rules, and simplification rules for formal program manipulation.
Summary updated