Knowledge Modeling and Reusability in ExClaim

Liviu BadeaIJCAI 1999

Abstract

ExClaim is a hybrid programming language designed to bridge the gap between Description Logics and Logic Programming for building reusable Knowledge-Based Systems (KBS). By implementing a three-level meta-architecture (Domain, Inference, and Task levels), ExClaim decouples domain-specific knowledge from problem-solving logic. This paper explores the architectural design of ExClaim, focusing on its reflection mechanisms and its ability to maintain consistency across non-monotonic and non-deterministic reasoning layers.

Introduction

Building Knowledge-Based Systems (KBS)—software designed to solve complex, data-intensive problems like medical diagnosis or resource scheduling—has historically been a challenge. Developers often faced a "Goldilocks" problem: expert system "shells" were too rigid and tied to specific data formats; general software tools lacked the intelligence to handle complex ontologies; and pure logic provers were often computationally intractable for real-world applications.

ExClaim, developed by Liviu Badea, addresses these issues by providing a "knowledge-level" development tool. The system is expressive enough for complex rules, capable of automated consistency checking, and modular enough to allow problem-solving methods (like scheduling algorithms) to be reused across different domains without rewriting core logic.

Methodology: The Three-Level Meta-Architecture

ExClaim achieves its goals through a three-level meta-architecture based on the KADS methodology. This structure separates the "what" of the domain from the "how" of the reasoning process.

The three-level meta-architecture of ExClaim showing the Task (T), Inference (I), and Domain (D/DL) levels.

1. The Domain Level (The "What")

The Domain Level uses Description Logics (DL) to define the system's ontology. It establishes the vocabulary and relationships between concepts (e.g., defining a "Professor" as a subtype of "Lecturer"). This level provides automated services such as consistency checking and classification.

Semantic interpretations of concepts and relations used at the Domain Level in ExClaim.

2. The Inference Level (The "How")

This level uses Logic Programming (similar to Prolog) to define primitive problem-solving steps. Crucially, these inferences do not interact with domain data directly. Instead, they operate on "Inference Roles"—abstract placeholders that act as a buffer between the logic and the data.

3. The Task Level (The Control)

The Task Level manages the high-level execution flow, determining which inferences to run and in what order to achieve a specific goal.

Reflection Rules

The integration between these levels is managed by Reflection Rules. When an inference requires data, an "upward reflection" fetches and translates domain objects into inference roles. Conversely, "downward reflection" pushes reasoning results back into the domain knowledge base.

Key Insights: Hybrid Reasoning and Reusability

ExClaim's primary innovation lies in its decoupling of logic from domain knowledge. By using abstract roles, the inference logic becomes domain-independent. For example, a resource allocation algorithm can be developed once and then applied to university scheduling or hospital staffing simply by swapping the underlying domain ontology.

Furthermore, ExClaim supports backtrackable knowledge. If a reasoning path leads to a contradiction, the system can backtrack. ExClaim ensures that any changes made to the domain knowledge base during the failed path are automatically undone, maintaining a clean state. The system also supports reification, allowing concepts (like "Secretary") to be treated as data instances of higher-level concepts ("Position"), which simplifies the creation of generic functions.

Results and Discussion

The effectiveness of the ExClaim architecture is demonstrated through a Lecturer Allocation scenario. In this example, the system assigns teachers to classes while adhering to complex constraints.

  • Automated Constraint Satisfaction: When a domain rule is defined (e.g., "Assistants can only teach seminars"), the Inference Level automatically respects this. The underlying DL engine filters invalid assignments during the reflection process without requiring manual coding of these checks in the inference logic.
  • Consistency Maintenance: The system successfully identifies when a partial solution becomes impossible based on domain constraints and triggers backtracking automatically. This proves the successful integration of the non-monotonic domain level with the non-deterministic inference level.

Limitations and Future Work

While ExClaim provides a robust framework, it faces challenges regarding the trade-off between expressivity and computational speed. Some complex constraints, such as "Role-Value Maps," are mathematically undecidable in Description Logics and must be handled manually at the Inference level.

Future work focuses on the development of a comprehensive, domain-independent library of Problem Solving Methods (PSMs). Additionally, further research is needed to evaluate how the system scales as the number of domain instances increases, a common bottleneck for Description Logic-based systems.

Summary updated

BibTeX
@inproceedings{badea1999knowledge,
  author = {Liviu Badea},
  title = {{Knowledge Modeling and Reusability in ExClaim}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1999},
}
Figure