Abstract
Linguistic terms are often used to represent the truth degree or credibility degree to manage the uncertainty or imprecision as one of popular ways of knowledge representation in the perception-based decision making problem. This present work represents the credibility of uncertain knowledge using linguistic values. The linguistic-valued credibility is then modeled based on a lattice ordered logical algebra -lattice implication algebra (LIA) and then used in the linguistic-valued based on approximated reasoning. This approach can better express both comparable and incomparable linguistic valued credibility information in knowledge base representation. And it can also fit to approximate reasoning under a fuzzy environment with both comparable and incomparable linguistic credibility. Examples illustrate that the proposed approach can retain the original information and can simulate people's way of thinking through reasoning with words.
Introduction
In real-world decision-making, information is rarely precise. Humans naturally communicate uncertainty using linguistic terms like "very credible" or "slightly incredible" rather than numerical probabilities. Traditional fuzzy logic systems often struggle with this because they typically require "quantification"—converting words into numerical fuzzy sets (0 to 1). This process can strip away the nuances of human perception.
A significant challenge in automated reasoning is handling incomparable terms—situations where one linguistic value cannot be strictly ranked as "greater" or "less" than another. Most existing models rely on a total linear ordering, which fails to capture the complexity of human thought. This paper proposes a mathematical foundation using Lattice Implication Algebra (LIA) to reason directly with linguistic credibility factors while preserving their original semantic meaning.
Methodology: Lattice-Ordered Linguistic Credibility
The core innovation of this research is the application of Lattice Implication Algebra (LIA) to model linguistic credibility.
1. Linguistic Scale and Lattice Structure
The authors define a set of ten credibility levels, ranging from "absolutely credible" to "absolutely incredible," including hedges such as "slightly," "somewhat," and "exactly." Unlike a simple list, these levels are organized into a lattice structure, visualized as a Hasse diagram.
This structure allows the system to mathematically define relationships between terms that are comparable (e.g., "very credible" > "slightly credible") and those that are incomparable in certain contexts.
2. Logical Operations
Using LIA, the authors define four essential operations for these linguistic terms:
- Join (∨) and Meet (∧): These function as "OR" and "AND" operators for linguistic terms.
- Inverse ('): This identifies the opposite of a credibility level (e.g., the inverse of "very credible").
- Implication (→): This serves as the engine of the reasoning process, enabling the system to derive conclusions from premises (IF-THEN rules).
3. Knowledge Representation
Knowledge is stored as 4-tuples: (Object, Attribute, Value, Credibility). For instance: (John, academic achievements, successful, very credible).
Case Studies and Validation
The framework was validated through illustrative case studies that simulate human reasoning chains.
Traffic Accident Risk Analysis
The system evaluated an individual's risk based on linguistic inputs such as "middle-aged," "very careful," and "somewhat tired." The LIA-based approach successfully aggregated these conflicting inputs to conclude that the risk was "at least somewhat incredible," reflecting a nuanced safety assessment that accounts for both the driver's skill and current state.
Social Risk Assessment System
A more complex five-input system was developed to evaluate social risk based on environmental factors and social vulnerability.
This case study demonstrated the system's ability to handle multiple layers of "IF-THEN" rules, propagating linguistic credibility from raw environmental data (e.g., floods, seismic hazards) up to a final aggregate "Social Risk" assessment.
Key Insights and Conclusion
The proposed approach offers several advantages over traditional numerical methods:
- Direct Reasoning: By using lattice theory, words themselves become mathematical objects, eliminating the need for translation into numbers.
- Handling Incomparability: The lattice structure represents complex human certainties where different opinions may be valid without a strict hierarchy.
- Human-Centric Output: The system outputs linguistic terms, making the reasoning process transparent and intuitive for human users.
While the current model uses a fixed scale of ten levels, future work may explore more dynamic or granular scales and address the computational complexity inherent in larger rule sets. Nonetheless, this research provides a robust theoretical foundation for simulating human-like reasoning with words.
Summary updated