TL;DR
This paper introduces a flexible framework called the "family of languages" approach to tackle the challenge of semantic interoperability between different knowledge representation languages. It proposes structuring these languages based on the properties of transformations between them, allowing for more adaptable and user-friendly integration of knowledge from diverse sources.
What Problem Does This Paper Address?
The Semantic Web aims to annotate web pages with formal knowledge, enabling machines to understand and process information. However, this formal knowledge is likely to be encoded in various languages due to factors like legacy systems, ease of use, and device heterogeneity. This creates a significant challenge: how can we ensure that knowledge expressed in one language can be understood and used by systems that operate in another? This problem, known as semantic interoperability, is crucial for the success of the Semantic Web. The paper focuses specifically on the mismatch between knowledge representation languages, rather than other interoperability issues like axiomatization mismatches.
Methods
The core of the paper is the "family of languages" approach, which organizes a set of knowledge representation languages into a structure based on the properties of transformations between them. Instead of a rigid hierarchy or a single pivot language, this approach uses a partial order. This means languages are related by transformations that preserve certain properties, such as logical consequence, interpretation, expressiveness, or consistency.
The paper defines several types of "coverage" relations that characterize these transformations:
- Language Inclusion ($\leq_{synt}$): A language L is included in L' if all expressions in L are also in L'. This is the simplest form of interoperability, where transformations are trivial.
- Interpretation Preservation ($\leq_{int}$): A transformation preserves interpretations if the meaning of an expression in the source language is the same as the meaning of its transformed expression in the target language.
- Expressibility ($\leq_{exprt}$): A language L is expressible in L' if theories from L can be transformed into L' while preserving models, potentially with renaming of predicates.
- Epimorphic Transformations ($\leq_{epi}$): This is a weaker form of expressibility, ensuring that each model of the transformed representation is related to a model of the source representation.
- Consequence Preservation ($\leq_{csq}$): This is a crucial property, ensuring that if a statement is a logical consequence in the source language, it remains a logical consequence in the target language after transformation.
- Consistency Preservation ($\leq_{sat}$): A transformation preserves consistency if an inconsistent knowledge base in the source language remains inconsistent in the target language.
The "family of languages" approach generalizes other architectures like the pivot approach (where all languages are translated to a single pivot language) and the layered approach (where languages form a strict hierarchy). It allows for more flexibility by enabling users to choose transformations that are "closer" to their needs and to compose simple transformations with known properties.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Flexibility over Rigidity: The "family of languages" approach offers a more flexible alternative to rigid pivot or layered architectures for semantic interoperability.
- Property-Driven Structure: The structure of the language family is defined by the properties of transformations between languages, rather than just syntactic relationships. This allows for a more nuanced understanding of interoperability.
- Choice and Composition: Users can select transformations based on the specific properties they need (e.g., preserving logical consequence vs. just consistency) and compose simpler transformations to build complex ones.
- Generalization of Existing Approaches: The framework elegantly generalizes existing approaches like pivot and layered architectures, showing how they fit within a broader structure.
- Practical Implementation: The paper demonstrates how this approach can be implemented using existing web technologies like XML and XSLT, making it practical for real-world applications.
Key Results
The paper defines a set of coverage relations that form a hierarchy of semantic relationships between languages based on the properties of transformations. It shows how these relations can be used to structure a "family of languages" and provides a framework for selecting and composing transformations to achieve semantic interoperability. The paper also presents a concrete example of importing ontologies from DAML-ONT and OIL into the SHIQ language, demonstrating the practical application of the approach. The transformations used in the example are described as consistency-preserving.
Limitations and Future Work
The authors acknowledge that while the approach is flexible, much work is still required for characterizing other useful properties, including properties on the syntax of languages. They suggest that further research could focus on developing more sophisticated semantic relations and exploring their compositionality. The paper also notes that some transformations, like those for eliminating domain restrictions, might not preserve complete semantics but are useful for specific reasoning tasks.
Summary updated