Representation of Semantic Mappings

Heiner Stuckenschmidt, Michael UscholdSemantic Interoperability and Integration 2005 · OpenReview

TL;DR

This paper summarizes a breakout session discussing the representation of semantic mappings between heterogeneous models. It highlights the lack of a clear definition for mappings, explores current approaches and their limitations, and outlines crucial open research questions for advancing semantic interoperability.

What Problem Does This Paper Address?

In an increasingly interconnected digital world, different systems and models need to communicate and share information. However, these models often use distinct languages and structures, creating a significant barrier to interoperability. Semantic mappings are crucial for bridging these differences by defining how elements in one model relate to elements in another. This paper addresses the fundamental challenge of how to effectively represent these mappings, exploring the underlying nature of mappings, the existing technologies, and the critical research gaps that need to be filled to enable robust and reliable semantic interoperability.

Methods

The paper doesn't present a novel algorithm or technique in the traditional sense. Instead, it's a summary of a discussion that explored the design space of mapping approaches. The "methods" discussed include:

  • Declarative vs. Operational Mappings: Distinguishing between mappings that define constraints on joint interpretations (declarative) and those that specify translation procedures (operational, e.g., rewrite rules).
  • Formal Properties: Identifying verifiable properties that mappings can satisfy, such as satisfiability of the overall model, preservation of inferences, or query answering.
  • Language-Specific vs. Language-Independent Mappings: Considering whether mappings should be tied to specific modeling languages or be more general.
  • Organization of Mappings: Discussing whether mappings are part of a model or specified independently, and whether they are uni- or bidirectional.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The Elusive Definition of "Mapping": The core takeaway is that a universally agreed-upon definition of what a "mapping" is remains elusive, underscoring the complexity of the problem.
  • Beyond Simple Equivalence: While mappings often aim to preserve "meaning," this is hard to verify. The focus shifts to more concrete, verifiable properties like inference preservation or query answering, often context-dependent.
  • The Semantic Web's Contribution (and Limitations): OWL offers some constructs for defining semantic relations (same, different, subclass), but it struggles with data transformations and has an "all or nothing" approach to importing external knowledge, potentially leading to inconsistencies.
  • The Need for Explicit Mappings: The paper advocates for treating mappings as "first-class citizens" in semantic modeling, requiring specialized tools for their creation, maintenance, and reasoning.
  • Richness of Semantic Relations: Real-world scenarios demand more than just simple equivalence or subclass relations; mappings need to capture richer semantic relationships like inconsistency, cause-effect, or overlap.

Key Results

This paper is a discussion summary, not an empirical study, so there are no quantitative results. The key "results" are the identification of:

  • A lack of a clear definition for semantic mappings.
  • Several existing approaches (e.g., in the Semantic Web with OWL, and database-inspired languages like Datalog, F-Logic, DLR) and their respective limitations.
  • A set of critical open research questions and directions.

Limitations and Future Work

Limitations:

  • Difficulty in Verifying "Meaning Preservation": The assumption that mappings preserve the intended meaning of models is practically impossible to verify.
  • OWL's "All or Nothing" Import: Importing entire ontologies in OWL can lead to inconsistencies when dealing with highly heterogeneous models.
  • Inability to Define Mapping Strength: Current approaches lack mechanisms to specify the confidence or strength of a mapping.
  • Limited Semantic Relations: Existing approaches often support only a restricted set of semantic relations.

Future Work:

  • Developing a Clear Definition of Mappings: Establishing a common understanding of what mappings are and what they should achieve.
  • Richer Semantic Relations: Exploring and defining more expressive semantic relations beyond simple equivalence and subclass.
  • Mappings as First-Class Citizens: Creating specialized tools and frameworks for managing, reasoning about, and composing mappings.
  • A Framework for Comparing Mappings: Designing a common framework to evaluate and compare different mapping approaches, possibly focusing on declarative mappings and logical frameworks.
  • Handling Language Heterogeneity: Developing robust translations between models encoded in different languages with guaranteed formal properties.
  • Measuring Relatedness: Developing methods to measure the degree of relatedness specified by a mapping, especially for automatically generated mappings.

Summary updated

Abstract

The aim of this breakout session was to chart the landscape of existing approaches for representing mappings between heterogeneous models, identify common ideas and formulate research questions to be addressed in the future. In the session, the discussion mainly concerned three aspects: The nature of mappings, existing proposals for mappings and open research questions.
BibTeX
@inproceedings{stuckenschmidt2005representation,
  author = {Heiner Stuckenschmidt and Michael Uschold},
  title = {{Representation of Semantic Mappings}},
  booktitle = {Semantic Interoperability and Integration},
  year = {2005},
  url = {https://openreview.net/forum?id=O7YdLkXa3m},
}
Figure