Formalizing Ontological Commitments

AAAI 1994

Sure, here's a blog-style summary of the paper "Formalizing Ontological Commitments":

TL;DR

This paper proposes a formal framework for defining "ontological commitments" – the implicit assumptions about reality that a logical language makes. By extending first-order logic with modal operators and mereo-topological primitives, the authors provide a semantic approach to specifying the intended meaning of vocabulary, enabling better knowledge sharing and reuse.

What Problem Does This Paper Address?

When we use logical languages to model aspects of the real world, we often rely on the implicit understanding of our words (e.g., "apple," "red") to convey meaning. However, this reliance on natural language can lead to ambiguity and make it difficult to share knowledge across different systems or users. The authors argue that current approaches to formalizing this implicit meaning, often through shared terminologies, are insufficient. They aim to provide a more robust, semantic definition of "ontological commitment" that captures the fundamental assumptions about reality embedded in a language's vocabulary, going beyond mere syntactic consistency.

Methods

The paper introduces a formal framework built upon first-order logic. Key methodological contributions include:

  • Modal Extension: The language is extended with modal operators (□ and ◇) to capture the notion of "ontological compatibility" between different possible worlds. This allows for reasoning about alternative states of affairs and the constraints they must satisfy.
  • Mereo-Topological Primitives: To represent the a priori structure of domains, the framework incorporates primitives for mereology (part-of relations) and topology (connection relations). This allows for formalizing concepts like "proper part" and "connection."
  • Ontological Commitment as a Set of Models: An ontological commitment is defined as a set of "compatibility models" for a modalized language. This set restricts the possible interpretations of the language's vocabulary, thereby specifying its intended meaning.
  • Classification of Unary Predicates: A significant part of the paper is dedicated to classifying unary predicates (predicates with one argument) based on their "discriminating power." This includes distinctions like:
  • Sortal vs. Non-Sortal: Sortals provide a principle for distinguishing and counting individuals (e.g., "apple"), while non-sortals describe properties of already distinguished individuals (e.g., "red").
  • Rigid vs. Non-Rigid: Rigid predicates are essential to an object's identity (e.g., "being a prime number"), while non-rigid predicates are contingent properties (e.g., "being red").
  • Substantial vs. Non-Substantial Sortals: Substantial sortals are rigid (e.g., "apple"), while non-substantial sortals are not (e.g., "student").
  • Characterizing vs. Pseudo-Sortal: Characterizing predicates are non-sortal and non-rigid (e.g., "red"), while pseudo-sortals are rigid but not sortal (e.g., "physical object").
  • Well-Founded Ontological Commitment: A stricter notion of commitment is introduced, requiring "categorial predicates" (like "physical object") to form a partition of the domain and imposing structural constraints on the relationships between sortals.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Ontological Commitment is Semantic, Not Syntactic: The authors strongly advocate for a semantic definition of ontological commitment, arguing that it should constrain models rather than just ensure syntactic consistency with axioms. This allows for different vocabularies to share the same underlying ontology.
  • Modality is Crucial for Ontology: Capturing the "a priori" nature of reality requires modal reasoning to express compatibility and constraints between different possible interpretations or states of affairs.
  • Mereology and Topology are Foundational: Formalizing the structure of reality necessitates primitives that can describe part-whole relationships and spatial/temporal connections.
  • Fine-Grained Classification of Predicates: The paper demonstrates that a rich classification of unary predicates (sortal, rigid, substantial, etc.) is essential for precisely specifying their intended meaning and ontological role. This goes beyond simple type hierarchies.
  • "Apple" vs. "Red": A Concrete Example: The paper uses the simple statement "a red apple exists" to illustrate how their framework can distinguish between the ontological commitments of "apple" (a substantial sortal) and "red" (a characterizing predicate).

Key Results

The paper presents a formal framework and a set of definitions for specifying ontological commitments. While no quantitative results are presented, the key outcome is the formalization of:

  • A modal logic extension for representing ontological compatibility.
  • Mereo-topological primitives for domain structure.
  • A detailed classification of unary predicates (sortal, non-sortal, rigid, substantial, characterizing, pseudo-sortal).
  • The concept of a "well-founded ontological commitment" for more structured knowledge representation.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Simplification of Formalization: The presented formalization is simplified, omitting aspects like temporal reasoning and a full treatment of functional symbols.
  • Need for Further Distinctions: More detailed distinctions are needed for meta-level categories of predicates and for binary relations.
  • Domain-Specific Axioms: While the framework provides general distinctions, specific axioms are required to characterize particular domain categories (e.g., "physical object," "event").
  • Computational Properties: The computational properties of the extended language (Lo) are not discussed, though the authors note that they do not affect the agent using the original language L.
  • "Natural Types": The concept of "well-founded ontological commitment" aims to capture something akin to "natural types," but further work is needed to fully formalize this.
  • A Priori Laws of Reality: The authors express a conviction that understanding and representing the a priori laws governing reality is a significant ongoing challenge.

Summary updated

BibTeX
@inproceedings{1994formalizing,
  title = {{Formalizing Ontological Commitments}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1994},
}

Loading related content...

Figure