TL;DR
This paper explores the definability of identity in higher-order logic. It shows that while identity is definable in standard second-order logic with set-theoretic semantics, it is not definable in a novel language called Quantified Naming Logic (QNL) or its variant with a different predication relation, even with higher-order quantification. This highlights that the structure of predication, not just the power of quantification, is crucial for identity's definability.
What Problem Does This Paper Address?
In classical first-order logic, the identity symbol '=' is typically taken as a primitive logical symbol. However, it's a well-known fact that identity can be defined in richer logical systems, specifically second-order logic, where quantifiers can range over predicates or sets. The paper starts by establishing this: in a first-order language without identity, we can't define a predicate that always behaves like identity. But, if we move to a second-order language where quantifiers can bind predicate variables that range over all subsets of the domain, identity is definable by a single formula.
The core problem this paper tackles is to investigate the limits of this definability. It asks: are there higher-order languages, even those with powerful quantification over sets, where identity is still not definable? The paper argues that the way predication is handled in a language is as crucial as the power of its quantifiers for the definability of identity.
Methods
The paper introduces and analyzes two main logical systems:
- Quantified Naming Logic (QNL): This system departs from the traditional distinction between singular terms and predicate symbols. Instead, it uses a single category of "name variables" and a special constant 'ɛ' (epsilon). The 'ɛ' symbol is interpreted as a predication relation, meaning "a ɛ b" is read as "a is one of b's" or "a is b." The semantics for QNL are explored using two types of models: S-models (where name variables map to subsets of the domain) and N-models (where a "naming relation" connects variables to objects). Crucially, quantifiers in QNL bind name variables and range over all subsets of the domain, similar to second-order quantification.
- Language Lɛ and Lf: The paper then considers variations of QNL. Lɛ uses the epsilon symbol 'ɛ' with specific satisfaction conditions. Lf replaces 'ɛ' with a new symbol 'f', which has the same syntactic role but a different semantic interpretation. The paper defines "Fs-models" and "Fn-models" for these languages, which are essentially semantic frameworks for S-models and N-models, respectively, but with the 'f' symbol.
The core method involves constructing "non-standard" models that satisfy a given set of formulas but do not interpret the relevant symbol (like 'ɛ' or 'f') in the standard, identity-preserving way. By showing that such non-standard models exist, the paper demonstrates that no set of formulas in the language can force the symbol to behave like identity.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Predication Matters as Much as Quantification: The most significant insight is that simply having higher-order quantification (over sets or analogous structures) is not sufficient to guarantee the definability of identity. The nature of the predication relation is a critical factor.
- Identity in QNL is Sensitive to Epsilon's Role: In QNL, the definability of identity relies heavily on the specific semantic interpretation of the epsilon symbol 'ɛ'. When 'ɛ' is interpreted in a way that captures singleton relationships (as in S-models) or direct reference (as in N-models), identity can be defined.
- The 'f' Symbol in Lf is Not Identity: The paper introduces a language Lf where a symbol 'f' plays the syntactic role of predication but is not definable as identity, even with powerful set-theoretic semantics for its models. This is demonstrated by constructing non-standard models where 'f' does not behave like identity.
- Distinction Between Singular Terms and Predicates is Key (Initially): The paper initially suggests that the distinction between singular terms and predicate symbols in classical second-order logic might be crucial for identity's definability. However, it later revisits this, indicating that introducing "quasi-individual variables" doesn't resolve the undefinability issue in the Lf framework.
- Identity of Singletons vs. Identity of Objects: The paper distinguishes between the identity of singletons (which can be captured in some models) and the identity of objects themselves. The latter is what proves elusive in the non-standard models.
Key Results
- Fact 2: Identity is definable in standard second-order logic (L₂) by a single formula, provided quantifiers range over all subsets of the domain.
- Fact 3: A specific relation between predicates, interpreted as identity between singletons, is definable in L₂.
- Fact 4: No set of formulas in the language Lf (with the 'f' symbol) can define the standard interpretation of 'f' in Fs-models. This means identity is not definable in Lf with Fs-semantics.
- Corollary 5: Identity between singletons is not definable in Lf with Fs-semantics.
- Fact 8: Identity is not definable in the extended language L (with quasi-individual variables) in a way that is robust across all models.
Limitations and Future Work
- Non-emptiness Assumption: The paper notes that standard semantics for QNL (and related logics) often assume a non-empty domain. While this is standard for comparison with second-order logic, allowing for empty models could be a point of future exploration.
- Focus on Specific Semantics: The paper primarily focuses on set-theoretic semantics for higher-order logic and specific interpretations of the epsilon and 'f' symbols. Exploring other semantic frameworks could yield different results.
- Diagnosing Undefinability: The paper offers a diagnosis that the undefinability of identity in Lf stems from taking the predication relation between a copula and its arguments as primitive, rather than a relation between a singular term and a predicate. Further work could explore this distinction more deeply.
- Generalizing Undefinability: The paper demonstrates undefinability for specific languages and symbols. A broader question is whether this undefinability extends to other higher-order languages with different predication mechanisms.
Summary updated