The Demarcation Problem and Logical Constants

1. The Problem and Its Significance

At first glance, the distinction might seem intuitive. We have a rough sense of which words do the 'logical' work in a sentence. However, when we press for a precise criterion, the issue becomes surprisingly complex.

To begin, let's get a firm grasp on why this problem is so central to the philosophy of logic and what makes it difficult.

As the article you've just read makes clear, a principled solution to the demarcation problem would help us understand the very nature and scope of logic. It would clarify what is special about logical truth compared to other forms of necessary or a priori truth.

Historically, the earliest attempts to delineate logical terms were tied to grammar. In medieval term logic, a distinction was made between categorematic terms (like 'Socrates' or 'runs'), which could function as subjects or predicates and were thought to have independent meaning, and syncategorematic terms ('every', 'and', 'if'), which signify nothing on their own but structure the combination of meaningful terms. These syncategorematic terms were the precursors to our modern logical constants.

However, as the SEP article briefly mentions (and you know from your study of Frege), the 19th-century revolution in logic replaced the old subject-predicate analysis with a more powerful function-argument structure. This shift made the simple categorematic/syncategorematic distinction problematic. Modern philosophers like Quine proposed a successor idea: identifying logical constants with grammatical particles. But this approach also faces significant hurdles. For instance, we could invent an artificial language with a quantifier-like operator for "there is at least one cat such that..." which would be a grammatical particle but would clearly not be a logical constant. The grammatical approach seems to depend on a prior, privileged choice of language, which pushes the problem back a step.

Given the difficulties with these grammatical approaches, philosophers have sought more robust, principled criteria. We will now examine the two most influential proposals.

2. The Semantic Criterion: Permutation Invariance

One of the most powerful and mathematically elegant ideas is that logical notions are uniquely characterized by their extreme generality or topic neutrality. Logic isn't about anything in particular; it applies to reasoning about physics, biology, or mathematics with equal force.

But "topic neutrality" is a vague philosophical notion. How can we make it precise? The key insight, developed by Tarski and others, is to define this neutrality as invariance under permutations. The idea is that logical operations don't care about the specific identity of the objects in a domain. You can swap the objects around, and the logical operation will yield a structurally identical result.

To understand the main proposals, it is helpful to see how they fit into the broader landscape of logical theories.

Now let's focus on the semantic approach based on permutation invariance, which van Benthem identifies as "the most successful account to date".

As you've read, an operation is logical if its extension is invariant under any permutation of the underlying domain of objects.

• Logical notions: The identity relation ($=$), the universal set (the extension of 'is a thing'), the empty set, and the standard truth-functional connectives and quantifiers are all permutation-invariant.
• Non-logical notions: Predicates like 'is a horse' are not invariant. If you permute the domain by swapping a horse with a non-horse, the set of horses changes. Relations like 'is a successor of' on the natural numbers are not invariant, because they rely on the specific structure of the numbers.

This approach has deep roots. Tarski saw it as a generalization of Felix Klein's Erlanger Programm for classifying geometries. In Klein's program, Euclidean geometry is defined by the properties that are invariant under rigid motions and scaling, while topology is defined by properties invariant under continuous transformations. Logic, on this view, is the most general science of all, studying those properties that are invariant under the most disruptive group of transformations possible: the group of all permutations, which destroys all structure except for cardinality.

Critique of the Invariance Criterion

Despite its elegance, the permutation invariance criterion faces serious philosophical objections.

1. Over-generation: The criterion is arguably too inclusive. It classifies many notions that we intuitively think of as mathematical rather than logical as being logical constants. For example, quantifiers like "there are infinitely many" or "there are an uncountable number of" are permutation-invariant. This suggests that the criterion may blur the line between logic and set theory.

2. Extensionality and Sense: The criterion operates on the level of extensions (the sets of objects or tuples that notions apply to), not on the level of senses or meanings. As the SEP article explains, we can define a connective '#' that happens to have the same truth-table as negation ($\neg$) because of some necessary but non-logical fact (e.g., #\phi is true iff $\varphi$ is false and water is H₂O). Since '#' and $\neg$ have the same extension (the same truth-function), if one is invariant, the other must be too. Yet, intuitively, '#' is not a logical constant because understanding it requires empirical knowledge about chemistry. The criterion seems to capture a necessary condition for logicality, but not a sufficient one, as it is blind to the way a constant's meaning is specified.

Johan van Benthem's paper offers a particularly sharp analysis of these issues.

Van Benthem's point about circularity is that the invariance account already presupposes crucial choices. To test for invariance, we must first decide on our domain of objects (e.g., individuals, not sets) and the transformations we allow (permutations). For instance, the criterion yields identity as logical because permutations are defined as bijections, which inherently respect identity and non-identity. The criterion doesn't discover that identity is logical so much as it is built to reflect it.

3. The Proof-Theoretic Criterion: Inferential Role

The perceived failure of the semantic criterion to account for the sense of logical constants motivated an alternative family of proposals. The proof-theoretic or inferentialist approach locates the essence of logical constants not in their semantic properties but in their role within a system of inference.

The central idea is that a term is a logical constant if and only if its meaning can be exhaustively defined by a set of introduction and elimination rules in a system of natural deduction.

For example, the meaning of 'and' ($\wedge$) is completely captured by its rules:

• Introduction: From $A$ and $B$, you can infer $A\wedge B$.
• Elimination: From $A\wedge B$, you can infer $A$. From $A\wedge B$, you can infer $B$.

A speaker understands 'and' not by grasping a semantic value, but by mastering this inferential practice. This approach seems to avoid the problem of co-extensional connectives like '#', because the rules for '#' would have to explicitly mention water and H₂O, revealing its non-logical character.

Critique of the Inferentialist Criterion

This approach also faces significant challenges.

1. The 'Tonk' Problem: As you read, A. N. Prior showed that not just any set of rules can define a meaningful connective. His connective 'tonk' had the introduction rule of 'or' (from $A$, infer $A\text{ tonk }B$) and the elimination rule of 'and' (from $A\text{ tonk }B$, infer $B$). This allows one to prove anything from anything, leading to triviality. Proponents of inferentialism must therefore provide criteria for "good" rules, such as principles of harmony or conservativeness, which state that the elimination rules should not let you infer more than the introduction rules justify. This adds another layer of complexity to the proposal.

2. Completeness of Rules: Critics question whether simple introduction and elimination rules truly exhaust the meaning of all logical constants. For instance, some argue that understanding the conditional ($\to$) or the universal quantifier ($\forall$) involves more than just mastering their formal rules; it may require grasping how they relate to practices of inductive reasoning or hypothesis formation.

Conclusion

We have analysed the demarcation problem by examining the two most prominent families of criteria for logical constancy. There is no consensus in philosophy on which, if any, is correct.

Key Takeaways:

• The Demarcation Problem: A central issue in the philosophy of logic is finding a principled criterion to distinguish logical constants from other expressions. The solution affects our understanding of the nature of logical truth and consequence.
• Semantic Criterion (Permutation Invariance): This approach identifies logical constants as those operations that are topic-neutral, defined precisely as being invariant under any permutation of the domain of objects.
  • Pros: Mathematically precise, philosophically motivated by generality, and extends Tarski's own thinking.
  • Cons: Arguably over-generates by classifying mathematical notions as logical, and fails to distinguish constants by their meaning (sense) rather than just their semantic value (extension). It may also be circular.
• Proof-Theoretic Criterion (Inferential Role): This approach identifies logical constants as those whose meaning is fully constituted by their introduction and elimination rules in a formal system.
  • Pros: Directly addresses the meaning or sense of constants, connecting logicality to inferential practice.
  • Cons: Requires additional constraints (like harmony) to rule out paradoxical connectives like 'tonk', and may not fully capture the meaning of all constants.

The persistence of this debate suggests that logicality might not be a single, monolithic property. As van Benthem hints, the core logical constants may be those special terms that satisfy multiple criteria at once—they are invariant, have harmonious inferential rules, and are computationally simple. The debate itself forces us to clarify what we expect from our logic and what makes it a unique and fundamental discipline.

Preview of the Next Lesson:

These two approaches to logical constants mirror the two great traditions in the study of logical consequence: the model-theoretic approach (concerned with truth in structures) and the proof-theoretic approach (concerned with formal derivability). In our next lesson, we will examine the relationship between them by analysing the philosophical significance of soundness and completeness theorems, which forge the crucial link between these two conceptions of logic.