Reasoning is grounded in First Order Logic - What is Logic?

Logic is stable and universal, and is identified with classical first order logic. Other logics are here considered to be first order theories, syntactically sugared in notationally convenient forms. From this point of view higher order logic is essentially first order set theory.

The most natural, intuitive and fruitful approach turned out to be the axiomatization, in the predicate calculus notation, of the concept of a set. Once this was done the way was cleared for a proper semantic foundation to be given to FOL itself. The rigorous mathematical concept of an interpretation, and of the denotation of a formula within an interpretation, was introduced by Alfred Tarski, who thus completed, in 1929, the fundamentals of the semantics of FOL.

The notion of logical consequence could now be defined by saying that a sentence S logically follows from a sentence P if there is no interpretation in which P is true and S is false.

This now made it possible to raise the following fundamental question about FOL: is the following Completeness Theorem provable: for all sentences S and P, S is formally derivable from P if and only if S logically follows from P?

The question was soon positively answered, independently, in their respective doctoral theses, by two graduate students. Kurt Gödel received his degree on February 6, 1930 at the University of Vienna, aged 23, and Jacques Herbrand received his at the Sorbonne on 11 June, 1930, aged 22.

The theorem is so important because the property of formal derivability is semidecidable: there is an algorithm by means of which, if a sentence S is in fact formally derivable from a sentence P, then this fact can be mechanically detected. Indeed, the detection procedure is usually organized in such a way that the detection consists of actually constructing a derivation of S from P. In other words, when looked at semantically, the syntactic detection algorithm becomes a proof procedure.

Turing's 1936 paper on Hilbert's Decision Problem was crucial. Ostensibly it was yet another attempt to characterize with mathematical precision the concept of computability, which he needed to do in order to show that there is no fixed computation procedure by means of which every definite mathematical assertion can be decided (determined to be true or determined to be false).

Alonzo Church independently reached the same result at essentially the same time, using his own quite different computational theory of lambda-definability. In either form, its significance for computational logic is that it proved the impossibility of a decision procedure for FOL. The semidecision procedure reflecting the deductive completeness of FOL is all there is, and we have to be content to work with that.

-- John Alan Robinson


There are a number of ways of approaching the issues, but for purposes here a *logic* can be thought of as a (typically rather large and inclusive) family of formal languages together with a semantic theory for those languages.

'''Semantic Theory'''

A *semantic theory* for a class of languages is simply a systematic method of assigning semantic values to the members of the various syntactic categories of those languages. The Semantic Web model distinguishes properties and classes considered as objects from their extensions - the sets of object-value pairs which satisfy the property, or things that are 'in' the class - thereby allowing the extension of a property or class to contain the property or class itself without violating the axiom of foundation.


An application of a semantic theory to a particular language yields an *interpretation* of the language, i.e., an assignment of appropriate semantic values to the basic lexical items of the language. Appropriate semantic values are then assigned by the theory to complex syntactic constructions of the language -- notably, truth and falsity to sentences -- recursively in terms of the semantic values of their component parts.

'''First Order Logic'''

What makes a logic *first-order* is a fairly technical matter, but it consists essentially in the possession of two semantic properties, compactness and the downward Löwenheim-Skolem property. These properties are critically connected to the fact that FOL is *complete*, that is, that it is possible to *axiomatize* logical consequence, i.e., the fundamental logical relation that holds between a set S of sentences and a given sentence A when any interpretation that makes all the members of S true also makes A true. It is (among other things) completeness that makes FOL so important to automated reasoning, as it means that, whenever A is a logical consequence of S in a first-order logic, there is actually a *proof* of A from some (finite) subset of S. (Alas, finding that proof is another kettle of fish entirely, which is why we have more constrained, less expressive logics like description logics and their ilk.)

ps: Re compactness and the downward L-S property, for those who might be interested: Say that a set S of sentences of a logic is *satisfiable* if there is *model* of S, that is, an interpretation of the language of the logic in which all the members of S are true, and that S is *unsatisfiable* if it is not satisfiable. Compactness says, roughly, that unsatisfiability is in a certain sense a finite property; more exactly, compactness says that if a set S of sentences is unsatisfiable, then some finite subset of S is as well. From the compactness of FOL follows the very interesting fact that FOL is incapable of expressing the concept of finitude.

The downward L-S property says, roughly, that it is not possible to force all of the models of any set of sentences to be larger than the set N of natural numbers; more exactly, a logic has the downward L-S property if, whenever, a set of sentences has an infinite model of any size, it also has a model the size of N.

The characterization of first-order logic in terms of compactness and the downward L-S property is known as Lindström's Theorem.

-- Pat Hayes