THE LOGIC NOTES

Definite descriptions Identity and existence

In 2020, when this page was written, whenever the Australian radio news carried an item about the doings of the Prime Minister, they would refer to him as "the Prime Minister, Scott Morrison". This may have been to make it clear which Scott Morrison they meant, or to inform Australians that Morrison was (still) their Prime Minister, but what is interesting from the logical point of view is that they used two referring expressions, "Scott Morrison" and "the Prime Minister" to denote the same individual. One refers to him by name; the other picks him out as the object satisfying a description. Expressions such as "the Prime Minister", "the man with the golden gun", "the natural satellite of the Earth" or "the square root of 36" are called definite descriptions, and their logic raises interesting questions.

Syntactically, a definite description behaves like a name, so it might seem quite natural to introduce a symbol ι and use this to create terms out of formulae. On this account, ι x Fx could be seen as a term able to occur in formulas just like a proper name, read as 'the thing, x, such that Fx' or more briefly as 'the F'. Using this notation, "The F is G" would be written

G( ι x Fx).

More generally, for any formula A containing a name m and for any variable v not occurring in it,

ι v Avm

would be a term. We might expect such terms to be semantically similar to names, since whether we use a name like "Scott Morrison" or a definite descrition like "the [first] Prime Minister of Australia in 2020" should make no difference to truth values, because they refer to the same person.

We shall see later that this way of forming terms is problematic, and suggest a better way of using the definite description operator, but for now we may continue to examine it, noting how useful it is to be able to express terms of this sort.

One of the main motivations for the early development of modern symbolic logic was the perceived need to provide logical foundations for mathematics. The logic of function symbols is clearly central to this enterprise, and a definite description operator seems to be exactly what we need to bring them within the scope of ordinary logic. Consider the addition function in ordinary arithmetic, for instance. Without using function symbols, we may represent the facts about addition by means of a 3-place predicate symbol S, writing Sabc to mean that a and b sum to c. Thus the formula S(7,5,12) is true, for instance, while the formula SOMEx (S(7,x,12) AND NOTS(x,3,8)) is false. Of course, we really want to write "7 + 5 = 12" rather than "S(7,5,12)", but how are we to construe the term on the left of this equation? Well, one idea is simply to define t + u as ι x S(t,u,x) and rely on our logical account of definite descriptions to do the work.

Any such logical account must allow for the fact that definite descriptions are semantically unlike names in that a name is an arbitrary (unstructured) symbol which gets to refer to an object as a result of an act of naming, whereas a definite description refers to it as a result of the facts about which things satisfy certain predicates and which do not. You cannot understand a name or find out its reference without being made party to a convention about it, but you can investigate which thing is picked out by a definite description in much the same way as you would investigate any facts expressed by ordinary indefinite descriptions.

The classical theory of definite descriptions was proposed by Bertrand Russell (On Denoting, 1905, reprinted in Russell). He starts by noting the need for such a theory, and presents some puzzles which it needs to be able to solve. We focus on two of these:

  • Puzzle 1: Definite descriptions threaten to violate laws of logic. Consider Russell's example: George IV wished to know whether Walter Scott was the author of the novel Waverley. That is, George wanted to know whether the identity statement "Scott = the author of Waverley" was true. Now in fact Scott was indeed the author of Waverley, so by the logical rule =E we should be able to derive that George wanted to know whether Scott was Scott. But this is not true: the First Gentleman of Europe had no interest in the natural deduction rule =I.

  • Puzzle 2: Sentences containing definite descriptions can be perfectly meaningful in cases where those descriptions do not refer to anyone. "The President of Australia is in Brisbane" is just as meaningful as "The Prime Minister of Australia is in Brisbane", although Australia has no president. Are we to say that the President of Australia is a non-existent object? Does that even make sense? Do we say that the sentence has no truth value? If we do, won't that contradict the law of the excluded middle? How can we even say (correctly) that the President does not exist without referring to someone who isn't there to be refered to?

Russell's idea is not to give a definition of the phrase "the F", but rather to give a systematic way of paraphrasing every sentence in which that phrase occurs into an equivalent sentence with only ordinary expressions of first order logic with identity. This kind of replacement is known as a contextual definition, because it allows the defined expression to be replaced only in context. Russell says that

The F exists

is really a conjunction. It means that there is one and only one F:

SOMEx Fx   AND   ALLxALLy ((Fx AND Fy) IMP x=y)

This formula asserts existence and uniqueness. It contains no part that directly translates "the F", so it does not get its truth value by referring to a possibly non-existent object but rather by quantifying over the existing ones.

Similarly,

The F is G

is a conjunction of three claims. It means that there is such a thing as the F, and whatever that is is G:

SOMEx Fx   AND   ALLxALLy ((Fx AND Fy) IMP x=y)   AND   ALLx (Fx IMP Gx)

A shorter equivalent formula is

SOMEx (Fx AND Gx)   AND   SOMExALLy (Fy IMP x=y)

or even more succinctly

SOMEx (ALLy (Fy IFF x=y) AND Gx)

Russell thus succeeds in bringing definite descriptions, and hence function symbols, within the scope of ordinary logic with quantifiers and identity, accounting for the semantic difference between them and names. He is also able to answer his own motivating puzzles.

  • Puzzle 1: Let us write Wx to mean that x wrote Waverley, and let s be Scott. Then Russell's analysis of what George wanted to know, i.e. the truth value of the identity statement s = ιx Wx, is the formula

    SOMEx (Wx AND s=x)   AND   SOMExALLy (Wy IMP x=y)

    By elementary logical moves, this is equivalent to

    Ws   AND   SOMExALLy (Wy IMP x=y)

    and indeed to the even simpler formula

    ALLx (Wx IFF x=s)

    That is: Scott wrote Waverley all by himself. Presumably George already knew the second conjunct of the above conjunction, that Waverley was not written by a committee, so all he was really demanding was the truth value of Ws. That is, he just wanted to know whether Scott wrote Waverley, which is surely right. On analysis, according to Russell, what George wanted to know is not really an identity statement at all, so there is no failure of =E or any other rule of ordinary logic.

  • Puzzle 2: The claim that the President of Australia is in Brisbane is a conjunction of three shorter claims: that a President exists, is unique and that whoever is a President is in Brisbane. It is false because its first conjunct is false, so the location of the nonexistent president is not an issue. Again, there is no failure of excluded middle or any other part of standard logic, and no need for a semantic theory involving non-existent objects or other strange things.

Russell at the time thought of his theory as analysing what is really said when people use definite descriptions in languages like English. He saw it as allowing the term-forming operator ' ι' into logic. My preference is a slight modification of Russell's view, on which ' ι' does not form terms at all, but instead is the quantity indicator of a quantifier. That is instead of writing

G( ι x Fx) AND NOT H( ι x Fx)

to represent the claim that the F is G but not H, we write

ι (x: Fx) (Gx AND NOT Hx).

An expression like ι (x: Fx) can count as a quantifier on our general theory of quantification. We could read it "For the thing x such that Fx...". Note that the above formula does not mean that exactly one F is G and not H; it means that there is exactly one F and it is both G and not H. The introduction and elimination rules for a definite description quantifier in the natural deduction calculus will not be specified here, as they are a little complicated (though not deep); their effect will be that the above formula is provably equivalent to Russell's three-way conjunction. A real advantage of this construal over the one taking definite descriptions to be terms is that it is able to represent the ambiguity of negation in the case of non-referring descriptions. On the present view, as Russell points out, a sentence like "The President of Australia is in Brisbane" has two different negations: one could deny the whole sentence, or one could assert that the President is elsewhere than Brisbane. The two formulae

NOT ι (x: Fx) Gx
ι (x: Fx) NOTGx

are not equivalent, yet either could be used to deny that the F is G. This ambiguity is clear and important, but it cannot even be expressed in the vocabulary of ι as a term-forming operator.

Part of the significance of Russell's contribution arose from the way in which it suggests that logical analysis can solve philosophical problems. It seems that the apparent or "grammatical" form of a sentence, the form that we see on the surface, may be different from its deeper "logical" form. The grammatical form may mislead us into philosophical problems or into mistaken world views like those postulating a universe of non-existent objects with nonsensical properties. These philosophical errors disappear, says Russell, when logic shows us the true form of our language.