Semantics
Semantics
(Version 0.3, revised 6:15pm , April 14, 2005. Please report typos to [email protected].)
The study of formal logic is based on the fact that the validity of an argument
depends only on its logical form; for the purposes of logical investigation, we
can ignore that argument’s content. In particular, if two arguments have the same
logical form, then either they are both valid, or they are both invalid.
Our main goal in this course is to figure out what it means to say that an argument
is valid; or, in other words, to define what a valid argument form is. As an aid
in achieving this goal, we have developed two ways for representing argument
forms: First we developed the propositional calculus, which is supposed to give a
perspicuous representation of the role of (truth functional) sentence connectives in
sentences and arguments. Then we developed the predicate calculus, which also
represents the role of quantifiers such as “all” and “some.”
Now, since the sentences of the propositional and predicate calculi have no content,
they are not, by themselves, either true or false. But our definition of validity makes
use of the notion of truth-preservation:
We now need a notion of interpretation for sentences in the predicate calculus. But
truth tables will not do the trick: in the predicate calculus, we cannot arbitrarily as-
sign truth values to truth-functionally simple sentences. For example, the sentence
(∃x)(F x & − F x) is truth-functionally simple, but we do not want to say that this
sentence could be true! So, an interpretation of a predicate calculus sentence must
start at a more primitive level than the level of truth-functionally simple sentences.
Rather, it must start at the level of names and predicates.
1 Informal Counterexamples
(x)(F x ∨ Gx)
(x)F x ∨ (x)Gx
is not a valid argument form. And if asked to justify your claim, you would most
likely think of examples of predicates to replace F and G so that the premise comes
out as true, and the conclusion comes out as false. For example, you might present
the following argument as a counterexample:
Since this argument has true premises and a false conclusion, it certainly cannot
have a valid form.
What you have done here is to give an interpretation of the argument form. Here
are the details of your interpretation:
(a.) The domain is the natural numbers: 1, 2, . . . (“Domain” is just a fancy name
for the collection of things that you are talking about.)
(b.) The predicate letter F is interpreted as the predicate “is even”, and the pred-
icate letter G is interpreted as the predicate “is odd.”
Here are three numbers: 2, 6, 17. The set of those three numbers is {2, 6, 17}.
That is, we use brackets around the objects to indicate the set that consists of those
objects.
The number 2 is a member of the set {2, 6, 17}. We can indicate this fact by writing
2 ∈ {2, 6, 17}, where “∈” is shorthand for “is a member of.”
Every member of the set {2, 17} is also a member of the set {2, 6, 17}. So, we
say that {2, 17} is a subset of {2, 6, 17}. And if we don’t feel like writing out the
words, we might just write {2, 17} ⊆ {2, 6, 17} as a shorthand.
Notice that is does not make sense to write 2 ⊆ {2, 6, 17}, because 2 is not itself a
set. It does make sense to write {2} ⊆ {2, 6, 17}.
Two sets A and B are equal, written A = B, just in case they have the same
members (and order does not matter). So, for example, {2, 6, 17} and {17, 2, 6}
are the same set.
There is one (and only one) set that does not contain anything at all. This set is
called the empty set, and it is sometimes denoted by ∅.
Question. Is the empty set a subset of {2, 6, 17}?
If A and B are sets, we use A ∩ B to denote the set that consists of those things
that are in both A and B. We use A ∪ B to denote the set that consists of those
things that are in either A or B.
Now we revisit the concept of counterexamples and interpretations, this time with
an eye toward precision and rigor.
I have not yet given a precise definition of when a sentence is true relative to an
interpretation. I have not done so because the definition turns out to be quite com-
plicated, and your intuitions will be sufficient for many cases.
Example. Consider the interpretation with domain X = {1, 2, 3, 4} and
• (∃x)(F x & Gx) is true relative to this interpretation, since the element 3 is
in both Ext(F x) and Ext(Gx).
Solution: Let X = {1, 2}, let Ext(F x) = {1} and Ext(Gx) = {2}. Since all
elements of X are either in Ext(F x) or Ext(F x), (x)(F x ∨ Gx) is true relative to
this interpretation. But since Ext(F x) is not equal to X, (x)F x is not true relative
to this interpretation. Similarly, (x)Gx is not true relative to this interpretation,
and so (x)F x ∨ (x)Gx is not true relative to this interpretation.
First we list the predicate letters in the top row of the table, and we list the elements
of the domain in the first column of the table. We then write “+” in cell (i, j) if
the predicate in column j applies to the object on row i. Otherwise, we write “−”
in cell (i, j).
Solution:
When giving an interpretation, you must be clear and explicit about which subsets
are assigned to which predicate letters. But given that you satisfy the requirements
of clarity and explicitness, you can use whatever method you want to specify your
interpretation.
Note. An interpretation must assign the same subset of the domain to both F x and
F y, even though they have different variables. In other words, the interpretation
gives the extension of the predicate letter F , and doesn’t care about what variable
we put after F . We will deal later with the tricky case of sentences that involve
the same predicate letter with different variables — e.g., (x)(y)(F x → F y) and
(∃x)(F x → (y)F y).
Recall that when we were doing the propositional calculus, we defined a bunch
of special kinds of sentences (e.g., tautologies, inconsistencies), and a bunch of
special logical relationships between sentences (e.g., implies, is subcontrary to).
Actually, each of those concepts was defined in terms interpretations, and so they
can be naturally extended to predicate calculus sentences. For example, a predicate
calculus sentence is tautologous just in case it is true relative to every interpreta-
tion.
Definition.
• Two sentences φ, ψ are logically equivalent if they have the same truth value
relative to every interpretation.
Solution:
Problem. Show that the sentence from the previous problem is consistent (true
relative to some interpretation).
Solution:
Solving a semantic problem (e.g., “is the argument with premises Γ and conclusion
φ valid?”) requires one to check all possible interpretations. But there are infinitely
many interpretations. So, it seems that your next homework assignment will take a
long time to finish!
Joking aside, it is generally quite difficult to figure out if a predicate calculus ar-
gument is valid. Moreover, the difficulty cannot be fully resolved — as it is in the
propositional calculus, where truth tables providing a completely reliable test for
validity. Indeed, logicians have proven that the task of deciding if a predicate cal-
culus argument is valid cannot be reduced to a routine algorithm. (This interesting
result would be discussed in an advanced logic course, such as PHI 312 or PHI
321.)
However, in the special case where our sentences contain only monadic predicates
(i.e., no “relation symbols”), we can transform this hard task into a routine algo-
rithm. That will be our next topic.
7.1 Algorithm A
The truth table method is an algorithm for testing the consistency of sentences.
In the spirit of inflationary terminology, we shall henceforth call the truth table
method “Algorithm A.”
7.2 Algorithm B
Definition. Suppose that we put together some predicate letters with the same
variable using truth functional connectives. For example:
Suppose that we then put a quantifier in front, yielding a sentence. For example:
The result is called a simple monadic sentence. So, a simple monadic sentence is a
sentence with only one quantifier at the very front.
φa11 , . . . , φamm
a1
φm+1 , . . . , φam+1
m
..
.
φan1 , . . . , φanm
If the truth table test says consistent, then output Yes. If the truth table test
says inconsistent, then output No.
7.3 Algorithm C
Before you read this section, you need to know how to transform a sentence into a
“disjunctive normal form” equivalent. Please read Appendix A of Lemmon’s book.
3. Test each disjunct using Algorithm B. If Algorithm B answers Yes for one
of the disjuncts, then output Yes. If Algorithm B answers No for each of
the disjuncts, then output No.
Although Algorithm C always gives the correct answer, it has the drawback that
it does not match the way that we normally reason when we try to decide if some
sentences are consistent. So, here I give you another method to test for consistency.
The small domain method is based on the following fact that has been proven by
paid logicians:
Thus, to test a pure monadic sentence for consistency, do the following: Find a
quantifier-free sentence that is equivalent to the original sentence relative to a do-
main with one object; test this resulting sentence for consistency using ordinary
truth tables. If the resulting sentence is consistent, you are done — the original
sentence is true in that domain. If the resulting sentence is inconsistent, then start
over again with a domain with two individuals. If the resulting sentence is consis-
tent, you are done — the original sentence is true in that domain. If the resulting
sentence is inconsistent, then repeat the procedure in a domain with three individ-
uals, etc., until either you find an interpretation relative to which the sentence is
true, or you conclude that there is no such interpretation. For when you are entitled
to draw the latter conclusion, see the last section.
Suppose that the domain X has only three objects 1, 2, 3. In this case, the universal
statement “(x)F x” is equivalent to a conjunction: “1 is an F , 2 is an F , and 3 is
an F .” Similarly, the existential statement “Something is a F ” is equivalent to the
disjunction “Either 1 is an F , or 2 is an F , or 3 is an F .” In sum, when there are
only finitely many things, we can (by naming each object) translate every simple
monadic sentence into a sentence without quantifiers.
The same sort of equivalences also hold for truth-functional combinations of simple
monadic sentences (i.e., pure monadic sentences). For example, if we take the
standard association of names with numbers:
Solution: Since the main operator is a disjunction “∨”, we separate the original
sentence into (x)(F x → Gx) and (∃x)(Hx & M x). Since these two statements
are quantified statements, we expand them as follows:
Problem. For each of the sentences (x)F x, (∃x)F x → (∃x)Gx, (x)(Gx ∨ Gx),
find a quantifier-free sentence that is equivalent relative to a domain containing two
individuals.
Solution:
(x)F x ≡ Fa&Fb
(∃x)F x → (∃x)Gx ≡ (F a ∨ F b) → (Ga ∨ Gb)
(x)(Gx ∨ F x) ≡ (Ga ∨ F a) & (Gb ∨ F b)
Relative to domains with a finite number of individuals, we can test the consistency
of a sentence of monadic PC using ordinary truth tables: Just translate the sentence
into an equivalent quantifier-free sentence.
Problem. Could (∃x)F x & (∃x) − F x be true relative to a domain with only one
object?
Solution: If there were only one thing in the universe, then (∃x)F x & (∃x) − F x
would be equivalent to F a & − F a, which is a contradiction. So, no; it couldn’t
be true if there were only only object in the domain.
Problem. Could (∃x)F x & (∃x) − F x be true if there were exactly two objects in
the domain?
A truth table test shows that this sentence is consistent — e.g., choose v(F a) = T
and v(F b) = F. So, yes; (∃x)F x & (∃x) − F x could be true if there were two
things in the domain.
Problem. Use the small domain method to show that (x)(F x → −F x) is consis-
tent.
X = {1}, Ext(F x) = ∅.
Problem. Use the small domain method to determine if the following argument is
valid:
1. Ha → (F a → Ga)
2. F a // Ha → Ga
A truth table test shows that if the two premises are true, then the conclusion must
also be true. So, we now try a domain with two members. Relative to a domain
with two members, we get the argument:
v(Ha) = T v(Hb) = F
v(Ga) = T v(Gb) = F
v(F a) = T v(F b) = F
Suppose that you need to determine whether a sentence φ is consistent. You check
a domain with one individual, and φ is false. You check a domain with two indi-
viduals and φ is false. You check a domain with three individuals and φ is false.
Surely you cannot check domains of all sizes! When, if ever, are you entitled to
conclude that there is no interpretation relative to which φ is true? Amazingly, it
has been shown that:
So, to take a specific case, if φ has two predicate letters, then for φ to be consistent,
it must be true in some domain with at most 4 objects! It follows that the small
domain method is a “decision procedure” for the consistency of pure monadic sen-
tences: it will answer any question you have about consistency in a finite amount
of time.
1
Compare with Boolos and Jeffrey, Computability and Logic, p. 250.