Lecture 8 of Artificial Intelligence

First Order Predicate Logic

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/1

Topics of this lecture
• Predicate logic • 述語論理
• Term and logic formula • 項と論理式の定義
• Clausal form/Conjunctive • 節形式/連言標準形
canonical form • 論理式の標準化
• Standardization of logic • 単一化と導出
formula • ホーン節とプロログ
• Unification and resolution
• Horn clause and Prolog

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/2

Predicate logic (述語論理)
• In propositional logic, the smallest unit is a proposition.
• Using propositional logic, it is difficult to represent
knowledge like “For anyone, if he/she is a Japanese,
he/she is a human”.
• In predicate logic, the subject and predicate are further
divided. For example,
– Daisuke is a Japanese  J(Daisuke)
– Daisuke is a human  H(Daisuke)
• This division is convenient to represent knowledge that is
generally or sometime useful.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/3

1st order predicate logic (第一階述語論理)

• Example of generally useful (valid) knowledge

– ∀x J(x)⇒H(x)：
For any x, if x is a Japanese, x is a human.
• Example of knowledge that is sometimes useful
– ∃x H(x)⇒J(x)
For some x, x is a human, x is a Japanese.
• In the 1st order predicate logic, variables can only be
“individuals” in the universe of discourse (対象領域).
• For high order predicate logic, predicates and functions
can also be variables.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/4

 Terminologies
• Domain or universe of discourse
– Denoted by 𝐷. Examples include
– Sets of students, human, living being.
– Sets of prime numbers, integers, real numbers.
– Sets of machines, robots, systems.
• Individual constant (個体定数) in 𝐷：
– a, b, c, Taro, Hanako
– Used in the same way as in propositional logic
• Individual variable(個体変数) in 𝐷：
– x, y, z
– Used to represent something not yet fixed.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/5

Terminologies
• Functional symbol (関数記号):
– Map from 𝐷 to 𝐷: find individual(s) from given individual(s).
– Ex1: y=mother(x): Find the mother of x, and put the result into y.
– Ex2: brother(x): Find the brother(s) of x.
• Predicate symbol (述語記号):
– Map from 𝐷 to {True, False}: from individuals to a logic value
– Ex1: Human(x): True if x is a human.
– Ex2: Mother(x, y): True if y is the mother of x.
• Example 3.5 (p. 49)
– a=“Chieko”, b=“Makoto”;
– Known: Chieko is the mother of Makoto
– Mother(b,a) is True;
– mother(b) is a=“Chieko”.
Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/6
 Terminologies
• Logical symbol (論理記号)：
– ∧、∨、￢、 ⇒,⇔
– The same as used in propositional logic.
• Quantifier (限量記号)：
– Here, we consider only the universal quantifier (全称
記号)”∀” and the existential quantifier (存在記号)”∃”.
– Ex1: ∀x H(x)⇒C(x):
• For any x, if x is a human, x is “cybernetic”.
– Ex2: ∃x H(x)⇒G(x):
• There exists an x, if x is a human, x is “genius”.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/7

Definition of term
(1) Any individual (constant or variable) in 𝐷 is a term (項).
(2) If t1,t2,…,tn are terms and f is an n argument function,
f(t1,t2,…,tn) is also a term.
(3) Only expressions obtained using (1) and (2) are terms.

• Examples of term:
– Individual constants: Taro, Jiro, Chieko
– Individual variables: x, y, z in 𝐷
– Function: mother(Makoto), brother(Jiro)

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/8

Definition of atomic formula (素式)
• If t1,t2,…,tn are terms, and P is a predicate symbol
representing the relation between these terms,
P(t1,t2,…,tn) is an atomic logic formula.
• Examples of atomic formulas
– Human(a): True if a is human
– Parent（b, mother(a）): True if mother(a) is (also) the
parent of b.
– Mother(a, mother(a)) is always True.
• Atomic formulas are the smallest units in the predicate
logic.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/9

Definition of logical formulas
(1) An atomic formula is a logic formula.
(2) If P and Q are logic formulas, ￢P, P∧Q, P∨Q, P⇒Q,
and P⇔Q are logic formulas.
(3) If P is a logic formula, and x is an individual variable,
∀x P and ∃x P are logic formulas.
(4) Only expressions defined by (1), (2), and (3) are logic
formulas.

As in propositional logic, logic formulas defined above

are called well-formed formulas (wff).

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/10

The clausal form
• An atomic formula and its negation are called literals.
The disjunction (OR) of the literals is called a clause.
• If Ci (i=1,2,…,n) are clauses, the closed formula (閉式)
given below is called a clausal form:
Qx1Qx2…Qxm[C1∧C2∧…∧Cn]

• Here, Q is a quantifier symbol, and x1,x2,…,xm are

individual variables contained in the logic formula.
• A “closed formula” is a formula in which all variables
are bounded (束縛される) by the quantifiers. That is,
there is no free variables (自由変数).
• The logic formula in the bracket is called a matrix (母式).

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/11

Process for converting a logic
formula to the clausal form
• Any predicate logic formula can be converted to the
clausal form as follows:
– Remove the equivalence and implication symbols;
– Move the negation symbol immediately before the atomic
formula;
– Remove ambiguities of the variables;
– Remove the existential quantifier ∃ using Skolem constant or
Skolem function;
– Remove all universal quantifiers;
– Use distributive laws.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/12

Process for converting a logic
formula to the clausal form
• Remove the equivalence and implication symbols
– P⇒Q = ￢P∨Q

• Move the negation symbol immediately before the atomic formula

– ￢(￢P) = P
– ￢(P∧Q) = ￢P∨￢Q, ￢(P∨Q) = ￢P∧￢Q
– ￢∀xP(x) = ∃x(￢P(x)), ￢∃xP(x) = ∀x(￢(P(x))

• Remove ambiguities of the variables

– ∀xP(x)∨∀xQ(x) = ∀xP(x)∨∀yQ(y)

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/13

Process for converting a logic
formula to the clausal form
• Remove the existential quantifier ∃ using Skolem
constant or Skolem function
– ∃x P(x) = P(s1)
– ∀x∃y P(x,y) = ∀x P(x,s2(x))

– Ex１：∃x Fly(x) = Fly(X)

• There is an x, x can fly  For some X, X can fly.

– Ex２：∀x∃y Love(x,y) = ∀x Love(x,love(x))

• For any x, there is a y, y loves x 
For any x, Y=love(x) loves x
• Pay attention to the difference between Love() and love().

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/14

Set of clauses
• A logic formula in clausal form is equivalent to a set of clauses. In
turn, this set can be considered a knowledge base in which each
clause is a “rule”.
1 ￢H(x1)∨M1(x1)
2 ￢B1(x2)∨M1(x2)
3 ￢M1(x3)∨￢E(x3)∨C(x3)
4 ￢M1(x4)∨￢S1(x4)∨￢S2(x4)∨C(x4)
5 ￢C(x5)∨￢B2(x5)∨￢B3(x5)∨L(x5)
6 ￢C(x6)∨￢B2(x6)∨￢M2(x6)∨F(x6)
7 H(a)
8 B2(a)
9 B3(a)
Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/15
A generalized rule for reasoning
• Suppose there are two clauses C1 and C2
which contain a literal P and its negation ￢P,
respectively.
• From C1 and C2, we can derive a new clause R
which contains all literals of C1 and C2 except P
and ￢P.
• C1 and C2 are called the parent clauses, and
R is called the resolvent clause.
• The above process is called resolution (導出）.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/16

Example 3.8 p. 55
• C１=￢S(a)∨H(a); S(x): x is a student;
H(x): x is a human;
• C2=￢H(x)∨M(x) M(x): x is mortal

• R=￢S(a)∨M(a) The left resolution process is actually

equivalent to syllogism （三段論法）.

Here, we have substituted a into x. Through this substitution,

we can convert H(a) and H(x) into the same form. This is
called unification (単一化）.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/17

More about unification
• The purpose of unification is to enable “pattern matching”
between two compliment literals.
• If the two compliment literals are the same from the
beginning, it is not necessary to conduct unification.
• In the process of automatic inference (reasoning), one of
the literals is often more “general” than its negation.
• Unification often makes the literals more specific.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/18

Examples 3.9 and 3.10 pp. 55-56
• Unification of F(x,q) and F(p,y):
– F(x,q) and F(p,y) can be unified by defining x=p and y=q.
– This unification can be denoted by {x/p,y/q}, and it is called a
unifier.

• Unification of Q(f(b),x) and Q(y, a)

– Q(f(b),x) and Q(y, a) can be unified by defining x=a and y=f(b)
– The unifier is {x/a,y/f(b)}.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/19

Prove by refutation (反駁証明）
• To prove whether a given formula is a theorem
or not, we will have to show that the formula is
true for all possible individuals in 𝐷. This is
usually very difficult.
• A more efficient way is to prove by refutation.

That is, if we add the negation of the formula into the

set of clauses, and if we can derive a contradiction, the
formula cannot be refused and it must be true.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/20

 Example 3.11 pp. 56-57
• Suppose that a set of clauses is given in Table 3.8.
Prove E(a)⇒L(a).

• Proof: The formula to prove is actually P=￢E(a)∨L(a),

and its negation is ￢P=E(a)∧￢L(a).
• Since ￢P is the AND of two formulas, we should add
both E(a) and ￢L(a) to the set of clauses, in order to
conduct proof through refutation.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/21

 Example 3.11 pp. 56-57
• C1：￢L(a) (added clause)
• C2：￢C(x5)∨￢B2(x5)∨￢B3(x5)∨L(x5) (the 5th clause)
• R1: ￢C(a)∨￢B2(a)∨￢B3(a) unifier: {x5/a}
• C１: R1 obtained above
• C2： B3(a) （the 9th clause）
• R2：￢C(a)∨￢B2(a) (unification not needed)
• C1： R2 obtained above
• C2： B2(a) （the 8th clause）
• R3: ￢C(a) (unification not need)
• C１：R3 obtained above
• C２：￢M(x3)∨￢E(x3)∨C(x3) （The 3rd clause）
• R4：￢M(a)∨￢E(a) unifier：{x3/a}

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/22

 Example 3.11 pp. 56-57
• C１：R4 obtained above
In fact, this proof is
• C２：E(a) （added clause or fact) similar to backward
• R5：￢M(a) (unification not needed) reasoning. That is, to
• C１：R5 obtained above verify that an animal is
• C２：￢H(x1)∨M(x1) （the 1st clause） lion if it eats meat,
based on the set of
• R6：￢H(a) unifier: {x1/a}
clauses, which include
• C1：R6 obtained above both the knowledge
• C2：H(a) （The 7th clause） and the observations.
• R7: F (unification not needed)

F is the empty clause, means a contradiction!

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/23

Horn Clause (ホーン節)
• A clause can be generally represented as follows:

￢A1∨ ￢A2∨…∨￢Am∨B1∨B2∨…∨Bn

• If n <= 1, the clause is called a Horn clause; if n=1, the clause is a

definite clause; if n=0, the clause is a goal clause; if m=0, the
clause is a fact.
• Using implication, a Horn clause equals to
A1∧A2∧…∧Am ⇒ B

• To simplify the proof process, it is convenient to put the

consequence to the left side (to improve the efficiency of formal
proof).

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/24

Horn Clause
• If the premise is the disjunction of several literals, the formula can be
separated to form several Horn clauses. For example:

A1∨A2 ⇒ B  A1 ⇒ B and A2 ⇒ B

• On the other hand, if there are several consequences for a premise,

that is, the consequence is the conjunction of several literals, the
formula can be separated into several Horn clauses. For example:

A ⇒ B1∧B2  A ⇒ B1 and A ⇒ B2

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/25

Selective linear resolution for definite clause
(SLD:選択的線形導出)
1) Assign the goal clause (to be proved) to the first parent
clause C1.
2) Find a C2 from the set of clauses that contains a
positive literal unifiable with the left literal of C1.
3) Obtain a resolvent clause R from C1 and C2. If R is the
empty clause, stop; otherwise, put R to C1 and return to
2).

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/26

Capability of the linear approach
• In general proof based on the above linear strategy is
not complete.
• That is, even if a goal clause is not satisfiable, we may
not get the empty clause when the program stops.
• To solve the problem, it is necessary to use more
general search algorithms like depth-first or breadth-first
search.

The process of formal proof is

actually search. SLD is nothing but
a heuristic for efficient search.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/27

Programing based on Prolog
• Prolog is a non-procedural programming language. It
belongs to logic programing language.
• In fact, Prolog is the short for programming in logic.
• Prolog was created around 1972 by A. Colmerauer with
Philippe Roussel.
• In Prolog, a Horn clause is represented by
– B:-A1,A2,…,An
– This is equivalent to A1∧A2∧…∧An⇒B
• The reason to put B on the left is to improve the
reasoning efficiency.
• “⇒” is replaced by “:-”, and “∧” is replaced by “,” to
make it easier to input the symbols using keyboard.
Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/28
 An example of Prolog program
Table 3.10, p. 60
Program Meaning
parent(sofu,otosan). sofu is the parent of otosan.
parent(msofu,okasan). msofu is the parent of okasan.
parent(otosan,miho). otosan is the parent of miho.
parent(otosan,taro). otosan is the parent of taro.
parent(otosan,jiro). otosan is the parent of jiro.
parent(X,Y):-married(Z,X),parent(Z,Y). If Z and X are husband and wife, AND Z
is the parent of Y, X is the parent of Y.
married(otosan,okasan). otosan and okasan are husband and
wife.
married(msofu,msobo). msofu and msobo are husband and wife.
married(sofu,sobo). sofu and sobo are husband and wife.
ancestor(X,Y):-parent(X,Y). If X is the parent of Y, X is the ancestor of
Y.
ancestor(X,Y):- If X is the parent of Z, AND Z is the
parent(X,Z),ancestor(Z,Y). ancestor of Y, X is the ancestor of Z.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/29

To run the test program
swipl
% omitted several lines here

1 ?- [test].
Save the prolog program given in the
true.
previous page to ‘test.pl’, and then run
2 ?- parent(X,taro). ‘swipl’ in the university environment.
X = otosan ; If you use your own PC, please download
X = okasan ; the free software first.
false.
URL: http://www.swi-
3 ?- ancestor(sofu,X). prolog.org/pldoc/man?section=quickstart
X = otosan ;
X = miho ;
X = taro ;
X = jiro ;
false.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/30

 Homework for lecture 8 (1)
(submit to the TA during the exercise class)
• Solve Problem 3.5 given in p. 59 in the textbook.
• Table 3.8 is a set of Horn clauses.
• Prove L(a) using the SLD strategy.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/31

 Homework for lecture 8 (2)
• Write a Prolog programming based on the following set of clauses:

p(X,Y) ∨ ¬q(X) ∨ ¬r(Y) Note 1: A ⇒ B = (¬A ∨ B)

q(X) ∨ ¬s(X) Note 2:
r(Y) ∨ ¬t(Y) g(X)∧h(Y) ⇒ f(X, Y) becomes
f(X, Y):- g(X), h(Y) in Prolog.
s(a)
t(b)

• Save your program into “prog_08_1.pl”, and prove p(a, b) using this
program.
• Try the program given in p. 29 by saving the program into
“prog_08_02.pl”.

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/32

 Quizzes of today
• In 1st order predicate logic, [________] • We can use [__________________]
can be a variable. constant to remove existential
quantifiers in a formula.
• In predicate logic, the return value of a
function is [_____________________]. • Unification is used to make two literals
the same form. After unification, we
• To specify the scope of a variable, we can derive a resolvent clause from two
use quantifiers. The universal [_________________] clauses.
quantifier is denoted by [________].
• A Horn clause is a clause with
• In predicate logic, we have “terms” and [_____________________________]
“atomic formulas”. The smallest unit of positive literal.
a logic formula is [________________].

• Prolog is a short of
[_____________________________].

Produced by Qiangfu Zhao (Since 2008), All rights reserved © AI Lec08/33