First-Order Logic (FOL)

aka. predicate calculus

Laura Dillon
Computer Science & Engineering
Michigan State University

Adapted from:
Tuomas Sandholm
Carnegie Mellon University
Computer Science Department

Fall 2014, CSE 814 Overview of Predicate Logic & 1

Automated Reasoning
 Automated Inference in FOL
•  Harder than in PL:
–  Variables can take on potentially an infinite number of possible values
from their domain.
–  Hence there are potentially an infinite number of ways to apply the
Universal-Elimination rules of inference
•  Goedel's Completeness Theorem: FOL entailment is
semidecidable.
–  If a sentence is true given a set of axioms, there is a procedure that will
determine this.
–  If the sentence is false, the procedure may never halt.
•  Goedel's Incompleteness Theorem: In a slightly extended
language (that enables mathematical induction), there are
entailed sentences that cannot be proved.

Fall 2014, CSE 814 Overview of Predicate Logic & 2

Automated Reasoning
 Inference Rules for FOL
Generalized Modus Ponens (GMP)
•  Combines And-Introduction, Universal-Elimination, and Modus
Ponens (assumes free variables are universally quantified)
–  E.g.: from P(5), Q(5), and (P(x) ^ Q(x)) => R(x), derive
R(5)

•  Need notation: term[v1/t1, v2/t2, ..., vj/tj] means to

replace in term all occurrences of variable symbol vi by term ti,
for i = 1, 2, … j.
–  E.g.: P(x)[x/5] stands for P(5)
–  E.g.: eats(y,x)[x/IceCream, y/Ziggy] stands for
eats(Ziggy, IceCream)

Fall 2014, CSE 814 Overview of Predicate Logic & 3

Automated Reasoning
 Inference Rules for FOL
Generalized Modus Ponens (GMP)
•  From: Theta unifies
Qi and Pi
(Q1 ^ Q2 ^ … ^ QN) => R,
P1, P2,…, PN,
and Pi[Theta]= Qi[Theta] for i=1,...,N

where P1,…,PN,Q1,…,QN and R are quantifier-free sentences; Theta =

[v1/t1, …, vj/tj] ; v1,… vj are variables; and t1,… tj are terms.

Infer: R[Theta]. resolvent

•  E.g.: from P(c), Q(c), and (P(x) ^ Q(x) => R(x)),

P(x)[x/c] = P(c)[x/c], and Q(x)[x/c] = Q(c)[x/c], infer R(c)

Fall 2014, CSE 814 Overview of Predicate Logic & 4

Automated Reasoning
 Unification
•  Unification: a "pattern matching" procedure that,
given two literals, returns a substitution that results
in them matching, or "failure”.
–  I.e., unify(P,Q) = Theta
only if P[Theta] = Q[Theta].
–  Moreover, when it succeeds, unification returns a most
general unifier (mgu)
•  To make literals match, the algorithm recursively
replaces variables by terms

Fall 2014, CSE 814 Overview of Predicate Logic & 5

Automated Reasoning
 Unification Algorithm
procedure unify(p, q, theta = [])
Scan p and q left-to-right.
If p and q are equal, return theta (success)
Otherwise, let r and s be the terms in p and q,
respectively, where disagreement first occurs
If r is a variable that does not occur in s, then
assign theta = theta + [r/s] and call
unify(p[theta], q[theta], theta)
else if s is a variable that does not occur in r, then
assign theta = theta + [s/r] and call
unify(p[theta], q[theta], theta))
else return "failure”

Fall 2014, CSE 814 Overview of Predicate Logic & 6

Automated Reasoning
 Example
unify( “parents(x, father(x), mother(Bill))”,
“parents(Bill, father(y), z)”,
[ ] )

unify( “parents(Bill, father(Bill), mother(Bill))”,

“parents(Bill, father(y), z)”,
[“x”/“Bill”] )

unify( “parents(Bill, father(Bill), mother(Bill))”,

“parents(Bill, father(Bill), z)”,
[“x”/“Bill”, “y”/“Bill”] )

unify( “parents(Bill, father(Bill), mother(Bill))”,

“parents(Bill, father(Bill), mother(Bill))”,
[“x”/“Bill”, “y”/“Bill”,
“z”/ “mother(Bill)”] )
Fall 2014, CSE 814 Overview of Predicate Logic & 7
Automated Reasoning
 GMP in Horn FOL
•  Generalized Modus Ponens is complete for Horn clauses
–  A Horn clause is a sentence of the form:
(P1 ^ P2 ^ ... ^ Pn) => Q
where the Pi's and Q are positive literals (includes True)
–  We normally, True => Q is abbreviated Q
–  Horn clauses represent a proper subset of FOL sentences.
E.g., P(a) v Q(a) is a sentence but is not a Horn clause.

Backward chaining
Generalized Modus Ponens

Forward chaining
Fall 2014, CSE 814 Overview of Predicate Logic & 8
Automated Reasoning
 Example of forward chaining
Forward chaining: Start with the hypotheses; repeatedly use GMP to derive
new sentences until the conclusion is derived.
Example:
Given: All cats like fish, cats eat everything they like, and Ziggy is a cat.
Therefore: Ziggy eats fish
Hypotheses in Horn FOL:
1.  cat(x) => likes(x, Fish)
2.  (cat(x) ^ likes(x,y)) => eats(x,y)
3.  cat(Ziggy)
Conclusion in Horn FOL: eats(Ziggy,Fish)
FC GMP Proof:
4. Use GMP with 1, 3, and [x/Ziggy] to derive: likes(Ziggy,Fish)
5. Use GMP with 3, 4, 2 and [y/Fish, x/Ziggy] to
derive eats(Ziggy, Fish)

Fall 2014, CSE 814 Overview of Predicate Logic & 9

Automated Reasoning
 (Slightly) More Interesting Example
Premises:
cat(x) ^ fish(y) => likes(x, y)
cat(x) ^ rodent(y) => likes(x, y)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x)
cat(Ziggy)
fish(Sebastian)
rodent(Mickey)
flies(Robin)
feathers(Robin)

Conclusion:
likes(Ziggy, Robin)

Fall 2014, CSE 814 Overview of Predicate Logic & 10

Automated Reasoning
 Simple FC algorithm
For a set of horn clauses H and atom q
infer(H, q):
if q[theta] = r[theta] for some h in H
and unifier theta, return True;
let R be the set of resolvents
producible from H using GMP;
if R is empty, return False
else return infer(H U R, q);

Fall 2014, CSE 814 Overview of Predicate Logic & 11

Automated Reasoning
 1st call to (FC) infer
H: cat(x) ^ fish(y) => likes(x, y)
cat(x) ^ rodent(y) => likes(x, y)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x)
cat(Ziggy)
fish(Sebastian)
rodent(Mickey)
flies(Robin)
feathers(Robin)

R: likes(Ziggy, Sebastian)
likes(Ziggy, Mickey)
q: likes(Ziggy, Robin) bird(Robin)

Fall 2014, CSE 814 Overview of Predicate Logic & 12

Automated Reasoning
 2nd call to (FC) infer
H: cat(x) ^ fish(y) => likes(x, y)
cat(x) ^ rodent(y) => likes(x, y)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x)
cat(Ziggy)
fish(Sebastian)
rodent(Mickey)
flies(Robin)
feathers(Robin)
likes(Ziggy, Sebastian)
likes(Ziggy, Mickey)
bird(Robin)

q: likes(Ziggy, Robin) R: likes(Ziggy, Robin)

Fall 2014, CSE 814 Overview of Predicate Logic & 13

Automated Reasoning
 Example of backward chaining
Backward chaining: start with the goal query, repeatedly replace goals with
antecedents derived from known implications, working "backwards" until
left only with hypotheses.
Hypotheses in Horn FOL:
1.  cat(x) => likes(x, Fish)
2.  (cat(x) ^ likes(x,y)) => eats(x,y)
3.  cat(Ziggy)
BC GMP Proof of eats(Ziggy, Fish):
4. eats(Ziggy, Fish) matches RHS of 2 with [x/Ziggy,y/Fish], so
replace it with new goals:
cat(Ziggy) and likes(Ziggy, Fish)
5. likes(Ziggy, Fish) matches RHS of 1 with [x/Ziggy], so replace it
with: cat(Ziggy)
6. cat(Ziggy)is a hypothesis, so goals are all proved.

Fall 2014, CSE 814 Overview of Predicate Logic & 14

Automated Reasoning
 (Slightly) More Interesting Example
Premises:
cat(x) ^ fish(y) => likes(x, y)
cat(x) ^ rodent(y) => likes(x, y)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x)
cat(Ziggy)
fish(Sebastian)
rodent(Mickey)
flies(Robin)
feathers(Robin)

Conclusion:
likes(Ziggy, Robin)

Fall 2014, CSE 814 Overview of Predicate Logic & 15

Automated Reasoning
 Simple BC algorithm
H:set of Horn clauses (hypotheses), G:non-empty set of atoms (goals)
W:list of sets of atoms (working list of sets of goals)
infer(H, G, W):
if G = {True}: return True
else:
choose g ( ≠ True) in G; # choice 1
let H’ be the set of clauses in H
whose RHS unifies with g;
if H’ is empty and W is empty: return False
else if H’ is empty: return infer(H, W[0], W[1:])
else:
G’ = G\{g} U p1[th1] # choice 2
W’ = W + [G\{g} U p2[th2], …,G\{g} U pk[thk]
where H’ = { pi => ci }i=1…k and
g[thi]=ci[thi], for i = 1…k
return infer(H, G’, W’)
Fall 2014, CSE 814 Overview of Predicate Logic & 16
Automated Reasoning
 1st call to (BC) infer
H cat(x) ^ fish(y) => likes(x, y) G likes(Ziggy, Robin)
cat(x) ^ rodent(y) => likes(x, y)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x)
W [ ]
cat(Ziggy)
fish(Sebastian)
rodent(Mickey) G’ cat(Ziggy)
flies(Robin) fish(Robin)
feathers(Robin)

g likes(Ziggy, Robin)
W’ cat(Ziggy)
rodent(Robin)
H’ cat(x) ^ fish(y) => likes(x, y)
cat(x) ^ rodent(y) => likes(x, y) cat(Ziggy)
cat(x) ^ bird(y) => likes(x, y) bird(Robin)

Fall 2014, CSE 814 Overview of Predicate Logic & 17

Automated Reasoning
 2nd call to (BC) infer
H cat(x) ^ fish(y) => likes(x, y) G cat(Ziggy)
cat(x) ^ rodent(y) => likes(x, y) fish(Robin)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x) W
cat(Ziggy) cat(Ziggy)
fish(Sebastian) rodent(Robin)
rodent(Mickey) cat(Ziggy)
flies(Robin) bird(Robin)
feathers(Robin)

g cat(Ziggy)

H’ True => cat(Ziggy)

W’ cat(Ziggy)
rodent(Robin)

G’ True cat(Ziggy)
fish(Robin) bird(Robin)
Fall 2014, CSE 814 Overview of Predicate Logic & 18
Automated Reasoning
 3rd call to (BC) infer
H cat(x) ^ fish(y) => likes(x, y) G True
cat(x) ^ rodent(y) => likes(x, y) fish(Robin)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x) W
cat(Ziggy) cat(Ziggy)
fish(Sebastian) rodent(Robin)
rodent(Mickey) cat(Ziggy)
flies(Robin) bird(Robin)
feathers(Robin)

g fish(Robin)

H’

Fall 2014, CSE 814 Overview of Predicate Logic & 19

Automated Reasoning
 4th call to (BC) infer
H cat(x) ^ fish(y) => likes(x, y) G cat(Ziggy)
cat(x) ^ rodent(y) => likes(x, y) rodent(Robin)
cat(x) ^ bird(y) => likes(x, y)
flies(x) ^ feathers(x) => bird(x) W
cat(Ziggy) cat(Ziggy)
fish(Sebastian) bird(Robin)
rodent(Mickey)
flies(Robin)
feathers(Robin)

And so on …

Fall 2014, CSE 814 Overview of Predicate Logic & 20

Automated Reasoning
 Incompleteness of GMP

∀x P(x) => Q(x)

∀x ¬P(x) => R(x) ! Cannot be converted to Horn
∀x Q(x) => S(x)
∀x R(x) => S(x)

For any constant A, S(A)is entailed; but it cannot be

inferred using GMP

Incomplete: there are entailed sentences that the

procedure cannot infer.

Fall 2014, CSE 814 Overview of Predicate Logic & 21

Automated Reasoning
 Resolution Procedure
Recall Resolution Rule for PL:
From P1 v P2 v ... v Pn and
~P1 v Q2 v ... v Qm
infer the resolvent sentence:
P2 v ... v Pn v Q2 v ... v Qm
Examples
–  From P and ~P v Q, derive Q (Modus Ponens)
–  From (~P v Q) and (~Q v R), derive ~P v R (Disj. syl.)
–  From P and ~P, derive False (Contradication)
–  From (P v Q) and (~P v ~Q), derive True (Tautology)
Fall 2014, CSE 814 Overview of Predicate Logic & 22
Automated Reasoning
 Resolution Procedure…
•  Resolution Rule for FOL:
–  From sentences
P1 v ... v Pn and Q1 v ... v Qm
where each Pi and Qi is a literal,
–  If P1 and ~Q1 unify with substitution Theta,
i.e., if P1[Theta] = ~Q1[Theta]
–  Then infer the resolvent sentence:
(P2 v … v Pn v Q2 v … v Qm)[Theta]

•  Example
–  From P(x, f(a)) v P(x, f(y)) v Q(y,x) and
~P(z, w) v ~Q(z,w)
with substitution [x/z, w/f(a)]:
–  Infer:
P(z, f(y)) v Q(y,z) v ~Q(z,f(a))

Fall 2014, CSE 814 Overview of Predicate Logic & 23

Automated Reasoning
 Resolution Procedure…
•  Proof by contradiction method: Refutation Procedure
•  Given a consistent set of hypotheses H and goal sentence Q, we
want to show that H |= Q – ie., that every interpretation I
that satisfies H, satisfies Q.
•  We know that any interpretation I satisfies either Q or ~Q, but
not both.
•  Therefore H |= Q if and only if no interpretation satisfies both
H and ~Q.
•  In other words, (H |= Q) <=> (H ^ ~Q |= False)
•  What's the gain? If H union ~Q is unsatisfiable, then some finite
subset is unsatisfiable.
Fall 2014, CSE 814 Overview of Predicate Logic & 24
Automated Reasoning
 Resolution Procedure…
•  Resolution procedure cannot, in general, be used to generate all
logical consequences of a set sentences.
•  Resolution procedure cannot be used to prove that Q is not
entailed by H.
•  Resolution procedure won't always give an answer since
entailment is only semidecidable. (You can't just run two proofs
in parallel, one trying to prove Q and the other trying to prove
~Q, since H might not entail either one.)

Fall 2014, CSE 814 Overview of Predicate Logic & 25

Automated Reasoning
 Resolution Algorithm
procedure resolution-refutation(H, Q)
;; H is a set of consistent, true FOL sentences
;; Q is a goal sentence that we want to derive
;; return success if H |= Q, and failure otherwise
K = union(KB, ~Q)
while false not in K do
if 2 sentences, S1 and S2, in K contain
literals that unify, then:
resolvent = resolution-rule(S1, S2)
K = union(K, resolvent)
else:
return “failure”

return "success"

Fall 2014, CSE 814 Overview of Predicate Logic & 26

Automated Reasoning
Resolution example (using PL sentences)
•  From “Heads I win and tails you lose” prove that “I win”
•  First, define the hypotheses :
1.  "Heads I win, tails you lose."
(Heads => IWin), i.e. (~Heads v IWin)
(Tails => YouLose), i.e. (~Tails v YouLose)
2.  Add some axioms about coins, winning, and losing:
(Heads v Tails)
(YouLose => IWin), i.e. (~YouLose v IWin)
(IWin => YouLose), i.e. (~IWin v YouLose)

•  Goal: IWin

Fall 2014, CSE 814 Overview of Predicate Logic & 27

Automated Reasoning
 Resolution example (using PL sentences)
~IWin
False
~Heads
~Heads v IWin
Tails

Heads v Tails
YouLose
~Tails v YouLose
IWin
~YouLose v IWin

~IWin v YouLose
Fall 2014, CSE 814 Overview of Predicate Logic & 28
Automated Reasoning
 Problems yet to be addressed
•  Resolution rule requires sentences in clause form:
P1 v P2 v ... v Pn,
where each Pi is a literal (negated or nonnegated
predicate applied to functions, constants, or variables).

•  Resolution strategy
–  How to pick which pair of sentences to resolve?
–  How to pick which pair of literals to unify?

Fall 2014, CSE 814 Overview of Predicate Logic & 29

Automated Reasoning
 Converting FOL sentences to clause form
Steps to convert a sentence to clause form:
1.  Eliminate all <=> connectives:
replace (P <=> Q) by ((P => Q) ^ (Q => P))
2.  Eliminate all => connectives:
replace (P => Q) by (~P v Q)
3.  Reduce scope of negation to a single predicate:
–  replace ~~P by P
–  replace ~(P v Q) by (~P ^ ~Q)
–  replace ~(P ^ Q) by (~P v ~Q)
–  replace ~(Ax)P by (Ex)~P
–  replace ~(Ex)P by (Ax)~P
4.  Standardize variables:
rename so each quantifier has its own unique variable name.
Fall 2014, CSE 814 Overview of Predicate Logic & 30
Automated Reasoning
Converting FOL sentences to clause form…
5.  Eliminate existential quantification: Skolemization
–  If the existential quantifier is within the scope of a universally
quantified variable, introduce a new function that depends on the
universally quantified variable.
•  E.g., (Ax)(Ay)(Ez)P(x,y z) is converted to
(Ax)P(x, y, f(x, y)).
•  (Ax)(Ey)loves(x,y) is converted to (Ax)loves(x,g(x))
where in this case g(x) specifies a person that x loves.
•  f is called a Skolem function, and must be a brand new function name
that does not occur in any other sentence in the entire KB.
–  If not, introduce a new constant
•  E.g., (Ex)P(x) to P(c) where c is a brand new constant symbol that
is not used in any other sentence.
•  c is called a Skolem constant.

Fall 2014, CSE 814 Overview of Predicate Logic & 31

Automated Reasoning
Converting FOL sentences to clause form…
6.  Remove universal quantification symbols:
make scope of each be the entire sentence, so can just “drop” them
•  E.g., convert (Ax)P(x) V (Ay)Q(y) to P(x) V Q(y)

7.  Put in conjunctive normal form: Distribute V into ^

•  E.g., convert (P ^ Q) v R to (P v R) ^ (Q v R)
•  E.g., convert (P v Q) v R to (P v Q v R)

8.  Split each conjunct into a separate clause--a disjunction of literals

9.  Standardize variables apart again so that each clause contains

variable names that do not occur in any other clause

Fall 2014, CSE 814 Overview of Predicate Logic & 32

Automated Reasoning
 Converting FOL sentences to clause form…

Example: Convert the sentence

(Ax)( ~(Ay)( p(f(x,y)) ^ (Ez)(~q(y,z) => p(g(z))) ) ^ p(x) )

1.  Eliminate <=>

Nothing to do here.

2.  Eliminate =>

(Ax)( ~(Ay)( p(f(x,y)) ^ (Ez)(q(y,z) V p(g(z))) ) ^ p(x) )

3.  Reduce scope of negation

(Ax)( (Ey)( ~p(f(x,y)) V ~(Ez)(q(y,z) V p(g(z))) ) ^ p(x) )
(Ax)( (Ey)( ~p(f(x,y)) V (Az)(~q(y,z) ^ ~p(g(z))) ) ^ p(x) )

Fall 2014, CSE 814 Overview of Predicate Logic & 33

Automated Reasoning
 Converting FOL sentences to clause form…

4.  Standardize variables (nothing to do here)

(Ax)( (Ey)( ~p(f(x,y)) V (Az)(~q(y,z) ^ ~p(g(z))) ) ^ p(x) )

5.  Eliminate existential quantification

(Ax)( ( ~p(f(x,h(x))) V (Az)(~q(h(x),z) ^ ~p(g(z))) ) ^ p(x) )

6.  Drop universal quantification symbols

( ~p(f(x,h(x))) V (~q(h(x),z) ^ ~p(g(z))) ) ^ p(x)

7.  Convert to conjunction of disjunctions

( ~p(f(x,h(x))) V ~q(h(x),z) ) ^
( ~p(f(x,h(x))) V ~p(g(z)) ) ^ p(x)

Fall 2014, CSE 814 Overview of Predicate Logic & 34

Automated Reasoning
Converting FOL sentences to clause form…

8.  Create separate clauses

–  ~P(f(x,h(x))) V ~q(h(x),z)
–  ~P(f(x,h(x))) V ~p(g(z))
–  P(x)

9.  Standardize variables

–  ~P(f(x,h(x))) V ~q(h(x),z)
–  ~P(f(y,h(y))) V ~p(g(w))
–  P(v)

Fall 2014, CSE 814 Overview of Predicate Logic & 35

Automated Reasoning
 Example: A resolution graph
~P(f(x,h(x))) V ~q(h(x),z)

~P(f(y,h(y))) V ~p(g(w))
~p(g(w))
[v/f(y,h(y)]
False
P(v) [v/g(w)]

Thus, the original formula is unsatisfiable (its negation

is true).
Fall 2014, CSE 814 Overview of Predicate Logic & 36
Automated Reasoning
Example: Hoofers Club [example from Charles Dyer]

•  Problem Statement: Tony, Shi-Kuo and Ellen belong to

the Hoofers Club. Every member of the Hoofers Club is
either a skier or a mountain climber or both. No mountain
climber likes rain, and all skiers like snow. Ellen dislikes
whatever Tony likes and likes whatever Tony dislikes.
Tony likes rain and snow.

•  Query: Is there a member of the Hoofers Club who is a

mountain climber but not a skier?

Fall 2014, CSE 814 Overview of Predicate Logic & 37

Automated Reasoning
 Example: Hoofers Club…
Translation into FOL Sentences
•  Let S(x) mean x is a skier, M(x) mean x is a mountain climber, and L(x,y)
mean x likes y, where the domain of the first variable is Hoofers Club
members, and the domain of the second variable is snow and rain. We can
now translate the above English sentences into the following FOL wffs:
1.  (Ax) S(x) v M(x)
2.  ~(Ex) M(x) ^ L(x, Rain)
3.  (Ax) S(x) => L(x, Snow)
4.  (Ay) L(Ellen, y) <=> ~L(Tony, y)
5.  L(Tony, Rain)
6.  L(Tony, Snow)

7.  Query: (Ex) M(x) ^ ~S(x)

8.  Negation of the Query: ~(Ex) M(x) ^ ~S(x)

Fall 2014, CSE 814 Overview of Predicate Logic & 38

Automated Reasoning
 Example: Hoofers Club…
•  Conversion to Clause Form
1.  S(x1) v M(x1)
2.  ~M(x2) v ~L(x2, Rain)
3.  ~S(x3) v L(x3, Snow)
4.  ~L(Tony, x4) v ~L(Ellen, x4)
5.  L(Tony, x5) v L(Ellen, x5)
6.  L(Tony, Rain)
7.  L(Tony, Snow)
8.  Negation of the Query: ~M(x7) v S(x7)

Fall 2014, CSE 814 Overview of Predicate Logic & 39

Automated Reasoning
 Example: Hoofers Club…
8. ~M(x7) v S(x7)
[x7/x1]

S(x1)
[x1/x3]
1. S(x1) v M(x1))
L(x3, Snow)
[x4/Snow,
x3/Ellen]
3. ~S(x3) v L(x3, Snow)
~L(Tony, x4)
[x4/Snow,
x3/Ellen]

4. ~L(Tony, x4) v ~L(Ellen, x4)

False

7. L(Tony, Snow)

Fall 2014, CSE 814 Overview of Predicate Logic & 40

Automated Reasoning
 Example: Hoofers Club…
Recall original question:
Given the hypotheses below
1.  (Ax) S(x) v M(x)
2.  ~(Ex) M(x) ^ L(x, Rain)
3.  (Ax) S(x) => L(x, Snow)
4.  (Ay) L(Ellen, y) <=> ~L(Tony, y)
5.  L(Tony, Rain)
6.  L(Tony, Snow)

Can we infer the conclusion: (Ex)(M(x) ^ ~S(x))

When provable, resolution can find a “witness” for x!

Fall 2014, CSE 814 Overview of Predicate Logic & 41

Automated Reasoning
 Example: Hoofers Club…
8. ~M(x7) v S(x7)
[x7/x1]

S(x1)
[x1/x3]
1. S(x1) v M(x1))
L(x3, Snow)
[x4/Snow,
x3/Ellen]
3. ~S(x3) v L(x3, Snow)
~L(Tony, x4)
[x4/Snow,
x3/Ellen]

4. ~L(Tony, x4) v ~L(Ellen, x4)

False

To find a witness:
use the substitutions to rewrite original 7. L(Tony, Snow)
(unnegated) goal:
M(Ellen) ^
M(x7) ^ ~S(x7) M(x1) ^ ~S(x1) M(x3) ^ ~S(x3)
~S(Ellen)

Fall 2014, CSE 814 Overview of Predicate Logic & 42

Automated Reasoning
 Subsumption
•  Motivation: Helps keep K small => less
search
•  Never keep more specific (“subsumed”)
sentences in the K than more general ones
•  E.g., if the K has P(x), then eliminate
–  P(c)
–  P(a) v P(b)

Fall 2014, CSE 814 Overview of Predicate Logic & 43

Automated Reasoning
 Strategies to control the search
•  Breadth-First
–  Compute all level 1 clauses possible, then all possible level 2 clauses, etc.
–  Complete, but very inefficient.
•  Set-of-Support
–  At least one parent clause must be from the negation of the goal or one of the
"descendents" of such a goal clause (i.e., derived from a goal clause)
–  Complete (assuming all possible set-of-support clauses are derived)
•  Unit Resolution
–  At least one parent clause must be a "unit clause," (clause containing a single literal)
–  Not complete in general, but complete for Horn clauses
•  Input Resolution
–  At least one parent from the original hypotheses or the negation of the goal
–  Not complete in general, but complete for Horn clauses
•  Linear Resolution
–  P and Q can be resolved together if P is from the original hypotheses or the negation
of the goal or if P is an ancestor of Q in the proof tree
–  Complete

Fall 2014, CSE 814 Overview of Predicate Logic & 44

Automated Reasoning