Lecture 7 Slide 1/22

ICS/CSE 141: Programming Languages

Unification & Prolog

School of Information and Computer Science

University of California, Irvine

Chris Stork
cstork@ics.uci.edu

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Slide 2/22

Overview

1. Rules and Backtracking

2. Unification

3. Prolog

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Slide 3/22

Rules & Backtracking

All examples refer to the file infer.scm

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Rules & Backtracking Slide 4/22

Rules
We want to be able to say:
(and (parent ?x ?y) (male ?x)) ==> (father ?x ?y)
(and (parent ?y ?x) (male ?x)) ==> (son ?x ?y)
Rules as DB entries
(<- (father ?x ?y) ⇐= head of the rule
(and (parent ?x ?y) (male ?x))) ⇐= body of the rule

(<- (son ?x ?y)

(and (parent ?y ?x) (male ?x)))

Famous Syllogism by Socrates:

1. All men are mortal.
2. Socates is a man.
3. Therefore Socrates is mortal.

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Rules & Backtracking Slide 5/22

Backtracking (depth-first)

bt-ex test from infer.scm. Draw the tree!

(define bt-ex-rules
’(
(<- (a)
(and (b) (c) (d)))
(<- (a)
(and (e) (f)))
(<- (b)
(f))
(e)
(<- (f) ; Another way of stating (f) as a fact;
(and)) ; works since (and) is always true!
(<- (a)
(f))
))

(set! *db* bt-ex-rules)

(test bt-ex
((query-infer ’(and (a) (e)))
(() ()))
)

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Rules & Backtracking Slide 6/22

Recursion & Infinite Search Space

Replace the last rule

(<- (a) (f))

by

(<- (a) (and (f) (a)))

Lazy evaluation would return infinitely many solutions.
What happens if we make the last rule the first rule?

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Slide 7/22

Unification

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 8/22

Beyond Matching

Draw (father homer bart) example!

We need unification to “match” our goal against heads of rules!

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 9/22

Unification

Unification is like our mmatch on steroids. It can “match” terms

that both contain logical variables.
First we will take look at unification in isolation as presented in
Mitchell, Chapter 15. This should then give us the “confidence” that
our implementation works.

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 10/22

Terms
Unification operates on terms, e.g.,

f (x, a), f (g(x), y), h(u)

Terms consist of
• variables: x, y, z, . . . (our “logical variables”)
• function symbols: f, g, h, . . . (our “relations”)
• parethesis and comma
Terms are inductively defined as
• a variable
• if f is an n-ary function symbol and t1 , . . . tn are terms, then
f (t1 , . . . tn ) is a term
Constants (0-ary functions) are written as a, b, c, . . . (our “objects”
or (f))

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 11/22

Example: Unification

Examples of Unification:

• a=a ⇒ yes, []
• f (b) = f (b) ⇒ yes, []
• f (a) = f (b) ⇒ no
• f (x) = f (2a) ⇒ yes, [?x=a]
• f (x) = f (y) ⇒ yes, [?x=?y]
• f (x, y) = f (a) ⇒ no
• f (x, y) = f (a, x) ⇒ yes, [?x=a,?y=a]
A variable can only have one value for a given unification to
succeed.

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 12/22

Unification Algorithm

Given a set of equations to unify, repeat

1. f (s1 , . . . , sn ) = f (t1 , . . . , tn ) replace by s1 = t1 , . . . , sn = tn

2. f (s1 , . . . , sn ) = h(t1 , . . . , tn ), f 6= h halt with failure

3. x=x delete equation

4. t = x, t not a variable replace by x =t

5. x = t, x not in t and x elsewhere substitute [x/t] everywhere

6. x = t, x in t and x 6= t halt with failure

until failure or no match anymore!

This is proven to terminate and find the most general unifier (MGU). (So if
we implement this we can be sure to do the “right thing”.)

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 13/22

More Examples

Example: Consider {f (x, a) = f (b, y)} for unification

• Use (1) to get {x = b, a = y}
• Use (4) to get {x = b, y = a}
• Nothing else applicable, unifies using [x/b] and [y/a]
Example: Consider {f (x, a) = f (g(z), y), h(u) = h(d)}
• Use (1) on first to get {x = g(z), a = y, h(u) = h(d)}
• Use (1) on third to get {x = g(z), a = y, u = d}
• Use (4) on second to get {x = g(z), y = a, u = d}
• Nothing else applicable, unifies using [x/g(z)], [y/a], [u/d]

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 14/22

More Examples
Consider {f (x, a) = f (g(z), y), h(x, z) = h(u, d)}
• Use (1) on first to get {x = g(z), a = y, h(x, z) = h(u, d)}
• Use (5) with first to get
{x = g(z), a = y, h(g(z), z) = h(u, d)}
• Use (1) on third to get {x = g(z), a = y, g(z) = u, z = d}
• Use (5) with fourth to get
{x = g(d), a = y, g(d) = u, z = d}
• Use (4) on second to get
{x = g(d), y = a, g(d) = u, z = d}
• Use (4) on third to get {x = g(d), y = a, u = g(d), z = d}
• Nothing else applicable, unifies using
[x/g(d)], [y/a], [u/g(d)], [z/d]

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 15/22

Our Unification Implementation

Our implementation mirrors Apt’s, except that we put the equations

with a single logical variable on the left-hand-side in our frame.
(define (query-infer q)
(query-infer-rec q ’()))

;; Like query-rec except that it calls the infer versions and

;; infer-rec-db calles unify...
(define (query-infer-rec q frame)
(case (car q)
((and) (query-infer-and (cdr q) frame))
((or) (query-infer-or (cdr q) frame))
((not) (query-infer-not (cadr q) frame))
(else (infer-rec-db q frame))))

Looks like query! Difference lies in infer. Look at code: first

simpler-but-wrong-infer and then infer.

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Unification Slide 16/22

Interesting Stuff

Look at append-to test case. Surprise, we can reason

“backwards”!
(We can also deal with lists and recursion in our rules. How come
this works? Demonstrate improper list syntax for define and look
at unify’s implementation.)
There is many more interesting extensions to our implementation:

• lazy evaluation with delay and force (approximating

Prologs real behavior)

• add Scheme expressions to be allowed as terms in rules

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Slide 17/22

Prolog

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Prolog Slide 18/22

Prolog

We just constructed the core of a programming languages called

Prolog!
Stands for programming in logic
Not based on abstract machines or mathematical functions
But on a version of predicate logic instead
Prolog is a complete programming language with

• arithmetic
• lists
• special control facilities, most notably cut

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Prolog Slide 19/22

How we differ from Prolog

• Syntax:
– variable names: Prolog uses capitalized names for logical
variables
– facts written as:happy.
– implications as: happy :- moreFood. (called
clauses)
• we use Scheme lists instead of Prologs special [H|T] lists
Prolog interactive queries:
append([],Xs,Xs).
append([X|Xs],Ys,[X|Zs]) :- append(Xs,Ys,Zs).

> append([1,2],A,[1,2,3,4])?
A=[3,4].

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Prolog Slide 20/22

Logic Programming Languages

It’s not the same as Prolog!! Much has changed since then. You
only got a taste of what’s out there.

Check out:

• Jess (Lisp & expert system (Rete algorithm) written

in Java)

• Mercury (descendent of Prolog with types and more)

• Kanren (declarative logic programming system
embedded in Scheme!)

• Constraint-based programming . . .

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Prolog Slide 21/22

What makes a language declarative

Functions are equalities, we just chose to read them one way! Take
any of our purly functional examples, e.g. app.
Pure functions, rules, and relations define, or better, declare
relationships!

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork
Lecture 7 Prolog Slide 22/22

11 Example: Unification
List of Slides
12 Unification Algorithm
1 Lecture 7
13 More Examples
1 Title
14 More Examples
2 Overview
3 Rules & Backtracking 15 Our Unification Implementation

4 Rules 16 Interesting Stuff

5 Backtracking (depth-first) 17 Prolog

6 Recursion & Infinite Search Space
18 Prolog
7 Unification
19 How we differ from Prolog
8 Beyond Matching
20 Logic Programming Languages
9 Unification
10 Terms 21 What makes a language declarative

ICS/CSE 141: Programming Languages Revision 1.5 (August 24, 2004)

Summer Session II, 2004 Copyright c Christian H. Stork