Announcements

Prolog Read Sections 10.2-10.3, and Chapter 11 in AI: A

Modern Approach for Monday

Homework #2 is due on Monday

– Handin directories are already set up, in case you’re
Burr H. Settles ready to turn it in
CS-540, UW-Madison
www.cs.wisc.edu/~cs540-1 Homework #3 will go out on Monday
Summer 2003 (but it won’t be due until after the midterm)

1 2

Logic Programming Logic Programming

Computation as inference on logical KBs Logic programs are also called expert systems
Ordinary Programming Logic Programming – Problem domain experts sit down and encode lots of
1. Identify problem Identify problem
information into the KB
2. Assemble information Assemble information – System then reasons using that “expert” knowledge
3. Plan out your algorithms Go have a beer! – Used for medical diagnosis, Q/A systems, etc.
4. Program the solution Encode information in the KB – KB must be in datalog format: FOL with no functions
5. Encode problem as data Encode problem as logical facts
6. Run program on data Ask queries
7. Debug errors Find/correct false facts Fall loosely under the “think rationally” quadrant
of AI research
Should be easier to debug capital(NewYork,USA) than find that pesky x += 2!
3 4

Prolog Prolog Execution

Prolog is probably the most common logic To Solve a Goal (i.e. answer a query)
programming language Try to unify:
A program is: – First with each of the ground facts
– A set of logic sentences in HNF (definite clauses) – When all facts fail, then with each of the consequents
• Called the database (DB, basically the KB) of the rules in the order in which they occur in DB
• Ordered by programmer (top to bottom)
Successful unification with a fact:
– Executed by specifying a query to be proved – Solved, pop goal from stack
• Backward-chaining with GMP
• Uses DFS on the ordered facts and rules Successful unification with a rule:
• Searches until a solution is found (or times out) – Solve the sub-goals in DFS manner (i.e. recursively
– Can find multiple solutions attempt to solve each of the rule’s premises)
5 6

1
Prolog Execution Prolog Execution

Backtracking During DFS Efficient implementation

While solving a rule: – Unification use “open coding”
– Retrieval and matching of clauses by direct linking
– If antecedent (premise) fails to be proved true
– Sophisticated memory management
– Then try to re-prove it using different facts or rules
Uses a closed world assumption
When a rule fails: – Negation as failure
– If an antecedent can’t be solved at all, the rule fails – e.g. given ¬dead(X) alive(X)
– Go on to the next rule in the program and try again alive(elvis) succeeds if dead(elvis) fails
(try to unify current goal with a different consequent)
Widely used in Europe and Japan

7 8

Basic Prolog Syntax Prolog Example

Database: Example prolog DB that encodes our “criminal” example

– Fact: a positive literal (atom) from last time (note that variables are capitalized, and
FOL: F(x) Prolog: f(X).
constants in lower-case):
FOL: ¬F(x) Prolog: not f(X). missile(m).
variables are capitalized and universally quantified owns(nono,m).
enemy(nono,america).
– Rules: 1 positive literal (the consequent, or head), and american(west).
at least 1 negative (the antecedents) weapon(X) :- missile(X).
FOL: A1 ∧ A2 ∧ … ∧ An C Prolog: C:- A1, A2, … , An. sells(west,X,nono) :- missile(X),owns(nono,X).

hostile(X) :- enemy(X,america).
Query: criminal(X) :-
– FOL: Q1 ∧ Q2 ∧ … ∧ Qn Prolog: ?- Q1, Q2, … , Qn. american(X), weapon(Y),
sells(X,Y,Z), hostile(Z).
9 10

Prolog Example Prolog Example

The implementation of prolog that we’ll use on the TUX Once YAP is running and the criminal KB is


machines is called YAP loaded, we can start by asking simple queries we

– “Yet Another Prolog”
clearly already know:
– Freeware implementation, downloadable from
www.ncc.up.pt/~vsc/Yap/ ?- missile(m). ?- american(west).
To run prolog on a TUX machine, type: Then we can move on to more complex queries:


% yap
To end prolog, type:


?- halt. ?- weapon(X). ?- owns(X,m).



To load a file, type: ?- [file]. ?- criminal(west).



The prolog extension is *.pl To view the entire BC search, YAP has a
– Try not to confuse it with perl programs debugging feature called “spy”:
– Note that you don’t need the extension when loading a program – Type spy(predicate). to turn it on
into prolog, it knows to look for the file with a *.pl extension
11
– And nospy(predicate). to turn it off 12

2
Another Prolog Example Another Prolog Example

Let’s consider a simple KB that expresses facts How should we define the relation sibling?
about a certain family: – Two people are siblings if they have the same mother
father(tom,dick). mother(tom,judy). and the same father (ignoring half-siblings, step-
father(dick,harry). mother(dick,mary). siblings, etc.)
father(jane,harry). mother(jane,mary).
How about this:
Now let’s also think about creating some FOL
sibling(X,Y) :- mother(X,M), mother(Y,M),
rules for defining family relations: father(X,M), father(Y,M).
– Parent? Let’s run this and see what happens!
parent(X,P) :- mother(X,P).
parent(X,P) :- father(X,P).
– Oops! Need to make sure X ≠ Y!
sibling2(X,Y) :- mother(X,M), mother(Y,M),
– Grandmother? father(X,M), father(Y,M), X\=Y.
granny(X,G) :- parent(X,Y), mother(Y,G).
13 14

More Prolog Syntax List Processing in Prolog

Prolog has built-in operators (predicates) for mathematical Suppose we want to define an “append” operator for
 

functions and equalities: lists… that is to take two lists L1 and L2, and merges their
– x = 2×(y+1) X is 2*(Y+1). elements together into a new list L3
– d < 20 D < 20. – Usually this is done with a function
– 1≤2 1 @=< 2. • e.g. L3 = append(L1,L2)
– x=y X = Y. – But prolog programs are datalog: no functions allowed!
– x≠y X \= Y. • Create make-shift functions by defining predicates with the return
value included as a parameter
The major data structure for Prolog is the list


• e.g. append(L1,L2,L3)
– [] denotes an empty list
How about defining a simple predicate that takes the first


– [H|T] denotes a list with a head (H) and tail (T)

two L1 and L2, and returns a new list [L1|L2]?
• The head is the first element of the list
– e.g. app(L1,L2,[L1|L2]).
• The tail is the entire sublist after it
• e.g. for the list [a,b,c,d]… H=[a] and T=[b,c,d] – Nope! Let’s try again…
15 16

List Processing in Prolog List Processing in Prolog

What we need to do is take one list and recursively Now we can ask the queries:
add one element at a time from the other list, until ?- append([1,2,3], [a,b,c], [1,2,3,a,b,c]).

we’ve added them all • Result: yes

?- append([1,2,3], [a,b,c], X).
Let’s assume that we start with L2 and want to add • Result: X = [1,2,3,a,b,c]
the elements from L1 one at a time to the front ?- append(A, B, [1,2]).
– Makes things easier: with [H|T], H is the front element • Result: A=[] B=[1,2]
A=[1] B=[2]
– What is our base case? A=[1,2] B=[]
• append([],L2,L2).
– Now how do we deal with the recursive aspect? Recall that, since prolog uses BC, we can try to
• append([H|T],L2,[H|L3]) :- append(T,L2,L3). find any single solution, or find all solutions
– After each result, type “;” to view another
17 18

3
Partitioning Lists Partitioning Lists

Another useful application might be how to First, let’s think of the base case for our
recursively sort prolog lists partitioning predicate
partition(E,[],[],[]).
Most sorting algorithms utilize some partitioning • That is, an empty list gets split into two empty lists
method, where the list L is split into two sublists
Second, we must consider the recursive aspect:
L1 and L2 based on a particular element E – Upon considering a new element H at the head of the
– e.g. splitting list [1,5,3,9,7,4,1] on element [5] would list, what conditions must we account for?
yield the lists [1,3,4,1] and [5,9,7] • If H < E, or if H ≥ E (to determine which sublist)
This would be a useful method to define first – Since we have two different cases, each with a different
desired result, we need two recursive definitions
partition(E, L, L1, L2).

19 20

Partitioning Lists Sorting in Prolog



If H < E, then we want to add H to the first list L1: Now that we know how to partition one list into
partition(E,[H|T],[H|T1],L2) :- two, and also how to append two lists together, we
H < E,
partition(E,T,T1,L2). have all the tools we need to sort a list!


However, if H ≥ E then we’ll add it to the second list L2: Let’s consider insertion sort:
partition(E,[H|T],L1,[H|T2]) :- – Walk through each position of the list
H @>= E,
partition(E,T,L1,T2).
– For each position, insert the list item i that belongs in
that position, relative to other items in the list
These predicates, together with the base case, will partition


– Recursively, we can achieve the same effect by walking

all the list items less than E in the first list, and all greater through each i, partitioning a pre-sorted list on i, and
or equal in the second list then appending the partitions on either side

21 22

Sorting in Prolog Parsing with Prolog

As always, we will need a base case for insertion sort A lot of early natural language processing (NLP) research
 

(assume that an empty list is sorted): was historically done using logic systems, because HNF
isort([],[]). rules are analogous to grammar productions


For the recursive aspect, we can walk through the whole – e.g. A simple English grammar: S → NP VP (NP)
list, and backtrack, inserting each element where it belongs • S means “sentence,” NP means “noun phrase,”
and VP means “verb phrase”
in the pre-sorted list:
isort([H|T], F) :-
– In prolog:
s(Input) :- np(Input, Mid), vp(Mid, []).
isort(T,L), s(Input) :- np(Input, Mid1), vp(Mid1, Mid2), np(Mid2, []).
partition(H,L,L1,L2),
Once we add definitions for np and vp (and ultimately


append(L1,[H|L2],F).
noun, verb, prep, det, etc.), we will have a full-blown
We can do something similar to implement quicksort, but


deterministic English parser!

I’ll leave that up to you to work out on your own!
23 24

4
Ordering Prolog Rules Other Logic Systems

The rules in a prolog program are searched depth- Production Systems



– Proposed by E.L. Post in 1943

first, exploring the potential rules from top down – Equivalent in computational power to a Turing machine.
Imagine that we are designing a knowledge-based • Rules are unordered, unlike Prolog
– Have been developed for a wide variety of problems, ranging from
reflex agent that has multiple rules which which it algebra word problems, mathematical and logical proofs, physics
can unify problems, and games.
– We want to make more specific rules toward the top, – Newell and Simon (1960s) used production systems to define
model human cognition
and more general rules toward the bottom • Production rules represent problem-solving skills stored in a person’s
– For recursive “functions,” that means making sure the long-term memory.
– Many other groups have tried to develop similar models of human
base case comes before the recursive case(s) cognition using production systems
– Harder to do inference than BC in Prolog
25 26

Other Logic Systems Summary



Semantic Networks Logic programs are a agent programs that use

– Used widely in computational linguistics facts and rules in a KB to answer questions about
– Developed to aid in machine translation and natural language a particular domain
understanding (“WordNet” is a famous SN)
– Represent knowledge in a hierarchy of semantic classes to be able
Such programs arrive at conclusions (or decide on
to deduce and disambiguate meaning actions) in a logical way
• e.g. “Jane looked for her keys. She needed milk.” Prolog is one of the most common logic
• The query: Why was Jane looking for her keys? programming languages
milk a cow
– Uses first-order definite clauses to encode the KB
walk
need $$ – Searches for proofs recursively with BC and GMP
need milk buy it drive car keys
go to the market – Can answer yes/no queries, or find bindings
steal it take bus
27 28