Professional Documents
Culture Documents
Manuscript
here to
view linked
Click here to Click
view linked
References
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Abstract
Answer set programming (ASP) has been extended to possibilistic ASP
(PASP), in which the notion of possibilistic stable models is defined for possibilistic logic programs. However, possibilistic inferences that correspond to
the three inferences in ordinary possibilistic logic have not been explored in
PASP yet. In this paper, based on the skeptical reasoning determined by
possibilistic stable models, we define three inference relations for PASP, provide their equivalent characterisations in terms of possibility distributions,
and develop algorithms for these possibilistic inferences. Our algorithms are
achieved by generalising some important concepts (Clarkes completion, loop
formulas, and guarded resolution) and properties in ASP to PASP. Besides
their theoretical importance, these results can be used to develop efficient
implementations for possibilistic reasoning in ASP.
Keywords: Possibilistic inference, Loop formula, Guarded resolution,
Uncertainty, Inconsistent management
1. Introduction
Answer set programming (ASP) [3, 27] is currently one of the most widely
used nonmonotonic reasoning systems due to its simple syntax, precise semantics and importantly, the availability of efficient ASP solvers, such as
clasp [25, 26], dlv [31], and smodels [43, 39]. On the other hand, a variety
of practical applications require the ability of ASP to represent and reason
Email addresses: yifan.jin@griffithuni.edu.au (Yifan Jin),
Preprint
submitted to International
Journal
of Approximate (Zhe Wang)
k.wang@griffith.edu.au
(Kewen Wang),
zhe.wang@griffith.edu.au
ReasoningSeptember 28, 2015 yifan.jin@grifthuni.edu.au
k.wang@grifth.edu.au
(Kewen
Wang),
zhe.wang@grifth.edu.au
Preprint submitted to International
Journal
of Approximate
ReasoningSeptember(Zhe
28, 2015
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
resolution [35], have not been adapted to PASP. Moreover, how can such
generalised concepts and algorithms be used in developing algorithms for
possibilistic inferences in ASP?
In this paper, we aim to provide a solution to these open problems. We
first introduce three possibilistic inferences in PASP. Moreover, we will show
that, these inferences can be fully characterized by possibility distribution
defined in [36]. This is not the case in many previous proposed semantics
like [36] and [6, 7]. We also develop algorithms for these possibilistic inferences. Our algorithms are achieved by generalising some important concepts
(Clarkes completion, loop formulas, and guarded resolution) and their properties in ASP to PASP.
Specifically, major contributions of this work are summarised as follows:
1. We define three forms of possibilistic reasoning in PASP and provide characterisations of these possibilistic inferences in terms of possibility distributions. These three forms of possibilistic reasoning are
inconsistency-tolerant thus can nontrivially infer conclusion even in inconsistent logic programs.
2. We provide a translation from possibilistic logic programs to possibilistic propositional theories. This is achieved by generalising Clarkes
completion and loop formulas from standard ASP to PASP. As a result, possibilistic completion and loop formulas allow us to compute
the least specific distribution of a possibilistic program by just computing that of the corresponding possibilistic theory. Moreover, this
result provides an algorithm for all of our three possibilistic inferences
in PASP through computing inconsistency degrees.
3. We propose possibilistic guarded resolution for PASP, which generalises
the guarded resolution for ordinary ASP in [35]. We show that the new
resolution algorithm is sound and complete with respect to possibilistic stable models. This result provides an algorithm for two of our
possibilistic inferences in PASP.
The rest of this paper is organized as follows. Section 2 first recaps some
basic concepts and notations in possibilistic logic and PASP. Section 3 introduces three forms of possibilistic reasoning in ASP and their characterisations
in terms of possibility distributions. A version of Clarkes completion and
loop formulas for PASP are provised in Section 4. Section 5 introduces a form
of guarded resolution for PASP. Section 6 discusses related work. Finally, the
work is concluded in Section 7.
3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
2. Preliminaries
2.1. Possibilistic logic
In possibilistic propositional logic, a (possibilistic) formula is a pair of
the form (, ) where is a propositional formula and the weight is an
element of a totally ordered set, which indicates the lower bound of the
necessity degree of and is denoted n() = . In this paper, we assume
that the weights are taken from a fixed finite subset of the interval [0, 1].
Informally, the formula (, ) expresses that is certain at least to the level
. A possibilistic knowledge base is a finite set of possibilistic formulas
= {(i , i ), i = 1, . . . , n}. Given a possibilistic knowledge base , we use
to denote its classical part that is obtained by ignoring weights of all
formulas in .
The semantics of possibilistic logic is based on the notion of possibility
distributions, which are mappings from the set of all propositional interpretations to [0, 1]. Given a possibility distribution and an interpretation
I , (I) represents the degree of compatibility of the interpretation I with
the available beliefs about the real world. By convention, (I) = 0 means
that I is impossible, and (I) = 1 means that nothing prevents I from being
the real world. When (I) > (I ), I is a preferred candidate to I for being
the real state of the world. Given two possibility distributions and , is
said to be less specific than if for all interpretation I, (I) (I).
Given a possibility distribution , two measures can be defined for possibilistic formulas. That is, the possibility degree () = max{(I) : I |= }
and the necessity degree N () = 1 max{(I) : I ! }. We say a possibility distribution satisfies a possibilistic formula (, ), denoted |= if
N () .
For each possibilistic knowledge base , the least specific possibility distribution, denoted , is defined as, for each I , (I) = 1 if I |= i for
all (i , i ) and (I) = 1 max{ | I ! , (, ) } otherwise.
In possibilistic logic, when a knowledge base is inconsistent, we are able
to extract certain consistent sub-knowledge bases based on the priority level
determined by weights of formulas and to use inconsistency degree to characterise the inconsistency of a possibilistic knowledge base. In possibilistic
logic, each priority level is referred to as a cut. Formally, given a weight , an
-cut (resp. strict -cut) of , denoted (resp. > ), is the set of formulas
in having a weight greater than (resp. strictly greater than) . The inconsistency degree of is defined as Inc() = max{ | is inconsistent }.
4
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
This inconsistency degree defines a plausibility level under which information is no more pertinent. The inconsistency degree of is actually determined by its least specific possibility distribution. Specifically, Inc() =
1 max{ (I) | I }.
There are three major inferences in possibilistic logic:
1. A formula is said to be a plausible consequence of , denoted |=p ,
if >Inc() |= .
2. A possibilistic formula (, ) is a consequence of , denoted by |=
(, ), if > Inc() and |= .
3. A formula is said to be a possibilistic consequence of to degree ,
denoted by |= (, ), if the following conditions are satisfied:(1)
is consistent, (2) |= , and (3) > , ! .
Here ( ) is denoted . The second and third inferences attach to
the consequence degree that is at least equal to the inconsistency degree of
the possibilistic knowledge base , yet they deal with different consequences.
The second inference checks if a possibilistic formula (, ) can be inferred
from , whereas the third checks whether a formula can be inferred from
and to what degree it can be inferred. Clearly, |= (, ) implies |=
(, ), which in turn implies |=p . In [22], it is proven that |=p
if and only if N () > Inc(), |= (, ) if and only if N ()
and > Inc(), and |= (, ) if and only if N () = and >
Inc(). We can see that meaningful consequences can be inferred from
an inconsistent possibilistic knowledge base and thus possibilistic logic is
inconsistency tolerant.
2.2. Possibilistic ASP (PASP)
In this subsection, we briefly introduce Nicolas et als approach to extending ASP by allowing possibilistic reasoning proposed in [36]. In their
PASP framework, each rule is associated with a level of priority like formulas in possibilistic logic. Formally, an ASP rule r is of the form a
a1 , . . . , am , not b1 , . . . , not bn where not is the default negation, a, ai s and
bj s are propositional atoms. A rule in PASP is a pair (r, ) where r is an ASP
rule and is the weight of r. Informally, the possibilistic rule (r, ) expresses
that the rule r is certain at least to the level n(r) = . A possibilistic logic
program is a finite set of possibilistic rules. Rules with zero degree are not
explicitly represented in the knowledge base. Given a possibilistic program
5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
P , we use P to denote its ASP rules part that is obtained by ignoring all
weights in rules.
Similar to the case of ordinary logic programs, the stable models of possibilistic logic programs can be first defined for possibilistic programs without
default negation, and then for possibilistic programs with default negation
through a generalised version of Gelfond-Lifschitz reduct.
Nicolas et al. defined an immediate consequence operator for possibilistic
programs without default negation as follows.
Let r be a possibilistic rule of the form a a1 , . . . , an , and A be a set of
possibilistic atoms. We say that r is -applicable in A if {(a1 , 1 ), . . . , (an , n )}
A and = min{, 1 , . . . , n }. Given an atom a, define
AP(P, A, a) = {r P | head(r) = a, r is -applicable in A, > 0}.
The immediate consequence operator TP for possibilistic program P is then
defined by, for each set A of possibilistic atoms,
"
#
"
" a head(P ), AP(P, A, a) = ,
TP (A) = (a, )"
" = max{ | r is -applicable in A, r AP(P, A, a)}
(1)
This operator is monotonic and thus has the least fixpoint L(TP ).
Given a possibilistic program P and a set of possibilistic atoms A, A
is a possibilistic stable model of P if A = L(TP A ), where P A = {rA | r
P, body (r)A = } where rA is the possibilistic rule head (r) body + (r), n(r).
A possibilistic program P is consistent if P has at least one possibilistic
stable model; otherwise P is inconsistent.
!
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
(3)
I
(4)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
is
|=s a, and for all > , P
is consistent, P
P |= (a, ), if P
inconsistent or P !s a.
3. A possibilistic atom (a, ) is a consequence of P , written P |= (a, ) if
P
is consistent and P
|=s a.
It is obvious that P |= (a, ) implies P |= (a, ), and the latter implies
P |=p a.
Example 2. Consider possibilistic program P = {( b, 0.9), (b not c, 0.8), (c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Proof
only if direction: Suppose P |= (a, ), then by definition, P
is consistent. To prove |= (a, ), it is sufficient to prove N (a) . We
note that N (a) if and only if 1 max{(I) | I ! a} . Moreover,
max{(I) | I ! a} 1 if and only if (I) 1 for each I ! a. Thus
we only need to show that (I) 1 for each I ! a.
If I " head(app((P )I , I)) or app((P )I , I) is not grounded, then (I) =
if direction: By assumption, P
is consistent. We show P |= (a, )
by contradiction. On the contrary, suppose there exists a stable model I of
P
but I ! a. Then I head(app((P )I , I)). Since (P )I (P )I , we
have I head(app((P )I , I)). Moreover, since app((P )I , I) is grounded,
for all rules r app((P )I , I) \ app((P )I , I), we have body(r) I
head(app((P )I , I)), thus app((P )I , I) is grounded. Then for any |= P ,
Proof
Suppose P |=p a, then there must be an such that P
is
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
While the above two properties are similar to those in possibilistic logic
in form, the characterisation of possibilistic inference in terms of possibility
distributions is quite different from that in possibilistic logic. In possibilistic
logic, if a formula is a possibilistic inference in possibilistic knowledge base
to the degree , then = N (). This is not the case in possibilistic
programs as the following example shows.
thus P |= (c, 0.7). But, NP (c) = 0.8 and P0.8 is not consistent.
assume that P
is consistent for some s.t. < . By Proposition 1
we would have that P |= (a, ), a contradiction.
): Since P
is consistent and NP (a) = > , by Proposition 1 we
have P |= (a, ). Since P is the least specific possibility distribution, we
have N (a) for each possibility distribution with |= P . On the
In possibilistic logic, reasoning services can be easily reduced to computing the inconsistency degree of a possibilistic knowledge base. However,
this is not straightforward in PASP due to its nonmonotonicity. Moreover,
inconsistency degree in possibilistic programs is now only defined under possibilistic distribution [36, 40], there is no syntactic definition of inconsistency
degree in possibilistic programs. We will address these issues in subsequent
sections.
10
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
11
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
ing formulas:
Gr11 p1 , . . . , Gr1k1 p1 ,
..
.
Grn1 pn , . . . , Grnkn pn .
Then the possibilistic loop formula LF (L, P ) associated with L is
$
p, 1).
([Gr11 Gr1k1 Grn1 Grnkn ]
pL
Let F (P ) = Comp(P ) LF where LF is the set of loop formulas associated with the loops of P . Then the least specific possibilistic distributions
of P and F (P ) coincide.
The major result of this section is stated as follows.
Theorem 1. If P is a possibilistic program and A is a set of atoms, then
P (A) = F (P ) (A).
Proof
Given a set of atoms A, assume that A " head(app(P A , A)). Then
there exist an atom a A such that a
/ head(app(P A , A)). It is easy to
prove that there exists a formula (a G(r1 ) G(rn ), 1) in Comp(P )
such that A ! a G(r1 ) G(rn ). Thus (A) = 0.
On the contrary, assume that app(P A , A) is not grounded. Then there
would exist a loop formula LF (L, P ) : ([ G(ri ) G(rj ) . . . ]
aL a, 1), which means (A) = 0.
Last, given a set A of atoms, assume A head(app(P A , A)) and app(P A , A)
is grounded. Then A satisfies both Comp (P ) and LF . So, (A) = 1 if A
is a model of Comp+ (P ) and (A) = 1 max{n(r) | A ! r, r Comp+ (P )}
otherwise. Thus, P (A) = F (P ) (A).
By Theorem 1, we have the following useful result.
Corollary 1. If P is a possibilistic program, then
1. Inc(P )=Inc(F (P )).
2. Inc(P ) = max{ | F (P ) (, )}.
12
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
is consistent. Moreover,
. Thus, P
if and only if A is a stable model of P
by by Inc(F (P )) = 0 and Inc(F (P ) {(a, 1)}) > 0, it follows that
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
By Corollary 1, a straightforward approach to computing the inconsistency degree of a possibilistic program is to compute its loop formulas. Yet
similar to the case of ASP, such a naive approach may not be practical, as
there can be an exponential number of loops in a possibilistic program. For
this reason, it would be more efficient if only necessary loop formulas are
included into the completion. This motivates the following method of computing its inconsistency degree when a possibilistic program is inconsistent.
Algorithm 1 Computing Inconsistency Degree
1: procedure
input: Possibilitic logic program P .
output:Inconsistency degree of P .
2:
Set TP := Comp(P ).
3:
If (TP ) has no models, then terminate and return Inc(TP ).
4:
Find a model M of (TP ) .
5:
If M is a stable model of P , then terminate and return 0.
6:
If M is not a stable model of P , then find a loop L of P such that
its loop formula LF (L, P ) is not satisfied by M .
7:
Set TP := TP {LF (L, P )} and go back to step 2.
8: end procedure
It is obvious that the above algorithm will terminate in finite number of
steps. We can show that the algorithm is sound.
Theorem 2. Let P be an inconsistent possibilistic program P , then Algorithm 1 outputs Inc(P ).
%
Proof
Let M LF (L, P ) denote the union of all LF (L, P ) added in
Algorithm 1.% If P is inconsistent, then F (P ) = Comp(P ) LF and TP =
Comp(P ) M LF (L, P ). Since P is inconsistent, for a set of atoms I, there
are three possible cases: (1) I ! Comp(P ) and I |= LF ; (2) I |= Comp(P )
and I ! LF ; (3) I ! Comp(P ) and I ! LF .
For the first case, it is easy to see F (P ) (I) = TP (I). For the second case,
I is a model of Comp(P ) but I does not satisfy some loop formula. From the
definition of TP , it follows that F (P ) (I) = TP (I). For the last case, since I
does not satisfy some loop formula, there exists a set L of atoms such that I
does not satisfy&the possibilistic formula ([Gr11 Gr1k1 Grn1
Grnkn ] pL p, 1), which means I does not satisfy any Griki but I
14
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
(5)
If a possibilistic rule r has no positive atoms in its body (that is, m = 0),
then pg(r) is also denoted (a : {b1 , . . . , bn }, ), called a possibilistic guarded
atom and {b1 , . . . , bn } is its guard. For a possibilistic program P , define
pg(P ) = {pg(r) : r P }.
The possibilistic guarded resolution rule is defined as the following inference rule:
(a A : B, ). (aj : C, ).
(6)
(a A \ {aj } : B C, min{, })
where A = {a1 , . . . , aj , . . . , am }, B = {b1 , . . . , bn }, C = {c1 , . . . , ch }
The guarded resolution rule naturally leads to the notion of a guarded
resolution proof P of a possibilistic guarded atom (a : S, ) from the program
15
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
pg(P ). A guarded resolution proof of (a : S, ) is a labelled tree such that every node that is not a leaf has two parents, the two parents are the upper part
of equation (6). Each leaf is either (a a1 , . . . , am : {b1 , . . . , bn }, ). such
that (a a1 , . . . , am : not b1 , . . . , not bn , ) is in P or (a : {b1 , . . . , bm }, )
such that (a not b1 , . . . , not bn , ) is in P . Note that in a guarded resolution proof, guards only grow as proceeding down the tree. Thus, the root
of the proof contains the guards of every label in the tree. In the following
we use (a : S, ) to denote both a possibilistic guarded atom as well as the
possibilistic guarded resolution proofs associated with it.
We say a set A of possibilistic atoms admits a possibilistic guarded atom
(a : S, ) if A S = , and that A admits a guarded resolution proof P if
it admits the label of the root of P.
The following result shows that possibilistic guarded resolution is sound
and complete with respect to possibilistic stable models.
Proposition 7. Let P be a possibilistic program and A be a set of possibilistic atoms. Then A is a possibilistic stable model of P iff the following two
conditions are satisfied:
(1) for every (a, ) A, for some set S of atoms, there is a guarded
resolution proof of (a : S, ) from pg(P ) that is admitted by A and for
all other guarded resolution proof of (a : S , ) from pg(P ) admitted by A.
(2) for every a
/ A and each S, there is no guarded resolution proof of
(a : S, ) from pg(P ) that is admitted by A.
Proof
As we already have A is a stable model of P as shown in [35],
for the if direction, we only need to proof given a possibilistic stable model
A, for all (a, ) A there exists a guarded resolution proof of (a : S, ) from
pg(P ) such that (a, ) L(TP A ) where L(T ) is the least fixpoint defined
after the immediate possibilistic consequence operator in section 2.
Let Q = P A , then for all (a, ) L(TQ ) we have (a, ) TQk where
TQ is the immediate possibilistic consequence operator defined in equation
(1). We claim that we can prove by induction on n N that whenever
(a, ) TQn , there exists a set of atoms S such that (a : S, ) possesses a
guarded resolution proof from pg(P ) admitted by A and for all other guarded
resolution proof of (a : S , ) from pg(P ) admitted by A, we have > .
If n = 1, then a , belongs to Q. That means, for some b1 , . . . , bn ,
we have a not b1 , . . . , not bn , belongs to P and {b1 , . . . , bn } A =
. Therefore the guarded atom (a : {b1 , . . . , bn }, ) is admitted by A and
16
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
it possesses a guarded resolution proof from pg(P ), namely, the one that
consists of the root labeled by (a : {b1 , . . . , bn }). Moreover, since (a, ) TQ1 ,
by the definition of TQ operator, is the maximum value among all guarded
atom (a : {b1 , . . . , bn }, ) admitted by A that has a guarded resolution proof
from pg(P ).
Now, let us assume (a, ) TQn+1 . Then there exists a rule (a
a1 , . . . , am , ) in Q such that (ai , i ) TQn for i = 1, . . . , m, Thus by induction, there are sets of atoms Si , 1 i n, such that (ai : Si , i ) possesses a
guarded resolution proof from pg(P ) admitted by A. As (a a1 , . . . , am , )
belongs to Q, there must exists atoms b1 , . . . , bm
/ A such that (p
a1 , . . . , am , not b1 , . . . , bn , ) is a rule in P . It is easy to combine the guarded
resolution proofs of (ai : Si , i ), 1 i m and the guarded clause (a
a1 , . . . , am : {b1 , . . . , bn }, ) to obtain a guarded resolution proof from pg(P )
of the following possibilistic guarded atom: (a : S1 Sn {b1 , . . . , bn }, )
where = min(, 1 , . . . , n ). In addition, for all other guarded clause (a
c1 , . . . , cm : {b1 , . . . , bn }, ) and (ci : Si , i ) that can obtain a guarded resolution proof from pg(P ) of (a : S1 Sn {b1 , . . . , bn }, min(, 1 , . . . , n )) we
have min(, 1 , . . . , n ) according to the definition of TQ operator. As
all the set occurring in the guard of this possibilistic guarded atom are disjoint from A, the resulting guarded resolution proof is admitted by A. This
means, for all possibilistic atom (a, ) in a possibilistic stable model, there
exists a guarded resolution proof of (a : S, ) from pg(P ) admitted by A and
for all other guarded resolution proof of (a : S , ) from pg(P ) admitted by
A, > .
Conversely, suppose (a : S, ) has a guarded resolution proof P from
pg(P ) admitted by A, and for all (a : S, ) that has a guarded resolution
proof from pg(P ) admitted by A, we have . All the guards occurring
in P are disjoint from A. We can prove by induction on the height of the
tree P that (a, ) L(TQ ).
If the height of P is 0, then it must be the case that (a not b1 , . . . , not bn , )
belongs to P where S = {b1 , . . . , bn }. Since A S = , the rule (a , )
belongs to Q and for all other rule (a , ) belongs to Q we have > ,
hence (a, ) L(TQ ).
Now assume that whenever (a : S, ) has a guarded resolution proof from
pg(P ) that is admitted by A of height less than or equal to n, and is the
largest value that satisfies this condition, then (a, ) L(TQ ). We now show
that the same property holds for all guarded atoms (a : S0 , ) which have
a guarded resolution proof from pg(P ) that is admitted by A of the height
17
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
(7)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
If every Si is empty, eqP (a) is defined as the possibilistic atom (a, max(1 , . . . , n ))
and if there is no possibilistic guarded resolution proof for a, eqP (a) is (a, 1).
Let E (P ) denote the possibilistic propositional theory consisting of all
eqP (a) for a in P . Then E (P ) has the following property that is useful for
reducing the tasks of computing plausible consequences and consequences to
the task of computing inconsistency degrees of E (P ) and its cuts.
We first present the following lemma.
Lemma 1. Let P be a possibilistic program and A be a set of atoms. Then
P (A) = 1 if and only if E (P ) (A) = 1.
Proof
As we already know A is a stable model of P if and only if A is
a model of EP , and P (A) = 1 if and only if A is a stable model of P , it is
easy to see P (A) = 1 if and only if E (P ) (A) = 1.
The task of computing consequences of a possibilistic program can be
reduced to that of computing inconsistency degrees of the corresponding
possibilistic theory and its cuts.
Proposition 8. Given a possibilistic program P , P |= (a, ) if and only if
Inc(E (P )) = 0 and Inc(E (P ) {(a, 1)}) > 0.
Similarly, the task of deciding plausible consequence can also be reduced
to that of computing inconsistency degrees as the following property shows.
Proposition 9. Given a possibilistic program P , P |=p a if and only if there
exists an such that Inc(E (P )) = 0 and Inc(E (P ) {(a, 1)}) > 0.
However, it is unclear to us whether the task of computing possibilistic
consequences of P can be reduced to the problem of computing inconsistency
degrees of E (P ) and its cuts.
6. Related Work
The idea of incorporating possibility theory into logic programming has
come a long way since it was first proposed in [17]. However the work in
[17] was restricted to classical formulas without default negation, thus the
reasoning ability is limited. Latter, Wagner et al. in [44] combined stable
models with a compositional version of possibilistic logic and proposed a new
framework. However, as shown in [19], the classical boolean tautologies are
19
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
not preserved in this logic. Another framework that combines defeasible logic
and possibility theory was proposed in [12], which allows to resolve conflicts
between contradictory goals.
The work that combining possibility theory and ASP was first presented
in [36]. In their work, Nicolas et al. introduced possibilistic stable models
as well as extended concepts like possibility distribution and inconsistency
degree from possibilistic logic to PASP. Their approach keeps the one-to-one
relationship between possibilistic stable models and classical stable models.
Their work latter extended in [40, 41] to capturing possibilistic nested logic
programs. The authors also explored the concept of strong equivalence in
PASP. Moreover, in [15], Nicolas approach was combined with ordered disjunction programs. It allows for a rich approach to capturing preferences in
ASP. Latter, an alternative semantics for PASP was proposed in [4], this approach takes a different intuition behind how negation as failure is managed.
While in Nicolas work, not a is true whenever a is more certain than a, in
Bauters work, not a is true to a degree in which it is possible that a is true.
Rules in Bauters work are interpreted as constraints on possibility distribution and used to reason over epistemic states. This is related to the work in
[23] as the authors show that ASP can be seen as a form of meta-epistemic
reasoning. However, semantics proposed in [4] dose not have one-to-one relationship between possibilistic stable models and classical stable models.
Thus as the authors in [7, 8] argued, the choosing of one approach or the
other strongly depends on the application domain. Other semantics like [14]
extended the concept of pstalbe models [42] to PASP, which is a framework
characterized by a fusion of ASP and paraconsistent logic. They are more in
line with classical models and possibilistic logic than stable models.
From a practical point of view, representing a logic program using propositional formulas is a way to implementing practical systems. Thus this topic
was widely discussed in literatures like [29, 34, 33]. In addition, uncertainly
also plays an important role in dealing with preference, handling inconsistency, and so on. Preference is an important topic in ASP to representation
and reasoning with qualitative preferential information [16]. To some extend,
the use of preference among rules is related to using certainty weights, and
the resulting semantics are closer in spirit to the approach in [36] thus also
closer to our approach. Different approaches were proposed for handling inconsistency. Like highlighting inconsistencies [24], resolving inconsistencies
[2] and reasoning under inconsistency [7]. Our approach presented in this paper can be used to reason under inconsistent logic programs unlike semantics
20
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
[3] C. Baral.: Knowledge Representation, Reasoning and Declarative Problem Solving. Cambridge University Press (2003)
[4] K. Bauters, S. Schockaert, M. De Cock and D. Vermeir.: Possibilistic
Answer Set Programming Revisited. Proceedings of the 26th Conference
on Uncertainty in Artificial Intelligence (UAI). 2010
[5] K. Bauters, S. Schockaert, M. De Cock and D. Vermeir.: Towards possibilistic fuzzy answer set programming. Proceedings of the 13th International Workshop on Non-monotonic Reasoning, (2011)
[6] K. Bauters, S. Schockaert, M. De Cock and D. Vermeir.: Possible and
Necessary Answer Sets of Possibilistic Answer Set Programs. Proceedings of the 24th International Conference on Tools with Artificial Intelligence (ICTAI12). 836843 (2012)
[7] K. Bauters, S. Schockaert, M. De Cock and D. Vermeir.: Semantics
for possibilistic answer set programs: Uncertain rules versus rules with
uncertain conclusions. International Journal of Approximate Reasoning.
55(2), 739761 (2014)
[8] K. Bauters, S. Schockaert, M. De Cock, and D. Vermeir.: Characterizing
and extending answer set semantics using possibility theory. Theory and
Practice of Logic Programming. 15(1), 79116 (2015)
[9] S. Benferhat and Z. Bouraoui.: Possibilistic dl-lite. Scalable Uncertainty
Management. Springer Berlin Heidelberg. 346359 (2013)
[10] B. Bouchon-Meunier, D. Dubois, L. Godo and H. Prade.: Fuzzy set and
possibility theory in approximate and plausible reasoning. Fuzzy Sets in
Approximate Reasoning and Information Systems. 5, 15-190 (1999)
[11] D. Cayrac, D. Dubois, and H. Prade.: Handling uncertainty with possibility theory and fuzzy sets in a satellite fault diagnosis application.
IEEE Transactions on Fuzzy Systems. 4(3), 251269 (1996)
[12] C. Chesnevar, G.Simari, T. Alsinet and L. Godo.: A logic programming
framework for possibilistic argumentation with vague knowledge. Proceedings of the 20th Conference on Uncertainty in Artificial Intelligence.
7684 (2004)
22
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
[13] K. L. Clark.: Negation as failure. Logic and data bases. Springer US.
293322 (1978)
[14] R. Confalonieri, J.C. Nieves and J. Vazquez-Salceda.: Pstable semantics
for logic programs with possibilistic ordered disjunction. Proceedings of
the 11th International Conference on Advances in Artificial Intelligence.
5261 (2009)
[15] R. Confalonieri, J.C. Nieves, M. Osorio and J. Vazquez-Salceda.: Dealing with explicit preferences and uncertainty in answer set programming Annals of Mathematics and Artificial Intelligence. 65(2-3), 159198
(2012)
[16] J.P. Delgrande, T. Schaub, H. Tompits, K. Wang.: A classification and
survey of preference handling approaches in nonmonotonic reasoning.
Computational Intelligence. 20(2), 308-334 (2004)
[17] D. Dubois, J. Lang, and H. Prade.: Towards possibilistic logic programming. Proceedings of the 8th Joint International Conference and
Symposium on Logic Programming. 581595 (1991)
[18] D. Dubois, J. Lang, and H. Prade.: Possibilistic Logic. Handbook of
Logic in Artificial Intelligence and Logic Programming. vol 3 (1994)
[19] D. Dubois and H. Prade.: Can we enforce full compositionality in uncertainty calculi? Proceedings of the 12th National Conference on Artificial
Intelligence. 149154 (1994)
[20] D. Dubois, D. Le Berre, H. Prade, and R. Sabbadin.: Using possibilistic
logic for modeling qualitative decision: Atms-based algorithms. Fundamenta of Informatica. 37(1,2), 130 (1999)
[21] D. Dubois and H. Prade.: Possibility theory, probability theory and
multiple-valued logics: A clarification. Annals of Mathematics and Artificial Intelligence. 32(1-4), 3566 (2001)
[22] D. Dubois and H. Prade.: Possibilistic logic: a retrospective and
prospective view. Fuzzy Sets and Systems. 144(1), 323 (2004)
[23] D. Dubois, H. Prade and S. Schockaert.: Stable models in generalized
possibilistic logic. Proceedings of the 13th International Conference on
Principles of Knowledge Representation and Reasoning. 315320 (2007)
23
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
[44] G. Wagner.: Negation in fuzzy and possibilistic logic programs. Proceedings of the 2nd International Workshop on Logic Programming and
Soft Computing. 113128 (1998)
[45] L. A. Zadeh.: Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets
and Systems. 100, 934 (1999)
26