Manuscript

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

Some Results on Possibilistic Reasoning in Answer set

Programming
Yifan
Jin,Zhe
Kewen
Wang,
ZheofWang
Yifan Jin, Kewen
Wang,
Wang
School
Computing and
Information
University
170Griffith
Kessels
Road
School of Technology,
Computing andGrifth
Information
Technology,
University
1704111
Kessels
Road NATHAN QLD 4111 Australia
NATHAN QLD
Australia

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

about uncertain and (partially) inconsistent knowledge. As argued in [45],

when our main concern is with the meaning of information-rather than
with its measure, the proper framework for information analysis is possibilistic rather than probabilistic in nature. We are not going to discuss details
of this statement but point out that possibilistic information is more on
the side of representing priority of formulas and rules [45]. Moreover, it is
easier for the user to manage possibilistic information than probabilistic information since probability axioms are not enforced in possibilistic reasoning
[21]. Zadehs observation motivated the introduction of possibilistic logic [18],
which has been applied in several application domains [1, 9, 20, 11, 21, 28, 32].
According to Dubois and Prade, possibilistic approach has its advantages
over probabilistic approach in dealing with logical entailment.
Thus, it is natural and useful to combine possibilistic reasoning and ASP,
and as a result, a possibilistic extension of ASP (PASP) is proposed in [8,
37, 40]. A possibilistic program consists of a finite set of possibilistic rules,
each of which is pair of a standard ASP rule and a weight (a value in a finite
ordered set). For instance, the following example program encodes a simple
model to predict the priority that one should continue working (or not):
(tired , 0.1).
(working not tired, 0.9).
The first rule says that the weight that one feels tired is 0.1 (relatively
small), and the second rule has the weight 0.9 (relatively high), which specifies that one should continue working if he or she is not feeling tired.
The semantics of a possibilistic program in [36] is defined by its possibilistic stable models, which are obtained from a generalisation of GelfondLifschitz reduction. This semantics is a natural generalisation of stable models for normal logic programs in the sense that if each normal logic program
P is regarded as a possibilistic program P whose rules have the same weight
1, the possibilistic stable models of P are exactly answer sets of P . The skeptical reasoning determined by this semantics is actually not inconsistencytolerant like other possibilistic logics. For instance, given a simple possibilistic logic program {(a , 0.8), (b not b, 0.1)}. The semantics proposed in
[36] will not have any possibilistic stable model even the rule that causes
inconsistency has a low weight. We will discuss this further in Section 3.
In fact, several issues in PASP are still unexplored yet. First, possibilistic inferences (similar to those in ordinary possibilistic logic) have not been
investigated in the literature; Second, the counterparts of some important
concepts in ASP, such as Clarkes completion [13], loop formulas [34] and
2

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.
!

Example 1. Consider possibilistic program P = {(a , 0.9), (b not c, 0.7), (c

a, not b, 0.8)}. If A = {(a, 0.9), (b, 0.7)}, then A = {a, b} and P A = {(a
, 0.9), (b , 0.7)}. It is easy to see that L(TP A ) = A. Thus, {(a, 0.9), (b, 0.7)}
is a possibilistic stable model of P . Similarly, {(a, 0.9), (c, 0.8)} is another
possibilistic stable model of P .
Given a possibilistic program P , let be the collection of all Herbrand
interpretations of P (i.e., all subsets of the atoms in P ). For a Herbrand
interpretation I and a rule r, I |= r denotes that I is a model of r. r is
applicable w.r.t. I if body + (r) I and body (r) I = . The set of all
applicable rules in P w.r.t. I is denoted app(P , I). P is grounded if it
6

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

can be ordered as a sequence r1 , . . . , rn such that for each i (1 i n),

ri app(P , head ({r1 , . . . , ri1 })). (P )I is the GL-reduction of P w.r.t. I
defined by (P )I = {head (r) body + (r) | r P , body (r) I = }. A
possibilitic distribution on satisfies P , denoted |= P , if for each I ,
the following conditions are satisfied:
(I) = 0, if I " head(app((P )I , I)) or app((P )I , I) is not grounded (2)
(I) = 1, if I is a model of (P )I

(3)
I

(I) 1 max{ | I ! r and (r, ) P }, otherwise

(4)

Given two possibility distributions and on , we say that is less

specific than if (I) (I) for each I . The least specific possibility
distribution P for P satisfies for each I , the above conditions (2), (3),
and otherwise P (I) = 1 max{ | I ! r and (r, ) P I }. The inconsistency degree, -cut, possibility degree and necessity degree of an atom
a under the possibility distribution can all be similarly defined as in ordinary possibilistic logic. In particular, a possibility distribution satisfies a
possibilistic atom (a, ), denoted |= (a, ), if N (a) .
3. Possibilistic Inferences in ASP
Consider the possibilistic program P = {(tired , 0.1), (working
not tired, 0.9)} mentioned in Section 1. Under the possibilistic stable model
semantics defined above, P has a unique possibilistic stable model (tired, 0.1).
Thus, the necessity degrees of working and tired are N (working) = 0 and
N (tired) = 0.1, respectively. This implies that even one does not feel tired
he/she should still not work. Furthermore, when the possibilistic program
is inconsistent, no possibilistic stable model exists according to [36]. So,
the skeptical reasoning determined by possibilistic stable models is unable to
handle uncertainty or inconsistency well.
In this section, we first introduce three forms of possibilistic reasoning
for logic programs based on possibilistic stable models. Our possibilistic
inferences are inconsistency-tolerant. We then show that they can be characterised in terms of possibility distributions defined in [36].
Unlike in ordinary possibilistic logic, given a possibilistic logic program P
and a weight such that Inc(P ), P> may not be consistent as shown
in Example 9 in [36]. This difference is mainly due to the fact that PASP is
nonmonotonic in general. So, some changes have to be made when we adapt
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

the three major forms of possibility inferences from propositional possibilistic

logic to PASP.
Recall that an atom a is a skeptical consequence of a normal program Q
under stable model semantics, denoted Q |=s a, if a is in every stable model

of Q. For simplicity, we use P

to denote (P ) .
Definition 1. Let P be a possibilistic program.
1. An atom a is a plausible consequence of P , denoted P |=p a, if there

exists [0, 1] such that P

is consistent and P
|=s a.
2. An atom a is a possibilistic consequence of P to degree , written

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

not a, 0.7), (a , 0.1)}. Since P0.7 is consistent and P0.7

|=s c, we have
that P |= (c, 0.7) even the whole possibilistic program is inconsistent.
These inference services allow us to deal with uncertainty and inconsistency. For example, possibilistic inference with degree allows us to infer to
what degree an individual atom can be nontrivially inferred from a possibilistic program. These inference services are also heavily related to possibilistic
distribution, as we noted before, a possibility distribution satisfies (a, ),
denoted |= (a, ) if and only if N (a) . In the following, we will discuss the relation between possibility distribution and possibilistic inference
services.
The above three forms of possibilistic reasoning for ASP can be equivalently characterized in terms of possibility distributions. Unlike other semantics proposed before, in [36] and [40] the possibility distribution can only
be used to characterize possibilistic answer sets in definite programs and in
[6, 7] the possibility distribution is defined over a sub set of a program, not
over the interpretations of a program.

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

Proposition 1. Let P be a possibilistic program and a be an atom. Then

P |= (a, ) if and only if P

is consistent, and |= P implies |= (a, )
for each possibility distribution .

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) =

0 1 . Otherwise, since I ! a and P |= (a, ), we have I ! P

. So
there exists a rule r P such that n(r) = , and I ! r. Moreover, we
have rI P I . Since N (rI ) , for all I ! rI , (I) 1 1 . Thus
(I) 1 .

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 ,

we have (I) = 1 max{n(r) | I ! r, r P I }. Because I |= P

, we
I
have max{n(r) | I ! r, r P } < and (I) > 1 . It follows that
N (a) = 1max{(I) | I ! a} < , which implies ! (a, ), a contradiction.
Thus P |= (a, ).

By Proposition 1, it is straightforward to show the following result for

plausible consequences.
Proposition 2. Let P be a possibilistic program and a be an atom. Then

P |=p a if and only if there exists [0, 1] s.t. P

is consistent and for
each possibility distribution , if |= P , then |= (a, ).

Proof
Suppose P |=p a, then there must be an such that P
is

consistent and P |=s a. Thus P |= (a, ). By Proposition 1, for each

possibility distribution , if |= P , then |= (a, ). Conversely, suppose

there exists such that P

is consistent, for each possibility distribution ,
if |= P , then |= (a, ). By Proposition 1, P |= (a, ). It follows that
P |=p a.

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.

Example 3. (example 2 continued) Since for all > 0.7 we have P

! c,

thus P |= (c, 0.7). But, NP (c) = 0.8 and P0.8 is not consistent.

However, we have the following characterisation of possibilistic inference,

which is significantly different from its counterpart in possibilistic logic.
Proposition 3. Let P be a possibilistic program and a be an atom. Then

P |= (a, ) if and only if P

is consistent, NP (a), and there is no

> s.t. NP (a) and P is consistent.

is consistent, this case is easy. So we only

Proof
Let = NP (a). If P

consider the case when P is inconsistent.

): Suppose P |= (a, ), then P

is consistent. Since P |= P , by
Proposition 1, P |= (a, ). So NP (a) , which implies . Since
NP (a) and P is the least specific possibility distribution, we have
N (a) for each possibility distribution with |= P . On the contrary,

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

contrary, assume that P |= (a, ) for some > with P

being consistent.
So we would have P |= (a, ), that is, NP (a) > , a contradiction.
Therefore, P |= (a, ).

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

4. Completion and Loop Formulas for PASP

There are several reasons for studying the concept of loop formulas for
possibilistic programs. First, it is interesting to extend the definitions of
completion and loop formulas to possibilistic programs. Moreover, loop formulas provide a nice way to compute inconsistency degrees and the least
specific possibility distribution, which are important for developing inference
algorithms for possibilistic programs.
Given a possibilistic program rule r = (a a1 , . . . , am , not b1 , . . . , not bn , ).
Denote G(r) = a1 am b1 bn . The completion of possibilistic
program P is defined by Comp(P ) = Comp+ (P ) Comp (P ):
(i) G(r) a, n(r) is in Comp+ (P ), for each rule r = (a body(r), n(r))
in P .
(ii) For each atom a in P , let r1 = (a body(r1 ), n(r1 )), . . . , rn = (a
body(rn ), n(rn )) be all the rules with head a in P , then (a G(r1 )
G(rn ), 1) is in Comp (P ).
We note that due to the involvement of weights, the definition of completion for possibilistic programs looks different from the standard one but
it is obviously a generalisation of the latter. Given a possibilistic program
P , the positive dependency graph GP of P is the directed graph whose set
of nodes is the set of all atoms in P , and whose set of edges contains an arc
from p to q if there is a rule of the form r = (p body(r), n(r)) in P such
that q body(r), for each pair of nodes p and q. Recall that a directed graph
is said to be strongly connected if for any two nodes in the graph there is a
directed path from one node to the other node. Given a directed graph, a
strongly connected component is a set of nodes such that for any two nodes
u and v in that set, there is a path from u to v and that set is not a subset
of any other such sets.
Given a finite possibilistic program P , a non-empty subset L of all atoms
in P is called a loop of P if for any u and v in L, there is a path of length
> 0 from u to v in GP such that all the vertices in the path are in L. This
means if L is non-empty, then L is a loop if and only if the subgraph of GP
induced by L is strongly connected.
Given a possibilistic program P , and a loop L in it, R (L, P ) = {G(r)
p | r P, head(r) = p, p L, and there does not exist q s.t. q body(r)
L}. Without loss of generality, suppose that R (L, P ) consists of the follow-

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

The above corollary shows that the inconsistency degree of possibilistic

program P can be obtained from those of the corresponding possibilistic theory F (P ). This allows us to deal with inconsistency degree of a possibilistic
program in a syntactic way. As a result, the loop formulas can be used to
reduce possibilistic reasoning in ASP to the task of computing inconsistency
degree of the corresponding possibilistic theory.
Proposition 4. Given a possibilistic program P , then P |= (a, ) if and
only if Inc(F (P )) = 0 and Inc(F (P ) {(a, 1)}) > 0
Proof
If Inc(F (P )) = 0 then there exists a set of atoms A such that
(A) = 1 and A is a model of (F (P )) . Note that A is a model of (F (P ))

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

a A for all models A of (F (P )) , which means P

|=s a and P |= (a, ).
Plausible consequences of a possibilistic program can be obtained by computing inconsistency degrees of possibilistic theories.
Proposition 5. If P is a possibilistic program and A is a set of atoms,
then P |=p a if and only if there exists such that Inc(F (P )) = 0 and
Inc(F (P ) {(a, 1)}) > 0.
Moreover, possibilistic consequences of a possibilistic program can also
be obtained by computing inconsistency degrees of possibilistic theories.
Proposition 6. If P is a possibilistic program and A is a set of atoms, then
P |= (a, ) if and only if the following three conditions are satisfied:
1. Inc(F (P )) = 0,
2. Inc(F (P ) {(a, 1)}) > 0, and
3. If = NF (P ) (a), then < NF (P ) (a) and Inc(F (P )) > 0 for each
with < NF (P ) (a).
Proof
From Proposition 4, we have P |= (a, ) if and only if Inc(F (P )) =
0 and Inc(F (P ) {(a, 1)}) > 0. According to Proposition 3, every between and NF (P ) (a) should not lead to a consistent cut of P , thus for all
defined here we have Inc(F (P )) > 0. As a result, P |= (a, ).
13

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

satisfies some pi L. So, I does not satisfy (p1 Gr1k1 Grnkn , 1) in

Comp(P ). Thus in this case we also have F (P ) (I) = TP (I).
Thus, in all three cases we have F (P ) (I) = TP (I), and Inc(P ) =
Inc(TP ).
The notions of completion and loop formulas for possibilistic programs
can be used to compute possibility distribution and inconsistency degree of
a possibilistic program. In addition, it can be used to reduce the task of
computing the inconsistency degree, like the three inferences for PASP and
consistency restore investigated in [36]. In the next section, we will introduce
a resolution procedure for possibilistic programs, which provides a sound and
complete procedure with respect to possibilistic stable models.
5. Guarded Resolution for PASP
In this section, we first present a resolution procedure for possibilistic
reasoning in logic programs, which is a generalisation of the resolution rule
defined in [35], and then show that our procedure is sound and complete with
respect to possibilistic stable models [36]. Moreover, as a useful application
of the guarded resolution, we show that it can provide alternative way of
reducing an inference task in a possibilistic program to that of computing
inconsistency degrees of the possibilistic program and its cuts.
Given a possibilistic rule r = (a a1 , . . . , am , not b1 , . . . , not bn , ), a
guarded clause pg(r) for r is defined as
(a a1 , . . . , am : {b1 , . . . , bn }, ).

(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

n + 1. Then the last step of the resolution proof of (a : S0 , ) is of the form

(aa :Z1 ,1 ) (a :S1 ,1 )
. As Z1 A = , S1 A = and = min(1 , 1 ). Now
(a:S0 ,)
(a , S1 , 1 ) has a guarded resolution proof from pg(P ) that is admitted by A
of height at most n and (a , 1 ) L(TQ ) by inductive assumption. Moreover,
for (a a : Z1 , 1 ) there must exists a path like:
(a a , a1 , . . . , am : Zm+1 , m+1 )
...
(a a , a1 : Z2 , 2 )
(a a : Z1 , 1 )
Where Zm+1 Zm Z1 and for each i, there is a node in the tree of
the form ai : Zi , i such that the resolution of (a a , a1 , . . . , ai : Zi+1 , i+1 )
and (ai : Zi , i ) results in the clause (a a , a1 , . . . , ai1 : Zi , min(i+1 , i ))
where i = min(i+1 , i ) and i is the largest value that satisfies this condition. Now each (ai : Zi , i ) is the root of a guarded resolution proof from
pg(P ) that is admitted by A of height less than or equal to n and i is
the largest value that satisfies this condition, hence (ai , i ) L(TQ ) for
i = 1, . . . , m. Since (a a , a1 , . . . , am : Zm+1 , m+1 ) is in the resolution
proof, there must be a rule (a a , a1 , . . . , am , not b1 , . . . , not bn , m+1 )
in P and Zm+1 = {b1 , . . . , bn }. Since A admits the proof tree, it must
be the case that {b1 , . . . , bn } A = , hence (a a , a1 , . . . , am , m+1 )
is in Q. Moreover (a , 1 ), (a1 , 1 ), . . . , (am , m ) are in L(TQ ), it follows
(a, min(1 , m+1 , 1 , . . . , m )) L(TQ ). We have min(1 , m+1 , 1 , . . . , m ) =
min(1 , m , 1 , . . . , m1 ) = = min(1 , 1 ) = and is the largest number that satisfies this condition, that means if (a : S0 , ) has a guarded
resolution proof from pg(P ) and is the largest number that satisfies this
condition, then (a, ) A where A is a possibilistic stable model of P .
Based on the soundness and completeness of possibilistic guarded resolution, we are able to transform a possibilistic program P into possibilistic
propositional theory E (P ) such that both possibilistic and plausible consequences of P can be obtained from the inconsistency degrees of E (P ) and
its cuts.
Given an atom a, assume that (a : S1 , 1 ), . . . , and (a : Sn , n ) consist of all possibilistic guarded resolution proofs from pg(P ) for every atom
a appearing in P . Then eqP (a) is defined as conjunction of the following
possibilistic propositional formulas:
(a S1 . . . Sn , max(1 , . . . , n ))
18

(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

in [36, 40, 4].

There also exists another different approach in combining ASP with
weights called Fuzzy ASP [38], in which the truth of literals is graded. Fuzzy
ASP is a tool for reasoning with gradual information in the presence of complete information, it is not desiged to model uncertainty. However, it is
possible to combine this two works together as shown in [10, 5].
7. Discussions and Conclusion
Based on the framework of possibilistic answer set programming (PASP)
in [36], we have introduced three forms of possibilistic reasoning in ASP,
provided their characterizations in terms of possibility distributions. Our
possibilistic inferences are inconsistency-tolerant. In order to develop inference algorithms for these possibilistic inferences, the notions of possibilistic
loop formulas and possibilistic guarded resolution are proposed, which extends their counterparts in standard ASP. We have demonstrated how these
techniques can be used to perform possibilistic reasoning in a possibilistic
program by computing inconsistency degrees of the corresponding possibilistic theory and its cuts. Besides their theoretical importance, our results
provide promising inference algorithms for PASP.
Several issues are interesting for future research. First, an efficient implementation of the three forms of reasoning in PASP is still missing; Second,
it is unclear to us if both possibilistic loop formulas and guarded resolution
provide the same translation to a possibilistic program; Third, we will investigate the possibility of adapting our results to the framework of PASP
proposed in [8]; Finally, we will extend our result to possibilistic disjunctive
programs [40].
References
[1] T. Alsinet, C. Iv`an Ches
nevar, L. Godo, and G. R. Simari.: A logic programming framework for possibilistic argumentation: Formalization and
logical properties. Fuzzy Sets and Systems. 159(10), 12081228 (2008).
[2] M. Balduccini and M. Gelfond.: Logic programs with consistencyrestoring rules. International Symposium on Logical Formalization of
Commonsense Reasoning. AAAI 2004 Spring Symposium Series. 918
(2003)
21

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

[24] M. Gebser, J. Puhrer, T. Schaub and H. Tompits.: A meta-programming

technique for debugging answer-set programs. Proceedings of the 23rd
National Conference on Artificial Intelligence. 448453 (2008)
[25] M. Gebser, B. Kaufmann, and T. Schaub.: The conflict-driven answer
set solver clasp: Progress report. Proceedings of International Conference on Logic Programming and Nonmonotonic Reasoning. Springer
Berlin Heidelberg. 509514 (2009)
[26] M. Gebser, B. Kaufmann, R. Kaminski, M. Ostrowski, T. Schaub and
M. Schneider.: Potassco: The Potsdam answer set solving collection. Ai
Communications. 24(2), 107-124 (2011)
[27] M. Gelfond and V. Lifschitz.: The stable model semantics for logic programming. ICLP/SLP. 88, 10701080 (1988)
[28] B. Hollunder.: An alternative proof method for possibilistic logic and
its application to terminological logics. Proceedings of the Tenth International Conference on Uncertainty in Artificial Intelligence. Morgan
Kaufmann Publishers Inc. 327335 (1994)
[29] T. Janhunen.: Representing normal programs with clauses. Proceeding
of the 16th Eureopean Conference on Artificial INtelligence, ECAI 04.
358362 (2004)
[30] A. Kimmig, S. Bach, M. Broecheler, B. Huang, and L. Getoor.: A short
introduction to probabilistic soft logic. Proceedings of NIPS Workshop
on Probabilistic Programming: Foundations and Applications. (2012)
[31] N. Leone, G. Pfeifer, W. Faber, T. Eiter, G. Gottlob, S. Perri, and F.
Scarcello.: The dlv system for knowledge representation and reasoning.
ACM Transactions Computational Logic. 7(3), 499562 (2006)
[32] M.J. Lesot, O. Couchariere, B. B. Meunier, and J.L. Rogier.: Inconsistency degree computation for possibilistic description logic: an extension of the tableau algorithm. Fuzzy Information Processing Society.
NAFIPS. Annual Metting of the North American. IEEE. 16 (2008)
[33] F. Lin and Y. Zhao.: On tight logic programs and yet another translation
from normal logic programs to propositional logic. Proceedings of the
24

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

18th International Joint Conference on Artificial Intelligence. 853858

(2003)
[34] F. Lin and Y. Zhao.: ASSAT: Computing answer sets of a logic program
by SAT solvers. Journal of Artificial Intelligence. 157(1), 115137 (2004)
[35] V. W. Marek and J. B. Remmel.: Guarded resolution for answer set
programming. Theory and Practice of Logic Programming. 11(1), 111
123 (2011)
[36] P. Nicolas, L. Garcia, I. Stephan, and C. Lef`evre.: Possibilistic uncertainty handling for answer set programming. Annals of Mathematics and
Artificial Intelligence. 47(1-2), 139181 (2006)
[37] P. Nicolas, L. Garcia, and I. Stephan.: A possibilistic inconsistency
handling in answer set programming. Symbolic and Quantitative Approaches to Reasoning with Uncertainty. 402414 (2005)
[38] D. Van Nieuwenborgh, M.De Cock and D. Vermeir.: Fuzzy answer set
programming. Proceedings of the 10th European Conference on Logics
in Artificial Intelligence. 359372 (2006)
[39] I. Niemela, and P. Simons.: Smodels - an implementation of the stable
model and well-funded sematnics for normal logic programs. Proceedings
of the 4th International Conference on Logic Programming and NonMonotonic Reasoning. 420429 (1997)
[40] J.C. Nieves, M. Osorio, and U. Cortes.: Semantics for possibilistic disjunctive programs. Logic Programming and Nonmonotonic Reasoning.
Springer Berlin Heidelberg. 315320 (2007)
[41] J.C. Nieves and H. Lindgren.: Possibilistic nested logic programs and
strong equivalence. International Journal of Approximate Reasoning. 59,
119 (2015)
[42] M. Osorio, J.A.N. Perez, J.R.A. Ramirez and V.B. Macias.: Logics with
common weak completions. Journal of Logic and Computation. 16(6),
867890 (2006)
[43] T. Syrjanen and I. Niemela.: The smodels system. Logic Programming
and Nonmotonic Reasoning. Springer Berlin Heidelberg. 434438 (2001)
25

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