You are viewing a javascript disabled version of the site. Please enable Javascript for this site to function properly.
Go to headerGo to navigationGo to searchGo to contentsGo to footer
In content section. Select this link to jump to navigation

Assumption-based argumentation for extended disjunctive logic programming and its relation to nonmonotonic reasoning

Abstract

The motivation of this study is that Reiter’s default theory as well as assumption-based argumentation frameworks corresponding to default theories have difficulties in handling disjunctive information, while a disjunctive default theory (ddt) avoids them. This paper presents the semantic correspondence between generalized assumption-based argumentation (ABA) and extended disjunctive logic programming as well as the correspondence between ABA and nonmonotonic reasoning approaches such as disjunctive default logic and prioritized circumscription. To overcome the above-mentioned difficulties of ABA frameworks corresponding to default theories, we propose an assumption-based framework (ABF) translated from an extended disjunctive logic program (EDLP) since an EDLP can be translated into a ddt. Our ABF incorporates explicit negation and the connective of disjunction “|” to Heyninck and Arieli’s ABF induced by a disjunctive logic program. In this paper, first, we show how arguments are constructed from disjunctive rules in our proposed ABF. Then, we show the correspondence between answer sets of an EDLP P and stable extensions of the ABF translated from P with trivialization rules. After defining rationality postulates, we show answer sets of a consistent EDLP are captured by consistent stable extensions of the translated ABF with no trivialization rules. Finally, we show the correspondence between ABA and disjunctive default logic (resp. prioritized circumscription). The relation between ABA and possible model semantics of EDLPs is also discussed.

1.Introduction

1.1.Background

Disjunctive information is often required in reasoning and argumentation to solve problems in our daily life. In nonmonotonic reasoning, Gelfond et al. [16] proposed disjunctive default logic as a generalization of Reiter’s default logic [27] to overcome problems of default logic in handling disjunctive information. To this end, they use the symbol “|” as the connective of disjunction in a disjunctive default theory instead of the classical “∨” used in a default theory, where PQ means “PQ is known”, while P|Q means “P is known or Q is known”.11 They also showed that an extended disjunctive logic program (EDLP) [15] which may include disjunction using the connective “|” in rule head as well as two kinds of negation (i.e. classical negation “¬” and negation-as-failure “not”) can be translated into a disjunctive default theory. In contrast, in the context of formal argumentation, Beirlaen et al. [2,3] presented the extended ASPIC+ framework where disjunctive reasoning is integrated in structured argumentation with defeasible rules [25] by incorporating reasoning by cases inference scheme [2]. In their framework, an argument is allowed to have a disjunctive conclusion, while disjunction is expressed by using the classical connective “∨”. However, they did not show the relationship between their framework and other approaches in nonmonotonic reasoning such as a disjunctive default theory. Consider the following example [3] shown by them.

Example 1

Example 1(Kyoto protocol).

There are two candidates for an upcoming presidential election. The candidates had a debate in the capital. They were asked what measures are to be taken in order for the country to reach the Kyoto protocol objectives for reducing greenhouse gas emissions. The first candidate, a member of the purple party, argued that if she wins the election, she will reach the objectives by supporting investments in renewable energy. The second candidate, a member of the yellow party, argued that if she wins the election, she will reach the objectives by supporting sustainable farming methods. We have reasons to believe that, no matter which candidate wins the election, the Kyoto protocol objectives will be reached. If the purple candidate wins, she will support investments in renewable energy (pr), which would in turn result in meeting the Kyoto objectives (rk). Similarly, if the yellow candidate wins, she will support sustainable farming methods (yf), which would result in meeting the Kyoto objectives (fk). Since one of the two candidates is going to win (py), we can reason by cases to conclude that the Kyoto objectives will be reached (k).

Beirlaen et al. [3] represented the information shown above in terms of the knowledge base K1=({pr,rk,yf,fk},{py}) which consists of defeasible rules in ASPIC+ [24,25] and the classical formula py as facts. Their formulation derives the intended result k, however, a problem happens when K1 is translated into a default theory. In fact, ψϕ encodes the normal default ψ:ϕϕ, then K1 is expressed by the default theory D1:

D1={p:rr,r:kk,y:ff,f:kk,py}.
D1 has a single extension, consisting of the disjunction py and its logical consequences, where the four defaults “don’t work”. Thus the result is contrary to what we would expect.

Moreover, regarding ABA as another structured argumentation system, Lehtonen et al. [19] recently presented the ABA framework instantiated with a propositional default theory. However the ABA framework corresponding to D1 which is constructed according to Lehtonen et al.’s definition cannot derive the expected result k under the stable (resp. grounded) semantics as well (see details in Section 6).

Notice that the disjunctive information s.t. “one of the two candidates is going to win” is expressed by py in K1 and D1. In a precise sense, however, it means that the election will yield a result s.t. either “that the purple candidate wins is known” or “that the yellow candidate wins is known”. Hence it should be expressed by p|y rather than py, that requires an extension to contain one of the two disjunctive terms rather than the disjunction due to [16]. Therefore, to solve the aforementioned problem arising in D1, let us represent the information by the disjunctive default theory D2:

D2={p:rr,r:kk,y:ff,f:kk,p|y},
or the associated EDLP P1={rp,not¬r,kr,not¬k,fy,not¬f,kf,not¬k,p|y} [16]. Then P1 has two answer sets S1={p,r,k} and S2={y,f,k}, while D2 has two extensions E1 and E2 such that kSiEi(i=1,2), where Ei consists of atoms from Si as well as the logical consequences derived from them. These results agree with our expectation that the Kyoto protocol objectives will be reached no matter which candidate wins the election. □

In regard to the relation between assumption-based argumentation and (disjunctive) default logic, Bondarenko et al. firstly showed in [5, Theorem 3.16] that there is a one-to-one correspondence between extensions of Reiter’s default theory and stable extensions of the corresponding assumption-based framework (ABF,22 for short). This denotes that the expected result k of the Kyoto protocol problem shown above is never obtained from stable extensions of their ABF corresponding to D1 based on their theorem. Besides, Bondarenko et al. [5] showed nothing about the relationship between ABA and disjunctive default logic. Hence, to solve such problems of ABFs corresponding to default theories in handling disjunctive information, it is required to find the relationship between ABFs and disjunctive default theories.

Recently, Heyninck and Arieli [18] proposed a generalized assumption-based framework induced by a disjunctive logic program (DLP), where disjunction using the connective “∨” in rule head as well as one kind of negation, i.e. negation-as-failure (NAF) are allowed to appear. Hence though their ABF induced by a DLP has a contrariness operator ¯¯ such that notp=p for every atom p, its language does not contain explicit negation. Then they showed a one-to-one correspondence between the stable models of a DLP [26] and the stable assumption extensions of the ABF induced by a DLP. In their work, however, there are several open problems left to be explored. First, the semantic relationship between their ABF induced by a DLP and approaches of nonmonotonic reasoning such as a disjunctive default theory is not considered in [18]. Second, they did not show how to construct an argument from disjunctive rules in a DLP nor took account of argument extensions in their ABF.

In logic programming, EDLPs [15] were proposed by Gelfond and Lifschitz to extend DLPs for knowledge representation by not only allowing classical negation (i.e. explicit negation) along with negation-as-failure but also using “|” instead of “∨”. At the same time, it is shown in [16] that an EDLP can be embedded into a disjunctive default theory (ddt, for short) which uses “|” as the connective of disjunction. In contrast, in formal argumentation, generally an assumption-based framework is capable of containing explicit negation in its language [9,10,12], but the ABFs corresponding to Reiter’s default theories have difficulties in handling disjunctive information. To our best knowledge, however, there is no study to show the relationship between ABFs and EDLPs as well as the relationship between ABFs and ddts to overcome such difficulties of ABFs discussed above.

1.2.Purpose of this paper

The purpose of this paper is first to investigate the semantic relationship between ABFs and EDLPs as well as the relationships between ABFs and other approaches in nonmonotonic reasoning (e.g. disjunctive default logic [16], prioritized circumscription [20,23]) that have not been studied, and second to show how to construct an argument in ABFs from disjunctive rules in (E)DLPs. We use the answer set semantics [15] and the paraconsistent stable model semantics [28] of EDLPs neither of which is two-valued for characterizing stable extensions of ABFs translated from EDLPs.

Regarding ABFs whose languages contain explicit negation, however, we should pay attention to avoid consistency problems which sometimes have occurred in applications of argumentation due to explicit negation contained in the language. To this end, so far, rationality postulates [6] were proposed as principles which rule-based argumentation systems should satisfy to avoid anomalous outcomes. In fact, in structured argumentation systems such as ASPIC+ and ABA frameworks whose languages contain explicit negation, conditions under which each system satisfies rationality postulates were proposed [1,12,24,25].

As for recent ABA applications containing explicit negation, Schulz and Toni [31] proposed the approach of justifying answer sets of an extended logic program (ELP) using argumentation. In their approach, they used the ABA framework ABAP instantiated with an ELP P. Though an ELP may contain classical negation, they took account of neither rationality postulates [6,12] nor consistency in ABA, and they claimed in their theorems [31, Theorem 1 and Theorem 2] that there is a one-to-one correspondence between answer sets of a consistent ELP P and the stable extensions of ABAP. In other words, their theorems claim that any ABA framework instantiated with a consistent ELP will never produce inconsistent stable extensions. However, there exist counterexamples to their theorems as shown in [33]. This means that anomalous outcomes may be obtained in applications of argumentation based on their theorems (e.g., there exists the inconsistent stable extension in ABAP instantiated with the given consistent ELP P which expresses the knowledge of the slightly modified Married John Example [33, Example 1], [6]).

Therefore, to achieve the above-mentioned purpose of this study, this paper proposes an assumption-based framework translated from an EDLP, which incorporates explicit negation as well as the connective “|” instead of “∨” in Heyninck and Arieli’s ABF induced by a DLP while avoiding consistency problems that arise in Schulz and Toni’s approach. Contributions of this study are as follows:

First, we define an argument in the ABF translated from a given EDLP which is constructed from disjunctive rules of the EDLP based on three inference rules provided in our ABF. Second, we show not only a one-to-one correspondence between p-stable models of an EDLP P and stable argument extensions (resp. stable assumption extensions) of the ABF translated from P but also a one-to-one correspondence between answer sets of an EDLP P and stable argument extensions (resp. stable assumption extensions) of the ABF translated from P with trivialization rules. Third, since our ABF incorporates explicit negation ¬, we define rationality postulates and consistency in our ABFs to avoid anomalous outcomes. Then we show answer sets of a consistent EDLP can be captured by consistent stable extensions of the translated ABF with no trivialization rules. This is useful for ABA applications containing explicit negation. Fourth, we show the new results about the relationship between ABA and disjunctive default logic [16] which enables us to overcome the difficulties of the ABF corresponding to a default logic in handling disjunctive information as well as the relationship between ABA and prioritized circumscription [20,23]. These results have not been shown so far to the best of our knowledge. Finally, we show how the possible model semantics of EDLPs [30] which is different from answer set semantics [15] is also captured by our ABFs translated from EDLPs. Appendix shows the correspondence between the semantics of ELPs [15,28] and stable assumption extensions of the associated ABA frameworks.

This paper is an extended and revised version of the paper [34], where not only Theorems 13, 14 and 15 in Section 4.1 but also Section 4.2, Section 5 and Appendix are newly introduced. Every proof sketch in [34] is replaced with a full proof while introducing new propositions, lemmas and two tables in this paper. Revisions are made throughout the paper and new considerations33 are often added. The rest of this paper is organized as follows. Section 2 shows preliminaries where new propositions proved in Appendix are mentioned. Section 3 presents an ABF translated from an EDLP, the definition of its argument, the semantic relationship between an EDLP and the corresponding ABF, the notion of consistency in ABFs, and the semantic relationship between a consistent EDLP and the corresponding consistent ABF. Section 4 shows the relationships between our proposed ABFs and nonmonotonic reasoning formalisms such as disjunctive default logic and prioritized circumscription. Section 5 shows the correspondence between the possible model semantics of EDLPs and ABA semantics of our ABFs. Section 6 discusses related work. Section 7 presents conclusions. Appendix contains new propositions relating ELPs and ABFs.

2.Preliminaries

2.1.Assumption-based argumentation

An ABA framework [5,9,10] is a tuple L,R,A,¯¯, where (L,R) is a deductive system, consisting of a formal language (a set of sentences) L and a set R of inference rules of the form: b0b1,,bm(biLfor0im), AL is a (non-empty) set of assumptions, and ¯¯ is a total mapping from A into L, which we call a contrariness operator. α is referred to as the contrary of αA. For a rule rR of the form b0b1,,bm, let the head be head(r)=b0 and the body be body(r)={b1,,bm}. We enforce that ABA frameworks are flat, namely assumptions in A do not appear in the heads of rules in R. In an ABA framework, arguments and attacks are defined as follows.

Definition 1

Definition 1([10,11]).

Let L,R,A,¯¯ be an ABA framework. An argument for the conclusion (or claim) cL supported by KA (Kc in short) is a (finite) tree with nodes labelled by sentences in L or by the special symbol τL representing “true”, such that

  • the root is labelled by c,

  • for every node N

    • - if N is a leaf then N is labelled by an assumption in A or by τ;

    • - if N is not a leaf and b0 is the label of N, then there is an inference rule b0b1,,bm(m0) and either m=0 and the child of N is labelled by τ or m>0 and N has m children, labelled by b1,,bm respectively,

  • K is the set of all assumptions labelling the leaves.

attacks between arguments and attacks between sets of assumptions are defined respectively as follows.
  • An argument K1c1 attacks an argument K2c2 iff c1=α for some αK2.

  • For Δ,ΔA, and αA, Δ attacks α iff Δ enables the construction of an argument for conclusion α. Accordingly, Δ attacks Δ if Δ attacks some αΔ.

A set of arguments Args is conflict-free iff ∄A,BArgs such that AattacksB. Args defends an argument A iff each argument that attacks A is attacked by an argument in Args. On the other hand, a set of assumptions Δ is conflict-free iff Δ does not attack itself. Δ defends αA iff each ΔA that attacks α is attacked by Δ.

Let AFF=(AR,attacks) be the abstract argumentation (AA) framework [8] generated from an ABA framework F, where AR is the set of all arguments such that aAR iff an argument a:Kc is in F, and (a,b)attacks in AFF iff a attacks b in F [11].

Let σ{complete,preferred,grounded,stable,ideal} be the name of the argumentation semantics. The ABA semantics is given by σ argument extensions as well as by σ assumption extensions. A σ argument extension ArgsAR and a σ assumption extension ΔA are defined respectively as follows.

ArgsAR is:  admissible iff Args is conflict-free and defends all its elements;  a complete argument extension iff Args is admissible and contains all arguments it defends;  a preferred (resp. grounded) argument extension iff it is a (subset-)maximal (resp. (subset-)minimal) complete argument extension;  a stable argument extension iff it is conflict-free and attacks every argument in ARArgs; an ideal argument extension iff it is a (subset-)maximal complete argument extension that is contained in each preferred argument extension.  In contrast, ΔA is: admissible iff Δ is conflict-free and defends all its elements; a complete assumption extension iff Δ is admissible and contains all assumptions it defends; a preferred (resp. grounded) assumption extension iff it is a (subset-)maximal (resp. (subset-)minimal) complete assumption extension;  a stable assumption extension iff it is conflict-free and attacks every ψAΔ; an ideal assumption extension iff it is a (subset-)maximal complete assumption extension that is contained in each preferred assumption extension.

There is a one-to-one correspondence between σ argument extensions and σ assumption extensions.

Let claim(Ag) be the conclusion (or claim) of an argument Ag. The conclusion of a set of arguments E is defined as Concs(E)={cL|c=claim(Ag)for an argumentAgE}.

Rationality postulates [6] are stated in terms of ABA in [12] as follows.

Definition 2

Definition 2(Rationality postulates [12]).

Let L,R,A,¯¯ be a flat ABA framework, where ¯¯ has the property such that contraries of assumptions are not assumptions. A set XL is said to be contradictory iff X is contradictory w.r.t. ¯¯, i.e. there exists an assumption αA such that {α,α}X; or X is contradictory w.r.t. ¬, i.e. there exists sL such that {s,¬s}X if L contains an explicit negation operator ¬. Let CNR:2L2L be a consequence operator. For a set XL, CNR(X) is the smallest set such that XCNR(X), and for each rule rR, if body(r)CNR(X) then head(r)CNR(X). X is closed iff X=CNR(X). A set XL is said to be inconsistent iff its closure CNR(X) is contradictory. X is said to be consistent iff it is not inconsistent. A flat ABA framework F=L,R,A,¯¯ is said to satisfy the consistency-property (resp. the closure-property) if for each complete extension E of AFF generated from F, Concs(E) is consistent (resp. Concs(E) is closed).

We say that a set of arguments E is consistent if Concs(E) is consistent [33].

In [6], rationality postulates are defined using notions such as direct consistency, indirect consistency and closure-property. If we use the notions, it may be said that under σ argumentation semantics, AFF satisfies closure iff for each σ-extension E of AFF, Concs(E)=CNR(Concs(E)); AFF satisfies direct consistency iff for each σ-extension E, Concs(E) is not contradictory; and AFF satisfies indirect consistency iff for each σ-extension E, Concs(E) is consistent, that is, CNR(Concs(E)) is not contradictory.

Heyninck and Arieli [18] proposed (generalized) assumption-based frameworks as follows. Let L be a propositional language, p,piL be atomic formulas (or atoms, for short), ψ,ϕL be compound formulas and Γ,Γ,ΛL be sets of formulas. L contains conjunction “,” disjunction “∨”, implication “→”, a negation operator “∼”, and a propositional constant T for truth.

A (propositional) logic for a language L [17,18] is a pair L=(L,), where ⊩ is a (Tarskian) consequence relation for L [36], which is a binary relation between sets of formulas and formulas in L satisfying the conditions such that it is reflexive (if ψΓ then Γψ), monotonic (if Γψ and ΓΓ, then Γψ), and transitive (if Γψ and Γ{ψ}ϕ, then ΓΓϕ). The L-based transitive closure of a set Γ of L-formulas is CnL(Γ)={ψL|Γψ} [17]. Let (·) be the powerset operator.

Definition 3

Definition 3(Assumption-based frameworks [18]).

An assumption-based framework is a tuple ABF=L,Γ,Λ,, where L=L, is a (propositional) Tarskian logic for a propositional language L, Γ (the strict assumptions) and Λ (the candidate or defeasible assumptions) where Λ{} are distinct countable sets of L-formulas, and −:Λ(L) is a contrariness operator, assigning a finite set of L-formulas to every defeasible assumption in Λ.

In ABF, attacks is defined as follows: For Δ,ΘΛ and ψΛ, Δ attacks ψ iff ΓΔϕ for some ϕψ. Accordingly, Δ attacks Θ if Δ attacks some ψΘ.

The usual semantics in AA frameworks [8] is adapted to their ABFs as follows.

Definition 4

Definition 4(ABF semantics [5]).

Let ABF=L,Γ,Λ, be an assumption-based framework and let ΔΛ. A set of defeasible assumptions Δ is closed iff Δ={αΛ|ΓΔα}. ABF is said to be flat iff every set of defeasible assumptions Δ is closed. Then Δ is conflict-free iff there is no ΔΔ that attacks some ψΔ. Δ is stable iff it is closed, conflict-free and attacks every ψΛΔ.

2.2.Disjunctive logic programs

A disjunctive logic program (DLP) [26] is a finite set of rules of the form44

(1)p1pkpk+1,,pm,notpm+1,,notpn,
where nmk>0. Each pi (1in) is a ground atom. The symbol not55 is the negation-as-failure (NAF) operator. An atom preceded by not is called a NAF-atom. Let P be a DLP and HBP be the Herbrand base of P, i.e. the (finite) set of all ground atoms in the language of P. Let MHBP. A set M satisfies a ground rule of the form (1) if {pk+1,,pm}M and {pm+1,,pn}M= imply {p1,,pk}M. M is a model of P if it satisfies every ground rule in P. The reduct of P w.r.t. M is the DLP PM={p1pkpk+1,,pm|thereisaruleoftheformp1pkpk+1,,pm,notpm+1,,notpninPs.t.{pm+1,,pn}M=}. Then M is a stable model of P if it is a ⊆-minimal model of PM [14,26].

In [18], stable models of a DLP P are defined based on A(P)HBP, where A(P) is the set of atomic formulas that appear in P, and the symbol ∼ is used instead of not to denote NAF. Heyninck and Arieli [18] denoted by L=LDLP, the logic for the language LDLP consisting of disjunctions of atoms (p1pn, for n1), NAF-atoms (notp) and rules of the form (1) of a DLP, where ⊩ is constructed for LDLP by three inference rules: Modus Ponens (MP), Resolution (Res) and Reasoning by Cases (RBC). (The precise form of these inference rules will be described later in Remark 1.)

The ABF induced by a DLP P is defined by ABF(P)=L,P,A(P), [18], where L=LDLP,, A(P)={notp|pA(P)}, and notp={p} for every pA(P). All the ABFs induced by DLPs are based on the same core logic L=LDLP,. Heyninck and Arieli showed a one-to-one correspondence between stable models of a DLP P and stable assumption extensions of ABF(P) as follows.

Proposition 1

Proposition 1([18, Proposition 2 and Proposition 3]).

Let P be a finite DLP and MA(P). If M is a stable model of P, then M={notp|p(A(P)M)} is a stable (assumption) extension of ABF(P). Conversely if E is a stable (assumption) extension of ABF(P), then E_={pA(P)|notpE} is a stable model of P.

2.3.Extended disjunctive logic programs

We consider a finite propositional extended disjunctive logic program (EDLP) [15] in this paper. An EDLP is a finite set of rules of the form:

(2)L1||LkLk+1,,Lm,notLm+1,,notLn,
where nmk>0. Each Li (1in) is a ground literal, that is, either a ground atom A (i.e. a propositional variable) or ¬A preceded by classical negation “¬”, where ¬A is called a negative literal. not denotes the negation-as-failure (NAF) as before, and notL is called a NAF-literal. The left (resp. right) part of ← is the head (resp. the body). body(r) denotes the body of a rule r from an EDLP. The symbol “|” is used to distinguish disjunction in the head of a rule from disjunction “∨” used in classical logic. An EDLP is called a normal disjunctive logic program (NDP) if ¬ does not occur in it, while an EDLP is called an extended logic program (ELP) if it contains no disjunction (k=1). An ELP is called a normal logic program (NLP) if ¬ does not occur in it. Let LitP (resp. HBP) be the set of all ground literals (resp. all ground atoms) in the language of an EDLP P. When an EDLP P is an NDP (or NLP), LitP reduces to HBP.

The semantics of an EDLP is given by answer sets [15].

Definition 5

Definition 5(answer sets).

Let SLitP. First, let P be a not-free EDLP (i.e. m=n for each rule in P). Then, S is an answer set of P if S is a minimal set (w.r.t. ⊆) satisfying the two conditions (i), (ii):

  • (i) For each rule L1||LkLk+1,,Lm in P, if {Lk+1,,Lm}S, then LiS for some i (1ik).

  • (ii) If S contains a pair of literals L and ¬L, then S=LitP.

Second, let P be any EDLP. The reduct PS of P w.r.t. S is the not-free EDLP PS such that PS={L1||LkLk+1,,Lm| there is a rule of the form L1||LkLk+1,Lm,notLm+1,notLn inPs.t.{Lm+1,,Ln}S=}. Then S is an answer set of P if S is the answer set of PS.

An answer set S is consistent if SLitP; otherwise S is contradictory. An EDLP P having a consistent answer set is consistent; otherwise P is inconsistent under answer set semantics.

It is considered, for an answer set S and a literal L, (i) SL iff LS, (ii) SnotL iff LS.

A literal LLitP is interpreted “unknown” under the answer set semantics if LS and ¬LS for any answer set S of an EDLP P.

The following example illustrates the difference between “|” and “∨” in logic programming.

Example 2.

The EDLP {p|¬p} has two answer sets {p} and {¬p}. If we use ∨ instead of |, however, {p¬p} is logically equivalent to {pp} under classical logic, which has the minimal model (stable model) {}. Hence, | and ∨ have different meanings in general.

The semantics of an EDLP is also given by (four-valued) paraconsistent stable models (or p-stable models66) [28], which are regarded as answer sets defined without the condition (ii) in Definition 5, denoting that we don’t get the trivialization or deductive explosion of inconsistent (p-stable) models.

Definition 6

Definition 6(p-stable models).

Let MLitP. First, let P be a not-free EDLP (i.e. m=n for each rule). Then, M is a p-stable model of P if M is a minimal set (w.r.t. ⊆) satisfying the following condition: For each rule L1||LkLk+1,,Lm in P, if {Lk+1,,Lm}M, then LiM for some i (1ik). Second, let P be any EDLP. The reduct PM of P w.r.t. M is the not-free EDLP PM defined in Definition 5. Then M is a p-stable model of P if M is a p-stable model of PM.

In [28], ILitP is considered as an interpretation which is defined as a function I:LitP{t,f,,}, where t,f,, are the four-valued truth values denoting true, false, contradictory, and undefined. Then, the truth value I(L) is assigned to each literal LLitP such that:

  (i) I(L)=t   if LI and ¬LI,     (ii) I(L)=f  if LI and ¬LI,

 (iii) I(L)=  if both LI and ¬LI,   (iv) I(L)=  otherwise.

A p-stable model M is inconsistent if M contains a pair of complementary literals (in other words, there exists a literal LM s.t. M(L)=); otherwise M is consistent. An EDLP P is consistent if P has a consistent p-stable model; otherwise P is inconsistent under paraconsistent stable model semantics.

Example 3.

Consider the EDLP P={¬p,p,q}. P has the inconsistent answer set LitP which is contradictory both w.r.t. p and q. P has also the inconsistent p-stable model M={¬p,p,q}, where contradiction is localized w.r.t. p, and the consistent information about q is obtained (i.e. M(p)=, M(q)=t). This shows that a p-stable model is paraconsistent even if it is inconsistent.

Gelfond et al. [16] proposed a disjunctive default theory (ddt, for short), which is a set of disjunctive defaults whose form is α:β1,,βmγ1|γ2||γn. The semantics of a ddt is given by extensions which are generalization of Reiter’s extensions for a default theory [27]. Regarding logic programming, they showed in the following theorem that a propositional EDLP P can be translated into a disjunctive default theory emb(P) by replacing every rule in P of the form (2) with the disjunctive default

Lk+1Lm:¬Lm+1,,¬LnL1||Lk.

The intuition behind the disjunctive default can be “if each of Lk+1,,Lm is believed and if each of ¬Lm+1,,¬Ln can be consistently believed, then Li is believed for some i (1ik)”.

The disjunctive defaults of this form are used to compute extensions of the ddt emb(P). There exists the correspondence between answer sets of an EDLP P and extensions of emb(P) as follows.

Theorem 1

Theorem 1([16, Theorem 7.2]).

Let P be a propositional EDLP. Then S is an answer set of P iff S is the set of all literals from an extension of the disjunctive default theory emb(P).

Let F(P)=LP,P,AP,¯¯ be the ABA framework translated from an ELP P, where NAFP={notL|LLitP}, LP=LitPNAFP, AP=NAFP, and notL=L for each notLAP. Given MLitP, let ΔM={notL|LLitPM}.77 Let Ptr=defP{Lp,¬p|pLitP,LLitP} be the ELP obtained from an ELP P by incorporating the trivialization rules [28]. It was shown that answer sets (resp. paraconsistent stable models) of an ELP P can be captured by stable argument extensions of the ABA framework translated from the ELP Ptr (resp. P) as follows.

Theorem 2

Theorem 2([32, Theorem 3]).

Let P be an ELP and MLitP. Then M is a p-stable model of P iff there is a stable (argument) extension E of the ABA framework F(P) such that MΔM=Concs(E) (in other words, M=Concs(E)LitP).

Theorem 3

Theorem 3([32, Theorem 4]).

Let P be an ELP and SLitP. Then S is an answer set of P iff there is a stable (argument) extension Etr of the ABA framework F(Ptr) such that SΔS=Concs(Etr) (in other words, S=Concs(Etr)LitP).

For a consistent ELP, the following theorem holds.

Theorem 4

Theorem 4([33, Theorem 5]).

Let P be a consistent ELP and SLitP. Then S is an answer set of P iff there is a consistent stable (argument) extension E of the ABA framework F(P) such that SΔS=Concs(E).

Notice that ΔM in Theorem 2 (resp. ΔS in Theorem 3) is a stable assumption extension of the ABA framework F(P) (resp. F(Ptr)), while ΔS in Theorem 4 is a consistent stable assumption extension of the ABF F(P) due to Proposition 12, Proposition 13 and Proposition 14 as proved in Appendix respectively.

3.ABA for extended disjunctive logic programming

3.1.Assumption-based frameworks translated from EDLPs

We propose an assumption-based framework (ABF) translated from an EDLP, which incorporates explicit negation along with | instead of ∨ in Heyninck and Arieli’s ABF induced by a DLP to achieve our purpose shown in the introduction. An ABF translated from an EDLP is based on the logic constructed by three inference rules: Modus Ponens (MP), Resolution (Res) and Reasoning by Cases (RBC):

[MP]ψϕ1,,ϕnϕ1ϕ2ϕnψ[Res]ψ1||ψm|1||n|ψ1||ψknot1notnψ1||ψm|ψ1||ψk12n[RBC]ψψψ1||nψ
where | is the connective of a disjunction, i is a propositional literal, each ϕi{i,noti} is a propositional literal or its NAF-literal, and ψ, ψi are disjunctions of propositional literals using |. ψϕ1,,ϕn is a rule of the form (2). Since ψ is identified with ψT, [MP] implies Reflexivity: [Ref]ψψ.

Definition 7.

We denote by L=LEDLP, the logic for the language LEDLP which consists of disjunctions of propositional literals (1||n, for n1), NAF-literals (not) and rules of the form (2) of an EDLP, where ⊩ is constructed for LEDLP by three inference rules: Modus Ponens (MP), Resolution (Res) and Reasoning by Cases (RBC) above. In other words, ⊩ denotes derivability using three inference rules: [MP] (including [Ref]), [Res] and [RBC].

Remark 1.

Heyninck and Arieli’s ABF [18] is based on the logic L=LDLP,, where ⊩ is constructed for LDLP by three inference rules: [MP], [Res] and [RBC] having the restricted forms such that | (resp. a literal i) is replaced with ∨ (resp. an atom pi), each ϕi{pi,notpi} is an atom or a NAF-atom, and ψ, ψi are disjunctions of atoms (p1pn, for n1).

Σφ holds iff φ is either in Σ or is derived from Σ using the three inference rules above. According to Definition 3, Σφ iff φCnL(Σ), where CnL(Σ) is the L-based transitive closure of Σ (namely, the ⊆-smallest set that contains Σ and is closed under [MP], [Res] and [RBC]). Notice that for any φCnL(Σ), if φ is not of the form 1||n, then φΣ.

For a special ABF =L,Γ,Λ, such that each defeasible assumption from Λ has a unique contrary (i.e. |α|=188 for αΛ), we may denote such an ABF by a tuple L,Γ,Λ, ¯¯, where ¯¯ is a total mapping from Λ into L, and αL is the contrary of αΛ.

In what follows, let NAFP={not|LitP} and LP=LitPNAFP for an EDLP P. We are now ready to define an ABF translated from an EDLP.

Definition 8.

Let P be an EDLP. The assumption-based framework (ABF) translated from P is defined by ABF(P)=L,P,AP,¯¯, where L=LEDLP,, AP=NAFP={not|LitP}, and not= for every notAP.

In the following, we show the ABF translated from an EDLP P is always flat.

Proposition 2.

ABF(P)=L,P,AP,¯¯ translated from an EDLP P is flat.

Proof.

Let ΔAP. Since ⊩ is reflexive, it holds that (i) PΔα for αΔ. On the other hand, it holds that (ii) PΔβ for βAPΔ, since NAF-literals, which is the only form the defeasible assumptions in ABF(P) can take, do not occur in the heads of rules from P. Then due to (i),(ii), it holds that Δ={αAP|PΔα} for ΔAP. Thus ABF(P) is flat. □

In ABF(P), the semantics is given by assumption extensions as follows.

Definition 9.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P. Let ΔAP and αAP. Δ is conflict-free iff there is no ΔΔ that attacks some ψΔ. Δ defends α iff each Δ that attacks α is attacked by Δ. Then Δ is: admissible iff Δ is conflict-free and defends all its elements;  a complete assumption extension iff Δ is admissible and contains all assumptions it defends; a preferred (resp. grounded) assumption extension iff it is a (subset-)maximal (resp. (subset-)minimal) complete assumption extension; a stable assumption extension iff it is is conflict-free and attacks every ψAPΔ; an ideal assumption extension iff it is a (subset-)maximal complete assumption extension that is contained in each preferred assumption extension.

In ABF(P), consistency of a set of literals and NAF-literals XLP is defined using a consequence operator CNP which is adapted from the CNR operator found in standard ABA.

Definition 10.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P, where L=LEDLP,. For a set XLP=LitPNAFP, X is said to be contradictory iff X is contradictory w.r.t. ¯¯, i.e. there exists an assumption αAP s.t. {α,α}X; or X is contradictory w.r.t. ¬, i.e. there exists sLP s.t. {s,¬s}X. Let CNP:(LP)(LP) be a consequence operator such that for XLP,

CNP(X)=def{ϕLP|PXϕ}=CnL(PX)LP.

CNP(X) is said to be the closure of X. X is said to be closed w.r.t. CNP iff X=CNP(X). A set XLP is said to be inconsistent iff the closure CNP(X) is contradictory. X is said to be consistent iff it is not inconsistent.

3.2.Correspondence between answer sets of an EDLP and stable assumption extensions

Proposition 1 for a DLP [18] is generalized to Proposition 4 and Proposition 5 for an EDLP shown below. To this end, firstly as the extension of Proposition 1, we prepare the following corollary for a DLP whose stable models are defined based on HBP.

Corollary 1.

The extended ABF induced by a DLP P is defined by ABF(P)=L,P,HBP,, where L=LDLP,, HBP={notp|pHBP} and notp={p} for every pHBP. Let MHBP. (i) If M is a stable model of P, then Δ={notp|p(HBPM)} is a stable assumption extension of ABF(P). (ii) Conversely if Δ is a stable assumption extension of ABF(P), then M={pHBP|notpΔ} is a stable model of P.

Proof.

Let ΔA(P)={notp|p(HBPA(P))}.

  • (i) Let M be a stable model of P, where MA(P)HBP. Then for M={notp|p(A(P)M)}, it holds that MΔA(P)=Δ={notp|p(HPPM)}. Based on [18, Corollary 1], it holds that,

    for pA(P),pM iff pCnL(PM) iffpCnL(PΔ).Hence,

    for pHBP,pM iff pCnL(PΔ) iff PΔp.

    Thus Δ is conflict-free and attacks every notpΔ for pHBP, which means that Δ is a stable assumption extension of ABF(P).

  • (ii) Conversely, let ΔHBP be a stable assumption extension of ABF(P). Then for α{notpΔ|pHBP}, Δ attacks α. This means that, PΔp, that is, pCnL(PΔ) for notpΔ where pHBP.

     On the other hand, based on [18, Corollary 1], it holds that, for a stable model M of P,

    (3)pMiffpCnL(PM)for pA(P)iffpCnL(PMΔA(P))=CnL(PΔ)for pHBP.

    Thus it holds that, pM for notpΔ and pHBP where Δ is a stable assumption extension.  □

Next, Corollary 1 for a DLP is mapped to Proposition 3 for an NDP based on two lemmas about DLPs and NDPs as follows.

Lemma 1.

Let P be a propositional NDP and PD be the DLP translated from P which is obtained by replacing a rule: p1||pkpk+1,,pm,notpm+1,,notpn from P with the rule p1pkpk+1,,pm,notpm+1,,notpn, where HBPD=HBP. Let MHBP. Then M is an answer set of P iff M is a stable model of PD.

Proof.

Given MHBP, it holds that, {p1,,pk}M iff piM for some i (1ik).

This is used in the following proof.

⇒: Suppose that MHBP is an answer set of P. First, let P be a not-free NDP (i.e., m=n in (2)). Since M is an answer set of P, M is a ⊆-minimal set satisfying the condition such that if {pk+1,,pm}M, then some piM (1ik) for each rule p1||pkpk+1,,pm in P. Hence M is a ⊆-minimal set satisfying the condition such that if {pk+1,,pm}M, then {p1,,pk}M for each rule p1pkpk+1,,pm in PD, which means that M is a ⊆-minimal model of PD. Now since (PD)M=PD holds for the not-free PD, M is a stable model of PD. Second, let P be any NDP. Since M is an answer set of P, M is an answer set of the not-free NDP PM. Therefore, M is a stable model of the not-free DLP (PD)M, which means that M is a ⊆-minimal model of the DLP ((PD)M)M=(PD)M. Hence M is a stable model of PD.

⇐: Suppose that MHBP is a stable model of PD. First, let PD be the not-free DLP (i.e., m=n in (1)). Since M is a stable model of the not-free PD, M is a ⊆-minimal model of (PD)M=PD which satisfies the condition such that if {pk+1,,pm}M, then {p1,,pk}M for each rule p1pkpk+1,,pm in (PD)M=PD. Therefore, M is a ⊆-minimal set of P satisfying the condition such that if {pk+1,,pm}M, then some piM (1ik) for each rule p1||pkpk+1,,pm in P, which means that M is an answer set of the not-free NDP P. Second, let PD be any DLP. Since M is a stable model of PD, M is a ⊆-minimal model of the DLP (PD)M which is not-free. Hence M is a ⊆-minimal set satisfying the condition such that if {pk+1,,pm}M, then {p1,,pk}M for each rule p1pkpk+1,,pm in (PD)M, which means that, M is a ⊆-minimal set satisfying the condition such that if {pk+1,,pm}M, then some piM (1ik) for each rule p1||pkpk+1,,pm in PM. Therefore M is an answer set of PM. Hence M is an answer set of P. □

Though Lemma 1 holds, the difference between NDPs and DLPs appears when considering their relation to (disjunctive) default logic as shown in the following example. This implies that an NDP can be transformed to a disjunctive default theory but a DLP can never be done so.

Example 4.

Heyninck and Arieli considered the DLP π1={pq,qp,pq} in [18, Example 1], which has the unique stable model M={p,q}. By replacing ∨ with | in π1, we obtain the NDP π2={p|q,qp,pq} which has the unique answer set S={p,q}. Then thanks to Gelfond et al.’s Theorem 1 [16, Theorem 7.2], π2 can be translated into the disjunctive default theory d2 (namely, emb(π2)) below. Instead, π1 is expressed by the standard default theory d1 as follows since it uses the connective ∨ rather than | in rule head.

d1={p:q,q:p,pq},d2={p:q,q:p,p|q}.
As a result, d1 has the unique extension E1 consisting of pq and its logical consequences based on Reiter’s default theory [27], while d2 has the unique extension E2 consisting of p, q and their logical consequences based on Gelfond et al.’s disjunctive default theory [16]. Therefore there is no relationship between E1 and M, while there is the correspondence between E2 and S such that S=E2Litπ2. Thus the default theory d1 corresponding to the DLP π1 has the same difficulty as D1 addressed in Example 1, while the disjunctive default theory d2 corresponding to the NDP π2 avoids such a difficulty.

The following lemma is also needed to obtain Proposition 3 for an NDP.

Lemma 2.

Let MHBP be an answer set of an NDP P and let ΔM={notL|LHBPM}. Then pM iff pCnL(PΔM) for pHBP.

Proof.

Let PD be the DLP translated from an NDP P as defined in Lemma 1.

pM iff pCnL(PDΔM) for pHBP, where L=LDLP,,  (due to Lemma 1 and (3))

   iff pCnL(PΔM) for pHBP, where L=LEDLP,. □

Proposition 3.

Let P be an NDP, MHBP, and ABF(P)=L,P,AP,¯¯, where L=LEDLP,. (i) If M is an answer set of P, then Δ={notp|p(HBPM)} is a stable assumption extension of ABF(P). (ii) Conversely if Δ is a stable assumption extension of ABF(P), then M={pHBP|notpΔ} is an answer set of P.

Proof.

This follows from Corollary 1 for a DLP based on Lemma 1 and Lemma 2. □

Now, we show that p-stable models (resp. answer sets) of an EDLP P are captured by stable assumption extensions of the ABF translated from P (resp. Ptr). To this end, the notation “+” [15,28] is used.

A positive form of an EDLP P [28] is the NDP P+ which is obtained by replacing each negative literal ¬L in P (where L is an atom) with a corresponding newly introduced atom L (called the positive form of ¬L [15]) in P+. Let M+ be an answer set of such P+. Then the following lemma holds by definition.

Lemma 3.

Let P be an EDLP and P+ be its positive form. Let MLitP. Then M is a p-stable model iff M+ is an answer set of P+.

Proposition 4.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P and MLitP.

If M is a p-stable model of P, then Δ={not|(LitPM)} is a stable assumption extension of ABF(P). Conversely if Δ is a stable assumption extension of ABF(P), then M={LitP|notΔ} is a p-stable model of P.

Proof.

Let P+ be the NDP which is the positive form of an EDLP P. For a set SLitP, let S+ be the set obtained by replacing each negative literal ¬L in S with a newly introduced atom L. Then the Herbrand base HBP+ of P+ is (LitP)+. Moreover, if a set U+HBP+ is given, we denote by U the set obtained by replacing each newly introduced atom LU+ with the corresponding original literal ¬LLitP.

  • (i) Let M be a p-stable model of an EDLP P. Then due to Lemma 3, M+ is an answer set of the NDP P+. Hence ΔM+={notp|p(HBP+M+)} is a stable assumption extension of ABF(P+) due to Proposition 3. Thus Δ=ΔM={not|(LitPM)} is a stable assumption extension of ABF(P).

  • (ii) Conversely, let Δ=ΔM={not|(LitPM)} be a stable assumption extension of ABF(P). Then ΔM+={notp|p(HBP+M+)}={notp|p(LitPM)+)} is a stable assumption extension of ABF(P+). Thus due to Proposition 3, M+={pHBP+|notpΔM+} is the answer set of the NDP P+. Hence M={LitP|notΔM} where Δ=ΔM is a p-stable model of P due to Lemma 3. □

Proposition 5.

Let P be an EDLP, SLitP and ABF(Ptr)=L,Ptr,AP,¯¯ be the ABF translated from the EDLP Ptr=P{Lp,¬p|pLitP,LLitP}, where LPtr=LP. If S is an answer set of P, then Δ={not|(LitPS)} is a stable assumption extension of ABF(Ptr). Conversely if Δ is a stable assumption extension of ABF(Ptr), then S={LitP|notΔ} is an answer set of P.

Proof.

[28, Theorem 3.5] shows that S is an answer set of an EDLP P iff S is p-stable model of Ptr. Then this follows from Proposition 4 based on [28, Theorem 3.5]. □

CNP(Δ) gives us the conclusion of an assumption extension Δ. In Proposition 6 below, we show the relationship between answer sets (or p-stable models) of an EDLP and the conclusions of assumption extensions of the translated ABF.

Lemma 4.

Let M be a p-stable model of an EDLP P and ΔM={not|(LitPM)}.  Then

M iff CnL(PΔM) for LitP.

Proof.

This follows from Lemma 2 based on the renaming technique used in the proof of Proposition 4. □

Proposition 6.

Let P be an EDLP, M (resp. S) be a p-stable model (resp. an answer set) of P, and ΔM={not|(LitPM)} (resp. ΔS={not|(LitPS)}) be the stable assumption extension of ABF(P) (resp. ABF(Ptr)). Then it holds that, CNP(ΔM)=MΔM, and CNPtr(ΔS)=SΔS.

Proof.

 

  • 1. Let M be a p-stable model of an EDLP P. Then for LitP, it holds that M iff CnL(PΔM) iff CNP(ΔM) according to Lemma 4. Besides notΔM iff notCNP(ΔM). Hence CNP(ΔM)=MΔM holds.

  • 2. Let S be an answer set of P. Then S is a p-stable model of Ptr due to [28, Theorem 3.5]. Besides ΔS={not|(LitPS)} is the stable assumption extension of ABF(Ptr) due to Proposition 5. Then by applying the result of Item 1 to a p-stable model S of Ptr, we obtain CNPtr(ΔS)=SΔS. □

In all examples shown in this paper, we assume that LitP i.e. the set of all ground literals in the language of an EDLP P coincides with the set {L|Lor¬LappearsinP}99 where ¬¬L=L.

Example 5

Example 5(Cont. Example 1).

Consider Kyoto protocol problem. For the EDLP P1, LitP1={p,r,y,f,k,¬p,¬r,¬y,¬f,¬k}, and (P1)tr=P1{Lp,¬p|pLitP1,LLitP1}. Answer sets of (P1)tr coincide with those of P1, that is, S1={p,r,k} and S2={y,f,k}. According to Proposition 5, ABF((P1)tr) (resp. ABF(P1)) has two stable assumption extensions Δ1 and Δ2 such that

Δ1={noty,notf,not¬p,not¬r,not¬y,not¬f,not¬k},Δ2={notp,notr,not¬p,not¬r,not¬y,not¬f,not¬k}.
The conclusion of each assumption extension Δi defined as CNP1(Δi) (i=1,2) is obtained as follows:
CNP1(Δ1)={p,r,k,noty,notf,not¬p,not¬r,not¬y,not¬f,not¬k}=S1Δ1,CNP1(Δ2)={y,f,k,notp,notr,not¬p,not¬r,not¬y,not¬f,not¬k}=S2Δ2.
As a result, the expected result is successfully obtained since kCNP1(Δi) (i=1,2).

Example 6.

Consider logic programs P={a,¬a,bnotb} and Q={¬a,anotb} shown in [28, Example 3.4]. Both are inconsistent under the answer set semantics (resp. under the p-stable model semantics) since P has the unique answer set S=LitP (resp. no p-stable model), while Q has no answer set (resp. only the inconsistent p-stable model M={¬a,a}). Correspondingly, ABF(P)=L,P,AP,¯¯ has no stable assumption extension (though it has the unique complete, preferred and grounded assumption extension {not¬b} which is inconsistent since CNP({not¬b})={a,¬a,not¬b}), while ABF(Ptr)=L,Ptr,AP,¯¯ has the unique stable assumption extension Δ={} which is inconsistent since CNPtr(Δ)=LitP={a,¬a,b,¬b}=SΔ. In contrast, ABF(Q) has the unique stable (resp. complete, preferred and grounded) assumption extension Δ={notb,not¬b} which is inconsistent since CNQ(Δ)={a,¬a,notb,not¬b}=MΔ, while ABF(Qtr) has no stable assumption extension.

3.3.Arguments and argument extensions in ABFs translated from EDLPs

Given an (E)DLP, it is impossible to capture its semantics by using arguments constructed based on Definition 1. The reason is as follows. For example, consider the EDLP P={p|q} which has two answer sets, {p} and {q} (or the DLP PD={pq} which has two stable models, {p} and {q}). Then if we use Definition 1 in the ABA framework F(P) (or F(PD)) instantiated with P (or PD), we may construct the tree which has the root labelled by p|q (or pq) and the unique child labelled τ together with two one-node-trees whose roots are labelled by either notp or notq, but no other trees. As a result, since there exists no attacks among these three arguments, there is the unique argument extension E consisting of these arguments s.t. Concs(E)={p|q,notp,notq} (or s.t. Concs(E)={pq,notp,notq}) under any σ semantics in the ABF F(P) (or F(PD)). Thus two answer sets shown above can never be captured based on Definition 1. In contrast, in ABF(P) (or in ABF(PD)), we can use the inference rule [Res]. Hence, if there is the aforementioned tree whose root is labelled by p|q, thanks to the inference rule [Res], we can construct two trees furthermore, each of which has the root labelled by p (resp. q) as well as two children such that one is the child tree whose root is labelled by p|q and the other is the child labelled by the defeasible assumption notq (resp. notp). As a result, two arguments (i.e. trees) whose roots are labelled by either p or q attack each other, which enables us to capture two answer sets of P. The same goes for ABF(PD) to capture two stable models of PD. Thus, this example illustrates that Definition 1 has the difficulty to deal with disjunctive rules in the ABA framework F(P) (resp. F(PD)), whereas inference rules provided in ABF(P) (resp. ABF(PD)) are useful to construct arguments using disjunctive rules. (Furthermore, see details in Example 7 as shown below.)

In what follows, arguments and attacks in the ABF translated from an EDLP P are defined.

Definition 11.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P, where L=LEDLP,. Ψ belonging to LEDLP is said to be a defeasible consequence of P and KAP if PKΨ in which any assumption contained in K is used to derive Ψ. K is said to be a support for Ψ w.r.t. P.

PKΨ addressed above is represented by a tree structure TΨ(K) as follows.

Definition 12.

Let L=LEDLP, and ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P. Let TΨ(K) denote PKΨ where K is the support for a defeasible consequence Ψ w.r.t. P. In other words, TΨ(K) is a (finite) tree with a root node labelled by Ψ (having a support K) defined as follows.

  • 1. The cases using no inference rules:

    • (1) For notAP, there is a one-node tree TΨ(K) whose root node is labelled by Ψ=not and K={not}.

    • (2) For a rule rP, there is a one-node tree TΨ(K) whose root node is labelled by Ψ=r and K=.

  • 2. The cases using inference rules:

    • (1) i. For a rule ψP, by [Ref], there is a tree Tψ(K) whose root node N is labelled by ψ and N has a unique child node, namely a one-node tree Tr() where r=ψ. Then K=.

      ii. For a rule ψϕ1,,ϕn in P, if for each ϕi (1in), there exists a tree Tϕi(Ki) with the root node Ni labelled by ϕi, then by [MP], there is a tree Tψ(K) with the root node N labelled by ψ and N has a child N0 labelled by r=ψϕ1,,ϕn which is a one-node tree Tr() as well as n children Ni (1in) where Ni is the root of a tree Tϕi(Ki). Then K=iKi.

    • (2) Let Φ=ψ1||ψm|1||n|ψ1||ψk and Ψ=ψ1||ψm|ψ1||ψk,

      where iLitP (1in). If there is a tree TΦ(K) with the root node N0 labelled by Φ, then by [Res], there is a tree TΨ(K) with the root node N labelled by Ψ and N has a child N0 as well as n children N1,Nn each of which is a one-node tree Tϕi({ϕi}) where ϕi=noti (1in). Then K=Ki=1n{noti}.

    • (3) Let (iψ)1010 denote the reasoning for the case i and Ti() be a one-node tree whose root is labelled by i. Suppose that

      • there is a tree TΦ(K) whose root node N0 is labelled by Φ=1||n; and

      • for each i (1in), there exists reasoning for a case i such that (iψ), namely P{i}Kiψ for KiAP, which is represented by a tree Tψ(Ki) constructed by newly introducing a tree Ti() in this definition.

      Then by [RBC], there is a tree Tψ(K) with the root node N labelled by ψ and N has the child N0 as well as n children N1,Nn where each Ni (1in) is the root of a tree Tψ(Ki) for the case i. Thus K=Ki=1n{Ki}. □

Given ABF(P), the set of all trees TΨ(K) is uniquely determined based on Definition 12.

In ABF(P), an argument is defined as a special tree Tϕ(K) whose root is labelled by a literal or a NAF-literal ϕLP, and the attack relation attacks is defined as usual.

Definition 13.

Let ABF(P)=L,P,AP, ¯¯ be the ABF translated from an EDLP P and ϕLP=LitPNAFP. Then in ABF(P),

  • an argument for a conclusion (or claim) ϕ supported by KAP (Kϕ, for short) is a (finite) tree Tϕ(K) whose root node is labelled by ϕLP.

  • K1ϕ1 attacks K2ϕ2 iff ϕ1=α for some αK2.

Notation 1.

Given an EDLP P, we often use a unique name to denote an argument Kϕ in ABF(P), e.g. a:Kϕ is an argument with name a. With an abuse of notation [31, Notation 3], the name of an argument sometimes stands for the whole argument, for example, a denotes the argument a:Kϕ.

Let AF=(AR,attacks) be the abstract argumentation (AA) framework generated from ABF(P), where AR is the set of all arguments such that aAR iff an argument a:Kϕ is in ABF(P), and (a,b)attacks in AF iff aattacksb in ABF(P).

Fig. 1.

Arguments of ABF(P) for P={p|q} in Ex. 7

Arguments of ABF(P) for P={p|q←} in Ex. 7
Fig. 2.

AF=(AR,attacks) in Ex. 7

AF=(AR,attacks) in Ex. 7

The semantics is also given by argument extensions in ABF(P) as follows.

Definition 14.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P, AR be the set of arguments generated from ABF(P), and ArgsAR. Args is conflict-free iff ∄A,BArgs such that AattacksB. Args defends an argument A iff each argument that attacks A is attacked by an argument in Args. Then ArgsAR is: admissible iff Args is conflict-free and defends all its elements; a complete argument extension iff Args is admissible and contains all arguments it defends; a preferred (resp. grounded) argument extension iff it is a (subset-)maximal (resp. (subset-)minimal) complete argument extension; a stable argument extension iff it is conflict-free and attacks every argument in ARArgs; an ideal argument extension iff it is a (subset-)maximal complete argument extension that is contained in each preferred argument extension.

Example 7.

Consider the EDLP P={p|q}. Then ABF(P) has four arguments Ai (1i4):

A1:{notq}p,A2:{notp}q,A3:{notp}notp,A4:{notq}notq,
whose tree structures are shown in Fig. 1. Fig. 2 shows the AA framework AF=(AR,attacks) for the ABF. It has two argument extensions E1, E2 under stable and preferred semantics as follows:
E1={A1,A4},E1={A2,A3},where Concs(E1)={p,notq},Concs(E2)={q,notp}.

Let σ{complete, preferred, grounded, stable, ideal}. It is shown that there is a one-to-one correspondence between σ argument extensions and σ assumption extensions of ABF(P) for an EDLP P.

Definition 15.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P, and AR be the set of all arguments that can be generated from ABF(P). Asms2Args:(AP)(AR) and Args2Asms:(AR)(AP) are functions such that

Asms2Args(Asms)={KϕAR|KAsms},Args2Asms(Args)={αAP|αKforanargumentKϕArgs}.

Theorem 5.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P, and AR be the set of all arguments generated from ABF(P). The following holds.

  • 1. If AsmsAP is a σ assumption extension, then Asms2Args(Asms) is a σ argument extension.

  • 2. If ArgsAR is a σ argument extension, then Args2Asms(Args) is a σ assumption extension.

Proof.

In case σ=stable, this is proved as follows.

  • 1. Let AsmsAP be a stable assumption extension and Args=Asms2Args(Asms).

    • (i) Since Asms is conflict-free, every argument constructed from Asms (i.e. Kϕ where KAsms) does not attack any assumption in Asms. In other words, any argument in Args does not attack any argument constructed from Asms. This means that Args is conflict-free.

    • (ii) Since Asms is a stable assumption extension, Asms attacks any assumption in APAsms. Let B=(KBϕ) be any argument in ARArgs, where KBAsms and (APAsms)KB. Then Asms attacks (APAsms)KB. Therefore Asms attacks B, denoting that Args attacks B. Hence Args attacks any argument in ARArgs.

    Due to (i) and (ii), Args is conflict-free and attacks every argument in ARArgs. Thus Args is a stable argument extension.

  • 2. Let ArgsAR be a stable argument extension and Asms=Args2Asms(Args).

    • (i) Args is conflict-free since Args is a stable assumption extension. Suppose Asms is not conflict-free. That is, Asms attacks some assumption αAsms. Then it is possible to construct some argument based on Asms (say A=(KAϕ) where KAAsms) whose conclusion ϕ is the contrary of some assumption α in Asms, i.e. ϕ=α for αAsms. The argument A cannot be a member of Args (otherwise Args would not be conflict-free). Hence AARArgs. Besides for αKAAsms, Args does not attack α since Args is conflict-free. This means that Args cannot attack the argument AARArgs. Hence Args is not a stable argument extension. Contradiction. Therefore Asms is conflict-free.

    • (ii) Suppose there exists some assumption αAPAsms which Asms does not attack. Due to αAsms, there is no argument in Args whose support contains α. Hence ({α}α)ARArgs. Since Args is a stable argument extension, there exists some argument B=(KBϕ)Args with ϕ=α and KBAsms which attacks the argument {α}α. This means that Args attacks {α}α. In other words, Asms attacks αAPAsms. Contradiction.

    Due to (i) and (ii), Asms is conflict-free and it attacks every assumption αAPAsms. Therefore Asms is a stable assumption extension.

The other cases are proved in a similar way to the proof in [32, Theorem 2]. □

When an ELP P with no disjunction is given, an argument Kϕ in the ABA framework F(P) is the tree constructed in accordance with Definition 1, whereas in ABF(P), an argument Kϕ is the tree TΨ(K) defined by Definition 12, which is constructed by using only the inference rule [MP] (including [Ref]). Though the tree structure of Kϕ in the ABA framework F(P) is different from that of Kϕ in ABF(P), they are semantically equivalent as follows.

Proposition 7.

Let P be an ELP, ϕLP, and KAP=NAFP. Then Kϕ is an argument of the ABA framework F(P)=LP,P,AP,¯¯ iff there is an argument Kϕ of ABF(P)=L,P,AP, ¯¯, where L=LEDLP,.

Proof.

We denote by MP derivability using modus ponens on ← as the only inference rule in the ABA framework F(P). When used on PK for KNAFP, MP treats NAF literals purely syntactically [31]. On the other hand, we denote by MP derivability using only the inference rule [MP] in ABF(P). Then

Kϕ is an argument in the ABA framework F(P), where K is a support for ϕLP=LP

iff PKMPϕ w.r.t. a support KAP for ϕ   (due to [31, Lemma 2])

iff PKMPϕ w.r.t. a support KAP for ϕ in ABF(P)

iff PKϕ w.r.t. a support KAP for ϕ in ABF(P)

iff Kϕ is an argument in ABF(P). □

The following proposition denotes that, given an ELP, the same abstract argumentation framework is obtained regardless of whether arguments are constructed according to either Definition 1 or Definition 12.

Proposition 8.

Given an ELP P, the abstract argumentation (AA) framework generated from the ABA framework F(P) coincides with the AA framework generated from ABF(P).

Proof.

Given an ELP P, let AF1=(AR1,attacks1) be the AA framework generated from the ABF F(P)=LP,P,AP,¯¯ and AF2=(AR2,attacks2) be the AA framework generated from ABF(P)=L,P,AP,¯¯.

  • (1) Then, aAR1 iff there exists (a:Kϕ) in F(P) where KAP and ϕLP

    iff there exists (a:Kϕ) in ABF(P) where KAP and ϕLP due to Proposition 7

    iff aAR2.

  • (2) (a,b)attacks1 iff a attacks b in F(P) iff (a:Kϕ) attacks (b:Kϕ) in F(P)

    iff there exist (a:Kϕ) and (b:Kϕ) s.t. notϕK in F(P)

    iff there exist (a:Kϕ) and (b:Kϕ) s.t. notϕK in ABF(P) due to Proposition 7

    iff (a:Kϕ) attacks (b:Kϕ) in ABF(P)

    iff a attacks b in ABF(P) iff (a,b)attacks2.

Thanks to (1) and (2), AR1=AR2 as well as attacks1=attacks2 hold. Hence AF1=AF2. □

3.4.Correspondence between answer sets of an EDLP and stable argument extensions

In ABF(P) translated from an EDLP P, the conclusion of a set of arguments E is defined as:

Concs(E)={ϕLP|ϕis a conclusion (or claim) of an argumentcontained inE}.

First of all, we show there is a one-to-one correspondence between answer sets of an NDP P and stable argument extensions of the ABF translated from P.

Theorem 6.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an NDP P where L=LEDLP,. Let MHBP and EAR, where AR is the set of all arguments generated from ABF(P). Then M is an answer set of an NDP P iff there is a stable argument extension E of ABF(P) such that MΔM=Concs(E)=CNP(ΔM), where ΔM={notp|p(HBPM)} is a stable assumption extension of ABF(P).

Proof.

Let NAFP={notp|pHBP} and ΔMNAFP. Firstly we show that due to the form of inference rules, PΔMnotp iff notpΔM. In other words,

(4)notpCnL(PΔM)iff notpΔM,

which is needed below to prove the equivalences given in this theorem.

⇒: Let M be an answer set of an NDP P. Then there exists the stable assumption extension ΔM of ABF(P) such that ΔM={notp|p(HBPM)} due to Proposition 3(i). Hence due to Theorem 5, there exists the stable argument extension E=Asms2Args(ΔM)={Kϕ|KΔM,LP=HBPNAFP}. Thus for M, ΔM and E=Asms2Args(ΔM), it holds that

Concs(E)={ϕLP|Kϕis constructed fromΔM,whereKΔMand ϕLP=HBPNAFP}

={ϕLP|PΔMϕ}={ϕLP|ϕCnL(PΔM)}=MΔM. (due to Lemma 2 and (4))

⇐: Let E be a stable argument extension of ABF(P) translated from an NDP P. Then due to Theorem 5, there is the stable assumption extension Δ of ABF(P) such that Δ=Args2Asms(E)={α|αKforKϕinE}=iKi for KiϕiE. Moreover, corresponding to this stable assumption extension Δ, there is the answer set M of the NDP P such that M={pHBP|notpΔ)} due to Proposition 3(ii), while for this answer set M, ΔM={notpNAFP|pM} is the stable assumption extension of ABF(P) due to Proposition 3(i). Thus obviously ΔM=Δ. Then for a stable argument extension E, Δ=iKi s.t. KiϕiE, M={pHBP|notpΔ} and ΔM=Δ, it holds that

Concs(E)={ϕ|KϕEforϕLP=HBPNAFP}={ϕLP|PΔϕforΔ=iKiwhereKiϕiE}={ϕLP|PΔMϕforM={pHBP|notpΔM=Δ}}={ϕLP|ϕCnL(PΔM)}=MΔM.(due to Lemma 2 and (4))

Based on Theorem 6, we show that there is a one-to-one correspondence between answer sets (resp. p-stable models) of an EDLP P and stable argument extensions of ABF(Ptr) (resp. ABF(P)) as follows, though Propositions 5 and 4 show the similar correspondences for stable assumption extensions of the respective ABFs.

Theorem 7.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P. Let MLitP and EAR, where AR is the set of all arguments generated from ABF(P). Then M is a p-stable model of an EDLP P iff there is a stable argument extension E of ABF(P) such that MΔM=Concs(E)=CNP(ΔM), where ΔM is a stable assumption extension of ABF(P).

Proof.

Let P+ be the NDP which is the positive form [28] of an EDLP P. Like the proof of Proposition 4, for a set SLitP, let S+ be the set obtained by replacing each negative literal ¬L in S with the newly introduced atom L. Thus Herbrand base HBP+ of the NDP P+ is (LitP)+. Then for ABF(P) translated from an EDLP P and ABF(P+) translated from the NDP P+, it holds that, there is a stable argument extension E of ABF(P) if and only if there is a stable argument extension E+ of ABF(P+) such that

(5)Concs(E+)=Concs(E)+.

Moreover, due to Lemma 3,

(6)M is a p-stable model of an EDLP P iff M+ is an answer set of the NDP P+.

On the other hand, based on Theorem 6 as well as (5), it holds that,

M+ is an answer set of the NDP P+
iff there is a stable argument extension E+ along with a stable assumption extension ΔM+ of ABF(P+)
such that M+ΔM+=Concs(E+),where ΔM+={nota|aHBP+M+}={nota|aLitP+M+}=ΔM+,M+ΔM+=(MΔM)+forMLitP,andΔM+=ΔM+
iff there is a stable argument extension E along with a stable assumption extension ΔM of ABF(P)
(7)such that MΔM=Concs(E), where ΔM={notL|LLitPM}.

Hence from (6) and (7), it follows that, M is a p-stable model of an EDLP P iff there is a stable argument extension E along with a stable assumption extension ΔM of ABF(P) such that MΔM=Concs(E). □

Theorem 8.

Let P be an EDLP and ABF(Ptr)=L,Ptr,AP,¯¯ be the ABF translated from the EDLP Ptr=P{Lp,¬p|pLitP,LLitP}, where LitPtr=LitP, AP=NAFP={not|LitP}, and not= for every notAP. Let SLitP and EtrAR for the set AR of all arguments generated from ABF(Ptr). Then S is an answer set of an EDLP P iff there is a stable argument extension Etr of ABF(Ptr) such that SΔS=Concs(Etr)=CNPtr(ΔS), where ΔS is a stable assumption extension of ABF(Ptr).

Proof.

The following (i) and (ii) hold according to [28, Theorem 3.5] and Theorem 7 respectively.

(i) S is an answer set of an EDLP P iff S is p-stable model of Ptr.

(ii) S is a p-stable model of an EDLP Ptr iff there is a stable argument extension Etr of ABF(Ptr)

  such that SΔS=Concs(Etr), where ΔS is a stable assumption extension of ABF(Ptr).

Hence this theorem follows from both (i) and (ii). □

Theorems 7 and 8 for an EDLP are the generalization of Theorems 2 and 3 for an ELP respectively.

Example 8

Example 8(Cont. Example 1).

To solve the Kyoto protocol problem in argumentation, we construct ABF(P1) from P1={rp,not¬r,kr,not¬k,fy,not¬f,kf,not¬k,p|y}. Arguments and attacks in ABF(P1) are obtained as follows:

A1:{noty}p,A2:{notp}y,A3:{noty,not¬r}r,A4:{notp,not¬f}f,A5:{noty,not¬r,not¬k}k,A6:{notp,not¬f,not¬k}k,A7:{not¬r,not¬f,not¬k}k,A8:{notp}notp,A9:{noty}noty,A10:{notr}notr,A11:{notf}notf,A12:{notk}notk,A13:{not¬p}not¬p,A14:{not¬y}not¬y,A15:{not¬r}not¬r,A16:{not¬f}not¬f,A17:{not¬k}not¬k;attacks={(A1,A2),(A1,A4),(A1,A6),(A1,A8),(A2,A1),(A2,A3),(A2,A5),(A2,A9),attacks=(A3,A10),(A4,A11),(A5,A12),(A6,A12),(A7,A12)}.
Fig. 3 shows A3 which is constructed based on [Ref], [MP] and [Res]. Each Ai (1i6) uses [Res]. But A7 uses [RBC] instead of [Res]. Arguments in ABF(P1) coincide with those in ABF((P1)tr).

Then ABF((P1)tr) (resp. ABF(P1)) has two stable (resp. preferred) argument extensions:

E1={A1,A3,A5,A7,A9,A11}{Ai|13i17},E2={A2,A4,A6,A7,A8,A10}{Ai|13i17},

where Concs(E1)={p,r,k,noty,notf}U,  Concs(E2)={y,f,k,notp,notr}U,

    for U=Concs({Ai|13i17})={not¬p,not¬y,not¬r,not¬f,not¬k}.

Hence the expected result is successfully obtained since kConcs(Ei) for Ei (i=1,2).

On the other hand, under the grounded and ideal semantics as the skeptical semantics, ABF(P1) has the unique argument extension E3:

E3={A7}{Ai|13i17},where Concs(E3)={k}U.
Then as a skeptical consequence, we obtain the result such that kConcs(E3) due to A7E3 where the argument A7 is constructed using [RBC], that meets our expectation based on reasoning by cases addressed in Example 1.
Fig. 3.

A3:{noty,not¬r}r in Example 8.

A3:{noty,not¬r}⊢r in Example 8.

Example 9.

Consider the EDLP P2={¬p|q,q¬p,not¬q,¬pq,notp}, where LitP2={p,q,¬p,¬q}. P2 is not head-cycle-free1111 but a general EDLP [4,13]. This means that the knowledge expressed by P2 is unlikely to be expressed by ELPs in general due to complexity results shown in [4,13]. P2 has the unique answer set S={¬p,q}, which is its unique p-stable model.

In contrast, ABF(P2) translated from P2 has the unique stable assumption extension ΔS={notp,not¬q}, while it has the following arguments:

A1:{notp}¬p,A2:{not¬q}q,A3:{notq}¬p,A4:{not¬p}q,A5:{not¬q,notp}¬p,A6:{notp,not¬q}q,A7:{not¬p,notp}¬p,A8:{notq,not¬q}q,A9:{not¬p}not¬p,A10:{notq}notq,A11:{notp}notp,A12:{not¬q}not¬q,andattacks={(A1,A4),(A1,A7),(A1,A9),(A2,A3),(A2,A8),(A2,A10),(A3,A4),(A3,A7),attacks=(A3,A9),(A4,A3),(A4,A8),(A4,A10),(A5,A4),(A5,A7),(A5,A9),(A6,A3),attacks=(A6,A8),(A6,A10),(A7,A4),(A7,A7),(A7,A9),(A8,A3),(A8,A8),(A8,A10)}.
Fig. 4 shows the tree structure of the argument A1:{notp}¬p which is constructed based on the inference rules: [Ref], [MP] and [RBC]. In this case, arguments in ABF(P2) coincide with those in ABF((P2)tr). Then ABF(P2) (resp. ABF((P2)tr)) has the following unique stable argument extension:
E={A1,A2,A5,A6,A11,A12},whereConcs(E)={¬p,q,notp,not¬q}=SΔS
for the answer set S of P2 and the stable assumption extension ΔS of ABF(P2).

Fig. 4.

A1:{notp}¬p in Example 9.

A1:{notp}⊢¬p in Example 9.
Example 10.

Consider the EDLP P3={¬a|bnot¬b,anotc}.

It has the unique answer set S={a,b}, while it has two p-stable models M1={a,b}=S and M2={a,¬a}, where M1 is consistent but M2 is inconsistent due to M2(a)=.

Let ABF(P3) be the ABF translated from P3, which has arguments and attacks as follows:

A1:{notc}a,A2:{notb,not¬b}¬a,A3:{not¬a,not¬b}b,A4:{nota}nota,A5:{notb}notb,A6:{notc}notc,A7:{not¬a}not¬a,A8:{not¬b}not¬b,A9:{not¬c}not¬c,andattacks={(A1,A4),(A2,A3),(A2,A7),(A3,A2),(A3,A5)}.
Then ABF(P3) has two stable argument extensions E1,E2 as follows.
E1={A1,A3,A6,A7,A8,A9},E2={A1,A2,A5,A6,A8,A9},where Concs(E1)={a,b,notc,not¬a,not¬b,not¬c}=M1ΔM1,whereConcs(E2)={a,¬a,notb,notc,not¬b,not¬c}=M2ΔM2,
in which ΔM1 and ΔM2 are stable assumption extensions of ABF(P3) such that

  ΔM1={notc,not¬a,not¬b,not¬c},ΔM2={notb,notc,not¬b,not¬c}.

In contrast, ABF((P3)tr) has the arguments Ai (1i9) along with Aj (10j15) shown below:

A10:{notb,not¬b,notc}a,A11:{notb,not¬b,notc}b,A12:{notb,not¬b,notc}c,A13:{notb,not¬b,notc}¬a,A14:{notb,not¬b,notc}¬b,A15:{notb,not¬b,notc}¬c.
As a result, ABF((P3)tr) has the unique stable argument extensions as follows:

          Etr={A1,A3,A6,A7,A8,A9}=E1,

    where Concs(Etr)={a,b,notc,not¬a,not¬b,not¬c}=SΔS,

for the unique answer set S of P3 and the unique stable assumption extension ΔS of ABF((P3)tr).

Remark 2.

In Example 8, [Ref] is used to construct each Ai(1i7). Thus we need [Ref]. Without [MP], we cannot build Ai(3i7), which means we cannot infer k. Thus we need [MP]. In Example 9, [RBC] is used to construct A1,A2,A5,A6. Hence without [RBC], we cannot construct these arguments, which means ABF(P2) has no stable extension. Thus we need [RBC]. In Example 10, without [Res], we cannot construct A3 along with A2. Thus we need [Res].

3.5.Correspondence between answer sets of a consistent EDLP and consistent stable extensions

Rationality postulates are defined in ABF(P) translated from an EDLP P like Definition 2. In what follows, we show that such ABF(P) always satisfies the closure-property (or direct consistency postulate [6]) under the stable semantics.

Definition 16

Definition 16(Rationality postulates).

Given an EDLP P, ABF(P)=L,P,AP,¯¯ is said to satisfy the consistency-property (resp. the closure-property) under the σ semantics if for each σ argument extension E of the AA framework AFF generated from F=ABF(P), Concs(E) is consistent (resp. Concs(E) is closed w.r.t. CNP).

Theorem 9.

Let F be ABF(P)=L,P,AP,¯¯ translated from an EDLP P and E be a stable argument extension of AFF generated from F=ABF(P).

  • (1) F satisfies the closure-property under the stable semantics.

  • (2) F satisfies the consistency-property under the stable semantics

    iff for every E, Concs(E) is consistent

    iff for every E, Concs(E) is not contradictory w.r.t. explicit negation ¬.

Proof.

 

  • (1) Let M be a p-stable model of P and E be a stable argument extension of F=ABF(P) satisfying MΔM=Concs(E). Then

    (8)CNP(MΔM)={ϕLP|PMΔMϕ}=ΔM{LitP|PMΔM}.

    Due to Lemma 4 and the transitive closure property of CnL, it holds that, for LitP,

    MiffCnL(PΔM)iffCnL(PΔMM{})=CnL(PΔMM).

    (9)Hence,MiffPMΔMfor LitP.

    Then, (9) means that M={LitP|PMΔM}.

    As a result, (8) leads to CNP(MΔM)=MΔM, namely, CNP(Concs(E))=Concs(E).

    Thus F satisfies the closure-property under the stable semantics.

  • (2) F satisfies the consistency-property under the stable semantics

    iff for every E, Concs(E) is consistent  (due to Definition 16)

    iff for every E, CNP(Concs(E)) is not contradictory  (due to Definition 10)

    iff for every E, Concs(E) is not contradictory  (due to Item (1), CNP(Concs(E))=Concs(E))

    iff for every E, Concs(E) is not contradictory w.r.t. explicit negation ¬

     (since every stable argument extension is not contradictory w.r.t. ¯¯). □

Given an EDLP P, the notions of consistent argument extensions and consistency in ABF(P) are defined like [33, Definitions 6, 7] as follows.

Definition 17

Definition 17(Consistent argument extensions).

Given an EDLP P, let E be a σ argument extension of ABF(P)=L,P,AP,¯¯. Then E is said to be consistent if Concs(E) is not contradictory w.r.t. ¬; otherwise it is inconsistent.

Definition 18

Definition 18(Consistency in ABFs translated from EDLPs).

Given an EDLP P, ABF(P)=L,P,AP,¯¯ is said to be consistent under σ semantics if ABF(P) has a consistent σ argument extension (or a consistent σ assumption extension); otherwise it is inconsistent.

We show that there is a one-to-one correspondence between answer sets of a consistent EDLP P and the consistent stable argument extensions of ABF(P) translated from P, which is a generalization of Theorem 4 for a consistent ELP. To prove it, we provide the following lemma regarding a consistent EDLP.

Lemma 5.

Let P be an EDLP. M is a consistent answer set of P iff there is a consistent p-stable model M of P.

Proof.

⇐: Let S be a consistent p-stable model of P. Then S does not contain a pair of complementary literals. Since S is also a p-stable model of the reduct PS according to Definition 5, S is a minimal set satisfying the condition (i) for PS which is the not-free EDLP. Then since S does not contain a pair of complementary literals, S is also a minimal set satisfying both conditions (i) and (ii) for PS. This denotes that S is the answer set of PS which does not contain a pair of complementary literals. Thus S is the answer set of PS and it is not LitP. Hence since the answer set S of PS which is not LitP is the answer set of P, S is the consistent answer set of P.

⇒: The converse is proved in a similar way. □

Hereby given a consistent EDLP, we can obtain the following proposition and theorem.

Proposition 9.

Let P be a consistent EDLP and ABF(P)=L,P,AP,¯¯ be the ABF translated from P. Then S is an answer set of P iff there is a consistent stable assumption extension ΔS={not|(LitPS)}) of ABF(P).

Proof.

Suppose that S is an answer set of a consistent EDLP P. Then based on Definition 5, S is a consistent answer set of P, which means that S is a consistent p-stable model of P due to Lemma 5. Then corresponding to a p-stable model S of P which is consistent, there is a stable assumption extension ΔS={not|(LitPS)} of ABF(P) where CNP(ΔS)=SΔS based on Proposition 4 and Proposition 6. Now, since S is a consistent answer set, CNP(ΔS)=SΔS is not contradictory according to Definition 10. This means that ΔS is consistent. Therefore, there is a consistent stable assumption extension ΔS={not|(LitPS)} of ABF(P). The converse is also proved in a similar way. □

Theorem 10.

Let P be a consistent EDLP and ABF(P)=L,P,AP,¯¯ be the ABF translated from P. Then S is an answer set of P iff there is a consistent stable argument extension E of ABF(P) such that SΔS=Concs(E)=CNP(ΔS), where ΔS is the consistent stable assumption extension of ABF(P).

Proof.

Suppose that E is a consistent stable argument extension of ABF(P). Then Concs(E) is consistent, i.e. not contradictory w.r.t. ¬ due to Definition 10 since E is conflict-free. Moreover, based on Theorem 7, there is the p-stable model S of P corresponding to this E such that SΔS=Concs(E), where ΔS is a stable assumption extension of ABF(P). Since Concs(E) is not contradictory w.r.t. ¬, in other words, it does not contain a pair of complementary literals, Concs(E)=SΔS as well as the p-stable model S are consistent. Hence due to Lemma 5, S is the consistent answer set of P. As a result, S is the answer set of the consistent EDLP P. The converse is also proved in a similar way. □

Corollary 2.

Let P be a consistent EDLP. The following holds.

(1) E is a consistent stable extension of ABF(P) iff E is a stable extension of ABF(Ptr).

(2) ABF(Ptr) satisfies the rationality postulates under the stable semantics.

Proof.

(1) follows from Theorem 10 as well as Theorem 8 for a consistent EDLP P. (2) directly follows from (1). □

Example 11

Example 11(Innocent unless proved guilty).

Consider the EDLP P4 [26], which states that everyone is pronounced not guilty unless proven otherwise:

P4={innocent|guiltycharged,¬guiltynotproven,charged}.

Let i,g,c,p be the abbreviations for innocent, guilty, charged, proven respectively. P4 has the unique answer set S={c,i,¬g}, where p (i.e. proven) is interpreted “unknown” under the answer set semantics. In contrast, P4 has two p-stable models, M1={c,i,¬g} and M2={c,g,¬g}. M1=S is consistent whose truth values are M1(c)=t, M1(i)=t, M1(g)=f, M1(p)=, while M2 is inconsistent due to M2(g)=. To solve this problem in argumentation, we construct ABF(P4), which has arguments:

A1:{}c,A2:{notg}i,A3:{noti}g,A4:{notp}¬g,A5:{noti}noti,A6:{notg}notg,A7:{notc}notc,A8:{notp}notp,A9:{not¬i}not¬i,A10:{not¬g}not¬g,A11:{not¬c}not¬c,A12:{not¬p}not¬p,and attacks={(A1,A7),(A2,A3),(A2,A5),(A3,A2),(A3,A6),(A4,A10)},where |attacks|=6.
Then ABF(P4) has two stable argument extensions, E1 and E2:
E1={A1,A2,A4,A6,A8,A9,A11,A12},E2={A1,A3,A4,A5,A8,A9,A11,A12},
where Concs(E1)={c,i,¬g,notg,notp,not¬i,not¬c,not¬p} with Concs(E1)LitP4=S,

  Concs(E2)={c,g,¬g,noti,notp,not¬i,not¬c,not¬p} with Concs(E2)LitP4=M2.

Thus E1 is consistent but E2 is inconsistent. Hence ABF(P4) is consistent under the stable semantics.

In contrast, ABF((P4)tr) has the unique stable argument extension E1 due to six additionally introduced arguments to Ai(1i12) and |attacks|=26.

Using ABF(P4), we can decide that the attorney-at-law having the argument A4 for the claim ¬g wins and the prosecutor having A3 for g loses since A4E1 and A3E1 for its unique consistent extension E1. Therefore ¬g is decided.

4.Relation to nonmonotonic reasoning

4.1.Correspondence between disjunctive default logic and assumption-based argumentation

A disjunctive default theory (ddt, for short) [16] is a set of disjunctive defaults of the form:

α:β1,,βmγ1||γn,
where α,β1,,βm,γ1,,γn(m,n0) are quantifier-free formulas. Formula α is the prerequisite of the default, β1,,βm are its justifications, and γ1,,γn are its consequents. If the prerequisite α in the form is the formula true, it will be dropped; if, in addition, m=0, then we write the default as γ1||γn.

The semantics of a ddt is given by extensions defined as follows.

Definition 19

Definition 19([16, Definition 5.1]).

Let D be a disjunctive default theory, and let E be a set of sentences. E is an extension for D if it is one of the minimal deductively closed sets of sentences E satisfying the condition: For any ground instance having the above form of any default from D, if αE and ¬β1,,¬βmE then, for some i ((1in), then, γiE. A theorem is a sentence that belongs to all extensions.

Observe that for standard (nondisjunctive) default theories, this definition gives Reiter’s extensions [27].

The definition of an extension for a ddt are also described based on the concept of reduct [16]. To this end, a disjunctive rule of the form αγ1||γn is defined. Then it is said that a theory E is closed under a disjunctive rule if, whenever αE, then there exist i, 1in, γiE.

Definition 20

Definition 20([16, Definition 5.2]).

Let D be a ddt and let E be a set of sentences. The reduct of D w.r.t. E, denoted DE, is the set of inference rules defined as follows: An inference rule αγ1||γn is in DE if for some β1,,βm such that ¬βiE, 1im, the default α:β1,,βmγ1||γn is in D.

Using the reduct DE, another definition of an extension for a ddt is given by the following theorem.

Theorem 11

Theorem 11([16, Theorem 5.3]).

A set of sentences E is an extension for a ddt D if and only if E is a minimal set closed under propositional consequence and under the rules from DE.

In what follows, we show the semantic relationship between ddts and assumption-based frameworks.

Theorem 12.

Let P be an EDLP and ABF(Ptr) be the ABF translated from Ptr.

S is the set of all literals from an extension of a disjunctive default theory emb(P)

iff there is a stable argument extension Etr of ABF(Ptr) such that S=Concs(Etr)LitP

iff there is a stable assumption extension Δ of ABF(Ptr) such that S=CNPtr(Δ)LitP.

Proof.

Based on Theorem 1 [16, Theorem 7.2], this theorem directly follows from Theorem 8 for an argument extension Etr (resp. from Proposition 5 for an assumption extension Δ). □

For a consistent EDLP, the following theorem holds.

Theorem 13.

Let P be a consistent EDLP and ABF(P) be the ABF translated from P.

S is the set of all literals from an extension of a disjunctive default theory emb(P)

iff there is a consistent stable argument extension E of ABF(P) such that S=Concs(E)LitP

iff there is a consistent stable assumption extension Δ of ABF(P) such that S=CNP(Δ)LitP.

Proof.

Based on Theorem 1 [16, Theorem 7.2], this theorem directly follows from Theorem 10. □

Given a nondisjunctive EDLP, i.e. an ELP P (resp. a consistent ELP P), Theorem 12 (resp. Theorem 13) also holds for the disjunctive default theory emb(P) and ABF(Ptr) (resp. ABF(P)). For such a special case, however, we can show the relationship between a standard (nondisjunctive) default theory which gives Reiter’s extensions and a standard ABA framework translated from an ELP as follows.

Theorem 14.

Let P be an ELP and F(Ptr) be the ABA framework (ABF) translated from Ptr.

S is the set of all literals from an extension of a default theory emb(P)

iff there is a stable argument extension Etr of the ABF F(Ptr) such that S=Concs(Etr)LitP

iff there is a stable assumption extension Δ of the ABF F(Ptr) such that S=CNPtr(Δ)LitP,

where CNPtr is the consequence operator for the ABF F(Ptr) defined in Definition 2.

Proof.

Based on Theorem 1 [16, Theorem 7.2], this theorem directly follows from Theorem 3 for an argument extension Etr (resp. from Proposition 13 for an assumption extension Δ). □

Theorem 15.

Let P be a consistent ELP and F(P) be the ABA framework translated from P.

S is the set of all literals from an extension of a default theory emb(P)

iff there is a consistent stable argument extension E of the ABF F(P) such that S=Concs(E)LitP

iff there is a consistent stable assumption extension Δ of the ABF F(P) such that S=CNP(Δ)LitP

where CNP is the consequence operator for the ABF F(P) defined in Definition 2.

Proof.

Based on Theorem 1 [16, Theorem 7.2], this theorem directly follows from Theorem 4 for an argument extension E (resp. from Proposition 14 for an assumption extension Δ). □

The following example shows that we can successfully obtain the expected result k of the Kyoto protocol problem based on our assumption-based framework corresponding to the ddt D2 thanks to Theorem 12. It should be noted here that this solution of the problem can never be obtained based on the existing works of ABA [5,19] corresponding to the default theory D1 as discussed in the introduction.

Example 12

Example 12(Cont. Example 1).

The disjunctive default theory D2 has two extensions E1 and E2 such that kEi(i=1,2), where SE1={p,r,k}E1 (resp. SE2={y,f,k}E2) is the set of all literals from E1 (resp. E2). On the other hand, D2 is the ddt emb(P1) in which the EDLPP1 can be embedded. Then according to Example 8, ABF((P1)tr) coincides with ABF(P1), and it hods that kConcs(Ei)(i=1,2) for two stable argument extensions E1 and E2 of ABF(P1), where Concs(E1)LitP1=SE1={p,r,k} and Concs(E2)LitP1=SE2={y,f,k}. Similarly, according to Example 5, it hods that kCNP1(Δi) (i=1,2) for each stable assumption extension Δi of ABF(P1), where CNP1(Δi)LitP1=SEi.

4.2.Correspondence between prioritized circumscription and assumption-based argumentation

Circumscription [20,22,23] is a form of nonmonotonic reasoning, which was proposed to formalize the human commonsense reasoning under incomplete information. Commonsense knowledge including preferences is also often used in human argumentation. Then Bondarenko et al. showed in [5, Theorem 6.7] that Herbrand models of parallel circumscription can be captured by sets of assumptions of a corresponding assumption-based framework. Nonetheless, though preferences can be handled not in parallel circumscription but in prioritized circumscription, no study has shown a correspondence between the semantics of prioritized circumscription and the ABA semantics to the best of our knowledge. In what follows, we show new results about the relationships between them.

We first review the framework of circumscription. The following definition is due to [20]. Given a first order theory T, let P and Z be joint tuples of predicate constants from T, where P is a tuple of predicate constants P1,,Pm. Let T(P,Z) be a theory containing P and Z. The circumscription of P in T(P,Z) with variable Z is defined by a second order formula as follows:

Circum(T;P;Z)=defT(P,Z)¬pz(T(p,z)p<P),
where p,z are tuples of predicate variables similar to P,Z, and p<P denotes the following formula:
i=1mx(pi(x)Pi(x))¬i=1mx(Pi(x)pi(x)),
where P=P1,,Pm. The formula of circumscription expresses that P has a minimal possible extension under the condition T(P,Z) where Z is allowed to vary in the process of minimization [20]. Due to the respective roles in the process of minimization, each Pi (1im) is called a minimized predicate, while each predicate in Z is called a variable predicate. Q denotes the rest of the predicates occurring in T, called the fixed predicates. This version of circumscription is called parallel circumscription.

If P is decomposed into disjoint parts P1,,Pk, and the members of Pi are assigned a higher priority than the members of Pj for i<j, then prioritized circumscription of P1>>Pk in T with variable Z is denoted by Circum(T;P1>>Pk;Z), which is also defined by a second order formula [20,23]. Parallel circumscription coincides with prioritized circumscription for k=1.

The semantics of circumscription is given based on the preorder P1>>Pk;Z defined as follows.

For a structure M, let |M| be its universe and MC the interpretation of a predicate constant C. For a tuple P of predicate constants, M1PM2P denotes M1PiM2Pi for every Pi in P.

Definition 21

Definition 21([20]).

Let P1,,Pk be k disjoint parts of minimized predicates P. For any two structures M1, M2, we write M1P1>>Pk;ZM2 if

  • (i) |M1|=|M2|,

  • (ii) M1K=M2K, for every constant K in Q,

  • (iii)
    • a. M1P1M2P1,

    • b. For every ik, if for every 1ji1, M1Pj=M2Pj, then M1PiM2Pi.

P;Z stands for the preorder P1>>Pk;Z for k=1. Then a structure M is a model of Circum(T;P;Z) iff M is minimal in the class of models of T with respect to P;Z. A structure M is a model of Circum(T;P1>>Pk;Z) iff M is minimal in the class of models of T with respect to P1>>Pk;Z.

In a nutshell, the idea of the circumscriptive theory is that human nonmonotonic reasoning under incomplete knowledge (e.g. commonsense knowledge) with preferences is based on the most preferable models which are minimal ones w.r.t. P1>>Pk;Z among models of the knowledge base T.

In this paper, we consider a first order theory T without function symbols. We assume that T is given by a set of clauses of the form:

A1A¬B1¬Bm,
where Ai (1i;0) and Bj (1jm;m0) are atoms and every variable in the formula is assumed to be universally quantified at the front. We also restrict our attention to Herbrand models of T, which has the effect of introducing both the domain closure assumption and the unique name assumption into T, and then reasoning with T reduces to the propositional level. Let Σ be a set of clauses. Then Th(Σ) stands for a set of clauses which are theorems of Σ. A clause in Th(Σ) which is not properly subsumed1212 by any theorem in Th(Σ) is called a characteristicclause. μTh(Σ) denotes the set of all characteristic clauses in Th(Σ). t stands for a tuple of ground terms occurring in the Herbrand universe of T. Parallel circumscription is transformed into an EDLP [29] as follows.

Definition 22

Definition 22([29]).

Given Circum(T;P;Z), an EDLP Πα1313 is constructed as follows, where p1,,ps, z1,,zt, and q1,,qu are atoms whose predicates belong to P,Z, and Q respectively.

  • (1) For any clause in T of the form:

    p1pz1zmq1qn¬p+1¬ps¬zm+1¬zt¬qn+1¬qu,

    Πα has the rule:

    z1||zm|q1||qnp+1,,ps,zm+1,,zt,qn+1,,qu,notp1,,notp.

  • (2) For every clause in μTh(T) of the form:

    p1pq1qn¬qn+1¬qu,
    Πα has the rule:   p1||p|q1||qnqn+1,,qu.

  • (3) For any atom p,z, and q from P,Z, and Q respectively,

    Πα has the rule:

    ¬pnotp,z|¬z,q|¬q.

The following theorem presents that there is a one-to-one correspondence between models of parallel circumscription and answer sets of Πα.

Theorem 16

Theorem 16([29]).

Let Πα be the EDLP translated from Circum(T;P;Z). Then M is a model of Circum(T;P;Z) iff M is an answer set of Πα.

Proposition 10.

Any answer set of an EDLP Πα is consistent.

Proof.

In Πα, classical negation ¬ appears only in rules constructed according to item 3 in Definition 22. Thus any answer set of Πα never contains both a and ¬a for any atom a from P,Z and Q. □

In [20, Theorem 2], the following equivalence is shown:

(10)Circum(T;P1>>Pk;Z)=i=1kCircum(T;Pi;Pi+1,,Pk,Z).

Based on (10), prioritized circumscription is transformed into an EDLP [35] as follows.

Definition 23

Definition 23([35, Definition 3.11]).

Given prioritized circumscription:

Circum(T(P1Pk,Z,Q);P1>P2>>Pk;Z),
where Pr(1rk), Z, Q are tuples of predicate symbols such as (Pr)1,,(Pr)r, Z1,,Zm,

Q1,,Qn, an EDLP Πβ is constructed in two steps as follows:

  • (1) According to (10), a given prioritized circumscription is represented by the conjunction of k parallel circumscriptions. So, let every ith (1ik) parallel circumscription:

    Circum(T(P1Pk,Z,Q);Pi;Pi+1,,Pk,Z)
    be transformed in such a way that all predicate symbols occurring in it are renamed by using Pi1,,Pik, Zi,Qi instead of P1,,Pk, Z,Q, which leads to the ith renamed parallel circumscription as follows:
    (11)Circum(Ti;Pii;Pii+1,,Pik,Zi),
    where Ti denotes T(Pi1,,Pik,Zi,Qi), and Pir, Zi, Qi are tuples of predicate symbols such as (Pir)1,,(Pir)r, (Zi)1,,(Zi)m, (Qi)1,,(Qi)n.

  • (2) Πβ consists of all rules from (Πα)1,,(Πα)k and Πγ where

    • (a) each (Πα)i(1ik) is an EDLP which is constructed from the ith renamed parallel circumscription (11) according to Definition 22.

    • (b) Πγ is a set of rules (i.e. an ELP) as follows:

      For any predicate u from P,Z,Q and any t,

      u(t)ui(t),¬u(t)¬ui(t).(1ik)
      where u and ui stand for any predicate symbol of (Pr)f,Zg,Qh and any one of (Pir)f,(Zi)g,(Qi)h (1fr,1gm,1hn).

In what follows, HB denotes Herbrand base of T and HBe=HB{¬p|pHB}. The following theorem shows that there is a one-to-one correspondence between models of prioritized circumscription and answer sets of Πβ.

Theorem 17.

Let Πβ be the EDLP translated from Circum(T;P1>P2>>Pk;Z). Then M is a model of Circum(T;P1>P2>>Pk;Z) iff there is a consistent answer set S of Πβ such that M=SHBe.

Proof.

Let AS(Π) be a set of all answer sets of a finite ground EDLP Π. We use the following notations.

  • - Acircumi=defj=1iCircumj,  where Circumi=defCircum(T;Pi;Pi+1,,Pk;Z).

  • - (Πγ)i=def{iiΠγ},  where a literal in Circumi is renamed to the literal i in the ith renamed Circum(Ti;Pii;Pii+1,,Pik,Zi).

  • - (Παγ)i=def(Πα)i(Πγ)i.

  • - (Πβ)i=defj=1i((Παγ)j).

  • - Sαγi denotes an answer set of (Παγ)i, namely SαγiAS((Παγ)i).

  • - For a set S of literals, we say that S is consistent if S does not contain both p and ¬p for an atom p.

Then, Mi is a model of Circumi=Circum(T;Pi;Pi+1,,Pk;Z)(1ik)

 iff there is a consistent answer set Mi of (Πα)i translated from Circumi  (due to Theorem 16)

(12)iff there is a consistent answer set Sαγi of (Παγ)i such that Mi=SαγiHBe.

Hence, M is a model of Circum(T;P1>P2>>Pk;Z)

iff M is a model of Acircumk=j=1kCircumj     (due to (10))

iff there is a consistent set S=j=1kSαγj for SαγjAS((Παγ)j)(1jk) such that M=SHBe

  (due to (12))

iff there is a consistent answer set S=j=1kSαγj of (Πβ)k such that M=SHBe

iff there is a consistent answer set S of Πβ=(Πβ)k translated from

  Circum(T;P1>P2>>Pk;Z) such that M=SHBe. □

Therefore thanks to Theorem 10, the semantics of prioritized circumscription (resp. parallel circumscription) can be captured by argumentation based on Theorem 17 (resp. Theorem 16) as follows.

Theorem 18.

Let Πα be the EDLP translated from Circum(T;P;Z).

Then M is a model of Circum(T;P;Z) iff there is a consistent stable argument extension E of ABF(Πα)=L,Πα,AΠα,¯¯ such that M=Concs(E)HBe.

Proof.

Based on Theorem 16, Theorem 10, and Proposition 10,

M is a model of Circum(T;P;Z) iff M is a (consistent) answer set of Πα

iff there is a consistent stable argument extension E of ABF(Πα) such that M=Concs(E)HBe. □

Theorem 19.

Let Πβ be the EDLP translated from Circum(T;P1>P2>>Pk;Z).

Then M is a model of Circum(T;P1>P2>>Pk;Z) iff there is a consistent stable argument extension E of ABF(Πβ)=L,Πβ,AΠβ,¯¯ such that M=Concs(E)HBe.

Proof.

Based on Theorem 17 and Theorem 10,

M is a model of Circum(T;P1>>Pk;Z)

iff there is a consistent answer set S of Πβ such that M=SHBe

iff there is a consistent stable argument extension E of ABF(Πβ) such that SΔS=Concs(E) and

M=SHBe

iff there is a consistent stable argument extension E of ABF(Πβ) such that M=Concs(E)HBe

 (due to M=SHBe=(SΔS)HBe=Concs(E)HBe). □

Theorem 19 (resp. Theorem 17) indicates that reasoning in prioritized circumscription can be computed based on assumption-based argumentation (resp. answer set programming).

Example 13.

Consider prioritized circumscription given in [35, Example 3.9]:

Circum({pq,qr};p>q;r).
It has two models M1={¬p,q,r} and M2={¬p,q,¬r}. Based on Theorem 17, these models can be obtained from answer sets of the EDLP Πβ=(Πα)1(Πα)2Πγ translated from it. In this case,
(Πα)1={q1notp1,q1|r1,¬p1notp1,q1|¬q1,r1|¬r1},(Πα)2={p2notq2,r2notq2,p2|q2,¬q2notq2,r2|¬r2,p2|¬p2},Πγ={pp1,qq1,rr1,¬p¬p1,¬q¬q1,¬r¬r1,Πγ=pp2,qq2,rr2,¬p¬p2,¬q¬q2,¬r¬r2}.
  where HBe={p,q,r,¬p,¬q,¬r}.  Then Πβ has two answer sets S1 and S2 [35]:

S1={¬p,q,r,¬p1,q1,r1,¬p2,q2,r2},S2={¬p,q,¬r,¬p1,q1,¬r1,¬p2,q2,¬r2},

 where  S1HBe={¬p,q,r}=M1,  S2HBe={¬p,q,¬r}=M2.

Thus, models of Circum({pq,qr};p>q;r) are computed based on answer set semantics.

On the other hand, based on Theorem 19, these models are also obtained in argumentation as follows.

Consider ABF(Πβ)=L,Πβ,AΠβ,¯¯ translated from Πβ. It has arguments and attacks as follows:

A1:{notp1}q1,A2:{notr1}q1,A3:{notq1}r1,A4:{notp1}¬p1,A5:{notq1}¬q1,A6:{not¬q1}q1,A7:{notr1}¬r1,A8:{not¬r1}r1,A9:{notq2}p2,A10:{notq2}r2,A11:{notp2}q2,A12:{notq2}¬q2,A13:{notr2}¬r2,A14:{not¬r2}r2,A15:{not¬p2}p2,A16:{notp2}¬p2,A17:{notq2}p,A18:{not¬p2}p,A19:{notp1}q,A20:{notr1}q,A21:{not¬q1}q,A22:{notp2}q,A23:{notq1}r,A24:{not¬r1}r,A25:{notq2}r,A26:{not¬r2}r,A27:{notp1}¬p,A28:{notp2}¬p,A29:{notq1}¬q,A30:{notq2}¬q,A31:{notr1}¬r,A32:{notr2}¬r,Ai:{notx}notx,where x{p,q,r,¬p,¬q,¬r} for 33i38,Ami:{notx1}notx1,where x1{p1,q1,r1,¬p1,¬q1,¬r1} for 39mi44,Ani:{notx2}notx2,where x2{p2,q2,r2,¬p2,¬q2,¬r2} for 45ni50),and 
Fig. 5.

The graphic representation of arguments and attacks in ABF(Πβ).

The graphic representation of arguments and attacks in ABF(Πβ).

attacks={(A1,A3),(A1,A5),(A1,A23),(A1,A29),(A1,A40),(A2,A3),(A2,A5),(A2,A23),(A2,A29),(A2,A40),(A3,A2),(A3,A7),(A3,A20),(A3,A31),(A3,A41),(A4,A42),(A5,A6),(A5,A21),(A5,A43),(A6,A3),(A6,A5),(A6,A23),(A6,A29),(A6,A40),(A7,A8),(A7,A24),(A7,A44),(A8,A2),(A8,A7),(A8,A20),(A8,A31),(A8,A41),(A9,A11),(A9,A16),(A9,A22),(A9,A28),(A9,A45),(A10,A13),(A10,A32),(A10,A50),(A11,A9),(A11,A10),(A11,A12),(A11,A17),(A11,A25),(A11,A30),(A11,A46),(A12,A49),(A13,A14),(A13,A26),(A13,A50),(A14,A13),(A14,A32),(A14,A50),(A15,A11),(A15,A16),(A15,A22),(A15,A28),(A15,A45),(A16,A15),(A16,A18),(A16,A48),(A17,A33),(A18,A33),(A19,A34),(A20,A34),(A21,A34),(A22,A34),(A23,A35),(A24,A35),(A25,A35),(A26,A35),(A27,A36),(A28,A36),(A29,A37),(A30,A37),(A31,A38),(A32,A38)}.

Fig. 5 shows the graphic representation of the AA framework generated from ABF(Πβ).

ABF(Πβ) has six stable argument extensions Ei (1i6), in which both E1 and E2 shown below are consistent, but the others are inconsistent:

E1={A1,A4,A6,A8,A11,A14,A16,A19,A21,A22,A24,A26,A27,A28,A33,A37,A38,A39,A43,A44,A45,A47,A49},E2={A1,A2,A4,A6,A7,A11,A13,A16,A19,A20,A21,A22,A27,A28,A31,A32,A33,A35,A37,A39,A41,A43,A45,A47,A49},
where Concs(E1)HBe={¬p,q,r}=M1, Concs(E2)HBe={¬p,q,¬r}=M2. Thus, models of Circum({pq,qr};p>q;r) are computed based on assumption-based argumentation.

5.Relation to the possible model semantics of EDLPs

Minimality-based semantics interprets disjunctions as exclusive as possible as addressed in [30]. Instead, to freely specify both inclusive and exclusive interpretations of disjunctions, the possible model semantics of an EDLP was introduced by Sakama and Inoue [30].

In this paper, we restrict our attention to a consistent possible model which does not contain a pair of complementary literals L and ¬L. In the possible model semantics, a split program1414 of an EDLP P is defined as a ground ELP obtained from P by replacing every ground disjunctive rule r:L1||LΓ from P with the nondisjunctive rules in RSplitr, where

Splitr={LiΓ|Γ=body(r)forrP,i=1,,}
and R is a non-empty subset of Splitr. Each rule in Splitr is called a split rule of r. We denote by sp(P) a split program of an EDLP P as defined as follows: given an EDLP P=j{rj}, where rj is a rule,
sp(P)=jRrj, where each rj is replaced with Rrj s.t. RrjSplitrj and Rrj.

P has multiple split programs in general. Then, a consistent possible model of P is defined as a consistent answer set of a split program of P. It should be noticed that a possible model of P is not always minimal among possible models of P. Any consistent answer set of P is a consistent possible model of P but not vice versa [30]. An EDLP is consistent under the possible model semantics iff it has a consistent possible model; otherwise it is inconsistent.

The possible models of an EDLP can be captured by stable extensions of a standard ABA framework since each of its split programs is an ELP as follows.

Theorem 20.

Let P be an EDLP and sp(P) a split program of P. Then S is a consistent possible model of P iff there is a split program sp(P) which has a consistent stable argument extension E of the ABA framework F(sp(P))=LP,sp(P),AP,¯¯ such that SΔS=Concs(E) (in other words, S=Concs(E)LitP), where ΔS is a consistent stable assumption extension of the ABF F(sp(P)).

Proof.

S is a consistent possible model of P

iff  there is a split program sp(P) which has a consistent answer set S

  (due to the definition of a consistent possible model)

iff  there is a consistent stable argument extension E of the ABA framework F(sp(P)) such that

  SΔS=Concs(E),  (due to Theorem 4)

where ΔS is a consistent stable assumption extension of the ABF F(sp(P)) due to Proposition 14. □

In what follows, we define an argument possible-extension (an argument p-extension, for short) and an assumption possible-extension (an assumption p-extension, for short) of ABF(P) which introduces the idea of possible models of an EDLP P in ABA.

Definition 24.

Let P be an EDLP, sp(P) a split program of P and ABF(sp(P))=L,sp(P),AP,¯¯. We say that E is a stable argument possible-extension (or a stable argument p-extension, for short) of ABF(P)=L,P,AP,¯¯ if E is a stable argument extension of ABF(sp(P)) for some ELP sp(P); and E is a consistent stable argument possible-extension (or a consistent stable argument p-extension) of ABF(P) if E is a consistent stable argument extension of ABF(sp(P)) for some ELP sp(P).

Similarly, we say that Δ is a stable assumption possible-extension (or a stable assumption p-extension, for short) of ABF(P) if Δ is a stable assumption extension of ABF(sp(P)) for some ELP sp(P); and Δ is a consistent stable assumption possible-extension (or a consistent stable assumption p-extension) of ABF(P) if Δ is a consistent stable assumption extension of ABF(sp(P)) for some ELP sp(P).

Notice that only the inference rule [MP] is needed to compute p-extensions of ABF(P) because any sp(P) contains no disjunction. Regarding p-extensions, the following proposition and theorem hold.

Proposition 11.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P. Then S is a consistent possible model of P iff there is a consistent stable assumption p-extension ΔS of ABF(P) such that ΔS={not|(LitPS)} where CNsp(P)(ΔS)=SΔS for some split program sp(P).

Proof.

S is a consistent possible model of P

iff there is a split program sp(P) which has a consistent answer set S  (due to its definition)

iff there is a consistent stable assumption extension ΔS of ABF(sp(P)) for a split program sp(P) s.t.

ΔS={not|(LitPS)} where CNsp(P)(ΔS)=SΔS is consistent  (due to Proposition 9)

iff there is a consistent stable assumption p-extension ΔS of ABF(P) such that

ΔS={not|(LitPS)} where CNsp(P)(ΔS)=SΔS for some sp(P). □

Theorem 21.

Let ABF(P)=L,P,AP,¯¯ be the ABF translated from an EDLP P. S is a consistent possible model of P iff there is a consistent stable argument p-extension E of ABF(P) such that SΔS=Concs(E), where ΔS={not|(LitPS)} is a consistent stable assumption p-extension of ABF(P).

Proof.

S is a consistent possible model of P

iff there is a consistent stable argument extension E along with the consistent stable assumption

 extension ΔS={not|(LitPS)} of ABF(sp(P)) for some split program sp(P) such that

Concs(E)=SΔS=CNsp(P)(ΔS) which is consistent  (due to Proposition 11 and Theorem 10)

iff there is a consistent stable argument p-extension E of ABF(P) such that SΔS=Concs(E),

 where ΔS={not|(LitPS)} is a consistent stable assumption p-extension of ABF(P). □

For an EDLP P with no disjunction, Proposition 11 reduces to Proposition 14, while Theorem 21 reduces to Theorem 4 respectively.

The following example shows that there is a case that needs the inclusive interpretation of disjunction though Kyoto protocol problem in Section 1 needs the exclusive interpretation of the disjunction p|y.

Example 14

Example 14(Taxi fare problem).

Consider the following problem.

Each of Jack and Mary takes a taxi if it is available. Suppose that a taxi is available, where its fare is v dollars. If they share it, each may pay only the half of v (say hv) for a taxi fare; otherwise anyone who takes it should pay v dollars.

This situation can be represented by the following EDLP P5, where j (resp. m) denotes that Jack (resp. Mary) takes a taxi, tx denotes that a taxi is available, and v (resp. hv) denotes that a person who takes a taxi pays v (resp. hv) dollars:

P5={j|mtx,tx,hvj,m,vj,notm,vm,notj}.

P5 has two answer sets S1={tx,j,v} and S2={tx,m,v}, where S1 (resp. S2) shows the case that Jack (resp. Mary) pays v dollar due to taking a taxi alone.

On the other hand, P5 has three split programs: π1=π{jtx}, π2=π{mtx} and π3=π{jtx,mtx}, where π={vj,notm,vm,notj,hvj,m,tx}. Since each πi has the unique answer set Si (1i3), P5 has three possible models S1, S2 along with S3={tx,j,m,hv} which corresponds to the third case such that both of Jack and Mary pay hv dollars thanks to sharing a taxi. Notice that S3 is not the answer set of P5.1515

In contrast, the same goes for argumentation under the stable semantics. First we construct ABF(P5) translated from P5 (resp. ABF(πi) translated from πi (1i3)). In this case, every extension of the respective ABF is consistent since each language does not include explicit negation. Then, ABF(π1) (resp. ABF(π2), ABF(π3)) has the unique stable assumption extension Δ1={notm,nothv} (resp. Δ2={notj,nothv}, Δ3={notv}), where CNπ1(Δ1)=CNP5(Δ1)={tx,j,v}Δ1=S1Δ1, CNπ2(Δ2)=CNP5(Δ2)={tx,m,v}Δ2=S2Δ2 and CNπ3(Δ3)={tx,j,m,hv}Δ3=S3Δ3CNP5(Δ3). Similarly, ABF(π1) (resp. ABF(π2), ABF(π3)) has the unique stable argument extension E1 (resp. E2, E3), where Concs(Ei)=SiΔi (1i3). Hence ABF(P5) has three stable assumption p-extensions Δi (1i3) along with three argument p-extensions Ei (1i3), while it has two stable assumption extensions Δ1,Δ2 along with two argument extensions E1,E2.

6.Related work

Beirlaen et al.’s extended ASPIC+ framework [2,3] as well as Heyninck and Arieli’s ABF induced by a DLP [18] can handle disjunctive information in argumentation. Moreover, it is shown that Heyninck and Arieli’s ABF induced by a DLP can capture stable model semantics of a DLP as shown in Proposition 1 [18, Proposition 2 and Proposition 3]. However the semantic correspondence to a disjunctive default theory is not shown in both Beirlaen et al.’s work and Heyninck and Arieli’s work. Notice here that Example 4 illustrates that an NDP (i.e. an EDLP having no classical negation) can be transformed to a disjunctive default theory, whereas a DLP used in Heyninck and Arieli’s ABF can neither be done so nor has the semantic correspondence to a default theory. In contrast, our approach based on a translation from an EDLP can capture not only the answer set semantics of an EDLP but also the semantics of a disjunctive default theory as shown in Theorem 8 (or Proposition 5) and Theorem 12. These are written in Table 1 and Table 2 in which existing work is shown in italic letter with proper references. Regarding the problem of how to construct arguments under the given disjunctive information, Beirlaen et al. [2,3] allowed arguments to have disjunctive conclusions, while Heyninck and Arieli [18] neither defined arguments nor took account of argument extensions in their ABFs. In contrast, in our ABFs, arguments are defined as trees due to Definition 12 but they are not allowed to have disjunctive conclusions according to Definition 13. Notice here that Heyninck and Arieli’s ABF has also three inference rules: [MP], [Res] and [RBC] like our ABFs though the classical ∨ is used instead of |. Then, thanks to our approach, it is possible to construct trees and define arguments in their ABF based on the slightly modified Definition 12 by replacing | with ∨. As a result, we may obtain a new theorem for a DLP P showing the correspondence between stable models of P and stable argument extensions of their ABF(P). Though it is possible to fill in the open box of Table 1 with such a theorem for a DLP, its details are omitted in this paper.

Table 1

Correspondence between LP and ABA

LPABA


ClassSemanticsAssumption extensionsArgument extensions
ELPp-stable modelsProposition 12Theorem 2 [32]
answer setsProposition 13 Proposition 14Theorem 3 [32] Theorem 4 [33]
DLPstable modelsProposition 1 [18]
NDPanswer setsProposition 3Theorem 6
EDLPp-stable modelsProposition 4Theorem 7
answer setsProposition 5 Proposition 9Theorem 8 Theorem 10
possible modelsProposition 11Theorem 21
Table 2

Correspondence between NMR and ABA

NMREDLPABA



FormalismsSemanticsAnswer setsArgument extensions
disjunctive default theoryextensionsTheorem 1 [16]Theorem 12 Theorem 13
parallel circumscriptionmodelsTheorem 16 [29]Theorem 18
prioritized circumscriptionmodelsTheorem 17Theorem 19

Lehtonen et al. [19] presented algorithms for reasoning in a default logic instantiation of ABA, where they defined the assumption-based argumentation framework (ABF) corresponding to a propositional default theory. In [19, Example 1], they showed the ABF corresponding a default theory which contains disjunctive formulas. However when the ABF corresponding to the default theory D1 is constructed according to [19, Definition 1], the ABF has the same difficulty as D1 addressed in Example 1. The reason is as follows: In their approach, given (Lp,Rp) as a deductive system for propositional logic, the ABF corresponding to D1 is F=(L,R,W,A,¯¯), where L=Lp{MααLp}, R=Rp{rp,Mr,kr,Mk,fy,Mf,kf,Mk}, W=D1Lp={py}, A={Mr,Mk,Mf} and a mapping function ¯¯ defined by Mx=¬x for all MxA. Besides, AA attacks BA iff WAR¬b for some MbB, where R denotes derivability via R. Then since there is no attacks between any two sets of assumptions, the ABF F has a unique assumption extension A under the stable (resp. grounded, complete) semantics, from which the expected result k is never derived under each semantics due to [19, Proposition 1]. In contrast, in our approach based on the disjunctive default theory D2 which has the associated EDLP P1, Example 12 shows that we can obtain the expected result k from the stable extensions Ei of the ABF translated from the EDLP P1 which correspond to the extensions Ei(i=1,2) of the ddt D2=emb(P1) based on Theorem 12, Theorem 8 and Proposition 5. Moreover, interestingly, Example 8 shows that the skeptical result meant by the problem is obtained under the skeptical semantics (i.e. the grounded and ideal semantics) in our ABF.

Bondarenko et al. [5] showed a correspondence between Reiter’s default extensions and stable extensions of the corresponding assumption-based framework (ABF) in [5, Theorem 3.16]. This indicates that Kyoto protocol problem shown in Example 1 cannot be solved under the stable semantics in their ABF corresponding to the default theory D1. Disjunctive default theories are not considered in [5]. Moreover, in [5, Theorem 6.7], Bondarenko et al. showed a correspondence between models of parallel circumscription:Circum(T;P;Z) (i.e. P,Z-minimal models of T) and maximal conflict-free extensions of the corresponding ABF, assuming that T is a first order theory without function symbols, T satisfies unique names axioms and domain closure axioms, and every model of T is a Herbrand model of T. However, they showed nothing about a semantic correspondence between prioritized circumscription and assumption-based frameworks. In contrast, this paper shows a correspondence between Herbrand models of parallel circumscription and stable argument extensions of our ABF in Theorem 18 as well as a correspondence between Herbrand models of prioritized circumscription and stable argument extensions of our ABF in Theorem 19 (see Table 2). Regarding logic programming, they showed a correspondence between stable models of an NLP and the stable extensions of the corresponding ABF in [5, Theorem 3.13]. Their theorem is, however, the special case of Proposition 13 for an ELP, Proposition 1 for a DLP, Proposition 3 for an NDP, and Proposition 5 for an EDLP.

Caminada and Schulz [7] showed the equivalence between various ABA semantics and various semantics of NLPs. NLPs prohibiting both disjunction and classical negation are less expressive than DLPs and ELPs. Hence a faithful modeling of real world problems often becomes impossible in the scope of NLPs.

7.Conclusion

We proposed an assumption-based framework (ABF) translated from an extended disjunctive logic program (EDLP), which incorporates explicit negation as well as | rather than ∨ in Heyninck and Arieli’s ABF induced by a DLP. Thanks to our proposed ABFs, the new results about the semantic relationships between logic programming (LP) and ABA as well as nonmonotonic reasoning (NMR) and ABA are obtained. That is, as for LP, the answer set semantics of an EDLP can be captured by the stable extensions of the ABF translated from an EDLP with trivialization rules, while as for NMR, extensions of a disjunctive default theory (resp. models of prioritized circumscription) can be captured by the stable extensions of the ABF translated from the EDLP corresponding to a given disjunctive default theory (resp. the EDLP corresponding to a given prioritized circumscription). Moreover, as another relationship to LP, it is shown that the possible model semantics of an EDLP is captured by the possible extensions under stable semantics of the ABF translated from an EDLP (see Table 1 and Table 2).

In the study of nonmonotonic reasoning, disjunctive default logic [16] was proposed as a generalization of default logic [27] to overcome difficulties of default logic in handling disjunctive information. In fact, defaults do not work in the default theory D1 expressing the Kyoto protocol problem [3] as well as in d1 corresponding to the DLP π1 [18, Example 1] shown in Example 4 due to lack of the capability to reason by cases in these default theories. As a result, D1 as well as the ABFs instantiated with D1 [5,19] reveal difficulties such that the expected result cannot be obtained from them as discussed in Section 1 and Section 6. In contrast, in our approach, the expected result k is successfully obtained not only from the stable extensions of the ABF translated from the EDLP P1 which correspond to the extensions of the disjunctive default theory D2=emb(P1) as shown in Example 12 but also from the extension(s) under the skeptical semantics (i.e. the grounded and ideal) as well as under the preferred semantics in our proposed ABF as shown in Example 8. Thus based on our ABFs corresponding to disjunctive default theories via EDLPs, our approach can overcome difficulties of assumption-based frameworks corresponding to default theories (e.g. [5,19]) in handling disjunctive information.

To sum up, as for argument extensions, Theorem 2 [32, Theorem 3] and Theorem 3 [32, Theorem 4] for an ELP (resp. Theorem 4 [33, Theorem 5] for a consistent ELP) in standard ABA frameworks are broadened to Theorem 7 and Theorem 8 for an EDLP (resp. Theorem 10 for a consistent EDLP) in generalized ABA frameworks, i.e. ABFs translated from EDLPs. Similarly as for assumption extensions, Proposition 12 and Proposition 13 (resp. Proposition 14) for the standard ABA framework instantiated with an ELP (resp. a consistent ELP) as well as Proposition 1 [18, Proposition 2 and Proposition 3] for the ABF induced by a DLP are generalized to Proposition 4 and Proposition 5 (resp. Proposition 9) for the respective ABFs translated from EDLPs (resp. a consistent EDLP). These are summarized in Table 1.

As one of practical advantages of our approach, even if disjunctive information exists, we can directly use dialectic proof procedures [9,11] since the AA framework [8] can be generated from our ABF treating disjunctive information.

In (extended) disjunctive logic programming, the existence of disjunction generally increases the expressive power of logic programs while brings computational penalty [13]. By analogy, argumentation in ABFs translated from (E)DLPs increases the expressive power of ABF while it would introduce additional complexity. Hence, the analysis of complexity is left for future work. Moreover, our future work is to explore and find the more general correspondence between Assumption-based frameworks and disjunctive default theories without intervening EDLPs.

Notes

1 Semantically, the latter requires an extension to contain one of the two disjunctive terms, rather than the disjunction [16].

2 In this paper, we use the abbreviation “ABF” to primarily denote assumption-based framework but sometimes ABA framework in different approaches [5,10,18,19,32,33] though it may refer to the respectively defined framework based on ABA in each approach.

3 For example, the difference between DLPs and NDPs as shown in Eample 4, the interesting results under the skeptical semantics (e.g. grounded) in Example 8, and the relation between our ABF and a standard ABA as shown in Proposition 8, etc.

4 A disjunctive logic program (DLP) defined in this paper is different from a normal disjunctive logic program (NDP) defined in Section 2.3. This is because we later consider transformation to disjunctive default theories, in which not a DLP but an NDP can be transformed to a disjunctive default theory. Details about this are discussed in Example 4.

5 Gelfond and Lifschitz [15] use the symbol not to denote the negation-as-failure (NAF), while Przymusinski [26] uses the symbol ∼ to denote NAF. This paper uses Gelfond and Lifschitz’s notation due to the purpose of this study.

6 In this paper, the term “p-stable models” is used not as an abbreviation of partial stable model semantics by Przymusinski [26] but as that of paraconsistent stable model semantics by Sakama and Inoue [28]. In [28], a p-stable model is defined for an EDLP whose rule head uses ∨ rather than |. However notice that, a p-stable model is defined regardless the connective of disjunction used in rule head (e.g. “∨”, “|”).

7 In [32], ¬.CM is used rather than ΔM to refer to the set {notL|LLitPM}. However for notational convenience, ΔM is used instead of ¬.CM to refer to this set in this paper.

8 For a set S, |S| denotes the cardinality of S.

9 This assumption is also used in all examples given in [32,33] though it is not stated there.

10 This is depicted vertically in the inference rule of [RBC].

11 The dependency graph of an EDLP P is a directed graph where each literal is a node and where there is an edge from L to L iff there is a rule in which a literal L is not preceded by the not operator in the body and L appears in the head. An EDLP is head-cycle free iff its dependency graph does not contain directed cycles that go through two literals that belong to the head of the same rule [4].

12 It is said that a clause C1 subsumes a clauseC2 if C1θC2 holds for some substitution θ [29].

13 In this study, any rule from an EDLP has the form (2). So if a rule r whose head is empty, i.e. “body(r)” is generated in the transformation, then Πα has the rule αbody(r),notα” instead of r both of which are semantically equivalent each other under the answer set semantics, where α is a newly introduced atom that does not appear in the language.

14 A split program of an EDLP can be encoded by choice rules supported in recent ASP solvers (e.g. clingo) [21].

15 However, when the problem is expressed by the EDLP P5=P5{j|¬j,m|¬m} instead of P5, P5 has three answer sets, S1={tx,j,v,¬m}, S2={tx,m,v,¬j} and S3={tx,j,m,hv} corresponding to three cases.

Acknowledgements

The author would like to thank Chiaki Sakama and the anonymous reviewers of the paper for their valuable comments and suggestions.

Appendices

Appendix

AppendixNew theorems for an ELP

A.1.Correspondence between answer sets of an ELP and stable assumption extensions

We show the correspondence between p-stable models (resp. answer sets) of an ELP P and stable assumption extensions in the standard ABA framework F(P) (resp. F(Ptr)) as follows.

Lemma 6.

Let P be an ELP and MLitP. We denote byderivability using modus ponens. Then M is a p-stable model of an ELP P iff M={LitP|PΔM}, where ΔM={not|(LitPM)}.

Proof.

It is proved in [5, p.80] using the inference rule (i.e. modus ponens) in [5, p.72, line 1] that

M is a stable model of an NLP P iff M={pHBP|PΔMp}(13)where ΔM={notp|pM}.

Now let P+ be the NLP which is the positive form [28] of an ELP P, and for a set SLitP, let S+ be the set obtained by replacing each negative literal ¬L in S with a newly introduced atom L like in the proof of Proposition 4. Then the Herbrand base HBP+ of P+ is (LitP)+.

  • (i) Suppose that M is a p-stable model of an ELP P. Then M+ is an answer set of the NLP P+ due to Lemma 3. Hence M+ is a stable model of the NLP P+ due to Lemma 1. Then based on (13), it holds that M+={pHBP+|P+ΔM+p} where ΔM+={notp|pM+}={notp|p(HBP+M+)}, which leads to M={LitP|PΔM} where ΔM={not|(LitPM)}.

  • (ii) The converse is also proved in a similar way to (i). □

Proposition 12.

Let P be an ELP and MLitP. If M is a p-stable model of P, then Δ={not|(LitPM} is a stable assumption extension of the ABA framework F(P)=LP,P,AP,¯¯. Conversely if ΔAP is a stable assumption extension of the ABF F(P), then M={LitP|notΔ} is a p-stable model of P.

Proof.

Let M be a p-stable model of an ELP P. Due to Lemma 6, it holds that, for any Litp, M iff PΔM, where ΔM={not|(LitPM)}=Δ. This means that in the ABF F(P)=LP,P,AP,¯¯, the set of assumption ΔM attacks not where M. Besides ΔM implies M={LitP|not(APΔM)}. Hence ΔM attacks not where not(APΔM). Hence Δ=ΔM is conflict-free and stable. Thus Δ is a stable assumption extension of the ABA framework F(P).

Conversely let ΔAP be a stable assumption extension of the ABA framework F(P) where AP=NAFP. Then since Δ is stable, Δ is conflict-free, and Δ attacks every not(APΔ), where LitP. This means that, PΔ for notΔ, and

(14)PΔfornot(APΔ).
Based on (14), let us define the set M such that M={LitP|PΔfornotΔ}. Then,
(15)Δ={not|(LitPM)}
is derived using M. Hence, based on Lemma 6, M is a p-stable model of P. Moreover, (15) implies M={LitP|notΔ}. Thus M={LitP|notΔ} is a p-stable model of P. □

Proposition 13.

Let P be an ELP and SLitP. If S is an answer set of P, then Δ={not|(LitPS)} is a stable assumption extension of the ABA framework F(Ptr). Conversely if Δ is a stable assumption extension of the ABF F(Ptr), then S={LitP|notΔ} is an answer set of P.

Proof.

(i) Suppose S is an answer set of an ELP P. Then due to [28, Theorem 3.5], S is a p-stable model of Ptr. Hence based on Proposition 12, Δ={not|(LitPS)} is a stable assumption extension of the ABF F(Ptr). (ii) The converse is also proved in a similar way. □

For a consistent ELP, the following proposition holds.

Proposition 14.

Let P be a consistent ELP and SLitP. S is an answer set of P iff there is a consistent stable assumption extension Δ={not|(LitPS)} of the ABA framework F(P).

Proof.

For a p-stable model S of an ELP P, it holds that, S={LitP|PΔ} for Δ={not|(LitPS)} due to Lemma 6. Based on it, it is derived that,

(16)CNP(Δ)={xLP|PΔx}=SΔ,for Δ={not|(LitPS)},
where CNP is a consequence operator given in Definition 2. Then

S is an answer set of a consistent ELP P

iff S is a consistent answer set of P

iff S is a consistent p-stable model of P

iff there is a stable assumption extension Δ={not|(LitPS)} of the ABF framework F(P),

  where due to (16), it holds that CNP(Δ)=SΔ, which is not contradictory, i.e. consistent

  according to Definition 2

iff there is a consistent stable assumption extension Δ={not|(LitPS)} of the ABF F(P). □

References

[1] 

O. Arieli, A. Borg, J. Heyninck and C. Straßer, Logic-based approaches to formal argumentation, IfCoLog Journal of Logics and their Applications 8: (6) ((2021) ), 1793–1898.

[2] 

M. Beirlaen, J. Heyninck and C. Straßer, Reasoning by cases in structured argumentation, in: Proceedings of the 2017 ACM Symposium on Applied Computing, ACM, (2017) , pp. 989–994.

[3] 

M. Beirlaen, J. Heyninck and C. Straßer, A critical assessment of Pollock’s work on logic-based argumentation with suppositions, in: Proceedings of the 17th International Workshop on Non-monotonic Reasoning (NMR-2018), (2018) , pp. 63–72.

[4] 

R. Ben-Eliyahu and R. Dechter, Propositional semantics for disjunctive logic programs, Annals of Mathematics and Artificial Intelligence 12: (1–2) ((1994) ), 53–87. doi:10.1007/BF01530761.

[5] 

A. Bondarenko, P.M. Dung, R.A. Kowalski and F. Toni, An abstract, argumentation-theoretic approach to default reasoning, Artificial Intelligence 93: ((1997) ), 63–101. doi:10.1016/S0004-3702(97)00015-5.

[6] 

M. Caminada and L. Amgoud, On the evaluation of argumentation formalisms, Artificial Intelligence 171: (5–6) ((2007) ), 286–310. doi:10.1016/j.artint.2007.02.003.

[7] 

M. Caminada and C. Schulz, On the equivalence between assumption-based argumentation and logic programming, Journal of Artificial Intelligence Research 60: ((2017) ), 779–825. doi:10.1613/jair.5581.

[8] 

P.M. Dung, On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games, Artificial Intelligence 77: (2) ((1995) ), 321–358. doi:10.1016/0004-3702(94)00041-X.

[9] 

P.M. Dung, R.A. Kowalski and F. Toni, Dialectic proof procedures for assumption-based, admissible argumentation, Artificial Intelligence 170: (2) ((2006) ), 114–159. doi:10.1016/j.artint.2005.07.002.

[10] 

P.M. Dung, R.A. Kowalski and F. Toni, Assumption-based argumentation, in: Argumentation in Artificial Intelligence, Springer-Verlag, (2009) , pp. 199–218. doi:10.1007/978-0-387-98197-0_10.

[11] 

P.M. Dung, P. Mancarella and F. Toni, Computing ideal sceptical argumentation, Artificial Intelligence 171: ((2007) ), 642–674. doi:10.1016/j.artint.2007.05.003.

[12] 

P.M. Dung and P.M. Thang, Closure and consistency in logic-associated argumentation, Journal of Artificial Intelligence Research 49: ((2014) ), 79–109. doi:10.1613/jair.4107.

[13] 

T. Eiter and G. Gottlob, Complexity results for disjunctive logic programming and application to nonmonotonic logics, in: Logic Programming, Proceedings of the 10th International Symposium (ILPS-1993), MIT Press, (1993) , pp. 266–278.

[14] 

M. Gelfond and V. Lifschitz, The stable model semantics for logic programming, in: Logic Programming, Proceedings of the Fifth International Conference and Symposium (ICLP/SLP-1988), MIT Press, (1988) , pp. 1070–1080.

[15] 

M. Gelfond and V. Lifschitz, Classical negation in logic programs and disjunctive databases, New Generation Computing 9: ((1991) ), 365–385. doi:10.1007/BF03037169.

[16] 

M. Gelfond, V. Lifschitz, H. Przymusinska and M. Truszczynski, Disjunctive defaults, in: Proceedings of the 2nd International Conference on Principles of Knowledge Representation and Reasoning (KR-1991), (1991) , pp. 230–237.

[17] 

J. Heyninck and O. Arieli, On the semantics of simple contrapositive assumption-based argumentation frameworks, in: Proceedings of the Seventh International Conference on Computational Models of Argument (COMMA-2018), IOS Press, (2018) , pp. 9–20.

[18] 

J. Heyninck and O. Arieli, An argumentative characterization of disjunctive logic programming, in: Proceedings of the 19th EPIA Conference on Artificial Intelligence (EPIA-2019), (2019) , pp. 526–538.

[19] 

T. Lehtonen, J.P. Wallner and M. Järvisalo, Algorithms for reasoning in a default logic instantiation of assumption-based argumentation, in: Proceedings of the 9th International Conference on Computational Models of Argument (COMMA-2022), IOS Press, (2022) , pp. 236–247.

[20] 

V. Lifschitz, Computing circumscription, in: Proceedings of the Ninth International Joint Conference on Artificial Intelligence (IJCAI-1985), (1985) , pp. 121–127.

[21] 

V.W. Marek, I. Niemelä and M. Truszczynski, Logic programs with monotone abstract constraint atoms, Theory and Practice of Logic Programming 8: (2) ((2008) ), 167–199. doi:10.1017/S147106840700302X.

[22] 

J. McCarthy, Circumscription – a form of non-monotonic reasoning, Artificial Intelligence 13: ((1980) ), 27–39. doi:10.1016/0004-3702(80)90011-9.

[23] 

J. McCarthy, Applications of circumscription to formalizing commonsense knowledge, Artificial Intelligence 28: ((1986) ), 89–116. doi:10.1016/0004-3702(86)90032-9.

[24] 

S. Modgil and H. Prakken, The ASPIC+ framework for structured argumentation: A tutorial, Argument and Computation 5: (3) ((2014) ), 31–62. doi:10.1080/19462166.2013.869766.

[25] 

H. Prakken, An abstract framework for argumentation with structured arguments, Argument and Computation 1: (2) ((2010) ), 93–124. doi:10.1080/19462160903564592.

[26] 

T.C. Przymusinski, Stable semantics for disjunctive programs, New Generation Computing 9: ((1991) ), 401–424. doi:10.1007/BF03037171.

[27] 

R. Reiter, A logic for default reasoning, Artificial Intelligence 13: ((1980) ), 81–132. doi:10.1016/0004-3702(80)90014-4.

[28] 

C. Sakama and K. Inoue, Paraconsistent stable semantics for extended disjunctive programs, Journal of Logic and Computation 5: (3) ((1995) ), 265–285. doi:10.1093/logcom/5.3.265.

[29] 

C. Sakama and K. Inoue, Embedding circumscriptive theories in general disjunctive programs, in: Proceedings of the Third International Conference on Logic Programming and Nonmonotonic Reasoning (LPNMR-1995), (1995) , pp. 344–357. doi:10.1007/3-540-59487-6_25.

[30] 

C. Sakama and K. Inoue, Abductive logic programming and disjunctive logic programming: their relationship and transferability, Journal of Logic and Computation 44: (1–3) ((2000) ), 75–100. doi:10.1016/S0743-1066(99)00073-4.

[31] 

C. Schulz and F. Toni, Justifying answer sets using argumentation, Theory and Practice of Logic Programming 16: (1) ((2016) ), 59–110. doi:10.1017/S1471068414000702.

[32] 

T. Wakaki, Assumption-based argumentation equipped with preferences and its application to decision-making, practical reasoning, and epistemic reasoning, Journal of Computational Intelligence 33: (4) ((2017) ), 706–736. doi:10.1111/coin.12111.

[33] 

T. Wakaki, Consistency in assumption-based argumentation, in: Proceedings of the 8th International Conference on Computational Models of Argument (COMMA-2020), IOS Press, (2020) , pp. 371–382.

[34] 

T. Wakaki, Assumption-based argumentation for extended disjunctive logic programming, in: Proceedings of the 12th International Symposium on Foundations of Information and Knowledge Systems (FoIKS-2022), (2022) , pp. 35–54.

[35] 

T. Wakaki and K. Satoh, Compiling prioritized circumscription into extended logic programs, in: Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI-1997), (1997) , pp. 182–187.

[36] 

https://arxiv.org/pdf/2106.10966.pdf.