Content uploaded by Antonis C. Kakas
Author content
All content in this area was uploaded by Antonis C. Kakas on Nov 26, 2013
Content may be subject to copyright.
Abduction in Logic Programming
Marc Denecker
1
and Antonis Kakas
2
1
Department of Computer Science, K.U.Leuven,
Celestijnenlaan 200A, B-3001 Heverlee, Belgium.
Marc.Denecker@cs.kuleuven.ac.be
http://www.cs.kuleuven.ac.be/∼marcd
2
Department of Computer Science, University of Cyprus,
75 Kallipoleos St., Nicosia, Cyprus.
antonis@ucy.ac.cy
http://www.cs.ucy.ac.cy/∼antonis
Abstract. Abduction in Logic Programming started in the late 80s,
early 90s, in an attempt to extend logic programming into a framework
suitable for a variety of problems in Artificial Intelligence and other areas
of Computer Science. This paper aims to chart out the main develop-
ments of the field over the last ten years and to take a critical view of
these developments from several perspectives: logical, epistemological,
computational and suitability to application. The paper attempts to ex-
pose some of the challenges and prospects for the further development
of the field.
1 Introduct ion
Over the last two decades, abduction has been embraced in AI as a non-monotonic
reasoning paradigm to address some of the limitations of deductive reasoning in
classical logic. The role of abduction has been demonstrated in a variety of appli-
cations. It has been proposed as a reasoning paradigm in AI for diagnosis [8,90],
natural language understanding [8,39,4,93], default reasoning [81,29,25,50], plan-
ning [28,110,71,59], knowledge assimilation and belief revision [54,76], multi-
agent systems [7,64,102] and other problems.
In the context of logic programming, the study of abductive inference started
at the end of the eighties as an outcome of different attempts to use logic pro-
gramming for solving AI-problems. Facing the limitations of standard logic pro-
gramming for solving these problems, different researchers proposed to extend
logic programming with abduction. Eshghi [28] introduced abduction in logic
programming in order to solve planning problems in the Event Calculus [65]. In
this approach, abduction solves a planning goal by explaining it by an ordered
sets of events -a plan- that entails the planning goal. This approach was further
explored by Shanahan [110], Missiaen et al. [72,71], Denecker [21], Jung [48] and
recently in [59,60]. Kakas and Mancarella showed the application of abduction in
logic programming for deductive database updating and knowledge assimilation
[53,55]. The application of abduction to diagnosis has been studied in [10,11]
A.C. Kakas, F. Sadri (Eds.): Computat. Logic (Kowalski Festschrift), LNAI 2407, pp. 402–436, 2002.
c
Springer-Verlag Be rlin Heidelberg 2002
Abduction in Logic Programming 403
within an abductive logic programming framework whose semantics was defined
by a suitable extension of the completion semantics of LP.
In parallel to these studies of abduction as an inferential method, Eshghi and
Kowalski [29] and later Kakas and Mancarella in [52,54] and Dung in [25], used
abduction as a semantical device to describe the non-monotonic semantics of
Logic Programming (in a way analogous to Poole in [81]). In [18,14], abductive
logic programming was investigated from a knowledge representation point of
view and its suitability for representing and reasoning on incomplete information
and definitional and assertional knowledge was shown.
For these reasons, Abductive Logic Programming
1
(ALP) [50,51] was recog-
nized as a promising computational paradigm that could resolve many limita-
tions of logic programming with respect to higher level knowledge representation
and reasoning tasks. ALP has manifested itself as a framework for declarative
problem solving suitable for a broad collection of problems.
Consequently, at the start of the 90s, a number of abductive systems were
developed. In [54], the abductive procedure of [29] for computing negation as fail-
ure through abduction was extended to the case of general abductive predicates.
Another early abductive procedure was developed in [10] using the completion.
[17] proposed SLDNFA, an extension of SLDNF with abduction allowing non-
ground abductive hypotheses. [21] proposed an extension of SLDNFA with a
constraint solver for linear order and demonstrated that this system could be
applied correctly for partial order planning in the context of event calculus.
Later, the idea of integrating abduction and constraint solving was developed
more generally in the ACLP framework [56,61,60]; this procedure is the result
of incorporating CLP constraint solving in the abductive procedure of [54]. In
[37] an abductive procedure that can be regarded as a hybrid of SLDNFA and
the procedure of Console et al has been defined based on explicit rewrite rules
with the completion and equality. This has later [66] incorporated constraint
solving in a similar way to the ACLP procedure. A bottom up procedure, later
combined with some top down refinement, was given in [106] and [42]; the latter
system was an implementation using the Model Generator MGTP developed
on the multiprocessor machine developed at ICOT. Another recent abductive
procedure in LP is that of AbDual [1] which exploits tabling techniques from
XSB.
Despite these efforts and the many potential applications for abduction, it has
taken considerable time and effort to develop computationally effective systems
based on abduction for practical problems. The field has faced (and to some
extend continues to do so) a number of challenges at the logical, methodological
and implementational level. In the recent past, important progress has been
made on all these levels. The aim of this chapter is to give a comprehensive
1
The July/August 2000 volume (Vol. 44) of the journal of Logic Programming is a
special issue on Abductive Logic Programming. This contains several papers that
open new perspectives on the relationship between abduction and other computa-
tional paradigms.
404 Marc Denecker and Antonis Kakas
overview of the state of the art of Abductive Logic Programming, to point to
problems and challenges and to sketch recent progress.
Bob Kowalski has been one of the founders of Abductive Logic Program-
ming. Recently, he has, together with others, proposed [64] that ALP can be
used as a framework in which we can integrate an agent’s knowledge on how to
reduce its goal to subgoals and thus plan how to achieve this goal, described in
the program part of an abductive theory, together with the agent’s obligations,
prohibitions and other elements that determine its reactive behaviour, described
in the integrity constraints part of its abductive theory. In this suggestion ALP
plays a central role in capturing the behaviour of an autonomous agent that
feeds from and reacts to its environment.
This together with his view of its role in the way that Logic Programming
should evolve more generally as a programming language of the future is de-
scribed elegantly in his short position statement on the future of Logic Pro-
gramming in this volume.
The rest of the paper is organized as follows. Section 2 briefly reviews the
study of abduction in AI and philosophy and situates Abductive Logic Pro-
gramming within this broad context. Section 3 gives the formal definition of
abduction, and reviews the different formal semantics that have been proposed
in the literature. Section 4 reviews the different ALP frameworks that have been
developed so far analyzing their potential scope to applications and their links
to other extensions of LP. The paper ends with a discussion of future challenges
and prospects of development for the field of ALP.
2 What Is Abduction?
2.1 What Is an Explanation?
The term abduction was introduced by the logician and philosopher C.S. Pierce
(1839-1914) who defined it as the inference process of forming a hypothesis
that explains given observed phenomena [77]. Often abduction has been defined
broadly as any form of “inference to the best explanation” [47] where best refers
to the fact that the generated hypothesis is subjected to some optimality cri-
terion. This very broad definition covers a wide range of different phenomena
involving some form of hypothetical reasoning. Studies of “abduction” range
from philosophical treatments of human scientific discovery down to formal and
computational approaches in different formal logics.
In the context of formal logic, abduction is often defined as follows. Given a
logical theory T representing the expert knowledge and a formula Q represent-
ing an observation on the problem domain, abductive inference searches for an
explanation formula E such that:
– E is satisfiable
2
w.r.t. T and
– it holds that
3
T |= E→Q
2
If E contains free variables, ∃(E) should be satisfiable w.r.t. T .
3
Or, more general, if Q and E contain free variables: T |= ∀(E→Q).
Abduction in Logic Programming 405
In general, E will be subjected to further restrictions such as the aforementioned
minimality criteria and criteria restricting the form of the explanation formula
(e.g. by restricting the predicates that may appear in it). This view defines an
abductive explanation of an observation as a formula which logically entails the
observation. However, some have argued, sometimes with good reasons, that it
is more natural to view an explanation as a cause for the observation [47]. A
well-known example is as follows [92]: the disease paresis is caused by a latent
untreated form of syphilis. The probability that latent untreated syphilis leads
to paresis is only 25%. Note that in this context, the direction of entailment and
causality are opposite: syphilis is the cause of paresis but does not entail it, while
paresis entails syphilis but does not cause it. Yet a doctor can explain paresis
by the hypothesis of syphilis while paresis cannot account for an explanation for
syphilis.
In practice, examples where causation and entailment do not correspond are
rare
4
. It turns out that in many applications of abduction in AI, the theory T
describes explicit causality information. This is notably the case in model-based
diagnosis and in temporal reasoning, where theories describe effects of actions.
By restricting the explanation formulas to the predicates describing primitive
causes in the domain, an explanation formula which entails an observation gives
a cause for the observation. Hence, for this class of theories, the logical entailment
view implements the causality view on abductive inference.
2.2 Relationship to Other Reasoning Paradigms
As mentioned in the previous section, the definition of abduction is very broad
and covers a wide range of hypothetical reasoning inference that could otherwise
be formally distinguished. Not surprisingly, there are many different views on
what is abduction and how to implement it. Many philosophers and logicians
have argued that abduction is a generalization of induction [34]. Induction can be
defined as inference of general rules that explain certain data. A simple example
illustrating inductive inference is the following derivation:
human(socrates)
mortal(socrates)
∀x.human(x) → mortal(x)
Hence, induction can also be seen as a form of inference to the best explanation.
The term abduction as used in this paper, refers to a form of reasoning that
can be clearly distinguished from inductive inference. In most current applica-
tions of abduction the goal is to infer extentional knowledge, knowledge that is
specific to the particular state or scenario of the world. In applications of induc-
tion, the goal is to infer intentional knowledge, knowledge that universally holds
4
See [91] where the relation between causal and evidential modeling and reasoning is
studied and linked to abduction.
406 Marc Denecker and Antonis Kakas
in many different states of affairs and not only in the current state of the world.
For example, an abductive solution for the problem that a certain car does not
start this morning is the explanation that its battery is empty. This explanation
is extentional. On the other hand, an inductive inference is to derive from a set
of examples the rule that if the battery is empty then the car will not start. This
is intentional knowledge. As a consequence of this distinction, abductive answers
and inductive answers have a very different format. In particular, inductive an-
swers are mostly general rules that do not refer to a particular scenario while
abductive answers are usually simpler formulas, often sets of ground atoms, that
describe the causes of the observation in the current scenario according to a given
general theory describing the problem domain. This distinction in the form of the
answer induces in turn strong differences in the underlying inference procedures.
Abduction as a form of inference of extentional hypotheses explaining ob-
served phenomena, is a versatile and informative way of reasoning on incomplete
or uncertain knowledge. Incomplete knowledge does not entirely fix the state of
affairs of the domain of discourse while uncertain knowledge is defeasible in the
sense that its truth in the domain of discourse is not entirely certain. In the pres-
ence of uncertain knowledge, [81] demonstrated how abductive inference can be
used for default reasoning. In the presence of incomplete knowledge, abduction
returns an explanation formula corresponding to a (non-empty) collection of pos-
sible states of affairs in which the observation would be true or would be caused;
on the other hand deduction is the reasoning paradigm to determine whether a
statement is true in all possible states of affairs. As such, abduction is strongly
related to model generation and satisfiability checking and can be seen as a re-
finement of these forms of reasoning. By definition, the existence of an abductive
answer proves the satisfiability of the observation. But abduction returns more
informative answers; answers which describe the properties of a class of possible
states of affairs in which the observation is valid.
2.3 Abduction and Declarative Knowledge Representation
An important role of logic in AI is that of providing a framework for declarative
problem solving. In this, a human expert specifies his knowledge of the problem
domain by a descriptive logic theory and uses logical inference systems to solve
computational tasks arising in this domain. Although in the early days of logic-
based AI, deduction was considered as the unique fundamental problem solving
paradigm of declarative logic [70] in the current state of the art deduction has
lost this unique place as the central inferential paradigm of logic in AI. Indeed,
we argue that a declarative problem solving methodology will often lead to
abductive problems. Let us illustrate this with an example problem domain,
that of university time tabling.
The process of declarative problem solving starts with the ontology design
phase: during this step, an alphabet of symbols denoting the relevant objects,
concepts and relationships in the problem domain must be designed. This al-
phabet defines the ontology of the problem domain. It precedes the knowledge
description phase during which the expert expresses his knowledge using the
Abduction in Logic Programming 407
symbols of the alphabet. In the example domain of university timetables we have
three important types of objects: lectures, time slots and class rooms.Wecould
represent them using the predicates lecture/1, time
slot/1androom/1. Impor-
tant relevant relationships between them refer to when and where lectures take
place; these relationships could be represented by predicates time
of lecture/2
and room
of lecture/2.
A key observation is that even though at this stage the knowledge spec-
ification has not even started, the choice of the alphabet already determines
that certain tasks will be abductive. In particular, the task of computing a
correct time table will consist of computing tables for time
of lecture/2and
room
of lecture/2 that satisfy certain logical constraints imposed on correct
schedules. This task is not a deductive task: the “correct” tables will not be
deducible from the theory. Rather it is an abductive problem — or a model gen-
eration problem
5
— a problem of completing the problem description so that
the goal (or ”observation”) that all lectures are scheduled, holds.
The ontology design phase has a strong impact on the specification phase.
If the alphabet is complex and does not have a simple correspondence to the
objects, concepts, relations and functions of the domain of discourse, this will
complicate the knowledge description phase and lead to a more complex, more
verbose, less comprehensive and less modular specification. For example, one
simple constraint is that each lecture must be scheduled at some time slot and
room:
∀l : lecture(l) →∃t, r : time
slot(t) ∧ time of lecture(t, l)∧
room(r) ∧ room
of lecture(r, l)
Another constraint is that two lectures cannot take place in the same room at
the same time:
∀t, r, l1,l2. room
of lecture(r, l1) ∧ room of lecture(r, l2)∧
time
of lecture(t, l1) ∧ time of lecture(t, l2)
→ l1=l2
If we had represented the assignments of rooms and time slots to lectures by
balanced binary tree structures, this would have complicated significantly the
expression of these requirements which are now expressed directly. Thus, an
important aspect of the declarative problem solving methodology, is that the
ontology and alphabet is designed in a task independent way such that it nat-
urally matches with the types of objects and relationships that occur in the
problem domain.
The above example illustrates that the choice of the alphabet may enforce the
use of a specific type of inference to solve a specific computational task, and that,
when we follow a declarative approach this will often lead to the problem tasks
to be abductive in nature. Vice versa, the a-priori choice of a specific inferential
5
Recall the close relationship between abduction (as viewed in this paper) and model
generation, as explained in the previous section.
408 Marc Denecker and Antonis Kakas
system such as Prolog or a CLP system to solve a specific computational problem
has a strong impact on the choice of the alphabet for that problem domain.
In the university time tabling problem, a Prolog or CLP solution will not be
based on the use of predicates time
of lecture/2androom of lecture/2 but on
another, in this case more complex alphabet, typically one in which predicates
range over lists or trees of assignments of lectures, time slots and rooms. This
alphabet is more complex and is choosen in a task-dependent way, which results
in reduced readability and modularity of the problem specification and in a
reduced reusability of the specification to solve other types of tasks. The fact that
the alphabet in Prolog (and CLP) programming must be chosen in relation to
the provided inference system rather than by its match with the human experts
conception of the problem domain is one of the fundamental reasons why even
pure Prolog programs are rarely perceived as truly declarative.
There is clearly a trade-off here. On the one hand, the choice of an alphabet in
correspondence with the concepts and relations in the mind of the human expert
is a prerequisite to obtain a compact, elegant and readable specification. As a
result of this often the computational task of problem solving links tightly to ab-
duction. Putting this more directly we would argue that Declarative Knowledge
Representation comes hand in hand with abduction. Abduction then emerges as
an important computational paradigm that would be needed for certain problem
solving tasks within a declarative representation of the problem domain.
On the other hand, in practice the choice of a representation is not governed
merely by issues relating to a natural representation but also and sometimes
even more by issues of computational effectiveness. Although the use of a more
complex ontology may seriously reduce the elegance of the representation, it may
be necessary to be able to use a specific system for solving a problem; in some
cases this would mean that the pure declarative problem representation needs
to be augmented with procedural, heuristic and strategic information on how to
solve effectively the computational problem.
Current research on abduction studies how more intelligent search and infer-
ence methods in abductive systems can push this trade-off as far as possible in
the direction of more declarative representations.
3 Abductive Logic Programming
This section presents briefly how abduction has been defined in the context of
logic programming.
An Abductive Logic Programming theory is defined as a triple (P, A, IC)
consisting of a logic program, P , a set of ground abducible atoms A
6
and a set
of classical logic formulas IC, called the integrity constraints, such that no atom
p ∈ A occurs in the head of a rule of P .
In the field of Abductive Logic Programming, the definition of abduction is
usually specialized in the following way:
6
In practice, the abducibles are specified by their predicate names.
Abduction in Logic Programming 409
Definition 1. Given an abductive logic theory (P, A, IC), an abductive expla-
nation for a query Q is a set ∆ ⊆ A of ground abducible atoms such that:
– P ∪ ∆ |= Q
– P ∪ ∆ |= IC
– P ∪ ∆ is consistent.
Some remarks are in order. First, this definition is generic both in terms of
syntax and semantics. Often, the syntax is that of normal logic programs with
negation as failure but some have investigated the use of abduction in the context
of extended logic programming [43] or constraint logic programming [56,60,66].
At the level of semantics, the above definition defines the notion of an abductive
solution in terms of any given semantics of standard logic programming. Each
particular choice of semantics defines its own entailment relation |=, its own no-
tion of consistent logic programs and hence its own notion of what an abductive
solution is. In practice, the three main semantics of logic programming — com-
pletion, stable and well-founded semantics — have been used to define different
abductive logic frameworks.
A second remark is that an abductive explanation ∆ aims to represent a
nonempty collection of states of affairs in which the explanandum Q would hold.
This explains the third condition that P ∪ ∆ should be consistent.
Third, when integrity constraints IC are introduced in the formalism, one
must define how they constrain the abductive solutions. There are different views
on this. Early work on abduction in Theorist in the context of classical logic [81],
was based on the consistency view on constraints. In this view, any extension
of the given theory T with an abductive solution ∆ is required to be consistent
with the integrity constraints IC: T ∪ IC ∪ ∆ is consistent. The above defini-
tion implements the entailment view: the abductive solution ∆ together with P
should entail the constraints. This view is the one taken in most versions of ALP
and is stronger than the consistency view in the sense that a solution according
to the entailment view is a solution according to the consistency view but not
vice versa.
The difference between both views can be subtle but in practice the different
options usually coincide. E.g. it frequently happens that P ∪ ∆ has a unique
model, in which case both views are equivalent. In practice, many ALP systems
[20,61] use the entailment view as this can be easily implemented without the
need for any extra specialized procedures for the satisfaction of the integrity
constraints since this semantics treats the constraints in the same way as the
query.
The above definition aims to define the concept of an abductive solution for a
query but does not define abductive logic programming as a logic in its own right
as a pair of syntax and semantics. However, a notion of generalized model can
be defined, originally proposed in [52], which suggests the following definition.
Definition 2. M is a model of an abductive logic framework (P, A, IC) iff there
exists a set ∆ ⊆ A such that M is a model of P ∪ ∆ (according to some LP-
semantics) and M is a classical model of IC, i.e. M |= IC.
410 Marc Denecker and Antonis Kakas
The entailment relation between abductive logic frameworks and classical logic
formulas is then defined in the standard way as follows:
(P, A, IC) |= F iff for each model M of (P, A, IC), M |= F .
Note that this definition is also generic in the choice of the semantics of
logic programming. This way, abductive extensions of stable semantics [52], of
well-founded semantics [79] and the partial stable model semantics [111] have
been defined. Also the completion semantics has been extended [10] to the case
of abductive logic programs. The completion semantics of an abductive logic
framework (P, A,IC) is defined by the mapping it to its completion. This is the
first order logic theory consisting of :
– UN, the set of unique names axioms, or Clark’s equality theory.
– IC
– comp(P, A), the set of completed definitions for all non-abducible predicates.
A recent study [16] that attempts to clarify further the representational and
epistemological aspects of ALP, has proposed ID-logic as an appropriate logic
for ALP. ID-logic is defined as an extension of classical logic with inductive def-
initions. Each inductive definition consists of a set of rules defining a specific
subset of predicates under the well-founded semantics. This logic gives an epis-
temological view on ALP in which an abductive logic program is a definition of
the set of the non-abducible predicates and abducible predicates are open predi-
cates, i.e. not defined. The integrity constraints in an abductive logic framework
are simply classical logic assertions. Thus the program P represents the human
expert’s strong definitional knowledge and the theory IC represents the human
expert’s weaker assertional knowledge. Therefore in ID-logic, ALP can be seen
as a sort of description logic in which the program is a TBOX consisting of
one simultaneous definition of the non-abducible predicates, and the assertions
correspond to the ABOX [115].
4 Abductive Logic Programming Frameworks
The framework defined in the previous section is generic in syntax and seman-
tics. In the past ten years, the framework has been instantiated (and sometimes
has been extended) in different ways. In order to show the wider variety of mo-
tivations and approaches that are found in Abductive Logic Programming, this
section aims to present briefly a number of these alternative frameworks, imple-
mented systems and applications. These different instantiations differ from each
other by using different formal syntax or semantics, or sometimes simply because
they use a different inference method and hence induce a different procedural
semantics.
Abduction in Logic Programming 411
4.1 Approaches under the Com pletion Semantics for LP
Abduction through Deduction. One of the first ALP frameworks is that of
[10]. The syntax in this framework is that of hierarchical logic programs
7
with
a predefined set of abducible predicates. The formal syntax is an extension of
Clark’s completion semantics [9] in which only the non-abducible predicates
are completed. The main aim of this work was to study the relationship
between abduction and deduction in the setting of non-monotonic reasoning.
In particular, many characterizations of non-monotonic reasoning such as
circumscription, predicate completion, explanatory closure implement a sort
of closure principle allowing to extract implicit negative information out
of explicit positive information. What is shown in this work is that for a
restricted class of programs, the abductive explanations to a query with
respect to a set of (non-recursive) rules can be characterized in a deductive
way if we apply the completion semantics as a closure principle.
Formally, given a (hierarchical) abductive logic program P with abducibles
A, its completion P
C
consists of iff-definitions for the non-abducible predi-
cates. These equivalences allow to rewrite any observation O to an equivalent
formula F in the language of abducible predicates such that P
C
|= O ↔ F
where |= is classical logic entailment. The formula F , called the explana-
tion formula, can be seen as a disjunctive characterization of all abductive
solutions of O given P . The restriction to hierarchical programs ensures ter-
mination of a procedure to compute the explanation formula. The framework
has been extended to handle (a restricted form of)integrity constraints.
The above abductive framework has been used to formalize diagnostic prob-
lem solving and classification in nonmonotonic inheritance hierarchies [10,24],
and has been extended to characterize updates in deductive databases [12].
The completion semantics is also the basis for the ”knowledge compilation”
optimization of abductive problem solving described in [11].
The IFF Framework. The IFF framework is also based on the completion
semantics. It was initially developed as a unifying framework integrating ab-
duction and view updating [36,37]. The IFF proof procedure is defined by
a rewriting system in which an initial goal is rewritten to a disjunction of
answers. The main rewrite rules are unfolding, namely backward reasoning
with the iff definitions, and propagation, namely forward reasoning with the
integrity constraints. IFF produces answers to goals in the form of conjunc-
tions of abducible atoms and denial integrity constraints. An extension of it
with special treatment of built-in predicates and constraint logic program-
ming was proposed in [120,66]. Another modification of the IFF proof pro-
cedure was developed for applications modeling reasoning of rational agents
[64] and management of active rules in databases [96]. The main underlying
LP semantics used in this framework is Fitting’s three-valued completion
semantics but correctness results have been proven also for perfect model
semantics and under some restrictions for stable semantics.
7
A hierarchical program is one without recursion.
412 Marc Denecker and Antonis Kakas
Prototype implementations of the three instances of the IFF procedure
[37,97] exist and have been applied in many experiments. The original IFF
proof procedure has been implemented in Java and was applied within a
Voyager extension to the problem of interaction and communication amongst
multiple agents, as well as cooperative problem solving. It was also used for
information integration from multiple sources [95], to the management of in-
formation networks [112], and it has been integrated with PROGOL to learn
preconditions of actions in the frameworks of the event and situation calculi.
The extension presented in [120,66] has been applied to job-shop scheduling
[66] and semantic query optimization [121]. The procedure suggested in [96]
was implemented in April, and was used in the context of applications for
active databases and agents. Recently, it has been used to study the problem
of resource allocation in a multi-agent environment [102].
4.2 Approaches under Stable and Well-Founded Semantics
In Logic Programming other semantics have been proposed as refinements of the
completion semantics. These include the stable model semantics [38] and the
well-founded model semantics [116]. The following ALP frameworks use these
semantics for their underlying LP framework.
SLDNFA and ID-logic. SLDNFA [17,20] is an abductive extension of SLDNF-
resolution [68], suitable for abductive reasoning in the context of (possibly
recursive) abductive logic programs under the completion semantics. It was
proven sound and, under certain restrictions, complete with respect to the
3-valued completion and well-founded semantics. This procedure came out
of the early attempts to implement AI-planning using abductive reasoning in
the event calculus. It was one of the first procedures that correctly handles
non-ground abduction, i.e. abduction of atoms with variables. The procedure
was also used in one of the first experiments of integration of abduction and
constraint solving. [21] describes an extension of SLDNFA with a constraint
solver for the theory of total order and applies it for partial order planning
and in the context of temporal reasoning with incomplete knowledge.
At the logical level, the work evolved into a study of the role of ALP
for knowledge representation and of SLDNFA for abductive and deductive
reasoning. A number of subsequent experiments with ALP and SLDNFA
demonstrated the role of ALP for knowledge representation of incomplete
and temporal knowledge [19,113,114,22]. To explain and clarify the repre-
sentational and epistemological aspects of ALP, [16] proposed ID-logic, an
integration of classical logic with inductive definitions under well-founded
semantics.
At the computational level, efforts were done to improve the computational
performance and expressivity of the original implementation of the SLDNFA
procedure. The SLDNFAC system [117] is developed at the K.U.Leuven and
implements abduction in the context of ID-Logic, supporting directly gen-
eral first order classical axioms in the language and higher order aggregates.
Abduction in Logic Programming 413
The system integrates constraint solving with the general purpose abductive
resolution SLDNFA. It is implemented as a meta-interpreter on top of Sicstus
prolog and is available from http://www.cs.kuleuven.ac.be/ dtai/kt/systems-
E.shtml.
The SLDNFAC system has been used in the context of prototypical con-
straint solving problems such as N-queens, logical puzzles, planning prob-
lems in the blocks world, etc . . . for proving infinite failure of definite logic
programs [5], failure of planning goals [13] and for semantic interpretation
of temporal information in natural language [119]. An extension of the sys-
tem has also been used in the context of a scheduling application for the
maintenance for power units of power plants. This experiment involves the
use of higher order aggregates and is described in detail in section 5.1. Re-
cently, [78] compared SLDNFAC with different other approaches for solving
constraint problems including CLP, ACLP and the Smodels system [74] and
shows that in many problems the system is competitive.
Bottom up Abduction. This approach was proposed originally in [107] and
aims to develop efficient techniques for computing abductive solutions un-
der the generalized stable model semantics [52] by translating the abductive
logic program to a standard logic program and applying efficient bottom
up stable model generators to this translation. This approach is based on a
translation of Abductive logic programs into pure logic programs with stable
model semantics [107]. Abductive solutions w.r.t. the original abductive logic
program correspond to stable models of its translation
8
. To compute abduc-
tive solutions, [107] also proposed a procedure for bottom-up stable model
computation based on truth maintenance techniques. It is an extension of
the procedure for computing well-founded models of [33,94] and dynami-
cally checks integrity constraints during the computation of stable models
and uses them to derive facts. Later this bottom up procedure was integrated
with a procedure for top-down expectation [108,46]. This top-down proce-
dure searches for atoms and rules that are relevant for the query (and the
integrity constraints) and thus helps to steer the search into the direction of
a solution. This procedure has been used for a number of applications in the
following two domains.
Legal Reasoning: A dynamic notion of similarity of cases in legal reason-
ing is implemented using abductive logic programming. The input of this
system is legal factors, case bases and the current case and position of
user (defendant or plaintiff). The system translates the case bases and
the current case into an abductive logic program. Using the top-down
proof procedure the system then computes important factors and re-
trieves a similar case based on the important factors and generates an
explanation why the current case is similar to the retrieved case which
is preferable to user’s position [105]. The system has also been extended
8
The correctness of this transformation of abductive logic programs to pure logic
programs has been shown to be independent of the stable model semantics, and has
been extended to handle integrity constraints [111].
414 Marc Denecker and Antonis Kakas
so that legal rules and legal cases are combined together for statutory
interpretation [103].
Consistency Management in Software Engineering: This system
computes a minimal revised logical specification by abductive logic pro-
gramming. A specification is written in Horn clauses which is translated
into an abductive logic program. Given an incompatibility between this
specification and new information the system computes by abduction a
maximally consistent program that avoids this incompatibility [104].
ACLP: Abductive Constraint Logic Programming. The ACLP frame-
work grew as an attempt to address the problem of providing a high-level
declarative programming or modeling environment for problems of Artifi-
cial Intelligence which at the same time has an acceptable computational
performance. Its roots come from the work on abduction and negation as
failure in [29] and the early definitions of Abductive Logic Programming
[52,53,50]. Its key elements are (i) the support of abduction as a central in-
ference of the system, to facilitate declarative problem solving, and (ii) the
use of Constraint Logic Programming techniques to enhance the efficiency
of the computational process of abductive inference as this is applied on the
high-level representation of the problem at hand.
In an ACLP abductive theory the program, P , and the integrity constraints,
IC, are defined over a CLP language with finite domain constraints. Its se-
mantics is given by a form of Generalized Model semantics which extends
(in the obvious way) the definition 1 above when our underlying LP frame-
work is that of CLP. Negation in P is given meaning through abduction
and is computed in a homogeneous way as any other abducible. The gen-
eral computation model of ACLP consists of a cooperative interleaving be-
tween hypotheses and constraint generation, via abductive inference, with
consistency checking of abducible assumptions and constraint satisfaction
of the generated constraints. The integration of abductive reasoning with
constraint solving in ACLP is cooperative, in the sense that the constraint
solver not only solves the final constraint store generated by the abductive
reduction but also affects dynamically this abductive search for a solution.
It enables abductive reductions to be pruned early by setting new suitable
CLP constraints on the abductive solution that is constructed.
The framework of ACLP has also been integrated with Inductive Logic Pro-
gramming to allow a form of machine learning under incomplete information
[62].
The ACLP system [60,49], developed at the University of Cyprus, imple-
ments the ACLP framework of ALP for a restricted sub-language of the
full ACLP framework. Currently, the system is implemented as a meta-
interpreter on top of the CLP language of ECLiPSe using the CLP con-
straint solver of ECLiPSe to handle constraints over finite domains (integer
and atomic elements). The architecture of the system is quite general and
can be implemented in a similar way with other constraint solvers. It can be
obtained, together with information on how to use it, from the following web
address: http://www.cs.ucy.ac.cy/aclp/. Direct comparison experiments [61]
Abduction in Logic Programming 415
of ACLP with the underlying CLP system of ECLiPSe have demonstrated
the potential of ALP to provide a high-level modeling environment which is
modular and flexible under changes of the problem, without compromising
significantly the computational efficiency of the underlying CLP framework.
ACLP has been applied to several different types of problems. Initial ap-
plications have concentrated on the problems of scheduling, time tabling
and planning. Other applications include (i) optical music recognition where
ACLP was used to implement a system that can handle recognition under in-
complete information, (ii) resolving inconsistencies in software requirements
where (a simplified form of) ACLP was used to identify the causes of incon-
sistency and suggest changes that can restore consistency of the specification
and (iii) intelligent information integration where ACLP has been used as a
basic framework in the development of information mediators for the seman-
tic integration of information over web page sources. Although most of these
applications are not of ”industrial scale” (with the notable exception of a
crew-scheduling [57] application for the small sized company of Cyprus Air-
ways - see also below 5.2) they have been helpful in indicating some general
methodological guidelines that can be followed when one is developing ab-
ductive applications (see [57]). The air-crew scheduling application produced
solutions that were judged to be of good quality, comparable to manually
generated solutions by experts of many years on the particular problem,
while at the same time it provided a flexible platform on which the company
could easily experiment with changes in policy and preferences.
Extended and Preference Abduction. In order to broaden the applicabil-
ity of ALP in AI and databases, Inoue and Sakama propose two kinds of
extensions of ALP: Extended abduction [43] and Preference abduction [44].
An abductive program in the framework of extended abduction is a pair
K, A of logic programs possibly including negation as failure and disjunc-
tions. Each instance of element of A is abducible. An explanation of a ground
literal G consists of a pair of sets (I,O) of subsets of A such that (K \ O) ∪ I
is consistent and entails G.Ananti-explanation of G satisfies the same con-
ditions except that (K \ O) ∪ I does not entail G. Thus, abduction in this
framework extends standard abduction by defining not only explanation but
also anti-explanations, by allowing solutions in which rules from the program
are deleted and by allowing general rules to be abduced or deleted.
Several implementation methods have been proposed for computing extended
abduction. [45] proposed a model generation method with term rewriting.
In [99,41], transformation methods are proposed that reduce the problem of
computing extended abduction to a standard abductive problem. Extended
abduction has several potential applications such as abductive theory re-
vision and abduction in non-monotonic theories, view update in deductive
databases, theory update, contradiction removal, system repair problems
with model checking, and inductive logic programming (see [43,99,41]).
A framework for preference abduction is an abductive logic program K, A
augmented with a set Ψ of possible priorities between different literals of
the program. For a given goal G, preferred abduction computes a set of
416 Marc Denecker and Antonis Kakas
abducible atoms I and a subset ψ of Ψ representing some priority relation,
such that K ∪ I is consistent and K ∪ I |=
ψ
G, which means that G is
true in every preferred answer set of the prioritized logic program (K ∪
I,ψ) [98]. Hence, preferred abduction not only abduces atoms but also the
priority relationship. A procedure to compute preference abduction has been
proposed in [44].
Preference abduction can be used in resolution of the multiple extension
problem in non-monotonic reasoning, skeptical abduction, reasoning about
rule preference, and preference view update in legal reasoning [44].
ABDUAL: Abduction in extended LP [1] proposes the ABDUAL frame-
work, an abductive framework based on extended logic programs. An ab-
ductive logic program in this framework is a tuple <P,A,IC >,whereP
is an extended logic program (with both explicit and default negation), IC
a set of constraints and A a set of ground objective literals i.e. atoms or ex-
plicitly negated atoms. The declarative semantics of this formalism is based
on the well-founded semantics for extended programs.
The procedure presented in [1] integrates a tabling mechamism in the ab-
ductive inference procedure. The procedure solves an abductive query in
two stages. First, the program is transformed by grounding it and adding
for each non-abducible ground atom p arulenot(p) ← R where R expresses
that none of the rules for p applies. The resulting program is called the
dual program. In the second step, abductive solutions are computed by an
evaluation method that operates on the dual program.
The ABDUAL system is currently implemented on top of XSB-Prolog [122].
The system is available from http://www.cs. sunysb.edu/˜tswift. Work is
currently being done in order to migrate some of the tabling mechanisms
of ABDUAL, now taken care of the meta-interpreter, into the XSB-engine.
Work is also underway on the XSB system so that the co-unfounded set
removal operation can be implemented at the engine level.
The ABDUAL system has been applied in medical psychiatric diagnosis [31]
as a result of an investigation into the logical representation and automation
of DSM-IV (Diagnostic and Statistical Manual of Mental Disorders). The
current user interface of the Diagnostica system (http://medicinerules.com)
uses abduction in a simple but clinically relevant way to allow for hypothet-
ical diagnosis: when there is not enough information about a patient for a
conclusive diagnosis, the system allows for hypothesizing possible diagnosis
on the basis of the limited information available. This is one of the first
applications of abduction that is been commercialized.
ABDUAL has also been employed to detect specification inconsistencies in
model-based diagnosis system for power grid failure [6]. Here abduction is
used to abduce hypothetical physically possible events that might cause the
diagnosis system to come up with a wrong diagnosis violating the specifica-
tion constraints.
Probabilistic Horn Ab duction and Independence Choice Logic. Proba-
bilistic Horn abduction [84], later extended into the independent choice logic
[86], is a way to combine logical reasoning and belief networks into a simple
Abduction in Logic Programming 417
and coherent framework. Its development has been motivated by the Theo-
rist system [88] but it has been extended into a framework for decision and
game-theoretic agents that includes logic programs, belief networks, Markov
decision processes and the strategic form of a game as special cases. In par-
ticular, it has been shown that it is closely related to Bayesian networks [80],
where all uncertainty is represented as probabilities.
An independent choice logic theory is made up of two parts:
– a choice space consisting of disjoint sets of ground atoms. The elements
of a choice space are called alternatives.
– an acyclic logic program such that no element of an alternative unifies
with the head of a clause.
The semantics is model-theoretic. There is a possible world for each choice of
one element from each alternative. What is true in a possible world is given
by the stable model of the atoms chosen and the logic program. Intuitively
the logic program gives the consequences of the choices. This framework is
abductive in the sense that the explanations of an observation g provide a
concise description of the worlds in which g is true. Belief networks can be
defined by having independent probability distributions over the alternatives.
Intuitively, we can think of nature making the choice of a value for each
alternative. In this case Bayesian conditioning corresponds exactly to the
reasoning of the above framework of independent choice logic. This can also
be extended to decision theory where an agent can make some choices and
nature others [86], and to the game-theoretic case where there are multiple
agents who can make choices.
Different implementations of the ICL and its various special cases exist.
These include Prolog-style implementations that find explanations top-down
[83,89], bottom-up implementations (for the ground case) that use a proba-
bilistic variant of the conflicts used in model-based diagnosis [85], and algo-
rithms based on efficient implementations of belief networks that also exploit
the context-specific independent inherent in the rule forms [87]. Initial stud-
ies of application of ICL have centered around problems of diagnosis and
robot control.
5 Example Applications of Abduction
ALP as a paradigm of declarative problem solving allows us to formalize a wide
variety of problems. A survey of the field reveals the potential application of
abduction in areas such as databases updates, belief revision, planning, diagnosis,
natural language processing, default reasoning, user modeling, legal reasoning,
multi-agent systems, scheduling, and software engineering. In this section, two
relatively large-scale applications of ALP are presented in some detail in order
to illustrate the main features of declarativeness and modularity of an abductive
based approach that have been exposed in the previous sections.
418 Marc Denecker and Antonis Kakas
5.1 Scheduling of Main tenance
This experiment is based on a real life problem of a Belgian electricity provider.
The problem description is as follows. The company has a network of power
plants, distributed over different areas and each containing several power pro-
ducing units. These units need a fixed number of maintenances during the year.
The problem is then to schedule these maintenances so that a number of con-
straints are satisfied and the risk of power shortage (and hence, import from other
providers) is as low as possible. The problem was solved using the SLDNFAC
system extended with a restricted yet sufficient form of higher-order aggregates.
The system accepts first order constraints which are first compiled to rules using
the Lloyd-Topor transformation. Below we given an overview of the problem
solution. For a more complete description of the solution and the abductive
procedure for reasoning on aggregates, we refer the reader to [117].
The fact that a maintenance M lasts from week B till week E,isrepre-
sented by the predicate start(M, B,E). This is the only abducible predicate in
the specification. Other predicates are either defined or are input data and are
defined by a table. Some of the main constraints that need to be satisfied are
given below
9
.
– Maintenances (maint(M)) and their duration (duration(M,D)) are given
by a table. All maintenances must be scheduled, thus for each maintenance
there exists an according start relation. This is specified via a first order
logical formula i.e. an integrity constraint as follows:
∀M : maint(M) →∃B,E,D : week(B) ∧ week(E) ∧ duration(M, D)∧
E = B + D − 1 ∧ start(M, B,E).
– A table of prohibited(U, Bp, Ep) facts specify that maintenances M for unit
U are not allowed during the period [Bp,Ep]:
∀U,Bp, Ep,M,B, E :
prohibited(U, Bp, Ep) ∧ maint
for unit(M,U) ∧ start(M, B,E)
→ (E<Bp∨ Ep < B).
– For each week the number of the units in maintenance belonging to a plant P
should be less than a maximal number Max.Atableofplant
max(P, Max)
atoms defines for each plant the maximal number of units in maintenance
simultaneously.
∀P, Max, W e : plant(P ) ∧ plant
max(P, Max) ∧ week(We)
→∃OnMaint : card({U | (unit(U) ∧ unit
in plant(U, P )∧
in
maint(U, W e))}, OnMaint)∧
OnMaint ≤ Max.
Note that this constraint uses the cardinality aggregate card.Themeaningof
the above cardinality atom is that the set of units of plant P in maintenance
9
In this and the following section, variable names start with a capital, as standard in
logic programming.
Abduction in Logic Programming 419
in week We contains OnMaint elements. The predicate in maint is defined
by an auxiliary program rule specifying that a unit U is in maintenance
during a certain week W if a maintenance M of this unit is going on during
W :
in
maint(U, W ) ← maint for unit(M,U),start(M,B,E),B ≤ W, W ≤ E.
– Another constraint is that the capacity of the units in maintenance belonging
to a certain area should not exceed a given area maximum. To represent this,
the summation aggregate is needed. A table of capacity(U, C) describes for
each unit its maximum capacity.
∀A, M ax, W e, CM : area(A) ∧ area
max(A, Max) ∧ week(We)∧
sum({(U, C)| (unit(U) ∧ in
area(U, A) ∧ in maint(U, W e)∧
capacity(U, C))},λ(U, Cap)Cap,CM)
→ 0 ≤ CM ∧ CM ≤ Max.
In the above constraint, the meaning of the sum aggregate atom is that ”the
sum of the lambda function over the set expression is CM”. It defines CM
as the total capacity of area A in maintenance during week We.
The above specification describes some of the necessary properties of a correct
schedule. However, not all schedules satisfying these properties are desirable. In
particular, schedules that minimise the risk of power shortage are preferable.
To this end, the company maintains statistical data about the expected peak
load per week. Desirable solutions are those that maximise the reserve capacity,
that is the difference between the available capacity and the expected peak load.
This relation (reserve(W eek, R)) can then be defined as the difference between
available capacity (the sum of capacities of all units not in maintenance during
this week) and the estimated peak load:
reserve(We,R) ← peakload(We,Load),total
capacity(T ),
sum( {(U, Cap)| (unit(U) ∧ in
maint(U, W e)∧
capacity(U, Cap))},
λ(U, Cap)Cap,InMaint),
R = T − Load − InMaint.
in which total
capacity(T ) means the sum of all capacities of all units.
In the SLDNFAC system, the query for the optimal solution for the schedul-
ing problem is
? minimum(set([R],(exists(W) : reserve(W,R)),M), maximize(M).
It expresses that an optimal abductive solution is desired in which the minimal
reserve for one year is as high as possible.
The actual problem, given by the company, consists of scheduling 56 mainte-
nances for 46 units in one year (52 weeks). The size of the search space is of the
order of 52
46
. The current implementation reduces the goal and the integrity con-
straints to a large finite domain constraint store without backtracking points. In
420 Marc Denecker and Antonis Kakas
the current implementation of SLDNFAC, this reduction phase is completed in
less than one minute . Subsequently the CLP solver starts to generate solutions
of increasing quality. The current implementation was able to find a solution
which is 97% away from the optimal one in 20 minutes.
The same problem was also solved using a CLP system. A comparison be-
tween the CLP solution and the ALP solution clearly shows the trade-off be-
tween efficiency and flexibility. The pure (optimized) CLP solution will setup
its constraint store in several seconds (3 to 4 seconds), and find the same so-
lution as the above specification within 2 minutes (compared to 20 minutes for
the SLDNFAC-solver). On the other hand, the CLP solution is a much larger
program (400 lines) developed in some weeks of time in which the constraints
are hidden within data structures, whereas the above representation in ALP is a
simple declarative representation of 11 logical formulae, written down after some
hours of discussion.
5.2 Air-Crew Assignment
The second application of abduction that we present is also based on a real-life
problem, namely that of crew-assignment for Cyprus Airways. The problem of
air crew-assignment is concerned with the assignment of air-crews to each of the
flights that an airline company has to cover over some specific period of time.
This allocation of crew to flights has to respect all the necessary constraints (va-
lidity) and also try to minimize the crew operating cost (quality). The validity
of a solution is defined by a large number of complex constraints, which express
governmental and international regulations, union rules, company restrictions
etc. The quality of the schedule is specified, not only by its cost, but also by the
needs and preferences of the particular company or crew at that specific period
of time. In addition, an airline is also interested in the problem of re-assignment
or of adapting an existing crew assignment to changes in the application en-
vironment such as flight delays or cancellations, new flight additions or crew
unavailability etc. These changes often affect the quality of an existing solution
or even make an existing solution unacceptable.
This problem for (the pilot crew) of Cyprus Airways was solved within ALP
using the ACLP system. The problem was represented entirely as an ALP theory
T =(P, A, IC). The program part P describes basic data and defines a number
of concepts that allow for encoding particular strategies for decomposing the
overall goal to subgoals. Different strategies affect efficiency of the problem solv-
ing process and the quality of the solutions with respect to the criteria of cost
or fairness of assignment. The solution of the problem is captured via an ab-
ducible predicate assigns(Crew,Task) (the only member of A) which gives the
assignment of crew members to different types of duty tasks (eg. flights, stand-
bys, day-offs, etc.). For details of this and for a more complete description of
the problem and its abductive-based solution see [58]. Here we will concentrate
more on how the complex validity constraints of the problems are represented
in the IC part of the theory.
Abduction in Logic Programming 421
The problem of air crew-assignment has a large variety of complex constraints
that need to be respected. These contain simple constraints such as that a pi-
lot can not be assigned to two overlapping flights but also many other quite
complex constraints such as that during any period of 6 days (respectively 14
days) a pilot must have one day off (respectively 2 consecutive days off). Lets us
illustrate how some of these would be represented as integrity constraints in IC.
The following integrity constraint expresses the requirement that for any pilot
theremustbeatleastMinRest hours rest period between any two consecutive
duties. MinRest is greater than or equal to 12 and it is calculated according to
the previous assignments of the crew. (All variables in the integrity constraints
below are universally quantified over the whole formula).
¬assign(Crew,Flight) ←
on
new duty(Crew,Flight),
end
prev duty(Crew,Flight,EndOfDuty),
time
difference(EndOfDuty, F light, RestP eriod),
MinRest(Crew,MR), RestP eriod < M R.
Here on
new duty(Crew,Flight) defines whether the flight, Flight, is the begin-
ning of a new duty period for Crew and end
prev duty(Crew,Flight,EndOfDu-
ty) specifies the time of the end of the duty, EndOfDuty, for the crew member,
Crew, which is immediately before the departure time of the flight Flight.These
are defined in the program P of the theory.
The requirement that each pilot must have at least 2 consecutive days off
during any 14 day period is represented by the integrity constraint:
consec2
daysof f(Crew, DeptDate, 14) ←
assign(Crew,Flight),
dept
date(Flight,DeptDate)
where consec2
daysof f(Crew, DeptDate, 14) means that the Crew has two con-
secutive days off within a time window of 14 days centered around the date
DeptDate. This is given in the program P with the help of the definition of
dayoff as follows:
consec2
daysof f(Crew, Date, N) ←
consec
days(Date,N,DayA,DayB),
dayoff (Crew,DayA),
dayoff (Crew,DayB)
dayoff (Crew,Date) ←
not assign(Crew,flight(Id,Date)),
crew
at base(Date),
further
free hrs(Crew,Date)
422 Marc Denecker and Antonis Kakas
further free hrs(Crew,Date) ←
next
date(Date, NDate),
assign(Crew,flight(Id,NDate)),
departure(flight(Id,NDate),NDate,DeptTime),DeptTime>8
further
free hrs(Crew,Date) ←
next
date(Date, NDate),
assign(Crew,flight(Id,NDate)),
departure(flight(Id,NDate),NDate,DeptTime),
DeptT ime > 6,previous
date(Date, P Date),
assign(Crew,flight(Id,PDate)),
arrival(flight(Id,PDate),PDate,ArrTime), ArrTime < 22
.
This expresses the definition of a day-off as a non-working day (0:00 - 24:00), at
base, with one of the following additional requirements. Either the crew begins
his/her duty after 8am the next morning, or s/he begins work after 6am but
finishes before 10pm (22:00) the day before.
During the computation, the satisfaction of this integrity constraint means
that whenever a new assumption of assignment of a Crew to a Flight is made we
need to ensure that consec2
daysof f for this Crew member remains satisfied. In
some cases this would then dynamically generate extra assignments, of the Crew
member to day-offs, to ensure that his/her flight assignments are consistent.
Airlines also have their own requirements on the problem stemming from
particular policies of the specific company and crew preferences. The abductive
formulation, with its modular representation of the problem, facilitates in many
cases a direct representation of these with additional integrity constraints in
IC. As an example consider a requirement of Cyprus Airways which states that
flight managers should not have more than two duties per week. This can be
represented by the following integrity constraint:
¬assign(Crew,Flight) ←
rank(Crew,flight
manager),
on
new duty(Crew,Flight),
num
of duties(Crew, F light, week period, N Duties),
NDuties > 2.
Here num
of duties(Crew, F light, week period, N Duties) counts the number
of duties NDuties thatacrewmemberhaswithinaweek
period centered
around the date of the flight Flight.
With regards to the problem of re-assignment under some new information,
given an existing solution, a new module is added to the crew-assignment sys-
tem which exploits the natural ability of abduction to reason with a given set
of hypotheses, in this case the (partial) existing solution. This module follows
three steps: (1) remove from the old solution all hypotheses which are affected by
these changes. This step is in fact optional, helping only in the efficiency, since
hypotheses which make the existing solution inconsistent will be eventually re-
moved automatically by the re-execution of the abductive goal in step 3 below,
Abduction in Logic Programming 423
(2) add the new requirements (changes) of the problem. These may be in the
form of integrity constraints or simply as new information in the domain of the
application and (3) re-execute the (or part of the) abductive goal of the prob-
lem with the set of the hypotheses in step (1) as a given initial set of abducible
assumptions.
Given the set of flights which are affected by the change(s), the aim is to re-
establish the consistency, and preferably also the quality, of the old solution by
re-assigning crew to these flights, without having to recalculate a new solution
from the beginning but rather by making the fewest possible changes on the old
existing solution, within 48 hours from the time of the change.
The re-assignment module in this application is interactive in the sense that
the user can select a crew for a particular flight or decide whether to accept
a system proposed selection of crew. Having searched for a crew member, the
system informs the user about the particular selection, together with a list of
other assignments (secondary changes) on this crew in the old solution, that are
affected and would also need to be rescheduled. It then gives him/her the option
to reject this choice, in which case the system will look for another possibility.
When the selection of a crew is done directly by the user, the system will check
if this choice is valid and inform the user of the list (if any) of secondary affected
flights, that would also need to be rescheduled, resulting from this choice.
Although Cyprus Airways is a small size airline it contains the full complexity
of the problem. During the busy months the flight schedule contains over 500
flight legs per month. The ACLP system was able to produce solutions in a
few minutes which were judged by the airline’s experts on this problem to be
of good quality comparable (and with respect to balancing requirement often
better) to the manually generated ones. The system was also judged to be useful
due to the flexibility that it allowed to experiment easily with changes in policy
and preferences of the company. The re-assignment module was able to suggest
solutions on how to adapt the existing roster within at most 5 seconds. It was
chosen as the most useful module of the system as it could facilitate the operators
to develop and adjust a solution to meet the specific needs and preferences that
they have at the time.
6 Links of ALP to Other Extensions of LP
In parallel with the development of the above frameworks and systems for ALP
it has become clear that there exist strong links between some ALP frameworks
and other extensions of Logic Programming.
ALP has tight connections to Answer Set Programming [32]. Recall that
the ABDUAL framework [1] is an extension of Answer Set Programming with
abduction. Standard ALP (with one negation) is strongly related Stable Logic
Programming [69,75], the restriction of Answer Set Programming [32] to pure
logic programs. As mentioned in section 4, an abductive logic framework un-
der the generalized stable semantics can be translated in an equivalent logic
program under stable semantics. Consequently, current systems for computing
424 Marc Denecker and Antonis Kakas
stable models such as SMODELS [75] can be used to compute abduction under
the generalized stable semantics. Interestingly, there are significant differences
between in computational models that are developed in both areas. Whereas
ALP procedures such as SLDNFA, IFF and ACLP are extensions of SLDNF
and operate in a top down way on predicate programs, systems like SMODELS
are based on bottom up propagation in the propositional grounding of a logic
program. More experimentation is needed to assess the strengths and weaknesses
of these approaches.
Links have been shown also between ALP and Disjunctive Logic Program-
ming [100,101,124]. The hypothetical reasoning of ALP and the reasoning with
disjunctive information of DLP can be interchanged. This allows theories in one
framework to be transformed to the other framework and thus to be executed
in this other framework. For example, it is possible to transform an ALP theory
into a DLP one and then use a system such as the recently developed dlv system
[27] to answer abductive queries. Vice versa, [124] showed that abductive proof
procedures can be used as for reasoning on DLP programs.
Another type of extension of the LP paradigm is Inductive Logic Program-
ming. Currently, several approaches are under investigation synthesizing ALP
and ILP [2,73,123,35]. These approaches aim to develop techniques for knowl-
edge intensive learning with complex background theories. One problem to be
faced by ILP techniques is that the training data on which the inductive process
operates often contain gaps and inconsistencies. The general idea is that ab-
ductive reasoning can feed information into the inductive process by using the
background theory for inserting new hypotheses and removing inconsistent data.
Stated differently, abductive inference is used to complete the training data with
hypotheses about missing or inconsistent data that explain the example or train-
ing data using the background theory. This process gives alternative possibilities
for assimilating and generalizing this data. In another integration of ALP and
ILP, ILP is extended to learn ALP theories from incomplete background data
[62]. This allows the framework to perform Multiple Predicate Learning in a
natural way.
As we have seen in previous sections several approaches to ALP have recog-
nized the importance of linking this together with Constraint Logic Program-
ming. They have shown that the integration of constraint solving in abductive
logic programming enhances the practical utility of ALP. Experiments indicate
that the use of constraint solving techniques in abductive reasoning make the
abductive computation much more efficient. On the other hand, the integrated
paradigm of ALP and CLP can be seen as a high-level constraint programming
environment that allows more modular and flexible representations of the prob-
lem domain. The potential benefits of this paradigm are largely unexplored at
the moment.
Abduction in Logic Programming 425
7 Challenges and Prospects for ALP
In the past decade, many studies have shown that extending Logic Programming
with abduction has many important applications in the context of AI and declar-
ative problem solving. Yet, at this moment the field of ALP faces a number of
challenges at the logical, methodological and computational level. In this section
we attempt to chart out some of these challenges and point to some promising
directions.
7.1 Heterogeneity of ALP
As can be seen in section 4, ALP is a very heterogeneous field. On the one hand,
this heterogeneity stems from the fact that logic programming itself shows a
complex landscape. On the other hand, it stems from the fact the term abduction
is defined very broadly and covers a broad class of rather loosely connected
reasoning phenomena.
At the conceptual level, abduction is sometimes used to denote concepts at
different conceptual levels. For example, in many of the frameworks discussed
earlier, abduction is a concept at the inferential level: it is a form of logical
inference. In other contexts such as in the abductive semantics for negation as
failure [29], abduction is a concept used at the semantical level, as a specific way
of formalizing model semantics. This mismatch between different conceptual
levels is confusing and a potential hazard for the field.
At the logical level, there are many different formalisms and different seman-
tics. Various forms of abduction have been introduced in different formalisms
including pure logic programming, answer set programming and recently a con-
ditional logic programming formalism [30]. The advantage of this is that the
field may act as a forum for integrating and relating a wide variety of different
forms of logical reasoning in otherwise distant areas. A disadvantage is that this
heterogeneity may hide a lack of coherence in which efforts of researchers to
build effective systems are scattered in a wide variety of incompatible views and
approaches. To develop a computational logic, a focused effort at different levels
is needed: research on semantics to clarify the declarative meaning, research on
knowledge representation to clarify the applications of the logic, research to ex-
plore the relation with other logics, and research to investigate how to implement
efficient problem solvers. These efforts should link together in a constructive and
cross supporting way.
7.2 Epistemological Foundations of ALP
One of the underlying problems of the field is the lack of understanding of the
epistemological foundations of ALP. Epistemological questions are what kind of
knowledge can be represented by an abductive logic framework and vice versa,
what does an ALP theory tell us about the problem domain or equivalently,
what information about the domain of discourse is expressed by a given ALP
theory? Such questions are fundamental to the understanding of any logic. A
426 Marc Denecker and Antonis Kakas
clear answer is a prerequisite for developing a well-motivated methodology for
declarative problem solving using ALP.
The standard definition of ALP as presented in section 3 does not attempt
to answer the above questions. The definition 1 of an abductive solution defines
aformalcorrectness criterion for abductive reasoning, but does not address the
question of how the ALP formalism should be interpreted. Also the (generic)
definition 2 of the formal model semantics of ALP does not provide answers.
In fact, here ALP inherits the ambiguity of logic programming at the episte-
mological level, as demonstrated recently in [15]. Here are some fundamental
questions:
– To understand the meaning of an ALP framework, at the very least we need
to understand the meaning of its symbols. How is negation in ALP to be un-
derstood? The extended completion semantics defined for ALP by Console,
Thorasso and Theseider Dupr´e [10] maps negation as failure literals to classi-
cal negation. On the other hand, in the generalized stable semantics [52] and
in the ABDUAL framework [1], negation as failure literals are interpreted as
modal literals ¬Kp in autoepistemic logic or default logic [38].
– What is the relationship between ALP and classical logic? An ALP frame-
work may contain an arbitrary classical logic theory IC of constraints; in
ALP’s model semantics, models of an ALP framework satisfy the constraints
in IC in the standard way of classical logic. This suggests that ALP is an
extension of classical logic. On the other hand, ALP is defined as a study of
abductive reasoning while classical logic is normally viewed as the study of
deductive reasoning. How are these two views reconciled?
The lack of clear epistemological foundations for ALP is one of the causes of
ALP’s lack of coherence and is a factor blurring the role and status of ALP at
the knowledge representation level in the broader context of logic-based AI. An
epistemological study of ALP can contribute significantly to the understanding
of the field at the logical and methodological level.
7.3 Computational Challenges
The computational challenges of the paradigm are considerable. The challenge
of building abductive systems for solving a broad class of problems formalized
by high-level declarative representations, is extremely difficult to realise.
At the theoretical level of complexity, formal results show that in general the
problem of computing abduction is hard [26]. In the datalog case, the problem
of computing abductive solutions is in general intractable. In the general case of
ALP frameworks with function symbols, the existence of an abductive solution
is undecidable. On the implementational level, the problem of implementing
abductive reasoning can be seen as an extension of the implementation of CLP
systems in which we need to reason about constraints of general first order logic.
Current systems such as ACLP, SLDNFAC and IFF are based on the inte-
gration of CLP techniques in high level abductive procedures. These systems
Abduction in Logic Programming 427
operate by reducing the high level constraints in a, in general, nondeterminis-
tic process to a constraint store that can be handled efficiently by specialised
constraint systems. Recent experiments with the ACLP and SLDNFAC systems
have shown that in those cases where the reduction process is deterministic,
these procedures can be very performant. However, when the process is nonde-
terministic, these procedures can start to trash. The reason for this behaviour
is that a number of techniques are built in in the current procedures that de-
lay the creation of choice points and perform deterministic computation first. In
many applications such as scheduling, these techniques can avoid making choices
altogether. In other cases, such as in planning applications, the arsenal of tech-
niques does not suffice to manage these choice points and the current procedures
often make uninformed selections of choices leading to uncontrolled depth first
execution and trashing.
The above analysis suggests different ways to improve the computational
techniques of ALP. One way is to further improve the techniques to discover
deterministic subgoals and delay creation of choice points. A second way is to
incorporate techniques for smarter and better informed selection of the choice
points and choice of alternatives in the choice point. A third way is an improved
control to avoid unrestricted depth first reasoning using techniques similar to
loop detection and iterative deepening can be used. With respect to the first
two problems, different approaches can be followed. One is to further refine the
current integration of Constraint Solving in the abductive inference. In the cur-
rent systems, the CLP solver is a black box that interacts with the abductive
solver by returning a solution at the end, or by reporting consistency or incon-
sistency of the constraint store at different points during the execution. One
direction to be examined is how to exploit the information present in the con-
straint store to steer the search for an abductive solution and make a better
informed selection of goals. An alternative direction is to apply techniques from
heuristic search in Artificial Intelligence.
An interesting application domain to study the above techniques for ab-
ductive reasoning is AI-planning, due to the strong links between abduction
and planning and the fact that recently, techniques from constraint solving and
heuristic search have been successfully applied in this domain. What we can learn
here is how recent developments of constraint and heuristic methods of search in
planning could be applied to the more general case of abductive computation.
A complementary approach to address the computational hardness of ALP
would be to develop ALP systems in which the user has the facility to incre-
mentally refine her/his model of the problem in a modular way. Starting from a
purely declarative problem description, it should be possible to refine the model
by adding more and more additional knowledge about the problem, including
non-declarative heuristic and operational control knowledge. Again recent work
suggests that this is a promising line of development but there is no systematic
study of how such a modeling environment would be designed and build in ALP.
A completely different approach is to exploit the kind of techniques used in
bottom up abduction [46] (see section 4) based on the computation of stable
428 Marc Denecker and Antonis Kakas
models of a ground logic program. Techniques like those used by the smodels
system [75] which integrates methods from propositional constraint propagation
with bottom up application of semantic fixpoint operators of the 3-valued com-
pletion semantics and well-founded semantics. In the current state of the art,
it seems that while the latter techniques based on reasoning on propositional
theories are more robust, the abductive extensions of SLDNF with CLP may
outperform the first ones especially as they can take into account more easily
additional problem domain specific information. Therefore, extending the latter
procedures along the lines suggested above is a promising research direction.
7.4 Challenges at the Application Level
In the past decade, the potential of the different ALP frameworks have been
demonstrated in a wide variety of application domains. However, only a few of
the current running applications exceed the level of academic toy examples.
Like in many other areas of AI, this potential has not yet been realized in
realistic and industrial scale applications. One of the challenges of the domain
is to find interesting niche domains with industrial impact in which the current
systems can be evaluated and fine-tuned. Experimentation with and evaluation of
abductive systems in realistic domains could yield important information at the
levels of language constructs, methodology, computational control, integration
of heuristic information, etc..
Some prototypical classes of problems that seem good candidates for fine-
tuning ALP methods are Scheduling and Planning domains and Knowledge
Intensive Learning where machine learning with a rich background knowledge
can be performed only if the inductive methods are integrated with abduction
[123,73,35].
7.5 A Possible Approach to These Challenges
In this section, we briefly describe our own views on how to approach the above
logical and computational challenges.
The underlying logic for ALP is ID-logic [16,23] a logic which is appropriate
for ALP in the way that it extends classical logic with inductive definitions of a
generalized non-monotone kind. As mentioned earlier in section 3, an abductive
logic framework (P, A, IC) has a natural embedding in ID-logic. P represents
a definition of the non-abducible predicates while IC represents a set of classi-
cal logic assertions. In this view, ALP is the study of abduction in the context
of ID-logic. ID-logic was defined in an attempt to cope with the epistemolog-
ical challenges of logic programming and gives answers to the epistemological
questions raised in section 7.2.
At the computational level, we are currently developing a system called the
A-system [63,118] integrating features of ACLP and SLDNFAC with special at-
tention to the search in the abductive computation. During the computation,
the selection and evaluation of choice points is guided by information obtained
Abduction in Logic Programming 429
from a constraint store associated to the abductive solution. With this informa-
tion the high level search can avoid deadend branches before entering them. The
result is a more robust and modular system which is capable to solve effectively
a wider range of problems than the older systems. The application domain of the
experiments with the A-system are currently focused on scheduling and planning
applications. The A-system is built on top of Sicstus Prolog ( version 3.8.5 or
above) and is available at http://www.cs.kuleuven.ac.be/∼dtai/kt/.
8Conclusion
Abductive logic programming grew out of attempts to use logic programming
techniques for a broad class of problems from AI and other areas of Computer
Science. At present Abductive Logic Programming presents itself as a ”conser-
vative extension” of Logic Programming that allows more declarative represen-
tations of problems. The main emphasis till now has been on setting up different
frameworks for abduction and showing how they provide a general approach to
declarative problem solving.
ALP faces a number of challenges, at the logical, methodological and com-
putational level typical for a field in an initial stage of development. We are now
beginning to understand the contributions of this field and to develop solutions
for the problems that the field faces.
At the logical level, ALP aims to be suitable for declarative knowledge rep-
resentation, thus facilitating maintenance, reusability and graceful modifiability.
Yet, ALP retains from logic programming the possibility of embedding high
level strategic information in an abductive program which allows us to speed up
and fine tune the computation. In this respect, ALP is able to combine the ad-
vantages of declarative specification and programming to a greater extent than
standard logic programming.
The field has also started to recognize the full extent of the problem and the
complexity of developing effective and useable ALP systems. The overall task of
ALP of providing a high-level general purpose modeling environment which at
the same time is computationally effective is an extremely difficult one. But we
are beginning to learn how to analyze and break this task down to appropriate
subproblems that are amenable to study within our current understanding of
the field. The hope remains that within the high-level programming environ-
ment that ALP could provide, the programmer will be able to solve problems
effectively in a translucent way.
Acknowledgements This work was partly supported by the European Union
KIT project CLSFA (9621109).
430 Marc Denecker and Antonis Kakas
References
1. J. J. Alferes, L. M. Pereira, T. Swift. Well-founded Abduction via Tabled Dual
Programs. In Procs. of the 16th International Conference on Logic Programming,
Las Cruces, New Mexico, Nov. 29 - Dec. 4, 1999.
2. H. Ade and M. Denecker. Abductive inductive logic programming. In C.S. Mellish,
editor, Proc. of the International Joint Conference on Artificial Intelligence, pages
1201–1209. Morgan Kaufman, 1995.
3. J. J. Alferes, J. A. Leite, L. M. Pereira, P.Quaresma. Planning as Abductive
Updating. In D. Kitchin (ed.), Procs. of AISB’00, 2000.
4. Balsa, J., Dahl, V. and Pereira Lopes, J.G. Datalog Grammars for Abductive
Syntactic Error Diagnosis and Repair. In Proc. Natural Language Understanding
and Logic Programming Workshop, Lisbon, 1995.
5. M. Bruynooghe, H. Vandecasteele, D.A. de Waal, and Denecker M. Detecting
unsolvable queries for definite logic programs. The Journal of Functional and
Logic Programming, 1999:1–35, November 1999.
6. J.F.Castro, L. M. Pereira, Z.Vale. Power Grid Failure Diagnosis Certification.
Technical Report, University of Lisbon.
7. A. Ciampolini, E. Lamma, P. Mello and P. Torroni. Expressing Collaboration and
Competition Among Abductive Logic Agents. In AI*IA Notizie - Anno XIII(3),
Settembre 2000, pag. 19–24.
8. E. Charniak and D. McDermott. Introduction to Artifical Intelligence. Addison-
Wesley, 1985.
9. K.L. Clark. Negation as failure. In H. Gallaire and J. Minker, editors, Logic and
Databases, pages 293–322. Plenum Press, 1978.
10. L. Console, D. Theseider Dupr´e, and P. Torasso. On the relationship between
abduction and deduction. Journal of Logic and Computation, 1(5):661–690, 1991.
11. L. Console, L. Portinale and Theseider Dupr´e, D., Using Compiled knowledge to
guide and focus abductive diagnosis. IEEE Transactions on Knowledge and Data
Engineering, Vol. 8 (5), pp. 690-706, 1996.
12. L. Console, M.L. Sapino and Theseider Dupr´e,D. TheRoleofAbductionin
Database View Updating. Journal of Intelligent Information Systems, Vol. 4(3),
pp. 261-280, 1995.
13. D. de Waal, M. Denecker, M. Bruynooghe, and M. Thielscher. The generation
of pre-interpretations for detecting unsolvable planning problems. In Proceedings
of the Workshop on Model-Based Automated Reasoning (15th International Joint
Conference on Artificial Intelligence), pages 103–112, 1997.
14. M. Denecker. A Terminological Interpretation of (Abductive) Logic Programming.
In V.W. Marek, A. Nerode, and M. Truszczynski, editors, International Conference
on Logic Programming and Nonmonotonic Reasoning, Lecture notes in Artificial
Intelligence 928, pages 15–29. Springer, 1995.
15. M. Denecker. On the Epistemological foundations of Logic Programming and its
Extensions. In AAAI Spring Symposium on Answer Set Programming: Towards
Efficient and Scalable Knowledge Representation and Reasoning, volume techni-
cal report SS-01-01. American Association for Artificial Intelligence, AAAI Press,
2001.
16. M. Denecker. Extending classical logic with inductive definitions. In J.Lloyd et al.,
editor, First International Conference on Computational Logic (CL2000),volume
1861 of Lecture notes in Artificial Intelligence, pages 703–717, London, July 2000.
Springer.
Abduction in Logic Programming 431
17. M. Denecker and D. De Schreye. SLDNFA; an abductive procedure for normal ab-
ductive programs. In K.R. Apt, editor, Proc. of the International Joint Conference
and Symposium on Logic Programming, pages 686–700. MIT Press, 1992.
18. M. Denecker and D. De Schreye. Representing incomplete knowledge in abductive
logic programming. In Proc. of the International Symposium on Logic Program-
ming, pages 147–163. MIT Press, 1993.
19. M. Denecker and D. De Schreye. Representing Incomplete Knowledge in Abductive
Logic Programming. Journal of Logic and Computation, 5(5):553–578, September
1995.
20. M. Denecker and D. De Schreye. SLDNFA: an abductive procedure for abductive
logic programs. Journal of Logic Programming, 34(2):111–167, 1998.
21. M. Denecker, L. Missiaen, and M. Bruynooghe. Temporal reasoning with abductive
event calculus. In Proc. of the European Conference on Artificial Intelligence.John
Wiley and sons, 1992.
22. M. Denecker, K. Van Belleghem, G. Duchatelet, F. Piessens, and D. De Schreye.
Using Event Calculus for Protocol Specification. An Experiment. In M. Maher,
editor, The International Joint Conference and Symposium on Logic Programming,
pages 170–184. MIT Press, 1996.
23. Marc Denecker, Maurice Bruynooghe, and Victor W Marek. Logic programming
revisited : logic programs as inductive definitions. ACM Transactions on Compu-
tational Logic, 2001. accepted.
24. Theseider Dupr´e, D.. Characterizing and Mechanizing Abductive Reasoning. PhD
Thesis, Dip. Informatica, Universit`a di Torino, 1994.
25. P.M. Dung. Negations as hypotheses: an abductive foundation for Logic Program-
ming. In Proc. of the International Conference on Logic Programming, 1991.
26. Thomas Eiter, Georg Gottlob, Nicola Leone. Abduction from Logic Programs:
Semantics and Complexity. Theoretical Computer Science 189(1-2):129-177 (1997).
27. Thomas Eiter, Wolfgang Faber, Nicola Leone, and Gerald Pfeifer. Declarative
problem-solving using the dlv system. In Jack Minker, editor, Logic-Based Artificial
Intelligence. Kluwer Academic Publishers, 2000.
28. K. Eshghi. Abductive planning with Event Calculus. In R.A. Kowalski and K.A.
Bowen, editors, Proc. of the International Conference on Logic Programming.The
MIT press, 1988.
29. K. Eshghi and R.A. Kowalski. Abduction compared with negation as failure. In
Proc. of the International Conference on Logic Programming. MIT-press, 1989.
30. D. Gabbay, L. Giordano, A. Martelli, and M.L. Sapino. Conditional reasoning in
logic programming. Journal of Logic Programming, 44(1-3):37–74, 2000.
31. J. Gartner, T. Swift, A. Tien, C. V. Dam´asio, L. M. Pereira. Psychiatric Diagnosis
from the Viewpoint of Computational Logic. In G. Wiggins (ed.), Procs. of AISB,
2000.
32. Gelfond, M., Lifschitz, V. Classical negation in logic programs and disjunctive
databases. New Generation Computing, pp. 365-387, 1991.
33. Fages, F. A New Fixpoint Semantics for General Logic Programs Compared with
the Well-Founded and the Stable Model Semantics. Proc. of ICLP’90, pp. 442 –
458, 1990.
34. P. Flach and A. C. Kakas (Eds.). Abduction and Induction: Essays on their Rela-
tion and Integration. Kluwer Academic Press, 2000.
35. P. Flach and A. C. Kakas. Abductive and Inductive Reasoning: Background and
Issues. In Peter Flach and Antonis Kakas, editors, Abduction and Induction: essays
on their relation and integration. Kluwer, 2000.
432 Marc Denecker and Antonis Kakas
36. Fung, T.H. Abduction by deduction. Ph.D. Thesis, Imperial College, London,
1996.
37. T.H. Fung, R.A. Kowalski. The iff procedure for abductive logic programming. In
Journal of Logic Programming 33(2):151–165, Elsevier, 1997.
38. M. Gelfond and V. Lifschitz. The stable model semantics for logic programming. In
Proc. of the International Joint Conference and Symposium on Logic Programming,
pages 1070–1080. IEEE, 1988.
39. Hobbs, J.R. An integrated abductive framework for discourse interpretation. Sym-
posium on Automated Abduction,Stanford, 1990.
40. K. Inoue. Hypothetical reasoning in Logic Programs. Journal of Logic Program-
ming, 18(3):191–228, 1994.
41. K. Inoue. A simple characterization of extended abduction. In: Proceedings of the
First International Conference on Computational Logic, Lecture Notes in Artificial
Intelligence, 1861, pages 718–732, Springer, 2000.
42. K. Inoue, Y. Ohta, and R. Hasegawa. Bottom-up Abduction by Model Generation.
Technical Report TR-816, Institute for New Generation Computer Technology,
Japan, 1993.
43. K.Inoue and C. Sakama. Abductive framework for nonmonotonic theory change.
In: Proceedings of IJCAI-95, pages 204–210, Morgan Kaufmann, 1995.
44. K. Inoue and C. Sakama. Abducing priorities to derive intended conclusions. In:
Proceedings of IJCAI-99, pages 44–49, Morgan Kaufmann, 1999.
45. K. Inoue and C. Sakama. Computing extended abduction through transaction pro-
grams. Annals of Mathematics and Artificial Intelligence, 25(3,4):339–367, 1999.
46. Iwayama, N. and Satoh, K. Computing Abduction by Using TMS with Top-Down
Expectation. Journal of Logic Programming, Vol. 44 No.1-3, pp. 179 – 206, 2000.
47. J.R. Josephson and S.G. Josephson, editors. Abductive Inference: Computation,
Philosophy, Technology. New York: Cambridge University Press, 1994.
48. C.G. Jung, K. Fischer, and A. Burt. Multi-agent planning using an abductive
event calculus. Technical Report DFKI Report RR-96-04, DFKI, Germany, 1996.
49. A. C. Kakas. ACLP: Integrating Abduction and Constraint Solving. In Proceedings
of NMR2000, 2000.
50. Kakas, A. C., Kowalski, R. A., Toni, F., Abductive logic programming. Journal of
Logic and Computation 2(6) (1993) 719–770
51. A. C. Kakas, R.A. Kowalski, and F. Toni. The role of abduction in logic pro-
gramming. Handbook of Logic in Artificial Intelligence and Logic Programming 5,
pages 235-324, D.M. Gabbay, C.J. Hogger and J.A. Robinson eds., Oxford Univer-
sity Press (1998)
52. A.C. Kakas and P. Mancarella. Generalised Stable Models: a Semantics for Ab-
duction. In Proc. 9th European Conference on AI, ECAI90, Stockolm, 1990.
53. A.C. Kakas and P. Mancarella. Database updates through abduction. In Proc.
of the 16th Very large Database Conference, pages 650–661. Morgan Kaufmann,
1990.
54. A.C. Kakas and P. Mancarella. On the relation of truth maintenance and abduc-
tion. In Proc. 1st Pacific Rim International Conference on Artificial Intelligence,
PRICAI90, Nagoya, Japan, 1990.
55. A. C. Kakas and P. Mancarella. Knowledge assimilation and abduction. Inter-
national Workshop on Truth Maintenance, Stockholm, ECAI90, Springer Verlag
Lecture notes in Computer Science, Vol. 515, pp. 54-71, 1990.
56. Kakas, A. C., Michael, A. Integrating abductive and constraint logic programming.
In Proc. International Logic Programming Conference, pp. 399-413, 1995.
Abduction in Logic Programming 433
57. A.C. Kakas and A. Michael. Air-Crew Scheduling through Abduction. Proceedings
of IEA/AIE-99, pp. 600–612, 1999.
58. A.C. Kakas and A. Michael. An Abductive-based scheduler for air-crew assingment.
Journal of Applied Artificial Intelligence Vol. 15, pp. 333–360, 2001.
59. A.C. Kakas, A. Michael and C. Mourlas. ACLP: a case for non-monotonic reason-
ing. in Proceedings of NMR98,pp. 46–56, 1998.
60. A.C. Kakas, A. Michael and C. Mourlas. ACLP: Abductive Constraint Logic
Programming. Journal of Logic Programming (special issue on Abductive Logic
Programming), Vol. 44 (1-3), pp. 129-177, 2000.
61. A.C. Kakas and C. Mourlas. ACLP: Flexible Solutions to Complex Problems.
Proceedings of Logic Programming and Non-monotonic Reasoning, LPNMR97,
1997,
62. A.C. Kakas and F. Riguzzi. Abductive Concept Learning. New Generation Com-
puting, Vol. 18, pp. 243-294, 2000.
63. A.C Kakas, Bert Van Nuffelen, and Marc Denecker. A-system : Problem solving
through abduction. In Proceedings of IJCAI’01 - Seventeenth International Joint
Conference on Artificial Intelligence, pp. 591-597, 2001.
64. R.A. Kowalski and F. Sadri, From Logic Programming towards Multi-agent Sys-
tems. Annals of Mathemathics and Artificial Intelligence, Vol 25, pp. 391-419,
1999.
65. R.A. Kowalski and M. Sergot. A logic-based calculus of events. New Generation
Computing, 4(4):319–340, 1986.
66. Kowalski, R.A.; Toni, F.; Wetzel, G.; 1998. Executing suspended logic programs.
Fundamenta Informaticae 34(3):203–224, ISO Press.
67. J. A. Leite, F. C. Pereira, A. Cardoso, L. M.Pereira. Metaphorical Mapping Consis-
tency via Dynamic Logic Programming. In J. Lloyd et al. (eds.), Procs. of First Int.
Conf. on Computational Logic (CL 2000), London, UK, pages 1362-1376, LNAI
1861, Springer, 2000.
68. J.W. Lloyd. Foundations of Logic Programming. Springer-Verlag, 1987.
69. V.W. Marek and M. Truszczy´nski. Stable models and an alternative logic pro-
gramming paradigm. In K.R. Apt, V. Marek, M. Truszczynski, and D.S. Warren,
editors, The Logic Programming Paradigm: a 25 Years Perspective, pages pp. 375–
398. Springer-Verlag, 1999.
70. J. McCarthy. Situations, actions and causal laws. Technical Report AI-memo 1,
Artifical Intelligence Program, Standford University, 1957.
71. Lode R. Missiaen, Marc Denecker, and Maurice Bruynooghe. CHICA, an abduc-
tive planning system based on event calculus. Journal of Logic and Computation,
5(5):579–602, September 1995.
72. L.R. Missiaen, M. Bruynooghe, and M. Denecker. Abductive planning with event
calculus. Internal report, Department of Computer Science, K.U.Leuven, 1992.
73. S. Muggleton. Theory Completion in Learning. In Proceedings of Inductive Logic
Programming, ILP00, 2000.
74. I. Niemela and P. Simons. Smodels - an implementation of the stable model and
well-founded semantics for normal logic programs. Proceeings of the 4th Inter-
national Conference on Logic Programming and Non-monotonic Reasoning, pp.
420-429, 1997.
75. I. Niemel¨a. Logic programs with stable model semantics as a constraint program-
ming paradigm. Annals of Mathematics and Artificial Intelligence, 25(3,4):241–273,
1999.
76. M. Pagnucco. The role of abductive reasoning within the process of belief revision.
PhD Thesis, Department of Computer Science, University of Sydney, 1996.
434 Marc Denecker and Antonis Kakas
77. C.S. Peirce. Philosophical Writings of Peirce. Dover Publications, New York, 1955.
78. Nikolay Pelov, Emmanuel De Mot, and Marc Denecker. Logic programming ap-
proaches for representing and solving constraint satisfaction problems : a compar-
ison. In Proceedings of LPAR’2000 - 7th International Conference on Logic for
Programming and Automated Reasoning, 2000. accepted.
79. L.M. Pereira, J.N. Apar
´
icio, and J.J. Alferes. Nonmonotonic reasoning with Well-
Founded Semantics. In K. Furukawa, editor, Proc. of the eight international con-
ference on logic programming, pages 475–489. the MIT press, 1991.
80. J. Pearl. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible
Inference. Morgan Kaufmann, San Mateo, CA, 1988.
81. D. Poole. A Logical Framework for Default Reasoning. Artifical Intelligence,
36:27–47, 1988.
82. D. Poole. A methodology for using a default and abductive reasoning system.
International Journal of Intelligent Systems, 5(5):521–548, December 1990.
83. D. Poole. Logic programming, abduction and probability: A top-down anytime
algorithm for computing prior and posterior probabilities. New Generation Com-
puting, 11(3–4):377–400, 1993.
84. D. Poole. Probabilistic Horn abduction and Bayesian networks. Artificial Intelli-
gence, 64(1):81–129, 1993.
85. D. Poole. Probabilistic conflicts in a search algorithm for estimating posterior
probabilities in Bayesian networks. Artificial Intelligence, 88:69–100, 1996.
86. D. Poole. The independent choice logic for modelling multiple agents under uncer-
tainty. Artificial Intelligence, 94:7–56, 1997. special issue on economic principles
of multi-agent systems.
87. D. Poole. Probabilistic partial evaluation: Exploiting rule structure in probabilistic
inference. In Proc. 15th International Joint Conf. on Artificial Intelligence (IJCAI-
97), pages 1284–1291, Nagoya, Japan, 1997.
88. D. Poole, R. Goebel, and R. Aleliunas. Theorist: A logical reasoning system for
defaults and diagnosis. In N. Cercone and G. McCalla, editors, The Knowledge
Frontier: Essays in the Representation of Knowledge, pages 331–352. Springer-
Verlag, New York, NY, 1987.
89. David Poole. Learning, bayesian probability, graphical models, and abduction. In
Peter Flach and Antonis Kakas, editors, Abduction and Induction: essays on their
relation and integration. Kluwer, 2000.
90. Poole, D., Goebel, R.G., Aleliunas, Theorist: a logical reasoning system for default
and diagnosis. The Knowledge Fronteer: Essays in the Representation of Knowl-
edge, Cercone and McCalla eds, Springer Verlag Lecture Notes in Computer Science
331–352, 1987.
91. D. Poole, A. Mackworth, R. G. Goebel, Computational Intelligence: a logical
approach Oxford University Press, 1998.
92. Stathis Psillos. Ampliative Reasoning: Induction or Abduction. In ECAI96 work-
shop on Abductive and Inductive Reasoning, 1996.
93. Rochefort, S., Tarau, P. and Dahl, V. Feature Interaction Resolution Through
Hypothetical Reasoning. In Proc. 4th World Multiconference on Systemics, Cy-
bernetics and Informatics (SCI2000) and the 6th International Conference on In-
formation Systems Analysis and Synthesis (ISAS2000), Orlando, USA July 23-26,
2000.
94. Sacc`a, D., Zaniolo, C. Stable Models and Non-Determinism in Logic Programs
with Negation. Proc. of PODS’90, pp. 205 – 217, 1990.
Abduction in Logic Programming 435
95. F.Sadri, F. Toni, I.Xanthakos. A Logic-Agent based System for Semantic Inte-
gration. 17th International CODATA Conference- Data and Information for the
Coming Knowledge Millennium- CODATA 2000, Theme I-3, Integration of Het-
erogeneous Databases and Data Warehousing.
96. F. Sadri, F. Toni. Abduction with negation as failure for active databases and
agents. Proc. AI*IA 99, 6th Congress of the Italian Association for Artificial
Intelligence, pages 353–362, Pitagora Editrice Bologna, 1999.
97. F. Sadri, F. Toni. Abduction with Negation as Failure for Active and Reactive
Rules, In E. Lamma and P. Mello eds., Proc. AI*IA 99, 6th Congress of the Italian
Association for Artificial Intelligence, Springer Verlag LNAI 1792, pages 49-60,
2000.
98. C. Sakama and K. Inoue. Representing priorities in logic programs. In: Proceedings
of the 1996 Joint International Conference and Symposium on Logic Programming,
pages 82–96, MIT Press, 1996.
99. C. Sakama and K. Inoue. Updating extended logic programs through abduction.
In: Proceedings of LPNMR ’99, Lecture Notes in Artificial Intelligence, 1730, pages
147–161, Springer, 1999.
100. C. Sakama and K. Inoue. Abductive logic programming and disjunctive logic
programming: their relationship and transferability. Journal of Logic Programming
- Special issue on ALP 44(1-3):71–96, 2000.
101. C. Sakama and K. Inoue. An alternative approach to the semantics of disjunc-
tive logic programs and deductive databases. Journal of Automated Reasoning
13(1):145–172, 1994.
102. F. Sadri, F. Toni and P. Torroni. Dialogues for negotiation: agent varieties and
dialogue sequences. In Pre-proc. ATAL’01, special track on negotiation. Seattle,
WA, August 2001.
103. Satoh, K. Statutory Interpretation by Case-based Reasoning through Abductive
Logic Programming. Journal of Advanced Computational Intelligence, Vol. 1, No.2,
pp. 94 – 103, 1997.
104. Satoh, K. Computing Minimal Revised Logic Program by Abduction. Proc. of
the International Workshop on the Principles of Software Evolution, IWPSE98,
pp. 177 – 182, 1998.
105. Satoh, K. Using Two Level Abduction to Decide Similarity of Cases. Proc. of
ECAI’98 pp. 398 – 402, 1998.
106. K. Satoh and N. Iwayama. A Query Evaluation method for Abductive Logic
Programming. In K.R. Apt, editor, Proc. of the International Joint Conference
and Symposium on Logic Programming, 1992.
107. Satoh, K. and Iwayama, N. Computing Abduction by Using the TMS. Proc. of
ICLP’91, pp. 505 – 518, 1991.
108. Satoh, K. and Iwayama, N. A Query Evaluation Method for Abductive Logic
Programming. Proc. of JICSLP’92, pp. 671 – 685, 1992.
109. Satoh, K. and Iwayama, N., “A Correct Goal-Directed Proof Procedure for a
General Logic Program with Integrity Constraints”, E. Lamma and P. Mello (eds.),
Extensions of Logic Programming, LNAI 660, pp. 24 – 44, Springer-Verlag (1993).
110. M. Shanahan. Prediction is deduction but explanation is abduction. In Proc. of
the IJCAI89, page 1055, 1989.
111. F. Toni. A semantics for the Kakas-Mancarella procedure for abductive logic
programming. Proc. GULP’95, M. Alpuente and M. I. Sessa, eds., pages 231-242,
1995.
436 Marc Denecker and Antonis Kakas
112. F. Toni. Automated Reasoning for Collective Information Management. Proc.
LocalNets, International Workshop on Community-based Interactive Systems,in
conjunction with AC’99, the Annual Conference of the EC I
3
Programme
113. K. Van Belleghem, M. Denecker, and D. De Schreye. Representing continuous
change in the abductive event calculus. In Proc. of the International Conference
on Logic Programming. MIT-Press, 1994.
114. K. Van Belleghem, M. Denecker, and D. De Schreye. The abductive event calculus
as a general framework for temporal databases. In Proc. of the International
Conference on Temporal Logic, pages 301–316, 1994.
115. K. Van Belleghem, M. Denecker, and D. De Schreye. A strong correspondence
between description logics and open logic programming. In Lee Naish, editor, Proc.
of the International Conference on Logic Programming, 1997, pages 346–360. MIT-
press, 1997.
116. A. Van Gelder, K.A. Ross, and J.S. Schlipf.r The Well-Founded Semantics for
General Logic Programs. Journal of the ACM, 38(3):620–650, 1991.
117. Bert Van Nuffelen and Marc Denecker. Problem solving in ID-logic with aggre-
gates: some experiments. In M. Denecker, A. Kakas, and F. Toni, editors, 8th Int.
Workshop on Non-Monotonic Reasoning (NMR2000), session on Abduction, pages
1–15, Breckenridge, Colorado, USA, April 9-11 2000.
118. B. Van Nuffelen and A.C Kakas, A-System: Programming with Abduction. In
Proceedings of LPNMR2001, LNAI Vol. 2173, pp. 393-396, Springer Verlag, 2001.
119. Sven Verdoolaege, Marc Denecker, and Frank Van Eynde. Abductive reasoning
with temporal information. In Ielka van der Sluis Harry Bunt and Elias Thijsse,
editors, Proceedings of the Fourth International Workshop on Computational Se-
mantics, pages 351–366, 2001.
120. Wetzel, G. Abductive and Constraint Logic Programming. Ph.D. Thesis, Imperial
College, London, 1997.
121. G. Wetzel, F. Toni. Semantic Query Optimization through Abduction and Con-
straint Handling. Proc. of the International Conference on Flexible Query Answer-
ing Systems, T. Andreasen, H. L. Larsen and H. Christiansen eds., Springer Verlag
LNAI 1495 (1998).
122. The XSB Group. The XSB logic programming system, version 2.0. 1999. Available
from http://www.cs.sunysb.edu/~sbprolog.
123. A. Yamamoto. Using abduction for induction based on bottom up generalization.
In Peter Flach and Antonis Kakas, editors, Abduction and Induction: essays on
their relation and integration. Kluwer, pp. 267-280,2000.
124. J.H. You, L.Y. Yuan and R. Goebel. An abductive approach to disjunctive logic
programming. Journal of Logic Programming (special issue on Abductive Logic
Programming), Vol. 44 (1-3), pp. 101-128, 2000.
