The Object Role Modeling language (ORM2) is one of the main conceptual modeling languages. Recently, a translation has been proposed of a main fragment of ORM2 (ORM2zero) into the description logic ALCQI, allowing the use of logical instruments in the analysis of ORM schemas. On the other hand, in many ontological domains there is a need for the formalization of defeasible information and of nonmonotonic forms of reasoning. Here we introduce two new constraints in ORM2 language, in order to formalize defeasible information into the schemas, and we explain how to translate such defeasible information in ALCQI.
ORM2 (‘Object Role Modelling 2’) is a graphical fact-oriented approach for modelling,
transforming, and querying business domain information, which allows for a
verbalisation in a language readily understandable by non-technical users [
On the other hand, in the more general field of formal ontologies in the last years a
lot of attention has been dedicated to the implementations of forms of defeasible
reasoning, and various proposals, such as [
In what follows we propose an extension of ORM2 with two new formal constraints,
with the main aim of integrating a form of defeasible reasoning in the ORM2 schemas;
we explain how to translate such enriched ORM2 schemas into ALCQI knowledge
bases and how to use them to check the schema consistency and draw conclusions.
In particular, the paper presents a procedure to implement a particular construction in
nonmonotonic reasoning, i.e. Lehmann and Magidor’s Rational Closure (RC)[
According to the ORM2 design procedure, after the specification of the relevant object types (i.e. entity and value types) and predicates, the static constraints must be considered. The rest of this section is devoted to an informal introduction of the constraint graphical representation, together with their intended semantics. Fig. 1 shows an example of an ORM2 conceptual schema modelling the ‘academic domain’ (where the soft rectangles are entity types, the dashed soft rectangles are value types, and the sequences of one or more role-boxes are predicates). The example is not complete w.r.t. the set of all the ORM2 constraints but it aims at giving the feeling of the expressive power of the language. The following are among the constraints included in the schema (the syntax we devised for linearizing them is in square brackets): 1. Subtyping (depicted as thick solid and dashed arrows) representing ‘is-a’ relationships among types. A partition, made of a combination of an exclusive constraint (a circled ‘X’ saying that ‘Research&TeachingStaff, Admin, Student are mutually disjoint’), and a total constraint (a circled dot for ‘Research&TeachingStaff, Admin, Student completely cover their common supertype’). [O-SETTot(fResearch&TeachingStaff, Admin, Studentg,UNI-Personnel)] 2. An internal frequency occurrence saying that if an instance of Research&TeachingStaff plays the role of being lecturer in the relation isGivenBy, that instance can play the role at most 4 times [FREQ(isGivenBy.Research&TeachingStaff,(1,4))]. A frequency occurrence may span over more than one role, and suitable frequency ranges can be specified. At most one cardinalities (depicted as continuos bars) are special cases of frequency occurrence called internal uniqueness constraints [e.g. FREQ(hasWeight.Course,(1,1))]. 3. An external frequency occurrence applied to the roles played by Student and Course, meaning that ‘Students are allowed to enrol in the same course at most twice’. [FREQ(isIn.Course,isBy.Student,(1,2))] 4. An external uniqueness constraint between the role played by Course in isIn and the role played by Date in wasOn, saying that ‘For each combination of Course and Date, at most one Enrollment isIn that Course and wasOn that Date’. [FREQ(isIn.Course,wasOn.Date,(1,1))] 5. A mandatory participation constraints (graphically represented by a dot), among several other, saying that ‘Each Course isGivenBy at least one instance of the Research&TeachingStaff type’ (combinations of mandatory and uniqueness translate into exaclty one cardinality constraints). [MAND(isGivenBy.Research&TeachingStaff,Research&TeachingStaff)] 6. A disjunctive mandatory participation, called inclusive-or constraint (depicted as a circled dot), linking the two roles played by the instances of AreaManager meaning that ‘Each area manager either works in or heads (or both)’. [MAND(fworksIn.AreaManager,heads.AreaManagerg,AreaManager)] 7. An object cardinality constraint forcing the number of the Admin instances to be less or equal to 100 (role cardinality constraints, applied to role instances, are also part of ORM2). [O-CARD(Admin)=(0,100)] 8. An object type value constraint indicating which values are allowed in Credit (role value constraints can be also expressed to indicate which values are allowed to play a given role). [V-VAL(Credit)=f4,6,8,12g] 9. An exclusion constraint (depicted as circled ‘X’) between the two roles played by the instances of Student, expressing the fact that no student can play both these roles. Exclusion constraint can also span over arbitrary sequences of roles. The combination of exclusion and inclusive-or constraints gives rise to exclusive-or constraints meaning that each instance in the attached entity type plays exactly one of the attached roles. Exclusion constraints, together with subset and equality, are called set-comparison constraints. [R-SETExc(worksFor.Student,collaborates.Student)] 10. A ring constraint expressing that the relation reportsTo is asymmetric. [RINGAsym(reportTo.Admin,reportTo.AreaManager)]
A comprehensive list of all the ORM2 constraints, together with their graphical
representation, can be found in [
The ALCQI encoding of ORM2zero
With the main aim of relying on available tools to reason in an effective way on ORM2
schemas, an encoding in the description logic ALCQI for which tableaux-based
reasoning algorithms with a tractable computational complexity have been developed [
Now, the discrepancy between ORM2 and ALCQI poses two main obstacles that
need to be faced in order to provide the encoding. The first one, caused by the absence of
n-ary relations in ALCQI , is overcome by means of reification: for each relation R of
arity n 2, a new atomic concept AR and n functional roles (R:a1); : : : ; (R:an) are
introduced. The tree-model property of ALCQI guarantees the correctness encoding
w.r.t. the reasoning services over ORM2. Unfortunately, the second obstacle fixes,
once for all, the limits of the encoding: ALCQI does not admit neither arbitrary
set-comparison assertions on relations, nor external uniqueness or uniqueness involving
more than one role, or arbitrary frequency occurrence constraints. In other terms, it can
be proven that ALCQI is strictly contained in ORM2. The analysis of this inclusion
thus led to identification of the fragment called ORM2zero which is maximal with
respect to the expressiveness of ALCQI, and still expressive enough to capture the
most frequent usage patterns of the conceptual modelling community. Let ORM2zero =
fTYPE; FREQ ; MAND; R-SET ; O-SETIsa; O-SETTot; O-SETEx; OBJg be the
fragment of ORM2 where: (i) FREQ can only be applied to single roles, and (ii)
R-SET applies either to entire relations of the same arity or to two single roles. The
encoding of the semantics of ORM2zero shown in table 1 is based on the SALCQI
signature made of: (i) A set E1; E2; : : : ; En of concepts for entity types; (ii) a set
V1; V2; : : : ; Vm of concepts for value types; (iii) a set AR1 ; AR2 ; : : : ; ARk of concepts
for objectified n-ary relations; (iv) a set D1; D2; : : : ; Dl of concepts for domain
symbols; (v) 1; 2; : : : ; nmax + 1 roles. Additional background axioms are needed here
in order to: (i) force the interpretation of the ALCQI knowledge base to be correct
w.r.t. the corresponding ORM2 schema, and (ii) guarantee that that any model of the
resulting ALCQI can be ‘un-reified’ into a model of original ORM2zero schema.
The correctness of the introduced encoding is guaranteed by the following theorem
(whose complete proof is available at [
Let us conclude this section with some observation about the complexity of
reasoning on ORM2 conceptual schemas, and taking into account that all the reasoning tasks
for a conceptual schema can be reduced to object type consistency. Undecidability of the
ORM2 object type consistency problem can be proven by showing that arbitrary
combinations of subset constraints between n-ary relations and uniqueness constraints over
single roles are allowed [
Rational Closure in ALCQI
Now we briefly present the procedure to define the analogous of RC for the DL
language ALCQI. A more extensive presentation of such a procedure can be found in [
: : : ; Rk:a1; : : : ; Rk:amg; O) (A) R-SETSub(A; B) (A) R-SETExc(A; B) (B) R-SETSub(A; B) (B) R-SETExc(A; B) O-SETIsa(fO1; : : : ; Ong; O) O-SETTot(fO1; : : : ; Ong; O) O-SETEx(fO1; : : : ; Ong; O) OBJ(R; O) logical characterization, it generally maintains the same complexity level of the underlying monotonic logic, and it does not give back counter-intuitive conclusions; its main drawback is in its inferential weakness, since there could be desirable conclusions that we won’t be able to draw (see example 2 below).
As seen above, each ORM2zeroschema can be translated into an ALCQI TBox. A TBox T for ALCQI consists of a finite set of general inclusion axioms (GCIs) of form C v D, with C and D concepts. Now we introduce also a new form of information, the defeaible inclusion axioms C < D, that are read as ‘Typically, an individual falling under the concept C falls also under the concept D’. We indicate with B the finite set of such inclusion axioms.
Example 1. Consider a modification of the classical ‘penguin example’, with the concepts P; B; F; I; F i; W respectively read as ‘penguin’, ‘bird’, ‘flying’, ‘insect’, ‘fish’, and ‘have wings’, and a role P rey, where a role instantiation (a; b):P rey read as ‘a preys for b’. We can define a defeasible KB K = hT ; Bi with T = fP v B; I v :F ig and B = fP < :F , B < F , P < 8P rey:F i; B < 8P rey:I; B < W g.
In order to define the rational closure of a knowledge base hT ; Bi, we must first of
all transform the knowledge base hT ; Bi into a new knowledge base h ; i, s.t. while
T and B are sets of inclusion axioms, and are simply sets of concepts. Then, we
shall use the sets h ; i to define a nonmonotonic consequence relation that models the
rational closure. Here we just present the procedure, referring to [
steps.
i. Starting with hT ; Bi, we apply the following Step 1. Define the set representing the strict form of the set B, i.e. the set Bv = fC v D j C < D 2 Bg, and define a set AB as the set of the antecedents of the conditionals in B, i.e. AB = fC j C < D 2 Bg.
Step 2. We determine an exceptionality ranking of the sequents in B using the set of the antecedents AB and the set Bv.
Step 2.1. A concept is considered exceptional in a knowledge base hT ; Bi only if it is classically negated (i.e. we are forced to consider it empty), that is, C is exceptional in hT ; Bi only if
T [ Bv j= > v :C where j= is the classical consequence relation associated to ALCQI. If a concept is considered exceptional in hT ; Bi, also all the defeasible inclusion axioms in B that have such a concept as antecedent are considered exceptional. So, given a knowledge base hT ; Bi we can check which of the concepts in AB are exceptional (we indicate the set containing them as E(AB)), and consequently which of the axioms in B are exceptional (the set E(B) = fC < D j C 2 E(AB)g).
So, given a knowledge base hT ; Bi we can construct iteratively a sequence E0; E1; : : : of subsets of B in the following way: – E0 = B – Ei+1 = E(Ei) Since B is a finite set, the construction will terminate with an empty set (En = ; for some n) or a fixed point of E.
Step 2.2 Using such a sequence, we can define a ranking function r that associates to every axiom in B a number, representing its level of exceptionality: r(C < D) = i if C < D 2 Ei and C < D 2= Ei+1
1 if C < D 2 Ei for every i : Here we shall assume that every concept has a finite ranking value, and we shall deal with the possible occurrence of some concept with 1 as ranking value in the following section. Step 3. Now we build a new formalization of the information contained in the knowledge base hT ; Bi, translating each of the two sets of axioms into two sets of concepts, and respectively. The set will simply correspond to the materialization of the inclusion axioms, i.e. the concepts translating the axioms.
= fC D j C v D 2 T g In order to define the set , given the rank value of the sequents in B, we construct a set of default concepts = f 0; : : : ; ng (with n the highest rank-value in B), with i = lfC
D j C < D 2 B and r(C < D) ig : Hence we substitute the conceptual system hT ; Bi with the pair h ; i, where and are sets of concepts, the former containing concepts to be considered valid for every individual of the domain, the latter containing concepts to be considered defeasibly valid, i.e. we apply such default concepts to an individual only if they are consistent with the information in our knowledge base. It is not difficult to see that the concepts in are linearly ordered by j=, that is, for every i, 0 i < n 1, j= i v i+1. Rational Closure. Consider now = fC1 D1; : : : ; Cm Dmg and = f 0; : : : ; ng. We define a nonmonotonic consequence relation between the concepts j h ; i that determines what presumably follows from a finite set of concepts . Simply, a concept D is a defeasible consequence of if it classically follows from , the strict information contained in the knowledge base (i.e. ), and the first default concept i that in the sequence h 0; : : : ; ni results classically consistent with the rest of the premises.
D, where i is the first ( )Definition 1. j h ; iD iff j= d u d u i v consistent formula 3 of the sequence h 0; : : : ; ni.
You can find in [
Proposition 1 ([
Moreover, as deciding entailment in ALCQI is EXPTIME-complete (see Theorem 2), and since the decidability problem for the rational closure is reducible to a finite number of decision w.r.t. the classical ALCQI consequence relation, we obtain immediately that Proposition 2. Deciding Cj hTe; eiD in ALCQI is an EXPTIME-complete problem. Example 2. Consider the KB of Example 1. Hence, we start with K = hT ; Bi. The strict form of B is Bv = fP v :F , B v F , P v 8P rey:F i; B v 8P rey:I; B v W g, with AB = fP; Bg. Following the procedure at Step 2, we obtain the exceptionality ranking of the sequents: E0 = fP v :F , B v F , P v 8P rey:F i; B v 8P rey:I; B v W g; E1 = fP v :F , P v 8P rey:F ig; E2 = ;. Automatically, we have the ranking values of every sequent in B: namely, r(B v F ) = r(B v 8P rey:I) = r(B v W ) = 0; r(P < :F ) = r(P < 8P rey:F i) = 1. From such a ranking, we obtain a set of default concepts = f 0; 1g, with 0 = (B 1 = (P
F ) u (B :F ) u (P 8P rey:I) u (P 8P rey:F i) : :F ) u (P 8P rey:F i) u (B
W ) Now, referring to definition 1, we can derive a series of desirable conclusions, as :F j :B, , B ^ greenj F , P ^ blackj :F , P j 8P rey::I. Instead, other counterintuitive connections are not valid, such as B ^ :F j P , B ^ :F j :P , or P j F . Here we can notice the main weakness of the Rational Closure: even if it would be intuitive to conclude that penguins have wings (P j W ), we cannot conclude that a class that results atypical (as penguins) cannot inherit any of the typical properties of its superclasses (as having wings), even if such properties are not logically connected to the ones that determine the exceptionality (not flying and eating fish). 5
As seen above, in order to introduce defeasible reasoning in DL we introduce the defeasible inclusion axiom C < D, indicating that the elements of the concept C typically, but not necessarily, are elements of the concept D. We want to introduce in the ORM2zeroschemas constraints playing an analogous role, i.e. representing defeasible constraints in the ontological organization of a particular domain. With this goal in mind, two constraints aimed at representing forms of defeasible constraints between classes, and classes and their properties, are introduced.
– A defeasible subclass relation: we introduce an arrow ‘;’, where ‘C ; D’ indicates that each element of the class C is also an element of the class D, if not informed of the contrary. 3 That is, 6j= d u d v : i. Example 3 (Defeasible subclass relation). Consider figure 2. The schema on the left represents in ORM2 the classic penguin example: penguins are birds and do not fly (the class Penguin is a subclass, respectively, of the classes Birds and Non-FlyingObj), while birds fly and have wings (the class Birds is a subclass, respectively, of the classes FlyingObj and WingyObj). The translation procedure into ALCQI gives back the TBox T in table 2. From T we can derive that the schema is inconsistent, since we have T j= :Penguin, i.e. the concept Penguin must be empty. We can modify the knowledge base introducing defeasible information, in particular stating that birds typically fly and typically have wings, and penguins typically do not fly. In this way we obtain the schema on the right, and in ALCQI we obtain a set B = fBird < WingyObject; Bird < FlyingObject; Penguin < NonFlyingObjectg, substituting the corresponding classical inclusion axioms in the TBox. – A defeasible mandatory participation: we introduce a new mandatory participation constraint ‘ ’, to use instead of the classically mandatory constraint ‘ ’. If the connection between a class C and a relation R is constrained by a constraint , we read it as ‘each element of the class C participates to the relation R, if we are not informed of the contrary’.
Example 4 (Defeasible mandatory participation). Consider figure 3. The schema represents the organization of a firm: the class Manager is a subclass of the class Employee, and every employee must work for a project. while every project must have at least an employee working on it. The class Manager is partitioned into AreaManager and TopManager. Each top manager mandatorily manages a project. The translation procedure into ALCQI of the left version of the schema gives back the TBox T in table 3. Since managing and working for a project are not compatible roles, T implies that the class TopManager is empty, since a top manager would manage and would work for a project at the same time. Instead, if we declare that typically an employee works for a project, we can consider the top managers as exceptional kind of employees; hence we substitute the mandatory constraint between Employee and WorkFor with a defeasible constraint (i.e. the schema on the right in figure 3); from such
a change we obtain a knowledge base as the one above, but with the defeasible inclusion axiom Employee < 9f1 :WorksFor instead of the axiom Employee v 9f1 :WorksFor.
Introducing such constraints, we introduce the forms of defeasible subsumptions appropriate for modeling nonmonotonic reasoning. In particular: – A subclass relation, as the ones in example 3, is translated into an inclusion axiom C v D, and correspondingly we translate the defeasible connection C ; D into a defeasible inclusion axiom C < D. – Analogously, consider the strict form of the example 4. The mandatory participation of the class B to the role AN is translated into the axiom B v 9f1 :AN. If we use the defeasible mandatory participation constraint, we simply translate the structure using the defeasible inclusion <, obtaining the axiom B < 9f1 :AN.
Hence, from a ORM graph with defeasible constraints we obtain an ALCQI knowledge base K = hT ; Bi, where T is a standard ALCQI Tbox containing concept inclusion axioms C v D, while the set B contains defeasible axioms of the form C < D. Once we have our knowledge base K, we apply to it the procedure presented in the previous section, in order to obtain the rational closure of the knowledge base.
Consistency. In ORM2, and in conceptual modeling languages in general, the notion of consistency is slightly different from the classical form of logical consistency. That is, generally from a logical point of view a knowledge base K is considered inconsistent only if we can classically derive a contradiction from it; in DLs that corresponds to saying that K j= > v ?, i.e. every concept in the knowledge base results empty. Instead, dealing with conceptual modeling schemas we generally desire that our model satisfies a stronger form of consistency constraint, that is, we want that none of the classes present in the schema are forced to be empty.
atomic concepts present in its axioms are forced to be empty, that is, if T 6j= > v :A for every atomic concept A appearing in the inclusion axioms in T .
As seen above, the introduction of defeasible constraints into ORM2zero allows to build schemas that in the standard notation would be considered inconsistent, but that, once introducing the defeasible constraints, allow for an instantiation such that all the classes result non-empty. Hence it is necessary to redefine the notion of consistency check in order to deal with such situations.
Such a consistency check is not problematic, since we can rely on the ranking procedure presented above. Consider a TBox T obtained by an ORM2zero schema, and indicate with C the set of all the atomic concepts used in T . It is sufficient to check the exceptionality ranking of all the concepts in C with respect to T : if a concept C has an exceptionality ranking r(C) = n, with 0 < n < 1, then it represents an atypical situation, an exception, but that is compatible with the information conveyed by the defeasible inclusion axioms. For example, in the above examples the penguins and the top managers would be empty classes in the classical formalization, but using the defeasible approach they result exceptional classes in our schemas, and we can consider them as non-empty classes while still considering the schema as consistent. The only case in which a class has to be considered necessarily empty, is when it has 1 as ranking value: that means that, despite eliminating all the defeasible connections we can, such a concept still results empty. Then, the notion of strong consistency for ORM2zero with defeasible constraints is the following: Definition 3 (Strong consistency with defeasible constraints). A knowledge base K = hT ; Bi is strongly consistent if none of the atomic concepts present in its axioms are forced to be empty, that is, if r(A) 6= 1 for every atomic concept A appearing in the inclusion axioms in K.
Given a defeasible ORM2zeroschema , eliminate from it all the defeasible constraints (call strict the resulting schema). From the procedures defined above, it is immediate to see that if strict is a strongly inconsistent ORM2zeroschema, in the ‘classical’ sense, then is a strongly inconsistent defeasible schema: simply, if the negation of a concept is forced by the strict part of a schema, it will be necessarily forced at each ranking level, resulting in a ranking value of 1.
On the other hand, there can be also strongly inconsistent defeasible schemas in which inconsistency depends not only on the strict part of the schema, but also on the defeasible part. For example, the schema in figure 4 is inconsistent, since the class A results to have a ranking value of 1 (the schema declares that the class A is directly connected to two incompatible concepts). Now, we can check the results of the defined procedure in the examples presented.
Example 5. Consider example 3. From the translation of the defeasible form of the schema we conclude that the axiom Penguin v NonFlyingObject has rank 1, while Bird v WingyObject and Bird v FlyingObject have rank 0, that means that we end up with two default concepts: – – 0 := dfPenguin 1 := Penguin
NonFlyingObject; Bird NonFlyingObject
WingyObject; Bird
We can derive the same kind of conclusions as in example 2, and again we can see the limits of the rational closure, since we cannot derive the desirable conclusion that Penguinj WingyObject. Example 6. Consider the knowledge base obtained in the example 4. We have only a defeasible inclusion axiom Employee < 9f1 :WorksFor, and, since Employee does not turn out to be an exceptional concept, we end up with a single default concept in B: – 0 := fEmployee
Since TopManager is not consistent with all the strict information contained in the schema plus 0, we cannot associate 0 to TopManager and, despite we have the information that for non-exceptional cases an employee works for a project, we are not forced to conclude that for the exceptional class of the top managers. 6
In this paper we have presented a way to implement a form of defeasible reasoning into the ORM2 formalism. Exploiting the possibility of encoding ORM2zero, that represents a big portion of the ORM2 language, into the description logic ALCQI on one hand, and a procedure appropriate for modeling one of the main forms of nonmonotonic reasoning, i.e. rational closure, into DLs on the other hand, we have defined two new constraints, a defeasible subclass relation and a defeasible mandatory participation, that are appropriate for modeling defeasible information into ORM2, and that, once translated into ALCQI , allow for the use of the procedures characterizing rational closure to reason about the information contained into an ORM2zero schema.
The present proposal deals only with reasoning on the information contained in the
TBox obtained from an ORM2 schema, but, once we have done the rational closure of
the TBox, we can think also of introducing an ABox, that is, the information about a
particular domain of individuals. A first proposal in such direction is in [