Lightweight Description Logics: DL-Lite A and EL ++

Transcription

1 Lightweight Description Logics: DL-Lite A and EL ++ Elena Botoeva 1 KRDB Research Centre Free University of Bozen-Bolzano January 13, 2011 Departamento de Ciencias de la Computación Universidad de Chile, Santiago, Chile 1 Part of the slides is borrowed from Diego Calvanese Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 1/45

2 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 2/45

3 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 3/45

4 Description Logics formal languages for representing knowledge bases TBox represents implicit knowledge (a set of axioms) ABox represents explicit knowledge (a set of individual assertions) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 4/45

5 Description Logics formal languages for representing knowledge bases TBox represents implicit knowledge (a set of axioms) ABox represents explicit knowledge (a set of individual assertions) talk about concepts Professor Student Course, and roles teaches attends Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 4/45

6 Description Logics formal languages for representing knowledge bases TBox represents implicit knowledge (a set of axioms) ABox represents explicit knowledge (a set of individual assertions) talk about concepts Professor Student Course, and roles teaches attends variable free syntax for describing complex concepts Professor Student teaches.phdcourse haschild.male for asserting implicit knowledge teaches Course Professor Student for asserting explicit knowledge Student(john) attends(john, db) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 4/45

7 Why Description Logics? Decidable fragments of FOL ( Well-defined semantics). DLs provide sound and complete reasoning services: checking knowledge base consistency, checking logical entailment, answering conjunctive queries (unions of CQ). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 5/45

8 Why Description Logics? Decidable fragments of FOL ( Well-defined semantics). DLs provide sound and complete reasoning services: checking knowledge base consistency, checking logical entailment, answering conjunctive queries (unions of CQ). Modelling capabilities. Description Logics (DLs) can express, e.g.: Taxonomy of classes of objects, UML class diagrams, ER models, etc. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 5/45

9 Why Description Logics? Decidable fragments of FOL ( Well-defined semantics). DLs provide sound and complete reasoning services: checking knowledge base consistency, checking logical entailment, answering conjunctive queries (unions of CQ). Modelling capabilities. Description Logics (DLs) can express, e.g.: Taxonomy of classes of objects, UML class diagrams, ER models, etc. DLs are widely used nowadays: underly OWL 2, the Semantic Web standard, serve as conceptual layer in Ontology Based Data Access, for formalizing bio-medical domain, etc. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 5/45

10 Lightweight Description Logics The majority of studied DLs is intractable: Satisfiability of the basic DL ALC is ExpTime-complete. Satisfiability of SROIQ, the basis of OWL 2, is 2NExpTime-complete. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 6/45

11 Lightweight Description Logics The majority of studied DLs is intractable: Satisfiability of the basic DL ALC is ExpTime-complete. Satisfiability of SROIQ, the basis of OWL 2, is 2NExpTime-complete. Two families of DLs that provide tractable reasoning have been developed, DL-Lite family by Calvanese et al. [5], and EL family by Baader et al. [2]. A common feature: no disjunction and no universal restrictions Professor Student haschild.male Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 6/45

12 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 7/45

13 DL-Lite and DL-Lite A DL-Lite is a family of tractable logics [5] specifically tailored to efficiently deal with large amounts of data. Reasoning in DL-Lite are FOL-rewritable, i.e., we can reduce them to the problem of query evaluation in relational databases. AC 0 in data complexity. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 8/45

14 DL-Lite and DL-Lite A DL-Lite is a family of tractable logics [5] specifically tailored to efficiently deal with large amounts of data. Reasoning in DL-Lite are FOL-rewritable, i.e., we can reduce them to the problem of query evaluation in relational databases. AC 0 in data complexity. DL-Lite A is the most expressive member of this family. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 8/45

15 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 9/45

16 DL-Lite A Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 10/45

17 DL-Lite A Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept and role constructs: B ::= A R basic concept C ::= B B complex concept R ::= P P basic role Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 10/45

18 DL-Lite A Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept and role constructs: B ::= A R basic concept C ::= B B complex concept R ::= P P basic role TBox and ABox assertions: B 1 B 2 concept inclusion B 1 B 2 disjointness of concepts R 1 R 2 role inclusion Dis(R 1, R 2) disjointness of roles Funct(R) role functionality A(a) P(a, b) membership assertions Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 10/45

19 DL-Lite A Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept and role constructs: B ::= A R basic concept C ::= B B complex concept R ::= P P basic role TBox and ABox assertions: B 1 B 2 concept inclusion B 1 B 2 disjointness of concepts R 1 R 2 role inclusion Dis(R 1, R 2) disjointness of roles Funct(R) role functionality A(a) P(a, b) A DL-Lite A Knowledge Base K is a pair T, A where T is a finite set of TBox axioms and A is a finite set of membership assertions. membership assertions Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 10/45

20 DL-Lite A Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept and role constructs: B ::= A R basic concept C ::= B B complex concept R ::= P P basic role TBox and ABox assertions: B 1 B 2 concept inclusion B 1 B 2 disjointness of concepts R 1 R 2 role inclusion Dis(R 1, R 2) disjointness of roles Funct(R) role functionality A(a) P(a, b) A DL-Lite A Knowledge Base K is a pair T, A where T is a finite set of TBox axioms and A is a finite set of membership assertions. membership assertions Note: for simplicity attributes, value-domain expressions and identification constraints are not presented. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 10/45

21 DL-Lite A Syntax Syntactic restriction to ensure tractability: Functional roles cannot be specialized. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 11/45

22 DL-Lite A Syntax Syntactic restriction to ensure tractability: Functional roles cannot be specialized. I.e., it is not allowed to have things like: R R Funct(R) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 11/45

23 DL-Lite A Syntax Syntactic restriction to ensure tractability: Functional roles cannot be specialized. I.e., it is not allowed to have things like: R R Funct(R) Otherwise, the resulting logic is ExpTime-hard in the size of ontology[1]. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 11/45

24 DL-Lite A Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 12/45

25 DL-Lite A Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept and role constructs ( B) I = I \ B I ( R) I = {x I y I, (x, y) R I } (P ) I = {(y, x) I I (x, y) P I } Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 12/45

26 DL-Lite A Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept and role constructs TBox and ABox assertions ( B) I = I \ B I ( R) I = {x I y I, (x, y) R I } (P ) I = {(y, x) I I (x, y) P I } I = B C iff B I C I I = R 1 R 2 iff R I I 1 R 2 I = Dis(R 1, R 2) iff R I 1 R I 2 = I = Funct(R) iff (x, y 1) R I, (x, y 2) R I y 1 = y 2 I = A(a) iff a I A I I = P(a, b) iff (a I, b I ) P I Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 12/45

27 DL-Lite A Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept and role constructs TBox and ABox assertions ( B) I = I \ B I ( R) I = {x I y I, (x, y) R I } (P ) I = {(y, x) I I (x, y) P I } I = B C iff B I C I I = R 1 R 2 iff R I I 1 R 2 I = Dis(R 1, R 2) iff R I 1 R I 2 = I = Funct(R) iff (x, y 1) R I, (x, y 2) R I y 1 = y 2 I = A(a) iff a I A I I = P(a, b) iff (a I, b I ) P I I is a model of K = T, A if it satisfies all axioms of T and A. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 12/45

28 DL-Lite A Example boss 1..* 1..1 AreaManager Employee empcode: Integer salary: Integer {disjoint} Manager 1..* TopManager 1..* Project projectname: String worksfor 1..1 manages 1..1 Note: DL-Lite A cannot capture completeness of a hierarchy. This would require disjunction (i.e., OR). Manager Employee AreaManager Manager TopManager Manager AreaManager TopManager worksfor Employee worksfor Project Employee worksfor Project worksfor Funct(manages) Funct(manages ) manages worksfor. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 13/45

29 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 14/45

30 Reasoning Problems The Knowledge Base Satisfiability problem is to check, given a DL-Lite A KB K, whether K admits at least one model. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 15/45

31 Reasoning Problems The Knowledge Base Satisfiability problem is to check, given a DL-Lite A KB K, whether K admits at least one model. The Query Answering problem is to compute, given a DL-Lite A KB K and a query q (either a CQ or a UCQ) over K, the set ans(q, K) of certain answers. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 15/45

32 Reasoning Problems The Knowledge Base Satisfiability problem is to check, given a DL-Lite A KB K, whether K admits at least one model. The Concept Satisfiability problem is to decide, given a TBox T and a concept C, whether there exist a model I of T such C I. The Concept Subsumption problem is to decide, given a TBox T and concepts C 1 and C 2, whether for every model I of T it holds that C I 1 C I 2 (T = C 1 C 2 ). The Role Subsumption problem is to decide, given a TBox T and roles R 1 and R 2, whether for every model I of T it holds that R I 1 R I 2 (T = R 1 R 2 ). The Query Answering problem is to compute, given a DL-Lite A KB K and a query q (either a CQ or a UCQ) over K, the set ans(q, K) of certain answers. The Concept Instance Checking problem is to decide, given an object name a, a concept B, and a KB K = T, A, whether K = C(a). The Role Instance Checking problem is to decide, given a pair (a, b), a role R, and a KB K = T, A, whether K = R(a, b). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 15/45

33 First Order Logic Rewritability ABox A can be stored as a relational database in a standard RDBMS as follows: For each atomic concept A of the ontology: define a unary relational table taba populate taba with each c such that A(c) A For each atomic role P of the ontology, define a binary relational table tabp populate tabp with each c 1, c 2 such that P(c 1, c 2 ) A We denote with DB(A) the database obtained as above. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 16/45

34 First Order Logic Rewritability ABox A can be stored as a relational database in a standard RDBMS as follows: For each atomic concept A of the ontology: define a unary relational table taba populate taba with each c such that A(c) A For each atomic role P of the ontology, define a binary relational table tabp populate tabp with each c 1, c 2 such that P(c 1, c 2 ) A We denote with DB(A) the database obtained as above. Definition KB satisfiability (QA) in DL-Lite A is FOL-rewritable if, for every T (and every UCQ q) there exists a FO query q, such that for every nonempty A (and every tuple of constants a from A), T, A is satisfiable iff q () evaluates to false in DB(A) ( a ans(q, T, A ) iff a DB(A) q DB(A) ). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 16/45

35 First Order Logic Rewritability ABox A can be stored as a relational database in a standard RDBMS as follows: For each atomic concept A of the ontology: define a unary relational table taba populate taba with each c such that A(c) A For each atomic role P of the ontology, define a binary relational table tabp populate tabp with each c 1, c 2 such that P(c 1, c 2 ) A We denote with DB(A) the database obtained as above. Definition KB satisfiability (QA) in DL-Lite A is FOL-rewritable if, for every T (and every UCQ q) there exists a FO query q, such that for every nonempty A (and every tuple of constants a from A), T, A is satisfiable iff q () evaluates to false in DB(A) ( a ans(q, T, A ) iff a DB(A) q DB(A) ). We show that KB satisfiability and QA in DL-Lite A are FOL-rewritable. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 16/45

36 Knowledge Base Satisfiability Problem Given a KB K = T, A, check whether there exists an interpretation I such that I = T and I = A Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 17/45

37 Knowledge Base Satisfiability Problem Given a KB K = T, A, check whether there exists an interpretation I such that I = T and I = A Positive Inclusions (PIs) are inclusions of the form B 1 B 2, R 1 R 2 Negative Inclusions (NIs) are inclusions of the form B 1 B 2, Dis(R 1, R 2), or Funct(R) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 17/45

38 Satisfiability of KBs with only PIs Positive inclusions cannot introduce contradicting information: Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 18/45

39 Satisfiability of KBs with only PIs Positive inclusions cannot introduce contradicting information: Theorem Let K = T, A be a DL-Lite A KB such that T consists only of PIs. Then K is satisfiable. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 18/45

40 Satisfiability of KBs with only PIs Positive inclusions cannot introduce contradicting information: Theorem Let K = T, A be a DL-Lite A KB such that T consists only of PIs. Then K is satisfiable. We can always build a model by adding missing tuples to satisfy PIs. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 18/45

41 Source of Unsatisfiability However, negative inclusions can cause a KB to be unsatisfiable: Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 19/45

42 Source of Unsatisfiability However, negative inclusions can cause a KB to be unsatisfiable: T : Dis(teaches, attends) A : teaches(john, db), attends(john, db) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 19/45

43 Source of Unsatisfiability However, negative inclusions can cause a KB to be unsatisfiable: T : Dis(teaches, attends) A : teaches(john, db), attends(john, db) T : Funct(teaches ) A : teaches(john, db), teaches(david, db) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 19/45

44 Source of Unsatisfiability However, negative inclusions can cause a KB to be unsatisfiable: T : Dis(teaches, attends) A : teaches(john, db), attends(john, db) T : Funct(teaches ) A : teaches(john, db), teaches(david, db) T : Student Professor, teaches Professor A : Student(john), teaches(john, db) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 19/45

45 Source of Unsatisfiability However, negative inclusions can cause a KB to be unsatisfiable: T : Dis(teaches, attends) A : teaches(john, db), attends(john, db) T : Funct(teaches ) A : teaches(john, db), teaches(david, db) T : Student Professor, teaches Professor A : Student(john), teaches(john, db) Interaction of negative and positive inclusions has to be considered. calculate the closure of NIs w.r.t. PIs. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 19/45

46 Knowledge Base Satisfiability Given a DL-Lite A KB K = T, A, we check its satisfiability as follows: Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 20/45

47 Knowledge Base Satisfiability Given a DL-Lite A KB K = T, A, we check its satisfiability as follows: Algorithm for checking KB satisfiability 1 Calculate the closure of NIs. 2 Translate the closure into a UCQ q unsat asking for violation of some NI. 3 Evaluate encoding of q unsat into SQL over DB(A). if Eval(SQL(qunsat ), DB(A)) =, then the KB is satisfiable; otherwise the KB is unsatisfiable. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 20/45

48 Knowledge Base Satisfiability Given a DL-Lite A KB K = T, A, we check its satisfiability as follows: Algorithm for checking KB satisfiability 1 Calculate the closure of NIs. 2 Translate the closure into a UCQ q unsat asking for violation of some NI. 3 Evaluate encoding of q unsat into SQL over DB(A). if Eval(SQL(qunsat ), DB(A)) =, then the KB is satisfiable; otherwise the KB is unsatisfiable. Correctness of this procedure shows FOL-rewritability of KB satisfiability in DL-Lite. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 20/45

49 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

50 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). cln(t ) : Student Professor T : teaches Professor } Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

51 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

52 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

53 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends add to cln(t ): Professor registeredto Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

54 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends add to cln(t ): Professor registeredto } cln(t ) : Dis(teaches, attends) T : registeredto attends Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

55 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends add to cln(t ): Professor registeredto } cln(t ) : Dis(teaches, attends) T : registeredto attends add to cln(t ): Dis(teaches, registeredto) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

56 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends add to cln(t ): Professor registeredto } cln(t ) : Dis(teaches, attends) T : registeredto attends add to cln(t ): Dis(teaches, registeredto) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

57 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor (or Professor Student) T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends (or attends Professor) T : registeredto attends add to cln(t ): Professor registeredto } cln(t ) : Dis(teaches, attends) (or Dis(attends, teaches)) T : registeredto attends add to cln(t ): Dis(teaches, registeredto) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

58 Closure of Negative Inclusions Closure of NIs cln(t ) w.r.t. PIs every NI is in cln(t ). } cln(t ) : Student Professor T : teaches Professor add to cln(t ) : Student teaches } cln(t ) : Professor attends T : registeredto attends add to cln(t ): Professor registeredto } cln(t ) : Dis(teaches, attends) T : registeredto attends add to cln(t ): Dis(teaches, registeredto) Note: functionality does not interact with PIs and other NIs. Note: the closure is finite since there are polynomially many different NIs. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 21/45

59 Translation to FOL Queries Having calculated cln(t ) we translate it to a UCQ q unsat as follows. Each NI α correspond to a CQ, δ(α): Student teaches x.student(x) y.teaches(x, y). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 22/45

60 Translation to FOL Queries Having calculated cln(t ) we translate it to a UCQ q unsat as follows. Each NI α correspond to a CQ, δ(α): Student teaches x.student(x) y.teaches(x, y). Funct(teaches ) x 1, x 2, y.teaches(x 1, y) teaches(x 2, y) x 1 x 2. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 22/45

61 Translation to FOL Queries Having calculated cln(t ) we translate it to a UCQ q unsat as follows. Each NI α correspond to a CQ, δ(α): Student teaches x.student(x) y.teaches(x, y). Funct(teaches ) x 1, x 2, y.teaches(x 1, y) teaches(x 2, y) x 1 x 2. Then Dis(attends, teaches) x, y.attends(x, y) teaches(x, y). q unsat = α cln(t ) δ(α) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 22/45

62 Query evaluation Let q be a UCQ. We denote by SQL(q) the encoding of q into an SQL query over DB(A). We indicate with Eval(SQL(q), DB(A)) the evaluation of SQL(q) over DB(A). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 23/45

63 FOL-rewritability of satisfiability in DL-Lite A Theorem Let K = T, A be a DL-Lite A KB. Then, K is unsatisfiable iff Eval(SQL(q unsat, DB(A)) returns true. In other words, satisfiability of a DL-Lite A ontology can be reduced to FOL-query evaluation. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 24/45

64 Query Answering Problem Query answering over a KB K = T, A is a form of logical implication: find all tuples c of constants of A s.t. K = q( c) We are interested in so called certain answers, i.e., the tuples that are answers to q in all models of K = T, A : cert(q, K) = { c c q I, for every model I of K } Note: We have assumed that the answer q I to a query q over an interpretation I is constituted by a set of tuples of constants of A, rather than objects in I. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 25/45

65 Query Answering over Satisfiable KBs Given a CQ q and a satisfiable KB K = T, A, we compute cert(q, K) as follows: Algorithm for answering CQs over KBs 1 Using T, rewrite q into a UCQ r q,t (the perfect rewriting of q w.r.t. T ). 2 Encode r q,t into SQL and evaluate it over A managed in secondary storage via a RDBMS, to return cert(q, K). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 26/45

66 Query Answering over Satisfiable KBs Given a CQ q and a satisfiable KB K = T, A, we compute cert(q, K) as follows: Algorithm for answering CQs over KBs 1 Using T, rewrite q into a UCQ r q,t (the perfect rewriting of q w.r.t. T ). 2 Encode r q,t into SQL and evaluate it over A managed in secondary storage via a RDBMS, to return cert(q, K). Correctness of this procedure shows FOL-rewritability of query answering in DL-Lite. Query answering over DL-Lite ontologies can be done using RDBMS technology. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 26/45

67 Query Rewriting Consider the query q(x) Professor(x) Intuition: Use the PIs as basic rewriting rules: AssistantProf Professor as a logic rule: Professor(z) AssistantProf(z) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 27/45

68 Query Rewriting Consider the query q(x) Professor(x) Intuition: Use the PIs as basic rewriting rules: AssistantProf Professor as a logic rule: Professor(z) AssistantProf(z) Basic rewriting step: when an atom in the query unifies with the head of the rule, substitute the atom with the body of the rule. We say that the PI inclusion applies to the atom. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 27/45

69 Consider the query Query Rewriting q(x) Professor(x) Intuition: Use the PIs as basic rewriting rules: as a logic rule: Basic rewriting step: AssistantProf Professor Professor(z) AssistantProf(z) when an atom in the query unifies with the head of the rule, substitute the atom with the body of the rule. We say that the PI inclusion applies to the atom. In the example, the PI AssistantProf Professor applies to the atom Professor(x). Towards the computation of the perfect rewriting, we add to the input query above, the query q(x) AssistantProf(x) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 27/45

70 Consider the query and the PI Query Rewriting (cont d) q(x) teaches(x, y), Course(y) teaches Course as a logic rule: Course(z 2 ) teaches(z 1, z 2 ) The PI applies to the atom Course(y), and we add to the perfect rewriting the query q(x) teaches(x, y), teaches(z 1, y) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 28/45

71 Consider the query and the PI Query Rewriting (cont d) q(x) teaches(x, y), Course(y) teaches Course as a logic rule: Course(z 2 ) teaches(z 1, z 2 ) The PI applies to the atom Course(y), and we add to the perfect rewriting the query q(x) teaches(x, y), teaches(z 1, y) Consider now the query q(x) teaches(x, y) and the PI as a logic rule: Professor teaches teaches(z, f (z)) Professor(z) The PI applies to the atom teaches(x, y), and we add to the perfect rewriting the query q(x) Professor(x) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 28/45

72 Query Rewriting Constants Conversely, for the query and the same PI as before as a logic rule: q(x) teaches(x, databases) Professor teaches teaches(z, f (z)) Professor(z) teaches(x, databases) does not unify with teaches(z, f (z)), since the skolem term f (z) in the head of the rule does not unify with the constant databases. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 29/45

73 Query Rewriting Constants Conversely, for the query and the same PI as before as a logic rule: q(x) teaches(x, databases) Professor teaches teaches(z, f (z)) Professor(z) teaches(x, databases) does not unify with teaches(z, f (z)), since the skolem term f (z) in the head of the rule does not unify with the constant databases. In this case, the PI does not apply to the atom teaches(x, databases). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 29/45

74 Query Rewriting Constants Conversely, for the query and the same PI as before as a logic rule: q(x) teaches(x, databases) Professor teaches teaches(z, f (z)) Professor(z) teaches(x, databases) does not unify with teaches(z, f (z)), since the skolem term f (z) in the head of the rule does not unify with the constant databases. In this case, the PI does not apply to the atom teaches(x, databases). The same holds for the following query, where y is distinguished, since unifying f (z) with y would correspond to returning a skolem term as answer to the query: q(x, y) teaches(x, y) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 29/45

75 Query Rewriting Join variables An analogous behavior to the one with constants and with distinguished variables holds when the atom contains join variables that would have to be unified with skolem terms. Consider the query and the PI as a logic rule: q(x) teaches(x, y), Course(y) Professor teaches teaches(z, f (z)) Professor(z) The PI above does not apply to the atom teaches(x, y). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 30/45

76 Query Rewriting Reduce step Consider now the query q(x) teaches(x, y), teaches(z, y) and the PI as a logic rule: Professor teaches teaches(z, f (z)) Professor(z) This PI does not apply to teaches(x, y) or teaches(z, y), since y is in join, and we would again introduce the skolem term in the rewritten query. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 31/45

77 Query Rewriting Reduce step Consider now the query q(x) teaches(x, y), teaches(z, y) and the PI as a logic rule: Professor teaches teaches(z, f (z)) Professor(z) This PI does not apply to teaches(x, y) or teaches(z, y), since y is in join, and we would again introduce the skolem term in the rewritten query. However, we can transform the above query by unifying the atoms teaches(x, y) and teaches(z, y). This rewriting step is called reduce, and produces the query q(x) teaches(x, y) Now, we can apply the PI above, and add to the rewriting the query q(x) Professor(x) Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 31/45

78 Query Rewriting Algorithm Algorithm PerfectRef(Q, T P ) Input: union of conjunctive queries Q, set of DL-Lite A PIs T P Output: union of conjunctive queries PR PR := Q; repeat PR := PR; for each q PR do for each g in q do for each PI I in T P do if I is applicable to g then PR := PR { ApplyPI(q, g, I ) }; for each g 1, g 2 in q do if g 1 and g 2 unify then PR := PR {τ(reduce(q, g 1, g 2))}; until PR = PR; return PR Observations: Termination follows from having only finitely many different rewritings. NIs or functionalities do not play any role in the rewriting of the query. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 32/45

79 Query answering in DL-Lite Example TBox: Professor teaches teaches Course Query: q(x) teaches(x, y), Course(y) Perfect Rewriting: q(x) teaches(x, y), Course(y) q(x) teaches(x, y), teaches(, y) q(x) teaches(x, ) q(x) Professor(x) ABox: teaches(john, databases) Professor(mary) It is easy to see that evaluating the perfect rewriting over the ABox viewed as a database produces as answer {john, mary}. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 33/45

80 Query answering in DL-Lite Theorem Let T be a DL-Lite TBox, T P the set of PIs in T, q a CQ over T, and let r q,t = PerfectRef(q, T P ). Then, for each ABox A such that T, A is satisfiable, we have that cert(q, T, A ) = Eval(SQL(r q,t ), DB(A)). In other words, query answering over a satisfiable DL-Lite ontology is FOL-rewritable. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 34/45

81 Query answering in DL-Lite Theorem Let T be a DL-Lite TBox, T P the set of PIs in T, q a CQ over T, and let r q,t = PerfectRef(q, T P ). Then, for each ABox A such that T, A is satisfiable, we have that cert(q, T, A ) = Eval(SQL(r q,t ), DB(A)). In other words, query answering over a satisfiable DL-Lite ontology is FOL-rewritable. Notice that we did not mention NIs or functionality assertions of T in the result above. Indeed, when the ontology is satisfiable, we can ignore NIs and functionalities and answer queries as if they were not specified in T. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 34/45

82 Complexity of Reasoning in DL-Lite Theorem Checking satisfiability of DL-Lite A KBs is 1 PTime in the size of the KB (combined complexity). 2 AC 0 in the size of the ABox (data complexity). Theorem Query answering over DL-Lite A KBs is 1 NP-complete in the size of query and KB (combined comp.). 2 PTime in the size of the KB. 3 AC 0 in the size of the ABox (data complexity). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 35/45

83 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 36/45

84 EL and EL ++ EL is another family of tractable logics [2, 3]. it is expressive enough to model bio-medical ontologies like SNOMED; allows for horn inclusions and qualified existential restrictions: Heartdisease has-loc.heartvalve CriticalDisease Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 37/45

85 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 38/45

86 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 39/45

87 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept constructs: C, D ::= A {a} C D P.C Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 39/45

88 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept constructs: C, D ::= A {a} C D P.C TBox and ABox assertions: C D concept inclusion P 1 P n P complex role inclusion A(a) P(a, b) membership assertions Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 39/45

89 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept constructs: C, D ::= A {a} C D P.C TBox and ABox assertions: C D concept inclusion P 1 P n P complex role inclusion A(a) P(a, b) membership assertions An EL ++ Knowledge Base K is a pair T, A where T is a finite set of TBox axioms and A is a finite set of membership assertions. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 39/45

90 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept constructs: C, D ::= A {a} C D P.C TBox and ABox assertions: C D concept inclusion P 1 P n P complex role inclusion A(a) P(a, b) membership assertions An EL ++ Knowledge Base K is a pair T, A where T is a finite set of TBox axioms and A is a finite set of membership assertions. Note: the concrete domain constructor, which is a part of EL ++, is not presented here. Note: complex role inclusions allow expressing transitivity of roles (P P P) and role hierarchy (P 1 P 2). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 39/45

91 EL ++ Syntax Let N A, N P, N a be sets of concept, role and individual names, respectively. Let A N A, P N P, a N a. Concept constructs: C, D ::= A {a} C D P.C TBox and ABox assertions: C D concept inclusion P 1 P n P complex role inclusion A(a) P(a, b) membership assertions An EL ++ Knowledge Base K is a pair T, A where T is a finite set of TBox axioms and A is a finite set of membership assertions. Note: the concrete domain constructor, which is a part of EL ++, is not EL concept constructs and assertions. presented here. blabla Note: complex role inclusions allow expressing transitivity of roles blabla (P P P) and role hierarchy (P 1 P 2). blabla Botoeva Lightweight Description Logics: DL-LiteA and EL ++ 39/45

92 EL ++ Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 40/45

93 EL ++ Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept constructs ( ) I = I ( ) I = ({a}) I = {a I } (C D) I = C I D I ( P.C) I = {x I y C I, (x, y) P I } Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 40/45

94 EL ++ Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept constructs ( ) I = I ( ) I = ({a}) I = {a I } TBox and ABox assertions (C D) I = C I D I ( P.C) I = {x I y C I, (x, y) P I } I = C D iff C I D I I = P 1 P n P iff P I 1 P I n P I I = A(a) iff a I A I I = P(a, b) iff (a I, b I ) P I Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 40/45

95 EL ++ Semantics An interpretation I is a pair I, I : for every concept name A, A I I ; for every role name P, P I I I ; for every individual name a, a I I. Concept constructs ( ) I = I ( ) I = ({a}) I = {a I } TBox and ABox assertions (C D) I = C I D I ( P.C) I = {x I y C I, (x, y) P I } I = C D iff C I D I I = P 1 P n P iff P I 1 P I n P I I = A(a) iff a I A I I = P(a, b) iff (a I, b I ) P I I is a model of K = T, A if it satisfies all axioms of T and A. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 40/45

96 EL ++ : Example 2 Endocardium Tissue cont-in.heartwall cont-in.heartvalve HeartWall BodyWall part-of.heart HeartValve BodyValve part-of.heart Endocarditis Inflammation has-loc.endocardium Inflammation Disease acts-on.tissue Heartdisease has-loc.heartvalve CriticalDisease Heartdisease Disease has-loc.heart part-of part-of part-of part-of cont-in has-loc cont-in has-loc 2 taken from [4] Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 41/45

97 Outline 1 Description Logics 2 Description Logic DL-Lite A Syntax and Semantics of DL-Lite A Reasoning in DL-Lite A Knowledge Base Satisfiability Conjunctive Query Answering 3 Description Logic EL ++ Syntax and Semantics of EL ++ Reasoning in EL Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 42/45

98 Reasoning Problems The Concept Subsumption problem is to decide, given a TBox T and concepts C and D, whether for every model I of T it holds that C I D I. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 43/45

99 Reasoning Problems The Concept Subsumption problem is to decide, given a TBox T and concepts C and D, whether for every model I of T it holds that C I D I. The Conjunctive Query Entailment problem is to decide, given a DL-Lite A KB K and a boolean query q over K, whether K = q. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 43/45

100 Reasoning Problems The Concept Subsumption problem is to decide, given a TBox T and concepts C and D, whether for every model I of T it holds that C I D I. The Concept Satisfiability problem is to decide, given a TBox T and a concept C, whether there exist a model I of T such C I. The Knowledge Base satisfiability problem is to check, given a DL-Lite A KB K, whether K admits at least one model. The Conjunctive Query Entailment problem is to decide, given a DL-Lite A KB K and a boolean query q over K, whether K = q. The Instance Checking problem is to decide, given an object name a, a concept B, and a KB K = T, A, whether K = B(a). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 43/45

101 Theorem Complexity of Reasoning in EL Subsumption in EL ++ can be decided in polynomial time (a polytime tableax for deciding subsumption). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 44/45

102 Theorem Complexity of Reasoning in EL Subsumption in EL ++ can be decided in polynomial time (a polytime tableax for deciding subsumption). Theorem Entailment of conjunctive queries in EL ++ (already in EL + ) is undecidable. ([7],[6]). Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 44/45

103 Theorem Complexity of Reasoning in EL Subsumption in EL ++ can be decided in polynomial time (a polytime tableax for deciding subsumption). Theorem Entailment of conjunctive queries in EL ++ (already in EL + ) is undecidable. ([7],[6]). Theorem Entailment of unions of conjunctive queris in EL is: 1 PTime-complete with respect to data complexity; 2 PTime-complete with respect to KB complexity; 3 NP-complete with respect to combined complexity. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 44/45

104 Thank you for your attention! Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 45/45

105 A. Artale, D. Calvanese, R. Kontchakov, and M. Zakharyaschev. The DL-Lite family and relations. J. of Artificial Intelligence Research, 36:1 69, F. Baader, S. Brandt, and C. Lutz. Pushing the EL envelope. In Proc. of the 19th Int. Joint Conf. on Artificial Intelligence (IJCAI 2005), pages , F. Baader, C. Lutz, and B. Suntisrivaraporn. CEL a polynomial-time reasoner for life science ontologies. In Proc. of the 3rd Int. Joint Conf. on Automated Reasoning (IJCAR 2006), volume 4130 of Lecture Notes in Artificial Intelligence, pages Springer, F. Baader, C. Lutz, and B. Suntisrivaraporn. Efficient reasoning in EL+. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 45/45

106 In Proc. of the 19th Int. Workshop on Description Logic (DL 2006), volume 189 of CEUR Electronic Workshop Proceedings, D. Calvanese, G. De Giacomo, D. Lembo, M. Lenzerini, and R. Rosati. Tractable reasoning and efficient query answering in description logics: The DL-Lite family. J. of Automated Reasoning, 39(3): , A. Krisnadhi and C. Lutz. Data complexity in the EL family of description logics. In Proc. of the 14th Int. Conf. on Logic for Programming, Artificial Intelligence, and Reasoning (LPAR 2007), pages , R. Rosati. On conjunctive query answering in EL. Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 45/45

107 In Proc. of the 20th Int. Workshop on Description Logic (DL 2007), volume 250 of CEUR Electronic Workshop Proceedings, Botoeva Lightweight Description Logics: DL-Lite A and EL ++ 45/45