• Keine Ergebnisse gefunden

5 Reduction to SAT

5.4 Completeness of the Reduction

Given a ground unifier γ of Γ w.r.t.O, we can define a valuationτ that satisfies C(Γ,O) as follows.

Let L ∈ Left and D ∈ Attr and i ∈ {0, . . . ,|T |}. We set τ([L v D]i) := 1 iff s(γ(L))←−(i)O s(γ(D)). According to Corollary 8, we thus have τ([L v D]i) = 0 for all i∈ {0, . . . ,|T |} iff γ(L)6vO γ(D). Otherwise, there is an i∈ {0, . . . ,|T |}

such that τ([LvD]j) = 1 for all j ≥i, and τ([LvD]j) = 0 for all j < i.

To define the valuation of the remaining propositional variables [X > Y] with X, Y ∈ Nv, we set τ([X > Y]) = 1 iff X >γ Y, where >γ is defined as before, i.e., X >γ Y iff γ(X)vO ∃w.γ(Y) for somew∈NR+.

Lemma 29. Let C ∈ At, D ∈ Atnv, and E be a top-level atom of γ(C) with E vsO γ(D). Then Dec(C vD) is evaluated to 1 under τ.

Proof. Consider the following cases:

• If C=D, then Dec(C vD) =>.

• If both C and D are ground, then Dec(C v D) = [C v D]|T | and γ(C) = C =E vO D =γ(D). By Corollary 8, this impliesγ(C)←−−(|T |)O γ(D), i.e., τ([C vD]|T |) = 1.

• IfC is a concept name, then we again haveγ(C)vO E vO γ(D), and thus τ([C vD]|T |) = 1.

• If C = ∃r.C0, then γ(C) is an atom, and thus E = γ(C), D = ∃s.D0, rEO s, and either γ(C0) vO γ(D0) or γ(C0) vO γ(∃t.D0) for a transitive role name t with r EO t EO s. We obtain either τ([C0 v D0]|T |) = 1 or τ([C0 v ∃t.D0]|T |) = 1, i.e., Trans(C vD) is evaluated to 1 under τ.

Lemma 30. The valuation τ satisfies C(Γ,O).

Proof. We consider the formulae introduced in Definition 24.

(I) Since γ is a unifier of Γ w.r.t. O, for every Lv? D in Γ we have γ(L)vO

γ(D). By Corollary 8, this implies γ(L) ←−−(|T |)O γ(D), and thus τ([L v D]|T |) = 1.

(II) 1) If C vO D does not hold for two ground atoms C ∈ At, D ∈ Attr, then γ(C) vO γ(D) does not hold, and thus τ([C v D]i) = 0 for all i∈ {0, . . . ,|T |}.

2) If Y ∈ Nv, B ∈ Atnv, L ∈ Left, and i, j ∈ {0, . . . ,|T |} are such that τ([L v Y]i) = τ([Y vB]j) = 1, then γ(L) ←−(i)O γ(Y) ←−(j)O γ(B), and thus γ(L) ←−−−(i+j)O γ(B). By Lemma 7, there must be a derivation of γ(L) from γ(B) that uses at most min{|T |, i+j} root steps, and thus τ([LvB]min{|T |,i+j}) = 1.

3) LetL∈Left\Nv,D∈Attr, andi∈ {0, . . . ,|T |}satisfyτ([LvD]i) = 1, i.e., γ(L)←−(i)O γ(D).

a) IfDis a ground atom andLis not a ground atom, then by Lemma 11 we have one of the following cases:

• If there is C ∈ L such that E vsO D for some top-level atom E of γ(C), then by Lemma 29 Dec(C vD) is evaluated to 1.

• Otherwise, there must be a GCI A1 u · · · uAk v B in T with γ(L) ←−−−(i−1)O A1, . . . , γ(L)←−−−(i−1)O Ak, and B vsO D. Note that this is only possible withi >0. We then have τ([L vA1]i−1) =

· · ·=τ([LvAk]i−1) = 1 and B vO D.

b) IfDis a non-variable, non-ground atom, then by Lemma 11 we have one of the following cases:

• If there isC ∈ L such that E vsO γ(D) for some top-level atom E of γ(C), then by Lemma 29 Dec(C vD) is evaluated to 1.

• If there is a GCI A1u · · · uAk vB ∈ T with γ(L) ←−(i)O B and B vsO γ(D), then τ([LvB]i) = 1 and Dec(B vD) is evaluated to 1.

(III) 1) Assume that τ([X > X]) = 1 for some variable X ∈ Nv. By the definition of>γ, this implies that γ(X)vO ∃w.γ(X) for somew∈NR+. This contradicts the assumption that O is cycle-restricted.

Moreover, if τ([X > Y]) = τ([Y > Z]) = 1, then γ(X) vO ∃ww0.γ(Z) with w, w0 ∈NR+, and thusτ([X > Z]) = 1.

2) Ifτ([X v ∃r.Y]i) = 1, thenγ(X)←−(i)O ∃r.γ(Y), which impliesγ(X)vO

∃r.γ(Y). By the definition of >γ, we thus haveτ([X > Y]) = 1.

This completes the proof of correctness of the presented reduction.

Theorem 31. Unification w.r.t. cycle-restricted ELHR+-ontologies is an NP -complete problem.

Proof. NP-hardness follows from NP-hardness of unification in EL w.r.t. the empty ontology [5]. The other direction is provided by the presented reduction to the NP-complete SAT problem.

This also shows locality of unification w.r.t. cycle-restricted ELHR+-ontologies, i.e., in this setting a unification problem has a unifier iff it has a local unifier.

Lemma 32. If a flat unification problem Γ has a unifier w.r.t. a flat, cycle-restricted ELHR+-ontology O, then it has a local unifier w.r.t. O.

Proof. If Γ has a unifier w.r.t.O, thenC(Γ,O) is satisfiable by Lemma 30. Thus, Section 5.3 shows that there is a local unifier of Γ w.r.t. O.

6 Conclusions

We have shown that unification w.r.t. cycle-restricted ELHR+-ontologies can be reduced to propositional satisfiability. This improves on the results in [2] in two respects. First, it allows us to deal also with ontologies that contain transitivity and role hierarchy axioms, which are important for medical ontologies. Second, the SAT reduction can easily be implemented and enables us to make use of highly optimized SAT solvers, whereas the goal-oriented algorithm in [1], while having the potential of becoming quite efficient, requires a high amount of additional optimization work. The main topic for future research is to investigate whether we can get rid of cycle-restrictedness.

References

[1] Franz Baader, Stefan Borgwardt, and Barbara Morawska. Unification in the description logic EL w.r.t. cycle-restricted TBoxes. LTCS-Report 11-05, Chair for Automata Theory, Institute for Theoretical Computer Science, Technische Universit¨at Dresden, Dresden, Germany, 2011. See http://lat.inf.tu-dresden.de/research/reports.html.

[2] Franz Baader, Stefan Borgwardt, and Barbara Morawska. Extending uni-fication in EL towards general TBoxes. In Proc. of the 13th Int. Conf.

on Principles of Knowledge Representation and Reasoning (KR’12). AAAI Press, 2012. Short paper.

[3] Franz Baader, Sebastian Brandt, and Carsten Lutz. Pushing the EL enve-lope. In Leslie Pack Kaelbling and Alessandro Saffiotti, editors, Proc. of the 19th Int. Joint Conf. on Artificial Intelligence (IJCAI’05), pages 364–369.

Professional Book Center, 2005.

[4] Franz Baader, Sebastian Brandt, and Carsten Lutz. Pushing the EL enve-lope. LTCS-Report 05-01, Chair for Automata Theory, Institute for Theoret-ical Computer Science, Technische Universit¨at Dresden, Dresden, Germany, 2005. See http://lat.inf.tu-dresden.de/research/reports.html.

[5] Franz Baader and Barbara Morawska. Unification in the description logicEL.

In Ralf Treinen, editor,Proc. of the 20th Int. Conf. on Rewriting Techniques and Applications (RTA’09), volume 5595 ofLecture Notes in Computer Sci-ence, pages 350–364. Springer-Verlag, 2009.

[6] Franz Baader and Barbara Morawska. SAT encoding of unification in EL.

In Christian G. Ferm¨uller and Andrei Voronkov, editors, Proc. of the 17th Int. Conf. on Logic for Programming and Automated Reasoning (LPAR’10),

volume 6397 of Lecture Notes in Computer Science, pages 97–111. Springer-Verlag, 2010.

[7] Franz Baader and Barbara Morawska. Unification in the description logic EL. Logical Methods in Computer Science, 6(3), 2010. Special Issue: 20th Int. Conf. on Rewriting Techniques and Applications (RTA’09).

[8] Franz Baader and Paliath Narendran. Unification of concept terms in de-scription logics. Journal of Symbolic Computation, 31(3):277–305, 2001.

[9] Franz Baader and Tobias Nipkow. Term Rewriting and All That. Cambridge University Press, United Kingdom, 1998.

[10] Franz Baader and Wayne Snyder. Unification theory. In John Alan Robinson and Andrei Voronkov, editors, Handbook of Automated Reasoning, pages 445–532. The MIT Press, 2001.

[11] Armin Biere, Marijn Heule, Hans van Maaren, and Toby Walsh, editors.

Handbook of Satisfiability. IOS Press, 2009.

[12] Sebastian Brandt. Polynomial time reasoning in a description logic with existential restrictions, GCI axioms, and - what else? In Ramon L´opez de M´antaras and Lorenza Saitta, editors, Proc. of the 16th Eur. Conf. on Artificial Intelligence (ECAI’04), pages 298–302. IOS Press, 2004.

[13] James R. Campbell, Alejandro Lopez Osornio, Fernan de Quiros, Daniel Luna, and Guillermo Reynoso. Semantic interoperability and SNOMED CT:

A case study in clinical problem lists. In K.A. Kuhn, J.R. Warren, and T.-Y. Leong, editors, Proc. of the 12th World Congress on Health (Medical) Informatics (MEDINFO 2007), pages 2401–2402. IOS Press, 2007.

[14] Anatoli Degtyarev, Yuri Gurevich, Paliath Narendran, Margus Veanes, and Andrei Voronkov. Decidability and complexity of simultaneous rigid E-unification with one variable and related results. Theoretical Computer Sci-ence, 243(1-2):167–184, 2000.

[15] Anatoli Degtyarev and Andrei Voronkov. The undecidability of simultaneous rigid E-unification. Theoretical Computer Science, 166(1–2):291–300, 1996.

[16] Jean Gallier, Paliath Narendran, David Plaisted, and Wayne Snyder. Rigid E-unification: NP-completeness and applications to equational matings. In-formation and Computation, 87(1/2):129–195, 1990.

[17] Jean Goubault. Rigid E-unifiability is DEXPTIME-complete. In~ Proc. of the 9th Annual IEEE Symp. on Logic in Computer Science (LICS’94), pages 498–506. IEEE Computer Society Press, 1994.

[18] Yuri Gurevich and Andrei Voronkov. Monadic simultaneous rigid E-unification. Theoretical Computer Science, 222(1-2):133–152, 1999.

[19] Carsten Lutz and Frank Wolter. Deciding inseparability and conservative extensions in the description logic EL. Journal of Symbolic Computation, 45(2):194–228, 2010.

[20] Julian Seidenberg and Alan L. Rector. Representing transitive propagation in OWL. In David W. Embley, Antoni Oliv´e, and Sudha Ram, editors, Proc. of the 25th Int. Conf. on Conceptual Modeling (ER’06), volume 4215 ofLecture Notes in Computer Science, pages 255–266. Springer-Verlag, 2006.

[21] Viorica Sofronie-Stokkermans. Locality and subsumption testing in EL and some of its extensions. In Carlos Areces and Robert Goldblatt, editors,Proc.

of the 7th Conf. on Advances in Modal Logic (AIML’08), pages 315–339.

College Publications, 2008.

[22] Boontawee Suntisrivaraporn, Franz Baader, Stefan Schulz, and Kent Spack-man. Replacing SEP-triplets in SNOMED CT using tractable descrip-tion logic operators. In Riccardo Bellazzi, Ameen Abu-Hanna, and Jim Hunter, editors,Proc. of the 11th Conf. on Artificial Intelligence in Medicine (AIME’07), volume 4594 of Lecture Notes in Computer Science, pages 287–

291. Springer-Verlag, 2007.

[23] Margus Veanes. On Simultaneous Rigid E-Unification. PhD thesis, Uppsala University, 1997.