* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download PPT - LSDIS
List of first-order theories wikipedia , lookup
Analytic–synthetic distinction wikipedia , lookup
Fuzzy concept wikipedia , lookup
Jesús Mosterín wikipedia , lookup
Boolean satisfiability problem wikipedia , lookup
Junction Grammar wikipedia , lookup
Indeterminacy (philosophy) wikipedia , lookup
More on Description Logic(s) Frederick Maier Note Added 10/27/03 So, there are a few errors that will be obvious to some: Some of the symbols used in expressions are not in the right font (or even of the right type in some cases). Instance checking is not reducible to subsumption in every case (e.g., see this). (The) typical means of proof is based upon satisfiability (as the slides on semantic tableaux indicate); I should have pointed this out more explicitly. Again, most of the material is taken form Enrico Franconi’s course website (I believe he’s even the originator of the DL logo) . I’d like to take the presentation down, as it really offers nothing that couldn’t be found elsewhere just as readily, but I’ll wait until the end of the term. Overview The language basics Interpretations A Family of Languages Subsumption And other problems Complexity Assumption: We must understand the Syntax, Semantics, and Inference Mechanisms of these languages if we are to use them effectively. OWL The language in which our ontologies are going to be written in is likely going to be OWL, or something like it. And OWL is based in part on DL. What are DL’s? Key features: They are a family of Knowledge Representation languages with a formal semantics based largely on FOL. They attempt to discover “implicitly represented knowledge” using efficient inference mechanisms. The complexity of the inferences is an area of determined research. Basic Concepts of a DL Individuals (such as john and mary) Concepts (such as Man and Woman). Roles (such as hasChild). Basic Concepts of a DL Individuals are treated exactly the same as constants in FOL. Concepts are exactly the same as Unary Predicates in FOL. Roles are exactly the same as Binary Predicates in FOL. Descriptions Just Like in FOL, what we are dealing with (ultimately) are sets of individuals and relations between individuals. The basic unit of semantic significance is the Description. “We are describing sets of individuals” Defining Descriptions (ALC, a typical language) A description C or D can be: A T C C1 ∏ D1 C1 D1 R.C R.C an atomic concept. (top) the universal concept. (bottom) the null concept a negated concept the intersection of concepts. the union of two concepts. (restriction) (existential quantification). * * * * * * * [* present in AL. Only atomic concepts can be negated. restricted to R.T] Interpretations and Models Mostly, the formal semantics of a DL follows FOL: An individual is interpreted as an element from the universe of discourse. A concept is interpreted as the set of elements from the universe to which the concept applies. and and deserve special attention. Note that they only can come before a Role: HasChild.Girl isEmployedBy.Farmer Remember, they describe sets of individuals. and HasChild.Girl would be interpreted as: The set { x | (y)( HasChild(x,y) Girl(y) ) } [Note the conditional: Am I in that set?]. and isEmployedBy.Farmer would be: The set { x | (y)( isEmployedBy(x,y) & Farmer(y) ) } A family of languages The expressiveness of a description logic is determined by the operators that it uses. Add or Eliminate certain operators (e.g., , ), and the statements that can be expressed are increased/reduced in number. Higher expressiveness implies higher complexity. The Language AL A description C or D can be: A T C C1 ∏ D1 R.C R.T an atomic concept. (top) the universal concept. (bottom) the null concept a negated Atomic concept the intersection of concepts. (restriction) (Limited existential quantification). A family of languages Operation Notation Union (U) CB Complementation (C) C (Any Concept) Full Existential Quantification (E) R.C Cardinality (N) ≥ nR, ≤nR Qualified Cardinality (Q) ≥ nR.C, ≤nR.C Enumeration (O) {a,b,…} Selection (F) f:C Axioms We may assign names to complex descriptions: Bachelor ≡ Unmarried ∏ Male Or assert that one concept is subsumed by another: CD These are Axioms of the system. Subsumption A concept C subsumes a concept D iff I(D) I(C) on every interpretation I. This means the same as the assertion: (x)(D(x) C(x)) where D and C are complex statements The Subsumption Problem CD ? Determining whether one concept logically contains another is called the subsumption problem. Other Problems: Satisfiability of a Concept or KB {C, C} Instance Checking Father(john)? Equivalence CreatureWithHeart ≡ CreatureWithKidney Disjointness C∏D Retrieval Father(X)? X = {john, robert} Realization X(john)? X = {Father} Reduction These problems can be reduced to subsumption (for languages with negation). They can be reduced to the satisfiability problem, as well. Complexity The Subsumption Problem: It’s undecidable for reasonably expressive languages, It’s non-polynomial for fairly restricted languages. Complexity Language Subsumption Instance Checking FLAL ALE ALC P P NP PSPACE P P PSPACE PSPACE ALCO SHIQ KL-ONE PSPACE EXPTIME undecidable PSPACE EXPTIME undecidable OWL-Lite ? ? Inference Mechanisms ALC is equivalent to L2 and so, theoretically, we could translate all the expressions of the DL into L2 and then use resolution or some algorithm as a decision procedure. However, it is generally the case that Tableau algorithms are computationally less expensive. Tableau algorithms They work by systematically building up a tree of possible models to for a KB. If every branch of the tree possesses a contradiction, then the KB is unsatisfiable. Tableau proofs are sound and complete for many languages, including ALC. Complexity: Notes In complexity theory the class PSPACE is the set of decision problems that can be solved by a Turing machine using a polynomial amount of memory, and unlimited time. In complexity theory, EXPTIME is the set of all decision problems solvable by a deterministic Turing machine in O(2p(n)) time, where p(n) is a polynomial function of n. EXPTIME is known to be a subset of EXPSPACE and a superset of PSPACE, NP-complete, NP, and P. That is significant because it is currently unknown which (if any) of those four sets are equal to each other. It is known however that P is a strict subset of EXPTIME [From www.wikipedia.org] References The Description Logic Website: http://dl.kr.org/ Presentations from Enrico Franconi’s DL course*: http://www.inf.unibz.it/~franconi/dl/course/ Chapter 2 of the Description Logic Handbook: http://www.inf.unibz.it/~franconi/dl/course/dlhb/dlhb-02.pdf *Upon which this presentation is mostly based.