数据与知识工程3HornLogic.pptVIP

教师：常亮 E-mail: changl@guet.edu.cn 办公室电话: 2291071 手机: 办公室：7212 数据与知识工程 欢迎参加 Undecidability of FOL FOL is very powerful can be used as a full programming language just as there is no way to detect in general when a program is looping There can be no procedure that does this: Proc[Clauses] = If Clauses are unsatisfiable then return YES else return NO One approach: limit ourselves to a certain interesting subset of FOL, where the Resolution procedure becomes much more manageable, as well as this subset is still sufficiently expressive for many purposes. Horn clauses Clauses: as disjunctions: e.g. [Rain, Sleet, Snow] as implications: e.g. [?Child, ?Male, Boy] Here we focus on the second use. Horn clause: clause containing at most one positive literal. positive / definite clause: containing exactly one positive literal e.g. [?p1, ?p2, ..., ?pn, q] read as: If p1 and p2 and ... and pn, then q write as: p1 ? p2 ? ... ? pn ? q to emphasize the “if-then” reading. Sometimes called “rules”. negative clause: there are no positive literals Resolution with Horn clauses Only two possibilities: Neg Pos Neg It is possible to rearrange derivations of negative clauses so that all new derived clauses are negative. Pos Pos Pos Further restricting resolution Can also change derivations such that each derived clause is a resolvent of the previous derived one (negative) and some positive clause in the original set of clauses. Since each derived clause is negative, one parent must be positive (and so from original set) and one parent must be negative. Chain backwards from the final negative clause until both parents are from the original set of clauses. Eliminate all other clauses not on this direct path. SLD Resolution Write as: SLD means S(elected) literals L(inear) form D(efinite) clauses Note: SLD derivation is just a special form of derivation. SLD Resolution for Horn clauses Theorem. SLD-Resolution