Incremental Individual Reasoning in ProtoDL
暂无分享,去创建一个
1. Requirements This paper continues our explorations 1, 3, 2] into the implementation of extensible DL reasoners that try to maintain a model-like data-structure, rather than using a refutation technique 1 A DL-KBMS 2 typically manages a set of named concept deenitions (\classes"), including maintaining an up-to-date taxonomy of them organized by speciicity according to subsumption. In practice, the DL-KBMS must also manage information about individuals, which accumulate information in the form of (i) associated descriptions (Anna has all(heightInCM,min(0.90)), and (ii) llers for roles, describing the relationships between individuals (Anna has Mimi as sister). For this purpose, the KBMS ooers operations createInd, tellClass, tellFiller and tellClosed. (The latter is needed to deal with the open-world assumption: tellClosed(Anna,sister) says there are no more sisters to come. There are of course corresponding retrieval operations: askClass, askFiller, askClosed. More commonly, users wish to obtain the list of all individuals known to be llers for a role: getallFillers(Ind,role), and the lowest class(es) in the taxonomy to which an individual belongs: askParents(Ind). In order to detect inconsistencies as early as possible, and as a general principle, our system performs "eager reasoning", computing and caching the answers to the last three kinds of questions for each individual. A crucial property of individual processing is that the acquisition of information about individuals is incremental. Whenever any new fact is asserted (either explicitly by a user or as a result of some inference) some of the following kinds of inferences may result: Additional facts (descriptions or ller information) can be inferred about the individual (e.g., learning that a role has no llers implies the individual satisses every all restriction). New facts can be inferred about other individuals (e.g., after tellFiller(Anna,sibling,Mimi) we can infer that Mimi is an instance of FEMALE, if we already know that Anna has all(siblings,FEMALE)). An individual may become "inconsistent" either because the conjunction of the descriptions asserted about it are incoherent (this is detected by concept-level reasoning, not discussed here) or because of connict between the descriptor and ller information (e.g., more llers added to an attribute than permitted by an at-most restriction). As a result of a change in an individual, it or some other individual may need to be reclassiied with respect to the taxonomy of classes. The data structure for every individual includes both a concept (to be called Descriptor), obtained by conjoining and normalizing the descriptions successively asserted about the …
[1] Andrea Schaerf. Query Answering in Concept-Based Knowledge Representation Systems: Algorithms, Complexity, and Seman , 1994 .
[2] Deborah L. McGuinness,et al. Reasoning with Black Boxes: Handling Test Concepts in CLASSIC , 1996, Description Logics.