Translators for interoperating and porting object-relational knowledge

Abstract

With the growth of analytics for knowledge-from-data extraction, Knowledge Bases (KBs) have been increasingly represented using various, expressively overlapping, rule and ontology languages. We will focus on rule knowledge, which includes factual data as a special case. Relational rule KBs, on top of relational data, have been widely represented in languages centered on n-ary relationships. Knowledge – including rules – centered on objects and corresponding languages have been studied in Artificial Intelligence and Semantic Web as graph-based representations (frames). Combining relational and object-centered rules, object-relational rule languages have been proposed for leveraging the advantages of the two other rule paradigms: The heterogeneous approach allows both relationships and frames to be separately used while the homogeneous approach generalizes them to a uniform construct. This dissertation advances homogeneous object-relational combinations by studying and realizing translators into and out of the Positional-Slotted Object-Applicative (PSOA) RuleML language. Two translators interoperate from PSOA to the relational languages TPTP (PSOA2TPTP) and Prolog (PSOA2Prolog); another one, from the object-centered language N3 to PSOA (N3-to-PSOA). All three translators are used for interoperating between rule languages in different paradigms. Moreover, the PSOA2TPTP and PSOA2Prolog translators are used for porting PSOA KBs and queries. An interoperation and portation architecture is proposed. It consists of two frameworks whose instantiations share the translator components. Each translator interoperating from PSOA realizes a composition of (1) a normalization within PSOA followed by (2) a conversion from the normalized PSOA to the target language. The normalization is itself sequentially composed of modular transformation steps within PSOA, which are shared for different targets. The transformation steps are formalized mathematically and realized as Java modules, partly generated from ANTLR grammars. We characterize the PSOA sublanguages for which the transformation steps, the conversions to TPTP and Prolog, as well as their compositions are semantics-preserving. The proofs of semantics preservation are given. The PSOA2TPTP and PSOA2Prolog translators are combined with reasoning engines into two instantiations of the PSOATransRun portation framework, which provide two implementations of PSOA RuleML query answering. The practicality of PSOA and PSOATransRun is supported by two use cases and evaluations of several test cases. The efficiency of relational rules is fully retained and the efficiency of certain object-centered and object-relational rules increases through our PSOA RuleML and PSOATransRun technology. Based on the findings in translator development and PSOATransRun evaluation, the PSOA RuleML language has been revised to achieve Version 1.0.