The Perfect Match: RPL and RDF Rule Languages

Path query languages have been previously shown to complement RDF rule languages in a natural way and have been used as a means to implement the RDFS derivation rules. RPL is a novel path query language specifically designed to be incorporated with RDF rules and comes in three flavors: Node-, edgeand path-flavored expressions allow to express conditional regular expressions over the nodes, edges, or nodes and edges appearing on paths within RDF graphs. Providing regular string expressions and negation, RPL is more expressive than other RDF path languages that have been proposed. We give a compositional semantics for RPL and show that it can be evaluated efficiently, while several possible extensions of it cannot. Graph traversal operators play a crucial role in rule and query languages for semi-structured data and for RDF rule languages in particular. This need bas been acknowledged by the development of languages like Versa [11] SPARQLeR [7] and nSPARQL [12] and underlined in [2]. Moreover, the need for traversal of semi-structured data in general, and XML in particular is underscored by the huge success of XPath, arguably the most prominent XML query language. In [12] it has been shown that SPARQL augmented with conditional (in the sense of [9]) regular path expressions is expressive enough to query RDF graphs under the RDFS semantics without computing the closure of the graph under the RDFS entailment rules. Most path query languages proposed up until now are unfit for clean integration with RDF rule languages for the following reasons: (i) their use of variables interferes with the use of logical variables already present in rule languages, (ii) they do not always evaluate to pairs of nodes and thus cannot be safely used at the place of RDF predicates in query patterns and (iii) they lack negation, regular string expressions and often also conditional operators. We propose the RDF path language RPL, that is designed for easy integration with RDF rule languages such as SPARQL [14], Xcerpt [5, 13] and RDFLog [4, 3]. RPL is an orthogonal extension to RDF rule languages in that it sets out to extend RDF rule languages by features they lack, and in that it tries to avoid duplication of features they already provide. RPL expressions always evaluate to pairs of nodes within an RDF graph, and can thus be safely used at the place of predicates within the body of RDF rules. Despite of this restriction, SPARQL extended with RPL predicates is capable, just as nSPARQL, to query RDF graphs under the RDFS semantics without computing the closure of the queried graphs under the RDFS entailment rules. RPL is more expressive than previously proposed RDF query languages in that it provides regular string expressions and negation. RDF Path Expressions (RPEs) come in three flavors: node-restricting, edgerestricting and path-restricting, identified by the keywords NODES, EDGES, PATH, respectively. Node-restricting (edge-restricting) RPEs only place restrictions on the nodes (edges) appearing within a path. Path-restricting expressions may place restrictions on both, nodes and edges. RPEs evaluate to sets of pairs of nodes – i.e. binary relations over the set N of nodes of an RDF graph. The three unrestrictive RPEs [PATH ( )∗], [EDGES ∗] and [NODES ∗] evaluate to N ×N . This paper is organized as follows: Section 1 informally introduces the semantics of RPL by example, before its syntax and semantics is formally defined in Sections 2 and 3. Section 5 compares RPL to related path query languages and comes up with first complexity results. Section 6 shows the tractability of RPL as a whole, and the intractability of node and edge flavored path RPL expressions augmented with unordered paths. The contributions of this paper are as follows: (i) We formalize the syntax and semantics of RPL expressions, and (ii) show that RPL can express all relevant RDFS queries. (iii) We show that RPL can be evaluated efficiently, and (iv) that also nSPARQL could be extended by regular string expressions and negation without sacrificing tractability. (v) Finally we show that extensions of RPL and nSPARQL to unordered paths results in the loss of the tractability of both languages.