RDF Needs Annotations

While the current mechanism of reification in RDF is without semantics and widely considered inappropriate and cumbersome, some form of reification – speaking about triples themselves – is needed in RDF for many reasonable applications: in particular, reification allows for enhancing triples with annotations relating to provenance, spatio-temporal validity, degrees of trust, fuzzy values and/or other contextual information. In this position paper, we argue that – besides resolving the issue of how to syntactically represent reification in the future (i.e., whether to stick with the current reification mechanism or standardise a different mechanism such as Named Graphs) – it is time to agree on certain core annotations that are widely needed. We summarise existing work and provide a possible direction towards handling reification by means of a general annotation framework that can be instantiated for those major use cases we currently see arising. 1 A Need for RDF Annotations In this paper, we motivate and discuss the syntactic representation and semantic interpretation of generic RDF annotations. In particular, we propose the introduction of a generic semantic framework to represent annotations in RDF. Although these annotations can be used to represent arbitrary domains, we suggest a standard representation for some common domains such as time, space and provenance. Syntactically, RDF Annotations can be considered an extension of RDF by attaching to each triple, or to sets of triples, an annotation value. Thus, instead of the RDF triple composed of subject, predicate and object, we may refer to an RDF quad composed of the previous elements plus an annotation value. Currently, reification is the only standardised mechanism to add meta-statements about RDF triples; however, it remains without a semantic specification meaning that, e.g., reified statements are not affected by RDFS or OWL inferences. Despite the lack of standard semantics, several current activities in W3C’s Semantic Web activities would warrant an extension that enables one to annotate statements: • Web data represented in RDF will – in parts – be dynamic, be it the affiliation of a person changing (represented, e.g., by a foaf:workplaceHomepage triple), or more rapidly changing sensor data such as current temperature readings published online1. These use cases warrant annotations that talk about the temporal validity of triples. • Similarly, relying on the sensor example from before, a temperature reading may be valid for a certain location or area only. Likewise, semantically annotated observations made in a blog may be local. These examples warrant annotation that talk about geo-spatial validity of triples. • As discussed in the report [8] of the W3C’s Incubator group on “Uncertainty Reasoning for the World Wide Web”2 there are use cases where agents may want to represent and express vagueness of state∗Digital Enterprise Research Institute, National University of Ireland, Galway. [email protected] †Cisco Systems Inc. [email protected] ‡Istituto di Scienza e Tecnologie dell’Informazione (ISTI CNR), Pisa, Italy [email protected] 1W3C’s Semantic Sensor Network Incubator Group is currently taking first steps towards common formats for publishing sensor data, cf. http://www.w3.org/2005/Incubator/ssn/charter 2http://www.w3.org/2005/Incubator/urw3/