Making UIMA Truly Interoperable with SPARQL
Rafał Rak, Sophia Ananiadou · Research Explorer (The University of Manchester) · 2013
Unstructured Information Management Architecture (UIMA) has been gaining popularity in annotating text corpora. The architecture defines common data structures and interfaces to support interoperability of individual processing components working together in a UIMA application. The components exchange data by sharing common type systems—schemata of data type structures—which extend a generic, top-level type system built into UIMA. This flexibility in extending type systems has resulted in the development of repositories of components that share one or several type systems; however, components coming from different repositories, and thus not sharing type systems, remain incompatible. Commonly, this problem has been solved programmatically by implementing UIMA components that perform the alignment of two type systems, an arduous task that is impractical with a growing number of type systems. We alleviate this problem by introducing a conversion mechanism based on SPARQL, a query language for the data retrieval and manipulation of RDF graphs. We provide a UIMA component that serialises data coming from a source component into RDF, executes a user-defined, typeconversion query, and deserialises the updated graph into a target component. The proposed solution encourages ad hoc conversions, enables the usage of heterogeneous components, and facilitates highly customised UIMA applications.