Tackling interoperability issues within UIMA workflows

Nicolas Hernandez (1)

1 LINA - Laboratoire d'Informatique de Nantes Atlantique

One of the major issues dealing with any workflow management frameworks is the components interoperability. In this paper, we are concerned with the Apache UIMA framework. We address the problem by considering separately the development of new components and the integration of existing tools. For the former objective, we propose an API to generically handle TS objects by their name using reflexivity in order to make the components TS-independent. In the latter case, we distinguish the case of aggregating heterogeneous TS-dependent UIMA components from the case of integrating non UIMA-native third party tools. We propose a mapper component to aggregate TS-dependent UIMA components. And we propose a component to wrap command lines third party tools and a set of components to connect various markup languages with the UIMA data structure. Finally, we present two situations where these solutions were effectively used: Training a POS tagger system from a treebank, and embedding an external POS tagger in a workflow. Our approch aims at providing quick development solutions.

Mots clés: LR Infrastructures and Architectures

Domaines: Traitement du texte et du document

Nicolas Hernandez. Tackling interoperability issues within UIMA workflows. Language Resources and Evaluation (LREC'12), May 2012, Istanbul, Turkey. pp.3618-3625, 978-2-9517408-7-7. ⟨hal-00709572⟩