Towards a Better Understanding of Discourse: Integrating Multiple Discourse Annotation Perspectives Using UIMA

摘要

There exist various different discourse annotation schemes that vary both in the perspectives of discourse structure considered and the granularity of textual units that are annotated. Comparison and integration of multiple schemes have the potential to provide enhanced information. However, the differing formats of corpora and tools that contain or produce such schemes can be a barrier to their integration. U-Compare is a graphical, UIMA-based workflow construction platform for combining interoperable natural language processing (NLP) resources, without the need for programming skills. In this paper, we present an extension of U-Compare that allows the easy comparison, integration and visualisation of resources that contain or output annotations based on multiple discourse annotation schemes. The extension works by allowing the construction of parallel sub-workflows for each scheme within a single U-Compare workflow. The different types of discourse annotations produced by each sub-workflow can be either merged or visualised side-by-side for comparison. We demonstrate this new functionality by using it to compare annotations belonging to two different approaches to discourse analysis, namely discourse relations and functional discourse annotations. Integrating these different annotation types within an interoperable environment allows us to study the correlations between different types of discourse and report on the new insights that this allows us to discover.