The Evolution of Text Annotation Frameworks

Abstract

This chapter outlines the evolution of linguistic annotation frameworks. The aim is tutorial, describing older approaches that introduced basic ideas before showing how their various contributions have been combined and integrated into more modern frameworks. After a summary of typical annotation tasks and some open source tools that can perform them, we present two older ways to organize the tools into pipelines that ensure the annotation tasks are done in the correct order, first using traditional Linux scripts, then XML-based Ant buildfiles which give independence across operating systems. Manual and automatic annotation processes were integrated in WordFreak, which supported interactive visualization and editing of annotations through its graphical user interface, and also used a stand-off XML annotation format. These developments (pipeline configuration, platform independence, graphical interface, stand-off XML mark up) were successfully integrated into GATE and UIMA, the main large-scale modern annotation frameworks. UIMA added a type system that supports automatic validation of inputs and outputs between components in the pipeline. We present examples from both GATE and UIMA, and illustrate interoperability between frameworks with another older approach using XSLT transformations. The chapter ends by discussing the differences between annotation toolkits and annotation frameworks.