Annotation Tools for Large-Scale Corpus Development: Using AGTK at the Linguistic Data Consortium

Large-scale corpus development demands substantial infrastructure. As part of this infrastructure, the Linguistic Data Consortium (LDC) has adopted the Annotation Graph Toolkit (AGTK) as a primary resource for annotation tool development. This paper reports on LDC’s experiences using AGTK to develop and implement highly customized annotation tools for a variety of large-scale corpus creation efforts. We describe two primary tools that are currently in active use at LDC, one speechand one text-based, as well as other new AGTKbased annotation tools. We also describe the use of AGTK to develop tools for comparing and adjudicating divergent annotations in order to produce gold standard evaluation data and to measure inter-annotator consistency. Finally, we discuss various issues in creating AGTK-based tools across a wide range of annotation tasks and divergent research areas.