Large Scale Arabic Error Annotation: Guidelines and Framework

Abstract

We present annotation guidelines and a web-based annotation framework developed as part of an effort to create a manually annotated Arabic corpus of errors and corrections for various text types. Such a corpus will be invaluable for developing Arabic error correction tools, both for training models and as a gold standard for evaluating error correction algorithms. We summarize the guidelines we created. We also describe issues encountered during the training of the annotators, as well as problems that are specific to the Arabic language that arose during the annotation process. Finally, we present the annotation tool that was developed as part of this project, the annotation pipeline, and the quality of the resulting annotations.

Resources

Details