More, Faster: Accelerated Corpus Annotation with Statistical Taggers

Abstract

We present our experiments with annotating a Latin corpus using an assisted annotation procedure where the corpus to be annotated is preannotated by a statistical tagger. This assisted procedure gives a notable reduction in annotator error compared to the unassisted annotation of previous annotation efforts, even with a huge tagset (1 000 tags) and modest tagger accuracy due to limited training data and domain effects.