Dependency relation scores

For each UD dependency relation, LAS shows the percentage of gold tokens for which the system predicted both the correct syntactic head and the correct relation. The table is sorted by the number of examples in the manually annotated test set.

Part-of-speech tag scores

For each universal part-of-speech tag, Accuracy shows the percentage of gold tokens the system tagged correctly. The table is sorted by the number of examples in the manually annotated test set.

Every word whose labelled attachment is wrong falls into exactly one of three categories, so the three totals below account for all of them.

Where the predicted annotation differs from the gold annotation, by layer.

Complete output of the evaluation script for this run: every metric family the evaluator reports, with precision, recall, F1 and aligned accuracy.