Table 7. Accuracy on dependency parsing for fine-tuned models and Chinese BERT.

The values with the bold style are the maximum values in each column among these methods or models.

Methods	Chinese BERT	WS-BERT	POS-BERT	CQ-BERT	NLI-BERT
	UD
Attn	51.63	41.14	44.48	43.24	7.02
Attn + embeddings	67.68	51.35	53.98	53.09	10.88
	CDT
Attn	54.00	49.93	56.62	53.20	7.07
Attn + embeddings	68.24	64.21	69.59	66.80	10.17