MIT Press

Table 3:

XLNet+prop segmentation and recognition results for different label sets granularities; in MRDA: full (51), general (11), basic (5), and pure segmentation (1); in SWDA basic (42) and pure segmentation (1). DER and JointWER are not defined for pure segmentation. All experiments are performed using full dialog context, with identical hyperparameters, except for the output layer size. The asterisk (*) denotes the label sets typically used in other works.

Case	Dataset	Tagset	micro_f1	macro_f1	DSER	SegWER	DER	JointWER
lower	MRDA	51	91.90	30.94	32.93	39.15	58.62	63.90
		12	94.07	48.39	35.51	40.56	48.72	49.42
		5^*	94.89	75.82	32.87	38.32	43.61	43.76
		1	96.74	95.23	32.85	38.94	–	–

	SWDA	42^*	95.57	54.86	17.48	24.09	37.51	44.38
		1	98.20	97.45	17.51	24.32	–	–

nolower	MRDA	51	93.85	40.65	13.88	17.38	45.22	49.11
		12	96.57	64.51	14.21	17.42	27.62	26.96
		5^*	97.55	85.67	14.15	16.85	22.29	18.92
		1	98.76	98.21	14.55	16.52	–	–

	SWDA	42^*	96.65	58.17	8.39	12.34	30.21	35.90
		1	99.22	98.89	8.37	12.18	–	–

Case	Dataset	Tagset	micro_f1	macro_f1	DSER	SegWER	DER	JointWER
lower	MRDA	51	91.90	30.94	32.93	39.15	58.62	63.90
		12	94.07	48.39	35.51	40.56	48.72	49.42
		5^*	94.89	75.82	32.87	38.32	43.61	43.76
		1	96.74	95.23	32.85	38.94	–	–

	SWDA	42^*	95.57	54.86	17.48	24.09	37.51	44.38
		1	98.20	97.45	17.51	24.32	–	–

nolower	MRDA	51	93.85	40.65	13.88	17.38	45.22	49.11
		12	96.57	64.51	14.21	17.42	27.62	26.96
		5^*	97.55	85.67	14.15	16.85	22.29	18.92
		1	98.76	98.21	14.55	16.52	–	–

	SWDA	42^*	96.65	58.17	8.39	12.34	30.21	35.90
		1	99.22	98.89	8.37	12.18	–	–

View Large

This Feature Is Available To Subscribers Only