MIT Press

Skip to Main Content

Table 6:

Accuracy of supervised models. p ¡ {0.05^⋆, 0.001^‡} with paired bootstrap compared to BERT.

		Normative Arguments						Non-normative Arguments
		ACC	AUC	F1	F1_sup	F1_att	F1_neu	ACC	AUC	F1	F1_sup	F1_att	F1_neu
1	TGA Net	71.5	88.3	62.2	43.5	54.3	88.7	76.6	90.8	69.8	62.9	53.9	92.5
2	Hybrid Net	66.8	78.2	56.2	42.9	42.4	83.4	71.8	82.2	65.7	55.6	51.4	90.2

3	BERT	79.5	92.4	73.3	60.5	65.2	94.2	83.8	94.6	79.2	72.3	68.8	96.6
4	BERT+LX	79.2	92.1	72.7	58.7	65.6^⋆	93.8	83.7	94.6	79.2	70.8	69.9^‡	96.9^‡
5	BERT+MT	79.3	92.6^⋆	73.4	63.8^‡	63.6	92.7	83.6	94.7	79.2	71.8	69.7^‡	96.1

6	LogBERT	80.0^‡	92.8^‡	74.3^‡	63.6^‡	66.2^‡	93.2	84.3^‡	95.0^‡	80.2^‡	73.1^‡	71.4^‡	96.1
(a) Kialo

		Normative Arguments						Non-normative Arguments
		ACC	AUC	F1	F1_sup	F1_att	F1_neu	ACC	AUC	F1	F1_sup	F1_att	F1_neu
1	TGA Net	71.5	88.3	62.2	43.5	54.3	88.7	76.6	90.8	69.8	62.9	53.9	92.5
2	Hybrid Net	66.8	78.2	56.2	42.9	42.4	83.4	71.8	82.2	65.7	55.6	51.4	90.2

3	BERT	79.5	92.4	73.3	60.5	65.2	94.2	83.8	94.6	79.2	72.3	68.8	96.6
4	BERT+LX	79.2	92.1	72.7	58.7	65.6^⋆	93.8	83.7	94.6	79.2	70.8	69.9^‡	96.9^‡
5	BERT+MT	79.3	92.6^⋆	73.4	63.8^‡	63.6	92.7	83.6	94.7	79.2	71.8	69.7^‡	96.1

6	LogBERT	80.0^‡	92.8^‡	74.3^‡	63.6^‡	66.2^‡	93.2	84.3^‡	95.0^‡	80.2^‡	73.1^‡	71.4^‡	96.1
(a) Kialo