Table A1.

Performance of different rationalization approaches on the MovieReviews, SST, and FEVER datasets.

Dataset	Rationale	Approach	F1	Suff	Comp	References
MovieReviews	Extractive	Pipeline	0.77	0.88	0.10	Atanasova et al., 2024
	Extractive	Pipeline	0.84	0.89	0.09	Guerreiro and Martins, 2021
	Extractive	Pipeline	0.91	0.95	0.12	Chan A. et al., 2022
	Extractive	MT Unsupervised	0.91	0.93	0.11	Lei et al., 2016
	Extractive	MT Unsupervised	0.94	0.92	0.12	Paranjape et al., 2020
	Extractive	MT Unsupervised	0.90	0.91	0.15	Carton et al., 2020
	Extractive	MT Supervised	0.92	0.93	0.14	Lei et al., 2016
	Extractive	MT Supervised	0.96	0.91	0.16	DeYoung et al., 2020
	Abstractive	MT Text-to-Text	0.97	0.89	0.11	Narang et al., 2020
SST	Extractive	Pipeline	0.80	0.75	0.11	Guerreiro and Martins, 2021
	Extractive	Pipeline	0.93	0.89	0.11	Chan A. et al., 2022
	Extractive	MT Unsupervised	0.92	0.95	0.15	Carton et al., 2020
	Abstractive	Generative Pipelined	0.90	0.79	0.07	Zhao and Vydiswaran, 2021
FEVER	Extractive	MT Unsupervised	0.71	0.85	0.05	DeYoung et al., 2020
	Extractive	Pipeline	0.70	0.89	0.07	Guerreiro and Martins, 2021
	Extractive	MT Unsupervised	0.82	0.85	0.15	Carton et al., 2020
	Extractive	MT Supervised	0.85	0.87	0.14	DeYoung et al., 2020
	Extractive	MT Supervised	0.87	0.87	0.16	DeYoung et al., 2020
	Abstractive	Generative MT	0.84	0.87	0.11	Zhou et al., 2020