Table 12 Model performance on different test sets splits, comparison between aci dialogues with ASR and ASRcorr transcript.

Test set	Bart Fine-tuning	Test Split	ROUGE-1	ROUGE-2	ROUGE-L	medcon
1	train	ASR	54.03	24.19	48.09	45.97
	+train_ASRcorr	ASR	54.10	24.47	47.92	47.17
	train	ASRcorr	54.04	24.35	48.19	46.00
	+train_ASRcorr	ASRcorr	54.04	24.50	47.85	47.36
2	train	ASR	51.62	23.05	46.01	45.31
	+train_ASRcorr	ASR	53.14	24.47	47.43	45.34
	train	ASRcorr	51.56	23.13	45.97	45.98
	+train_ASRcorr	ASRcorr	53.44	24.51	47.66	45.47
3	train	ASR	53.07	23.53	47.73	45.27
	+train_ASRcorr	ASR	52.53	23.14	47.06	46.38
	train	ASRcorr	53.13	23.48	47.63	45.52
	+train_ASRcorr	ASRcorr	52.38	23.00	46.87	45.78

The model finetuned on the train set is the BART + FT_SAMSum (Division) fine-tuned with 10 epochs on the original train set, as in the baseline methods. The train + train_ASRcorr model refers to the BART + FT_SAMSum (Division) finetuned for 3 more epochs on the aci with ASRcorr split of the train set.

Quick links

Search