notesum.ai

Published at December 10

CoPrUS: Consistency Preserving Utterance Synthesis towards more realistic benchmark dialogues

cs.CL

Released Date: December 10, 2024

Authors: Sebastian Steindl¹, Ulrich Schäfer¹, Bernd Ludwig²

Aff.: ¹Ostbayerische Technische Hochschule Amberg-Weiden; ²University Regensburg

Arxiv: http://arxiv.org/pdf/2412.07515v1

		F1		BLEU		JGA
Model	Task	CoPrUS	MWZ	CoPrUS	MWZ	CoPrUS	MWZ
jointBERT	NLU	89.67	89.59	-	-	-	-
t5-small	NLU	83.51	83.45	-	-	-	-
SCGPT	NLG	-	-	23.48	24.63	-	-
t5-small	NLG	-	-	29.54	29.50	-	-
SetSUMBT	DST	92.18	92.27	-	-	50.83	51.12
t5-small	DST	91.81	91.76	-	-	53.04	52.67