Table 2.

Surface linguistic diversity.

N-gram overlap measure	Naïve LLaMAa	TH-VAEb	LLaMA	Human
ROUGE-1	0.36	0.47	0.48	0.28
ROUGE-2	0.10	0.17	0.17	0.05
ROUGE-3	0.04	0.08	0.07	0.01
ROUGE-L	0.23	0.38	0.38	0.22
ROUGE-S2	0.09	0.15	0.15	0.04
ROUGE-SU2	0.16	0.23	0.23	0.11
Average	0.17	0.25	0.25	0.12

^aLLaMA: large language model Meta AI.

^bTH-VAE: timeline-hierarchical variational autoencoder.