Table 5:

Flesch-Kincaid, ARI, and SciBERT masked probability scores for generated PLS. Differences wbetween abstracts and generated PLS are statistically significant; so are differences in FK and ARI between UL models and No-UL (p < 0.01, paired t-test).

	FK↓	ARI↓	SciBERT↓
Abstracts	14.42	15.58	0.57
PLS	13.11	15.08	0.53
No UL	13.44	15.09	0.55
UL-Cochrane	11.97	13.73	0.55
UL-Newsela	12.51	14.15	0.54
UL-Both	12.26	14.04	0.54