Skip to Main Content
Table 5: 

Results of abstractiveness-controlled models with different specified bins on Newsroom-b dataset. Bin 3 is the most abstractive bin. Our CMDP framework significantly improves the BERTScore, MoverScore, and bin % over all the bins (p < 0.003, approx. randomization test).

bin 1bin 2bin 3
MethodBERTS.MoverS.Bin %BERTS.MoverS.Bin %BERTS.MoverS.Bin %
ControlSum 26.53 16.15 99.40 24.50 13.11 6.42 20.53 9.87 24.30 
 
PG 26.49 15.99 99.36 22.59 11.15 8.49 17.67 7.45 26.03 
PG+CMDP 29.62 18.44 99.72 26.95 12.96 97.65 22.78 7.67 98.88 
 
D.GPT2 27.85 17.05 99.17 26.41 14.57 0.47 22.30 11.16 00.37 
D.GPT2+CMDP 30.12 18.45 99.72 31.21 17.52 72.48 25.75 13.18 87.09 
 
D.GPT2+MDP 29.77 18.27 99.77 30.40 17.06 72.37 25.25 13.06 80.95 
bin 1bin 2bin 3
MethodBERTS.MoverS.Bin %BERTS.MoverS.Bin %BERTS.MoverS.Bin %
ControlSum 26.53 16.15 99.40 24.50 13.11 6.42 20.53 9.87 24.30 
 
PG 26.49 15.99 99.36 22.59 11.15 8.49 17.67 7.45 26.03 
PG+CMDP 29.62 18.44 99.72 26.95 12.96 97.65 22.78 7.67 98.88 
 
D.GPT2 27.85 17.05 99.17 26.41 14.57 0.47 22.30 11.16 00.37 
D.GPT2+CMDP 30.12 18.45 99.72 31.21 17.52 72.48 25.75 13.18 87.09 
 
D.GPT2+MDP 29.77 18.27 99.77 30.40 17.06 72.37 25.25 13.06 80.95 
Close Modal

or Create an Account

Close Modal
Close Modal