| PEGASUS | CNN/DailyMail, XSum | Rouge-F1 ~0.43-0.46 | Transformer encoder and decoder | Gap Sentence Generation + Masked Language Model + Vocabulary | Rouge-F1 +0.03-0.05 | Rouge-F1 +0.05-0.1 |
| BART | CNN/DailyMail, XSum | Rouge-F1 ~0.40-0.45 | Bidirectional encoder and auto-regressive decoder | Noise + Sequence-to-Sequence Model + Vocabulary | Rouge-F1 +0.01-0.02 | Rouge-F1 +0.02-0.05 |
| Dimsum | CL-LaySumm 2020, ScisummNet | Rouge-F1 ~0.44-0.46 | BART + Multi-label Model | Tokens and Labels + Simultaneous (abstractive and extractrive) | Rouge-F1 +0.03-0.04 | - |
| T5 | CNN/DailyMail | Rouge-F1 ~0.42-0.43 | Transformer encoder and decoder + Prefix Model | Maximum Likelihood | Rouge-F1 +0.01-0.02 | - |
| Multi-View S2S | SAMSum | Rouge-F1 ~0.49 | Conversation encoder + Multi-view decoder | Multi-view sequence-to-sequence model | Rouge-F1 +0-0.01 | - |
| ConDigSum | SAMSum, MediaSum | Rouge ~0.54 | Transformer encoder and decoder + coherence regressor + log likelohood | Sub-summary selection + Coherence + Sensitivity test | Rouge-F1 +0.05 | Rouge-F1 +0-0.01 |
| ConvSumm | SAMSum | Rouge-F1 ~0.51-0.53 | Generative Language Model + Granularity Controller | Snorkel Module + Constituency Parser | Rouge-F1 +0-0.01 | - |
| TOASTS | Reddit TIFU, arXiv | Rouge ~0.28 | Sequential, Simultaneous and Continual MTL | Task Families + Training Strategies | Rouge-F1 +0.014-0.015 | Rouge-F1 +0.03-0.05 |
| MTL-KIGN | CNN/DailyMail | Rouge-F1 ~0.39 | Document and Key-info encoder + Multi-view attention + Decoder | Joint training + Pointer network + Key-info + Prediction guide | Rouge-F1 +0.02-0.03 | - |