DPO on human-ranked data modestly improves paraphrase-type generation accuracy and human rankings, but key claims are under-supported and some numbers conflict.
Dolan and Chris Brockett
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Enhancing Paraphrase Type Generation: The Impact of DPO and RLHF Evaluated with Human-Ranked Data
DPO on human-ranked data modestly improves paraphrase-type generation accuracy and human rankings, but key claims are under-supported and some numbers conflict.