Enhancing consistency with the fusion of paralleled decoders for text generation

Yaolin Li, Heyan Huang, Yu Bai, Yang Gao

Abstract

Highlights•MDFUT: concurrent plan-text generation with the fusion of both distributions.•Novel bridging loss for closer decoder alignment, boosting plan-text consistency.•Extensive tests show MDFUT excels in open-ended story tasks with coherent outputs.
BibTeX
@article{li2025enhancing,
  author = {Yaolin Li and Heyan Huang and Yu Bai and Yang Gao},
  title = {{Enhancing consistency with the fusion of paralleled decoders for text generation}},
  journal = {Information Fusion},
  year = {2025},
  url = {https://openreview.net/forum?id=bZuoyoa8pf},
}

Loading related content...

Figure