Search Results for author:
Found 2 papers, 2 papers with code
Benchmarking and Improving Compositional Generalization of Multi-aspect Controllable Text Generation
Compositional generalization, representing the model's ability to generate text with new attribute combinations obtained by recombining single attributes from the training data, is a crucial property for multi-aspect controllable text generation (MCTG) methods.
Air-Decoding: Attribute Distribution Reconstruction for Decoding-Time Controllable Text Generation
Then we design a novel attribute distribution reconstruction method to balance the obtained distributions and use the reconstructed distributions to guide language models for generation, effectively avoiding the issue of Attribute Collapse.
Cannot find the paper you are looking for? You can
Submit a new open access paper.
