nlg

This paper introduces an attribute-guided genre expansion framework to scale creative writing data beyond story-centric formats, creating a multi-genre corpus that improves language model performance on diverse creative tasks.

This paper proposes SPG, a structure-aware patent generation method that jointly predicts claim topology and content during autoregressive decoding using a pointer head and a preference optimization stage, achieving significant improvements in parent link recovery and antecedent consistency on the HUPD-DCG benchmark.