commonsense generation is a challenging task of generating a plausible
sentence describing an everyday scenario using provided concepts. Its
requirement of reasoning over commonsense knowledge and compositional
generalization ability even puzzles strong pre-trained language generation
本论文中,我们探讨如何运用常识知识图谱提高条件文本生成模型的综合性能,通过从 Conceptnet 中提取常识关系,将这些关系注入到 Unified Language Model (UniLM) 中,并通过输出约束强制实施词汇要求,以提高生成文本的语义正确性和符合人类理解,从而实现了匹配词性和完全概念覆盖的要求。