Towards Inter-character Relationship-driven Story Generation

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8970 - 8987 December 7-11, 2022 ©2022 Association for Computational Linguistics Towards Inter-character Relationship-driven Story Generation Anvesh Rao Vijjini1 Faeze Brahman2,3 Snigdha Chaturvedi1 1UNC Chapel Hill 2Allen Institute for Artificial Intelligence 3Paul G. Allen School of Computer Science & Engineering, University of Washington {anvesh,snigdha}@cs.unc.edu faezeb@allenai.org Abstract In this paper, we introduce the task of modeling interpersonal relationships for story generation. For addressing this task, we propose Relation- ships as Latent Variables for Story Generation, (RELIST). RELIST generates stories sentence by sentence and has two major components - a relationship selector and a story continuer. The relationship selector specifies a latent variable to pick the relationship to exhibit in the next sentence and the story continuer generates the next sentence while expressing the selected re- lationship in a coherent way. Our automatic and human evaluations demonstrate that RELIST is able to generate stories with relationships that are more faithful to desired relationships while maintaining the content quality. The relation- ship assignments to sentences during inference brings interpretability to RELIST. 1 Introduction Interpersonal relationships between characters are, in many ways, the glue that holds a story together. Almost every story revolves around at least one, if not more, inter-character relationships. Despite the importance of relationships in stories (Bochner et al., 1997), only few studies in NLP have ex- plored story generation from the perspective of character relationships. Recent story generation methods typically generate stories from a prompt (Fan et al., 2018) or a planner detailing events or keywords of the story (Martin et al., 2018; Rashkin et al., 2020; Goldfarb-Tarrant et al., 2020; Brahman et al., 2020). While these methods can generate stories based on open-ended prompts and plans they can neither encode character relationships nor can they give explicit control over the characters and their relationships. In this paper, we intro- duce Relationship-driven Story Generation where given a prompt sentence and a set of inter-character relationships, the goal is to generate a story follow- ing the prompt sentence which exhibits the desired The story starts with Amy going to the market. It starts to rain and Amy had forgotten her umbrella. Amy’s friend Julia lends Amy her umbrella. Amy is approached by Wisk who pressures her for money [...] Relationship Set Julia <positive> Amy Wisk <negative> Amy Prompt Sentence “The story starts with Amy going to the market.” Story Amy Wisk Julia Figure 1: Example of relationship-driven story gener- ation task: given a set of relationships and a prompt sentence, the goal is to generate a story continuing the prompt sentence and reflecting the input relationships. Positive and negative relationships are highlighted. relationships between the characters. While rela- tionships can be described in many ways, following previous works (Chaturvedi et al., 2016; Srivastava et al., 2016; Si et al., 2021), we represent relation- ships using relationship polarity. Specifically, we summarize the overall interaction between pairs of characters as being positive, neutral or negative. Figure 1 illustrates the task. …

Content truncated to 500 words