Psychology-guided Controllable Story Generation

Proceedings of the 29th International Conference on Computational Linguistics, pages 6480–6492 October 12–17, 2022. 6480 Psychology-guided Controllable Story Generation Yuqiang Xie Yue Hu† Yunpeng Li Guanqun Bi Luxi Xing Wei Peng Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China {xieyuqiang,huyue,liyunpeng,biguanqun,xingluxi,pengwei}@iie.ac.cn Abstract Controllable story generation is a challenging task in the field of NLP, which has attracted in- creasing research interest in recent years. How- ever, most existing works generate a whole story conditioned on the appointed keywords or emotions, ignoring the psychological changes of the protagonist‡. Inspired by psychology theories, we introduce global psychological state chains, which include the needs and emo- tions of the protagonists, to help a story gen- eration system create more controllable and well-planned stories. In this paper, we propose a Psychology-guIded Controllable Story Gen- eration System (PICS) to generate stories that adhere to the given leading context and desired psychological state chains for the protagonist. Specifically, psychological state trackers are employed to memorize the protagonist’s local psychological states to capture their inner tem- poral relationships. In addition, psychological state planners are adopted to gain the protago- nist’s global psychological states for story plan- ning. Eventually, a psychology controller is de- signed to integrate the local and global psycho- logical states into the story context representa- tion for composing psychology-guided stories. Automatic and manual evaluations demonstrate that PICS outperforms baselines, and each part of PICS shows effectiveness for writing stories with more consistent psychological changes. 1 Introduction Controllable Story Generation (CSG) is an impor- tant task in natural language processing (NLP) (Por- teous and Cavazza, 2009; Peng et al., 2018; Alab- dulkarim et al., 2021). It has also become one of the test methods for progress in artificial intelli- gence (AI). Most existing state-of-the-art works (Kong et al., 2021; Rashkin et al., 2020; Paul and †Corresponding author. ‡In this work, we define the protagonist as the most fre- quently occurring character in a story (Morrow, 1985). Mike complained to the server, then Tory comforted him. The server apologized with a gift and Mike replied with a smile. Love Physiologic Physiologic Physiologic Esteem Joy None Sadness Anger Joy Psychological State Chains Emotion Need Psychology-guided Story Events 1 2 3 4 5 Given Leading Context The service was slow and Mike began to feel uncomfortable. Mike delightedly went on a date with his girlfriend Tory. Mike was hungry and they went to a restaurant. Figure 1: Example of psychology-guided controllable story generation conditioned on dotted frames (global psychological state chains, i.e., need/emotion, as well as leading context). Each psychological state and its corresponding tokens are highlighted in the same color. Frank, 2021; Xu et al., 2020b) generate a story con- ditioned by the appointed keywords or emotions, with the help of remarkable pre-trained language models (PLM), like GPT-2 (Radford et al., 2019) and BART (Lewis et al., 2020). While most of these systems have been able to generate fluent stories, CSG still has many …

Content truncated to 500 words