Personality Understanding of Fictional Characters during Book Reading

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 14784–14802 July 9-14, 2023 ©2023 Association for Computational Linguistics Personality Understanding of Fictional Characters during Book Reading Mo Yu1∗ Jiangnan Li2∗ Shunyu Yao3 Wenjie Pang1 Xiaochen Zhou4 Xiao Zhou1 Fandong Meng1 Jie Zhou1 1Pattern Recognition Center, WeChat AI 2Institute of Information Engineering, Chinese Academy of Sciences 3Princeton University 4Syracuse University moyumyu@global.tencent.com lijiangnan@iie.ac.cn Abstract Comprehending characters’ personalities is a crucial aspect of story reading. As readers en- gage with a story, their understanding of a char- acter evolves based on new events and informa- tion; and multiple fine-grained aspects of per- sonalities can be perceived. This leads to a nat- ural problem of situated and fine-grained per- sonality understanding. The problem has not been studied in the NLP field, primarily due to the lack of appropriate datasets mimicking the process of book reading. We present the first la- beled dataset PERSONET for this problem. Our novel annotation strategy involves annotating user notes from online reading apps as a proxy for the original books. Experiments and human studies indicate that our dataset construction is both efficient and accurate; and our task heavily relies on long-term context to achieve accurate predictions for both machines and humans.1 1 Introduction Lively fictional characters with distinct personali- ties are the first drive of the plotline developments. The authors shape the characters with various per- sonality types, which distinguish a character from others and explain the motivations and behaviors of the characters. As a reverse process, the readers grasp the characters’ personalities during reading a story, which helps to understand the logics of a plot and predict its future developments. The NLP community has also recognized the values of personality understanding; and con- ducted studies (Bamman et al., 2013; Flekova and Gurevych, 2015; Sang et al., 2022b) along this di- rection. In the problem definition of the existing tasks, the input is an entire book. By construction, they ask for the general impression of character personalities. Also for this reason, they only focus on coarse-grained personality types, e.g., the four coarse MBTI types (Myers and McCaulley, 1988). ∗Authors contributed equally to this paper. 1Available at https://github.com/Gorov/personet_acl23. Figure 1: Illustration of the WeRead App interface enabling our dataset construction (user IDs masked for privacy); and an example of our task, situated personality prediction. To make a personality prediction task more prac- tical and useful, we consider two important aspects of character understanding in real life that have not been studied in the context of machine reading. First, we aim at predicting fine-grained personality types, with an exhaustive vocabulary of personality traits as the targets. Second and more importantly, we study the continuous-process nature of story reading — As people read, they form dynamic im- pressions of the characters and plots. We name this process situated comprehension. Specific to personality understanding, a character may have multi-faced personalities. In a certain point of the story, the character’s behaviors can reflect one of them …

Content truncated to 500 words