“where is this relationship going?”: Understanding Relationship Trajectories in Narrative Text

Proceedings of the Ninth Joint Conference on Lexical and Computational Semantics (*SEM), pages 168–178 Barcelona, Spain (Online), December 12–13, 2020 168 “where is this relationship going?”: Understanding Relationship Trajectories in Narrative Text Keen You and Dan Goldwasser Department of Computer Science, Purdue University {you54, dgoldwas}@purdue.edu Abstract We examine a new commonsense reasoning task: given a narrative describing a social in- teraction that centers on two protagonists, sys- tems make inferences about the underlying relationship trajectory. Specifically, we pro- pose two evaluation tasks: Relationship Out- look Prediction MCQ and Resolution Predic- tion MCQ. In Relationship Outlook Prediction, a system maps an interaction to a relationship outlook that captures how the interaction is ex- pected to change the relationship. In Resolu- tion Prediction, a system attributes a given rela- tionship outlook to a particular resolution that explains the outcome. These two tasks parallel two real-life questions that people frequently ponder upon as they navigate different social situations: “where is this relationship going?” and “how did we end up here?”. To facilitate the investigation of human social relationships through these two tasks, we construct a new dataset, Social Narrative Tree, which consists of 1250 stories documenting a variety of daily social interactions. The narratives encode a multitude of social elements that interweave to give rise to rich commonsense knowledge of how relationships evolve with respect to so- cial interactions. We establish baseline perfor- mances using language models and the accu- racies are significantly lower than human per- formance. The results demonstrate that mod- els need to look beyond syntactic and seman- tic signals to comprehend complex human re- lationships. 1 Introduction A relationship between two people is constantly being shaped by their social interactions (Duck, 1994). For example, if two people have a proper conversation after a heated argument, they may be- come more intimate as they understand each other This work is licensed under a Creative Commons Attri- bution 4.0 International License. License details: http: //creativecommons.org/licenses/by/4.0/. better. On the other hand, they may insist on their own views and part ways. For humans, being able to reason about relationship trajectories is crucial in achieving personal goals while avoiding conflicts with other people. This skill is formally termed as social competence, which humans easily acquire (Rubin et al., 1995). In contrast to humans’ innate ability to perceive social situations, machines struggle in develop- ing social understanding. For example, machines’ performance in reasoning about motivation and emotional reactions given a short context is sig- nificantly lower compared to human performance (Sap et al., 2019). Moreover, identifying the intents and reactions of characters in narratives poses an- other challenge for machines (Goyal et al., 2010; Chaturvedi et al., 2016; Rahimtoroghi et al., 2017; Rashkin et al., 2018a). Reasoning about relation- ship trajectories adds yet another layer of difficulty as in addition to the challenge of identifying the im- plicit factors that guide characters’ behavior, such as intents, reactions and mental states, systems also need to understand how these elements collectively contribute to the evolution of …

Content truncated to 500 words