Event-Location Tracking in Narratives: A Case Study on Holocaust Testimonies

This work focuses on the spatial dimension of narrative understanding and presents the task of event-location tracking in narrative texts, namely the extraction of the sequence of locations where the narrative is set. We present several architectures for the task that seek to model the global structure of the sequence, with varying levels of context awareness. We compare these methods to a number of strong baselines and ablated variants. We also develop methods for the generation of location embeddings and show that learning to predict a sequence of continuous embeddings is advantageous in terms of performance over predicting a string of locations. We focus on the test case of Holocaust survivor testimonies, motivated by the moral and historical importance of studying this dataset using computational means. The dataset further provides a unique case of a large set of narratives with a relatively restricted set of location trajectories. Our results show that models that are aware of the global context of the narrative can generate more accurate location chains. We corroborate the effectiveness of our methods by showing similar trends in an additional domain.