Learning Representations of Long Narratives for Summarization and Inference

Explore cutting-edge research on improving automatic understanding and modeling of long, complex narratives in this lecture by Lea Frermann. Dive into two projects aimed at enhancing NLP systems' ability to process extended texts and make inferences. Learn about topic-aware summarization models that leverage document structure for improved flexibility, especially in long documents. Discover a novel approach to incremental inference in multi-modal, evolving environments through a case study on identifying perpetrators in TV crime series episodes. Gain insights into the comparison between model and human predictions for this task. Understand the speaker's background in efficiency and robustness of human learning and inference, and how it applies to processing large corpora of child-directed speech and complex narratives like book and film plots.

Learning Representations of Long Narratives for Summarization and Inference -- Lea Frermann 2019