Picking Apart Story Salads
arXiv:1810.13391
Abstract
During natural disasters and conflicts, information about what happened is often confusing, messy, and distributed across many sources. We would like to be able to automatically identify relevant information and assemble it into coherent narratives of what happened. To make this task accessible to neural models, we introduce Story Salads, mixtures of multiple documents that can be generated at scale. By exploiting the Wikipedia hierarchy, we can generate salads that exhibit challenging inference problems. Story salads give rise to a novel, challenging clustering task, where the objective is to group sentences from the same narratives. We demonstrate that simple bag-of-words similarity clustering falls short on this task and that it is necessary to take into account global context and coherence.
arXiv commentAccepted at EMNLP 2018 (long paper)
References in corpus (10)
- Adam: A Method for Stochastic Optimization
- Teaching Machines to Read and Comprehend
- A Structured Self-attentive Sentence Embedding
- SQuAD: 100,000+ Questions for Machine Comprehension of Text
- Towards K-means-friendly Spaces: Simultaneous Deep Learning and Clustering
- A large annotated corpus for learning natural language inference
- What you can cram into a single vector: Probing sentence embeddings for linguistic properties
- Sentence Ordering and Coherence Modeling using Recurrent Neural Networks
- Implicit Argument Prediction with Event Knowledge
- Modeling Semantic Plausibility by Injecting World Knowledge