Across Multiple Sources
Explore cross-document event understanding with FAMuS and SEAMuS datasets.
FAMuS
Frames Across Multiple Sources
FAMuS pairs Wikipedia passages with their cited sources, providing FrameNet annotations for events and arguments across both documents. Built on MegaWika, the dataset contains over 1,255 report-source pairs covering 253 diverse event types.
- Source Validation: Determine if a source describes the same event as the report
- Cross-Document Argument Extraction: Extract event arguments from both report and source
SEAMuS
Summaries of Events Across Multiple Sentences
Built on expert reannotations of the FAMuS data, SEAMuS extends event-keyed summarization to the cross-document setting. It provides summaries of events that synthesize information from FAMuS report and source documents, along with template annotations on those summaries.
- Event-Keyed Summarization: Generate summaries focused on specific events
- Cross-Document Synthesis: Combine information from report and source documents