The event deduplication task is defined as a sentence-pair comparison task, where each sentence represents an event and the model predicts whether these two sentences are describing the same event.