Modeling Textual Cohesion for Event Extraction

Event extraction systems typically locate the role fillers for an event by analyzing sentences in isolation and identifying each filler independently of others. We argue that more accurate requires a view larger context to decide whether entity is related relevant event. propose bottom-up approach initially identifies candidate then uses information as well discourse properties model textual cohesion. The novel component architecture sequentially structured sentence classifier event-related story contexts. lexical associations relations across sentences, domain-specific distributions within sentences. This yields state-of-the-art performance on MUC-4 data set, achieving substantially higher precision than previous systems.