Abstract

Events are vital parts of natural language, reflecting the state changes of entities. The Event Extraction (EE) task aims to extract event triggers (the most representative words or phrases) and their arguments (participants in the event) from the given text. Most current works use sequence tagging models to solve the EE task. However, those methods only treat event and argument types as different class numbers, ignoring the semantics of those labels. However, label semantics are critical in the EE task. For example, the trigger word “fight” is semantically closer to the event type “Conflict:Attack” rather than “Life:Marriage”. To emphasize the label semantics in events, we formulate EE as a prototype matching task and propose a Prototype Matching framework for Joint Event Extraction (PMJEE). Specifically, prototypical embeddings for both event trigger and argument types are introduced to encode their label semantics and correlations. Then a dual-channel attention layer and extraction modules are used to jointly extract event triggers and arguments. Prototypical embeddings will be optimized during training to improve the event extraction performance. Extensive experiments indicate our method achieves better performance than strong baselines, especially in data-scarce scenarios. In the detailed analysis, we verify the effectiveness of each part of the model and explore the impact of different label semantic materials.

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.