SPEECH: Structured Prediction with Energy-Based Event-Centric Hyperspheres
This addresses event-centric structured prediction in NLP, offering a novel approach for handling complex event dependencies, but it appears incremental as it builds on existing energy-based and hypersphere methods.
The paper tackles the challenge of representing complex event structures with manifold dependencies in NLP by proposing SPEECH, which models dependencies with energy-based modeling and represents event classes with hyperspheres, achieving predominant performance in event detection and event-relation extraction tasks on two datasets.
Event-centric structured prediction involves predicting structured outputs of events. In most NLP cases, event structures are complex with manifold dependency, and it is challenging to effectively represent these complicated structured events. To address these issues, we propose Structured Prediction with Energy-based Event-Centric Hyperspheres (SPEECH). SPEECH models complex dependency among event structured components with energy-based modeling, and represents event classes with simple but effective hyperspheres. Experiments on two unified-annotated event datasets indicate that SPEECH is predominant in event detection and event-relation extraction tasks.