SPEECH: Structured Prediction with Energy-Based Event-Centric Hyperspheres

TL;DR

SPEECH introduces an energy-based hypersphere model to effectively capture complex event dependencies in structured prediction tasks, improving event detection and relation extraction in NLP.

Contribution

The paper proposes a novel energy-based hypersphere approach for modeling complex event structures, enhancing structured prediction in NLP.

Findings

Outperforms existing methods in event detection

Improves event-relation extraction accuracy

Effective modeling of complex event dependencies

Abstract

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.