Event extraction is fundamental to information extraction. Prior approaches often separate event detection and argument extraction or depend on dataset-specific designs, limiting scalability and cross-domain generalization. We propose a unified generative sequence-to-sequence framework that performs event extraction subtasks jointly and supports both pipeline and end-to-end configurations. We fine-tune pretrained language models on multiple event datasets across diverse domains, enabling a single model to retain domain-specific semantics while generalizing over large and evolving label spaces. We demonstrate these capabilities through a web-based application tailored for researchers and practitioners. The platform supports document upload, schema-aware event extraction, visualization of triggers and arguments, and comparison of different extraction configurations across domains.
Figures & tables
Figure 1: (1) Event Detection identifies event triggers and assigns event types, followed by (2) Event Argument Extraction, which extracts arguments and labels them with semantic roles.
Figure 2: We train a unified sequence-to-sequence model based on T5 across six diverse domains, using all available datasets. Each dataset is mapped to a task-specific prompt and a corresponding output pattern to standardize event extraction. This formulation allows a single model to learn cross-domain event representations and supports both pipeline and E2E modes.
Figure 3: Pipeline and E2E event extraction workflows in inference time.
Protester , Target , Place , Time , Reason , Organizer
GENIA 2013 ( Bio )
13
Theme , Agent , Site , Resulting Entity
M2E2 ( Geo )
8
Location , Victims , Casualties , Time
RAMS ( Newswire )
139
Participant , Location , Time , Topic , Outcome
WikiEvents ( Wiki )
50
Entity , Organization , Position , Place , Time , Reason
Table 1: Representative event types and their argument roles.
Figure 4: Two-stage event extraction interface. Stage 1 performs event detection using pipeline and E2E modes and merges the detected triggers. Stage 2 extracts arguments conditioned on the detected events and event-type-specific argument roles. Text highlighting supports result inspection.
Figure 5: View of complete E2E event extraction, where the model runs all event extraction subtasks in one step.
Figure 6: Event extraction on the same document with different dataset schemas.