INCEpTION - Rapid adaptation of event detection

Rapid adaptation of event detection

Use-case

Source: This example was kindly contributed by Ryan Gabbard and Marjorie Freedman, VISTA group at the Information Sciences Institute at the University of Southern California, USA

The VISTA group at the Information Sciences Institute at the University of Southern California is using INCEpTION to enable rapid adaptation of event detection and event argument attachment algorithms to new languages and ontologies. While past training sets for these tasks have largely depended either on exhaustively annotated corpora or active learning, their approach uses INCEpTION’s external search capabilities to allow annotators to rapidly seek out and select informative training examples. This effort aims to build event systems for an ontology of nearly 150 event types in English, Russian, and Ukrainian.

Features of INCEpTION that we consider of particular importance to our use-case are:

  • being able to search for documents over a large corpus and import them for annotation
  • span and relation annotations
  • monitoring of annotators’ progress

Additionally, INCEpTION’s open development model and extremely responsive developers allowed INCEpTION to easily be adapted for the particular needs of this task.

References
  • Gabbard, Ryan ; DeYoung, Jay ; Freedman, Marjorie : Events Beyond ACE: Curated Training for Events. In: CoRR, abs/1809.05576. (2018) [arXiv]