A

• DOI: 10.18653/v1/w19-1908
• Publication Date: 2019-07-21T17:37:55Z
• Authors: Chen Lin; Timothy Miller; Dmitriy Dligach; Steven Bethard; Guergana Savova
• Keywords: Relationship extraction

Classic methods for clinical temporal relation extraction focus on relational candidates within a sentence. On the other hand, break-through Bidirectional Encoder Representations from Transformers (BERT) are trained large quantities of arbitrary spans contiguous text instead sentences. In this study, we aim to build sentence-agnostic framework task CONTAINS extraction. We establish new state-of-the-art result task, 0.684F in-domain (0.055-point improvement) and 0.565F cross-domain (0.018-point improvement), by fine-tuning BERT pre-training domain-specific models instances with WordPiece-compatible encodings, augmenting labeled data automatically generated “silver” instances.