Temporal Annotation in the Clinical Domain

Area of Science:

• Natural Language Processing
• Clinical Informatics
• Computational Linguistics

Background:

• Clinical narratives contain complex temporal information crucial for patient care.
• Existing temporal annotation standards like ISO-TimeML require adaptation for the nuances of clinical text.
• The THYME corpus presents unique challenges due to inference-based reasoning demands.

Purpose of the Study:

• To define requirements for formal temporal annotation in clinical narratives.
• To extend ISO-TimeML for the THYME corpus, creating the THYME-TimeML guideline.
• To evaluate system adaptability to the clinical domain using the new resource.

Main Methods:

• Developed "THYME Guidelines to ISO-TimeML (THYME-TimeML)" based on the THYME corpus.
• Distinguished between linguistically-derived and inferentially-derived temporal relations.
• Applied a leading TempEval 2013 system to assess clinical domain adaptation.

Main Results:

• The THYME-TimeML guideline effectively addresses clinical temporal annotation needs.
• System performance evaluation highlights challenges in adapting to clinical temporal reasoning.
• The THYME corpus is a valuable resource for temporal information extraction research.

Conclusions:

• Formalizing temporal annotation in clinical narratives is essential for NLP advancements.
• The THYME-TimeML guideline provides a robust framework for clinical temporal data.
• The THYME corpus and guideline facilitate research and system development in clinical NLP.