The BioScope corpus: biomedical texts annotated for uncertainty, negation and their scopes

Area of Science:

• Biomedical Natural Language Processing
• Computational Linguistics
• Bioinformatics

Background:

• Accurate detection of negation and uncertainty is crucial for biomedical text mining.
• Existing resources for handling these linguistic phenomena in biomedical texts are limited.
• This work introduces the BioScope corpus, a freely available resource for this purpose.

Purpose of the Study:

• To create and release the BioScope corpus for research on negation and uncertainty detection in biomedical texts.
• To provide a standardized dataset for training, testing, and comparing biomedical Natural Language Processing (NLP) systems.
• To facilitate linguistic analysis of scientific and clinical texts.

Main Methods:

• A corpus annotation project involving medical free texts, biological full papers, and biological abstracts.
• Token-level annotation of negation and speculative keywords, and sentence-level annotation of their linguistic scope.
• Annotation by two independent linguists, with disagreements resolved by a chief linguist.

Main Results:

• The BioScope corpus comprises over 20,000 annotated sentences.
• More than 10% of the annotated sentences contain linguistic cues for negation or uncertainty.
• Annotations include token-level keywords and sentence-level scope, with reported statistics on corpus size and inter-annotator agreement.

Conclusions:

• The BioScope corpus is a valuable, freely accessible resource for the biomedical research community.
• It supports the advancement of NLP systems for information extraction from biomedical literature.
• The corpus also serves as a foundation for deeper linguistic investigations into scientific and clinical language.