Semi-automatic conversion of BioProp semantic annotation to PASBio annotation

Area of Science:

• Natural Language Processing
• Computational Biology
• Bioinformatics

Background:

• Semantic Role Labeling (SRL) is crucial for text analysis, representing sentences as predicate-argument structures (PAS).
• PropBank is standard for news, while PASBio is preferred in biomedicine.
• Lack of PASBio corpora hinders development of machine-learning (ML)-based SRL systems for biomedical applications.

Purpose of the Study:

• To develop a system for converting BioProp (PropBank-based) SRL annotations to PASBio format.
• To enable the use of ML-based SRL systems with PASBio annotations in the biomedical domain.

Main Methods:

• Developed a system combining an existing ML-based SRL system (BIOSMILE) with a novel BioProp-to-PASBio rule-based converter.
• Incorporated a semi-automatic rule generator to enhance the conversion process.

Main Results:

• The rule-based converter achieved an F-score of 85.29% independently.
• The combined system (BIOSMILE + converter) achieved an F-score of 69.08% for PASBio's 29 verbs.

Conclusions:

• The developed system facilitates PAS conversion between BioProp and PASBio annotations.
• This approach matches state-of-the-art domain-specific ML-based SRL systems and supports PASBio application development.