Approaches to verb subcategorization for biomedicine

Area of Science:

• Computational linguistics
• Bioinformatics
• Natural Language Processing (NLP)

Background:

• Verb subcategorization frames (SCFs) are vital for NLP tasks like information extraction.
• High-quality SCF lexicons are needed for the biomedical domain to process vast scientific literature.
• Current methods for creating biomedical SCF lexicons are less developed than for general language.

Purpose of the Study:

• Introduce subcategorization and acquisition techniques.
• Motivate the development of methods tailored for biomedical NLP.
• Address the critical issue of linguistic variation within biomedical subdomains.

Main Methods:

• Review traditional linguistic definitions of subcategorization.
• Describe existing approaches for learning SCF lexicons from large datasets (general and biomedical).
• Analyze linguistic variation across different biomedical subdomains.

Main Results:

• Demonstrate significant linguistic variation among biomedical subdomains.
• Show that this variation does not align with general lexical variation patterns.
• Identify key requirements for future research in biomedical SCF lexicon acquisition.

Conclusions:

• Biomedical NLP requires specialized approaches for SCF lexicon acquisition due to subdomain variation.
• Future research should focus on high-quality gold standards, exploring diverse subcategorization definitions, and developing minimally-supervised methods for subdomain-specific lexical learning.