Why discourse structures in medical reports matter for the validity of automatically generated text knowledge bases

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Current automatic analysis of medical full-texts often overlooks text coherence.
• This neglect results in inadequate descriptions within automatically generated medical knowledge bases.
• Existing knowledge structures are often fragmented, incomplete, and invalid due to this bias.

Purpose of the Study:

• To highlight the impact of neglecting text coherence on medical knowledge base generation.
• To identify specific textual phenomena that disrupt coherence in medical texts.
• To propose methodologies for incorporating coherence analysis into automatic medical text processing.

Main Methods:

• Discussing textual phenomena: pronominal anaphora, nominal anaphora, and textual ellipsis.
• Analyzing how these phenomena affect discourse unit reference relations.
• Outlining basic methodologies for detecting and resolving these coherence issues.

Main Results:

• Neglecting coherence leads to representation bias in medical knowledge bases.
• Fragmented, incomplete, and invalid knowledge structures are direct consequences.
• Identification of key coherence phenomena provides targets for improvement.

Conclusions:

• Incorporating coherence analysis, specifically addressing anaphora and ellipsis, is crucial for accurate medical knowledge base generation.
• Improved methodologies can mitigate representation bias and enhance the adequacy of medical knowledge bases.
• Future work should focus on robust computational models for these textual phenomena.