Text Pattern Visualization for analysis of biology full text and captions

Large textbanks comprised of thousands of full-textbiology papers are rapidly becoming available. Wedescribe an approach to characterize all major languagepatterns in biology text in terms of Frameworks.Frameworks are "containers" made up of commonphrases surrounding specific informational items such asgene and protein names. A Framework Viewer has beendeveloped that shows similar text Frameworks aligned onthe screen much as biosequence visualization tools do.Using the Viewer, it is evident that Frameworks have thepower to find the types of structures needed to developuseful information retrieval systems. As a simpleexample, one Framework was able to concisely select45,000 nouns from a corpus of 5 million words withouterror. This work points the way to highly automatedsystems that will be able to extract and index informationin biology textbanks. Work in progress includesextensions to characterize recursive structures in text,subsystems to retrieve figures in papers, and thediscovery of semantic relations to aid concept-basedretrieval.