Semantic annotation of biosystematics literature without training examples

Hong Cui, David Boufford, Paul Selden

Research output: Contribution to journalArticle

19 Scopus citations

Abstract

This article presents an unsupervised algorithm for semantic annotation of morphological descriptions of whole organisms. The algorithm is able to annotate plain text descriptions with high accuracy at the clause level by exploiting the corpus itself. In other words, the algorithm does not need lexicons, syntactic parsers, training examples, or annotation templates.The evaluation on two real-life description collections in botany and paleontology shows that the algorithm has the following desirable features: (a) reduces/eliminates manual labor required to compile dictionaries and prepare source documents; (b) improves annotation coverage: the algorithm annotates what appears in documents and is not limited by predefined and often incomplete templates; (c) learns clean and reusable concepts: the algorithm learns organ names and character states that can be used to construct reusable domain lexicons, as opposed to collectiondependent patterns whose applicability is often limited to a particular collection; (d) insensitive to collection size; and (e) runs in linear time with respect to the number of clauses to be annotated.

Original languageEnglish (US)
Pages (from-to)522-542
Number of pages21
JournalJournal of the American Society for Information Science and Technology
Volume61
Issue number3
DOIs
Publication statusPublished - Mar 2010

    Fingerprint

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence
  • Information Systems
  • Human-Computer Interaction
  • Computer Networks and Communications

Cite this