A Natural Language Processing Pipeline to extract phenotypic data from formal taxonomic descriptions with a focus on flagellate plants

Assembling large-scale phenotypic datasets for evolutionary and biodiversity studies of plants can be extremely difficult and time consuming. New semi-automated Natural Language Processing (NLP) pipelines can extract phenotypic data from taxonomic descriptions, and their performance can be enhanced by incorporating information from ontologies, like the Plant Ontology (PO) and the Plant Trait Ontology (TO). These ontologies are powerful tools for comparing phenotypes across taxa for large-scale evolutionary and ecological analyses, but they are largely focused on terms associated with flowering plants. We describe a bottom-up approach to identify terms from flagellate plants (including bryophytes, lycophytes, ferns, and gymnosperms) that can be added to existing plant ontologies. We first parsed a large corpus of electronic taxonomic descriptions using the Explorer of Taxon Concepts tool (http://taxonconceptexplorer.org/) and identified flagellate plant specific terms that were missing from the existing ontologies. We extracted new structure and trait terms, and we are currently incorporating the missing structure terms to the PO and modifying the definitions of existing terms to expand their coverage to flagellate plants. We will incorporate trait terms to the TO in the near future