Natural language descriptions of plant phenotypes present in databases and the scientific literature are a rich source of information for biological research that seeks to untangle relationships between genes and observable phenotypes, such as plant health or size. The volume and unstructured nature of these text descriptions however necessitates a computational approach for leveraging them to predict gene-to-phenotype associations. We computationally translated descriptions of plant phenotypes into structured representations that can be processed to identify biologically meaningful associations. These representations include numerical vectors generated using neural networks and a variety of additional techniques applied from the domain of natural language processing (NLP), as well as representations constructed by automatically mapping text to terms from biological ontologies. We compared phenotype similarities derived from these automated techniques to those derived from manually curated data, and evaluated each approach on predictive tasks such as categorizing genes by functional group or biochemical pathway and predicting protein-protein interactions. Computationally derived representations were comparably successful in recapitulating biological truth to representations created through manual curation, indicating that it is now possible to computationally and automatically produce and populate large-scale information resources that enable researchers to query phenotypic descriptions directly. We present a dataset of computationally-inferred phenotype similarity networks, and discuss tools aimed at the research community for querying and visualizing this information.