Answers are grounded in this paper and cite their sources.
Etymological classifications of species epithets provide insight into how taxonomists encode morphology, ecology and culture in scientific names, but such classifications are rarely available for large clades. We benchmarked a large language model (LLM) against expert labels for the etymology of species epithets in 989 zooplankton species (rotifers and microcrustaceans). The LLM assigned each epithet to one of five categories (Morphology, Ecology & Behavior, Geography, People, Culture), and we evaluated its predictions using a confusion matrix, overall accuracy, Cohen’s kappa, per-class precision, recall and F1-scores, and generalized additive models (GAMs) fitted to yearly and decadal frequencies. Overall accuracy was 0.84, with Cohen’s kappa of 0.74. F1-scores were very high for Morphology and People and high for Geography, but moderate for Ecology & Behavior and low for Culture. GAMs showed that LLM-based labels broadly reproduced long-term temporal trends in epithet usage, with clear discrepancies only for Culture-related names. These results indicate that LLMs can provide reliable first-pass etymology labels and support macro-scale analyses of naming practices, while culturally nuanced and context-dependent epithets remain under expert curation. We recommend using LLM-based classification as a complementary, human-in-the-loop tool for building and maintaining large etymology databases.
No takes yet. Share an insight, caveat, or question.
Sugeno et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: