Quantifying pollen and spore morphology using a language model.

In a typical systematic paper, specimen morphology is described with natural language, and taxonomic specialists use glossaries of technical terminology to establish and communicate taxon concepts, using morphology to delimit species. Textual descriptions of specimen morphology are therefore foundational to biological systematics, but they also serve as a primary data source for studies of morphological evolution. Typically, researchers extract attributes of taxon morphology from descriptions and analyse them to quantify how organismal form has evolved. Focussing on fossil pollen and spores as records of plant evolution, we outline a potential approach to quantifying and comparing morphology that involves using a language model (LM) to transform textual descriptions of specimen morphology into numerical embeddings and then analysing the embeddings statistically. Our results show how an LM can be used to quantitatively analyse the morphological information contained within the text of taxonomic descriptions, illustrating that first-order botanical groupings such as pollen versus spores can be distinguished in the text of a single monograph and potentially from multiple texts by different authors. Although our dataset is small, we tentatively suggest that the spore-plant dominated tropical vegetation of the Carboniferous was comparable in morphological diversity to the angiosperm dominated tropical rainforests of the Palaeogene.