Integrating scanned literature and online sources to build a data text collection for plant biodiversity

Plant taxonomic descriptions are essential for botany, but remain scattered in unstructured texts. Automatic trait extraction is difficult due to linguistic variation, specialised terminology and the need for large, structured datasets. Methods : We developed WordGen, an algorithm for species name recognition tolerant to OCR and typographical errors, using curated botanical dictionaries to reliably identify species sections in noisy and web-scraped texts. Results : We created a 345,000 species file dataset with natural language descriptions of global plant species from monographs, Wikipedia and databases. It achieved over 99% coverage and high recall on New Caledonia and Cameroon Floras datasets, enabling advanced AI-based plant taxonomy analysis.