Mining historical recipe collections
The early modern period witnessed a boom in how-to books. So-called books of secrets were large compendia of recipes that could benefit greatly from being analysed through computational methods.
The early modern period witnessed not only the expansion of general knowledge-organisation resources but also a growing demand for practical knowledge, which was, among other things, expressed through “how-to” books and so-called “books of secrets”. These compilations of alchemical, medical, and household recipes often contain hundreds of entries, many already presented in relatively standardised formats. Despite their richness, they have not yet been analysed as quantitative sources.
This project uses large language models to identify and annotate ingredients within these recipes, followed by human verification to create an intermediary data format. The data are then transformed into RDF, allowing related recipes to be grouped and compared.
Such analysis addresses the tendency of compilers to expand their collections by presenting minor variations as entirely new recipes, whether through the substitution of ingredients or alternative descriptions of chemically equivalent processes.
In a subsequent stage, selected recipes will be investigated through computational chemistry methods within the quantum chemistry subproject, enabling the exploration of their underlying chemical processes through in silico simulation. By identifying these relationships computationally, the project seeks to reconstruct patterns of transmission and transformation across large corpora of recipes.