Today, many projects focus on the application of linguistic technologies on modern medical corpora, especially in the field of Named Entity Recognition. Besides, ancient pharmacopoeias are being explored with manual data entry by specialists in history and biology in order to extract knowledge. These analyses are carried out without necessarily going through the automatic recognition of named entities which could accelerate the exploration of the manuscripts. Therefore, we propose here a link between the two practices by: (1) creating a named entity recognition dataset for English translations of medieval Arabic pharmacopoeias and (2) training and evaluating language models that are pre-trained on multiple domains.