This presentation aims to illustrate the results of the development of a reference dataset of lemmatised Sanskrit texts through the training of machine learning models adapted to the peculiar linguistic complexities of Sanskrit, particularly the subdivision of sandhi, a neural network-based model of which is presented. The creation of the dataset made it possible to compare some significant terms in the Sanskrit translation of the Nicene-Constantinopolitan Creed with the texts in the reference dataset: the occurrences found allow us to verify or illuminate the translation strategies present in W.H. Mill's translation. The differences and semantic shifts found provide stimuli for comparative investigations and perspectives in historical-religious, philosophical, and philological studies.