DEEP LEARNING E TRADUZIONE INTRALINGUISTICA: RIFORMULARE I TESTI DELLA PUBBLICA AMMINISTRAZIONE IN MODO INCLUSIVO

In this article, we present the experimentation of the E-MIMIC device (Empowering Multilingual Inclusive Communication), an application based on neural networks that aims at making administrative texts more inclusive. As result of the collaboration of data scientists and linguists, this device is giving particularly significant results for the Italian language of Italy and for the French language of France. E-MIMIC was designed as an intralingual translation device. This allows us to exploit, by analogy, previous knowledge related to interlinguistic automatic translation and the problems it raises (e.g. the “smoothing” of linguistic diatopia and biases produced by this type of translation). The novelty of this device lies on the one hand in the AI models adopted, which provide for supervised learning through the “humans in the analytics loop” system, and on the other in the linguistic-discursive criteria used to improve and optimize the deep learning of the networks. From this point of view, E-MIMIC also becomes an experiment to propose new professional profiles in the translation and linguistic fields, straddling linguistics and computational or computer sciences.