Terminologization of Anthropomorphic Metaphors in the Italian Discourse around AI

This paper investigates anthropomorphic language in Italian discourse on artificial intelligence (AI), asking whether and how lexical expressions attributing human traits to AI systems undergo terminolo- gization, the process by which a metaphorical expression is conventionalized as a technical term, with a corresponding loss of perceived figurativity. After building a targeted web corpus through a semi- automated bootstrapping procedure (BootCaT, 2004), a dataset of 162 anthropomorphic expressions was extracted from 68 accurately selected Italian sources (2023–2026) and annotated across four levels: semantic category (Inie, Zukerman, and Bender (2026)’s eight-category taxonomy), grammatical level, distancing markers (quotation marks, negation, hedges), and source register. Five hypotheses tested fre- quency, diachronic trajectory, grammatical level, register injection, and semantic category as proxies for terminologization. Results show a structural split rather than a uniform trend: nominal role-metaphors terminologize silently, as predicted, showing low distance marking despite high frequency, while ver- bal cognition-metaphors remain heavily marked and negated regardless of frequency, sustained by an ongoing public debate on machine cognition. Contrary to expectation, distancing markers increased corpus-wide between 2023 and 2026, driven by implicit rather than explicit hedging, suggesting that crit- ical metalinguistic discourse can itself act as a brake on terminologization. At the individual-lemma level, there are cases where the three tested proxies behave as expected: marking falls with frequency, over time, and as terms enter institutional use. The two findings are not contradictory: individual terms qui- etly terminologize even as the broader critical discourse keeps the metaphorical status visible, slowing terminologization as a corpus-wide phenomenon.