From Lexical Theory to AI Translation: Procomplement Verbs and ChatGPT

Procomplement verbs (PCVs) are a highly frequent yet understudied class of semantically idiosyncratic verb–clitic constructions found across Romance languages. Despite their relevance for everyday communication, they remain only marginally represented in grammars and dictionaries. Building on recent theoretical work on Italian PCVs and drawing on Generative Lexicon theory, this study investigates the relationship between the syntactic–semantic properties of PCVs and their treatment by large language models. A dataset of 160 highly frequent Italian PCVs, primarily extracted from GRADIT and comprising 256 usage examples, was submitted to ChatGPT 5.2 for translation from Italian into English. The analysis focuses on the model’s ability to account for the semantic, pragmatic, and phraseological complexity of these constructions. The results show that translation quality is influenced less by the specific subclass of PCVs than by transversal properties such as connotation and the pragmatic functions encoded by individual utterances. While ChatGPT generally succeeds in translating strongly conventionalised and highly idiomatic expressions, it frequently encounters difficulties with polysemous PCVs, context-sensitive pragmatic meanings, and morphologically complex verb–clitic combinations. The findings highlight the value of a fine-grained syntactic–semantic classification of PCVs for bilingual lexicography and suggest potential applications for the improvement of AI-assisted translation systems.