Object encoding in spoken language data and antipassives

This paper investigates object deletion in spoken Italian as a strategy for backgrounding the patient argument, drawing a parallel with antipassive (AP) constructions in a cross-linguistic perspective. Although Italian lacks a dedicated morphological AP, we examine whether the semantic and discourse-related factors known to condition AP derivation, such as agent affectedness, telicity, and contextual accessibility, also shape patterns of object realization in spontaneous discourse. Based on a corpus-driven analysis of 1.500 occurrences of 15 transitive verbs from the KIParla corpus, we combine a conditional inference tree and a random forest with a mixed-effects logistic regression model to assess the relative contribution of semantic, syntactic, and discourse-pragmatic predictors, while controlling for lexical variability. The results show that object deletion in Italian is primarily governed by contextual accessibility, which emerges as the strongest and most robust predictor across modelling techniques. Agent affectedness further modulates realization preferences, facilitating object deletion in event types that foreground changes in or consequences for the agent. By contrast, the other predictors do not exert independent effects once accessibility and lexical differences are taken into account. Overall, the findings highlight the importance of spoken data for a discourse-sensitive typology of argument realization.