The (in)felicity of a dog 'or' a reindeer. On the interpretation and processing of ignorance and scalar inferences triggered by disjunction

Disjunction can give rise to both scalar implicatures and ignorance inferences. While scalar implicatures yield an exclusive reading (e.g., A or B, but not both), ignorance inferences convey the speaker’s uncertainty about which disjunct holds. However, the relationship between these two types of inference and their processing dynamics remains largely unexplored. This study investigates how the speaker’s epistemic state and the number of visible objects modulate the interpretation and processing of disjunctive sentences in Italian. We employed a sentence-evaluation task in which participants (n = 141) judged statements produced by a dealer in a card game who either described (Description Mode) or guessed (Guess Mode) the dealt cards without seeing them (Guess Mode). The number of cards displayed was manipulated (2 vs. 4 cards). Results showed that participants were sensitive to scalar implicature violations, with lower acceptance rates for underinformative than for true disjunctions across all conditions. However, acceptance rates remained relatively high (> 74%), suggesting widespread tolerance of pragmatic violations. The number of objects in the visual scenario significantly affected acceptance rates of disjunctive statements, with the 4- card version eliciting higher agreement, particularly in Guess Mode. Response time analyses revealed asymmetric processing costs: accepting true disjunctions incurred a processing penalty, consistent with default derivation of ignorance inferences and their subsequent cancellation, whereas accepting underinformative disjunctions yielded lower response times, suggesting non-default computation of scalar implicatures. Crucially, a three-way interaction between Mode, Condition, and Response Type indicated that the processing cost of violating ignorance inferences was greater in Description Mode. These findings suggest that ignorance inferences are computed more systematically than scalar implicatures, can be cancelled at a cognitive cost, but do not consistently affect overt judgments.