Dotlacil, J.
University of Groningen
It is well-known that when interpreting, we consider what has been said, as well as what has not been said. Thus, if someone mentions that some boys are tired, it is common to assume that the fact that the speaker did not use "all" instead of "some" suggests that the universal statement is not true (implicated meaning; implicature). Recently, several frameworks have been developed that model the way humans arrive at implicatures (e.g., Bayesian models, RSA theory). However, such statistical models are atemporal -- they only predict whether an implicature is drawn, not what amount of time it would take.
I present an ACT-R model that aims to capture the data from Bott and Noveck (2004) (B&N), who studied implicatures in sentences like "Some elephants are mammals". The model starts by learning implicatures related to "some" (by learning what quantifiers are relevant alternatives). This process arguably happens as part of acquisition. The resulting model applied to B&N and supplemented with simple reasoning module for category relations (elephants are mammals) correctly predicts mean RTs for target sentences with and without implicatures, as well as several mean RTs for baseline conditions which use "all" instead of "some".