Ignorance implicatures of modified numerals

Abstract Modified numerals, such as at least three and more than five , are known to sometimes give rise ignorance inferences. However, there is disagreement in the literature regarding nature of these inferences, their context dependence, differences between . We present a series experiments which sheds new light on issues. Our results show that (a) inferences robust those (b) presence strength triggered by both depends question under discussion (QUD), (c) whether detected given experimental setting partly task participants asked perform (e.g., an acceptability versus inference task). offer Optimality Theoretic account findings. In particular, effect captured assuming performing task, take speaker ’s perspective order determine expression optimal certain epistemic state, while they addressee what most likely state expression. To execute latter fully rational manner, have higher-order reasoning about alternative expressions could used. Under assumption do not always but also often resort so-called unidirectional optimization finds natural explanation. This allows us relate our finding asymmetries comprehension production been found language acquisition.