Pedro Sol Santos

Attempts at accounting for so-called embedded conversational implicatures (CIs) come up against the basic problem of making sense of kinds of content that, despite being derived from bona fide conversational implicatures, seem to be part of the literal meaning of utterances. The problem resists the conventionalist way out (championed, among others, by Chierchia and Levinson), which construes generalized CIs as not meriting the status of CIs proper, this being one way of dealing with the puzzle. One decisive drawback of this strategy is that particularized CIs, which could not credibly be conventionalized, display the same kind of puzzling embedding behavior as generalized CIs. In this paper I discuss Mandy Simons’s and François Recanati’s approaches to the conundrum. Simons assigns to embedded CIs the status of bona fide Gricean inferences, building on the distinction between embedded pragmatic effects and embedded pragmatic computations and claiming that the puzzle is solved by the realization that only the latter phenomenon is really problematic and that embedded CIs are to be analysed as cases of the former rather than as cases of the latter. Recanati has (arguably justifiable) qualms about this sort of approach and attempts to account for the phenomenon by dismissing (somewhat in the vein of the conventionalists) the very notion of an embedded CI as intrinsically incoherent. He purports to do this on the basis of a globality criterion on Gricean inferences that he argues cases of alleged embedded CIs only partially fulfil, so that they count as cases of Gricean inferences latu sensu (as, for instance, are cases of what he calls modulation) but surely not as genuine CIs. I argue that Recanati’s take on the problem cannot be successful, as globality is a faulty criterion as far as distinguishing cases of (embeddable) modulation from CIs proper is concerned.