(1) (a) I'm not happy ==> (b) I'm unhappy.
Though strictly speaking (1a) merely denies that the speaker is happy, it is likely to be understood as implying (1b). Note that this implication is not an entailment: for having uttered (1a) the speaker might just as well continue by saying that he is not unhappy either.
A similar phenomenon is observed in so-called "neg-raising" contexts' where an external negation of an attitude verb is construed as if it had narrow scope:
(2) I don't believe it is raining ==> I believe it isn't raining
In this talk we want to focus on examples where the existence of morphological alternatives raises interpretive blocking effects, as in the following:
(3) I'm not unhappy =/=> I'm happy.
(4) I don't mistrust Fred =/=> I trust Fred.
In contrast to total blocking (as in "*unfoolish" or "*unsad"), blocking is partial in these examples: (3) implicates that the speaker is less than happy, and (4) would normally imply that the speaker's trust in Fred is less than complete. The pattern of partial blocking found in these examples exhibits a form of iconicity: given an expression with two interpretations, the conceptually more marked (less prototypical) interpretation is preferred in case there is a blocking expression that conveys the unmarked interpretation in a more efficient (less costly) manner.
In the second part of our paper the interaction of R-based strengthening and blocking is described within the framework of bidirectional optimization. This framework explains the iconicity pattern in a straightforward way, and thus accounts for the data under discussion.