Traditionally, researchers have claimed (e.g., Fillmore 198, Gillon 2006) that some verbs (such as eat and bake) lexically allow Implicit Objects (IOs) whereas other verbs (devour, kill) do not. This lexical idiosyncrasy is thought to explain why I at __ and I baked __ sound more natural than I devoured __ and I killed __. Other authors (e.g., Resnik 1993, Goldberg 2001, Scott 2006) have tried to explain such contrasts in a more principled way by exploring how IOs interact with discourse and information structure. In particular, they have observed that a verb’s object must be “recoverable” in the discourse in order to be omitted. In this paper, I explore two prongs of this “recoverability” criterion. As the first prong, I argue that “recoverability” is a matter of degree; some objects can be recovered with more precision than others. I show that a given context’s standard of “recoverability” is pegged to the speakers’ goals and interests, so that an IO can be only loosely recoverable when it does not bear on speakers’ goals, but must be more precisely recoverable when it is important. Turning to the second prong, I argue that an IO’s “recoverability” depends on the common ground of a particular community. For example, since athletes routinely lift weights, it’s part of their common ground that I lifted __ tends to mean I lifted weights. I report a simply corpus study showing that in communities where the action denoted by a given verb is associated with a routine action with a predictable object, as with lifting weights for athletes, the verb is more likely to appear with an IO. In both of these ways, I show that speakers’ interests and shared knowledge can help to explain the apparent idiosyncrasy surrounding English IOs.