The assumption an atelic usage of a inherently telic form is wrong finds its origins in what is called the classical model of categorization, which (usually implicitly) assumes that categories are fixed subject-external entities with well-defined boundaries and no ambiguity. This model has been common in linguistics since Roman Jakobson put forward his model of phonological features in the 1930’s and then developed into componential analysis by linguistics like Eugene Nida (cf. 1979). It is a structuralist model that assumes that the real world realization of categories (in this case: states and completives) are solely defined on the basis of binary feature assignment and that any extended usages should be derived from these features. So if completives are +telic then all usages of completives should also be +telic. The problem is that human categorization, including what we find in language, does not function like this.
To state it plainly in terms of the question of intensive state perfects and completive semantics, the idea of there being a contradiction in the stative (atelic) uses of a completive gram (telic in its central usage) is based upon the misguided structuralist assumption that meaning exists only as a result of the formal oppositions that exist in the language system, as described by Nida (1979) and more recently by Nida and Louw (1992).
In contrast to this approach, I take seriously the critique of classical categories put forward by proponents of prototype theory, which goes back to in assuming that the units we should compare across language are not features of a componential analysis (such as Jakobsonian distinctive features) but the semantic content of each gram, which may be thought of as focal points in conceptual space. Grams do not derive their meaning from the oppositions they enter into in a language, but rather have semantic content of their own which contributes to the formation of the conceptual system of the language (Bybee, Perkins, and Pagliuca’s (1994 45-6; see also Dahl 1985). Linguistics categories have their own inherent meaning independent of the linguistic system.
Prototype theory takes a different view of how categories work and how they are defined. Rosch’s (1978) emphasizes that the prototypical instantiations of a given category are maximally distinct from each other. She states,
To increase the distinctiveness and flexibility of categories, categories tend to become defined in terms of prototypes or prototypical instances that contain the attributes most representatives of items inside and least representative of terms outside the category (1978, 30).
This is essentially a double characterization in that prototypicality is here defined both positively (most representative attributes) and negatively (least representative attributes). One implication of this definition is the fact that when we are dealing with two or more contrastive categories, non-prototypical usage of one of those categories will likely involve some of the prototypical attributes of that category, but also some prototypical attributes of another category. The logical result of this fact is that it may be entirely possible for two contrasting categories to reflect near synonymy in some discourse contexts—contexts where the most representative attributes of their given category are dramatically downplayed. These sorts of non-contrasts between grammatical categories are prototype effects derived from human cognition.
These facts are derived not merely from how categorization works for a limited set of items (e.g. lexemes), but how it functions for human cognition and reasoning in general. That is to say, all categorization is prototypical categorization. On this basis it makes perfect sense that a prototypically telic category can and does have a non-prototypical usage that is not telic. The non-prototypical usage exists comfortably on the boundary between the prototypical atelic and the prototypical telic sharing aspects of each. Thus, intensive state usage of the completive perfect shares aspects of stative semantics and aspects of copmletive semantics, while also not full adhering to either.
Given what we know about how language (and human categorization in general) works, this situation should not be a surprise or viewed as contradictory. It should viewed as expected from the outset and we should anticipate find such language data from the very first.
Somehow this blog post got much longer than I intended and somehow there is still far, far more that could be said. A fuller discussion of my views on the nature of semantics can be found in Widder et al. (forthcoming) in the chapter on Greek and linguistics.
Rosch, Eleanor. 1975. Cognitive Representations of Semantic Categories. Journal of Experimental Psychology 104 (3): 192–233.