Michael Devitt offered some thoughts at the ECAP on why referential uses of definite descriptions should not be thought of as conventional implicatures. Were they such, correct referential uses of 'The F is G' would convey both a proposition about whatever is uniquely F and (by implicature) a proposition about a particular thing. But (according to Devitt) in many cases the former is not conveyed at all. Consider for instance 'The book is on the desk'. This usually isn't supposed to convey any claim about some thing which is uniquely a book.
It might be argued in response that the speaker's quantifiers are supposed to be suitably restricted so that there is only one book in their range, or that 'the book' is elliptical for some longer description (such as 'the book in front of us'). But Devitt thinks this can't always be right, because in many cases, an uninformed or misinformed speaker would be unable to supply the required restriction on the quantifier or the required non-elliptical description.
I asked whether it is really fair to demand that the speaker be able to supply these things. Devitt replied that it must be facts about the speaker that determine the range of her quantifiers or the full form of her elliptical descriptions. But I wondered why it needs to be facts which are accessible to the speaker (as opposed to, say, facts about the speaker's causal relation to the world and/or other speakers of her language).