We show how in the hierarchies F AE of Fieldian truth sets, and Herzberger's H AE revision sequence starting from any hypothesis for F 0 (or H 0 ) that essentially each H AE (or F AE ) carries within it a history of the whole prior revision process.As applications (1) we provide a precise representation for, and a calculation of the length of, possible path independent determinateness hierarchies of Field's construction in [4] with a binary conditional operator.(2) We demonstrate the existence of generalised liar sentences, that can be considered as diagonalising past the determinateness hierarchies definable in Field's recent models. The 'defectiveness' of such diagonal sentences necessarily cannot be classified by any of the determinateness predicates of the model. They are 'ineffable liars'. We may consider them a response to the claim of [4] that 'the conditional can be used to show that the theory is not subject to "revenge problems".'
The ScopeThe purpose of this note is to investigate more closely the hierarchies of truth sets produced by the revision sequence process. The first hierarchy, the one produced by Herzberger, [12], [11], was invented to test how various self-referential sentences in a language containing names for elements of a ground model M , and sufficient to define such diagonalising sentences, would behave under repeated applications of the Tarskian definability scheme which produced repeatedly truth sets. Herzberger allowed this process to proceed into the transfinite by using a liminf rule (all of which we specify in more detail below). This revision process has been the subject of various investigations and extensions, notably by Gupta and Belnap in a series of papers, but also in the book [9].
1More recently Field in e.g. [4], has used such a liminf revision process, to analyse the consequences of adding a binary operator°! to a language similar to the above, with Tr a truth predicate. Field takes at each successive stage not just a new level of definability in the Tarskian sense, but a strong Kleenean fixed point (à la Kripke [15]).The two sequences of sets we shall dub here hH AE | AE 2 Oni (the "H -sets") and hF AE | AE 2 Oni (the "F -sets") (where On denotes the class of ordinals). When defined over the same model, such as M = they are, mathematically at least, surprisingly similar. Indeed we showed in [22] the stability sets consisting of those sentences that are in all the H -sets from some point on, and Field's ultimate truth sets are recursively isomorphic -that is there is a pencil and paper algorithm for converting members of one set into the other, and conversely. Of course this is not to say that the members of the final sets are the same or have the same intended meaning. The phenomenon we are seeing here is that the liminf rule is acting as some kind of very powerful infinitary logical rule. One can show that whatever one does (within some considerably wide bounds) at successor steps will be swamped in effect by the limit rule. This is why the two ultimate sets a...