Kahneman et al. (1992) introduced a technique for examining how the visual system maintains a record of currently relevant objects. Fundamental to their account of this process is the notion of an object file: defined as a ‘temporary “episodic” representation’ (of a real world object; p. 176). Briefly, object files are created for currently relevant objects to maintain records of appropriate object features. Such a temporary record is said to be ‘addressed by its location at a particular time’ (p. 178) and is said to collect sensory information about a particular object at a given location. As the optic array changes, the information in the object file is modified accordingly. With the passage of time, an object file will be discarded as a given object loses its relevance for the observer. The central point is that object constancy is mediated via object files.

In putting forward a more general account of visual information processing, Kahneman et al. discussed a three component process model. By this account, in order to accommodate change, the visual system must first establish a correspondence between current objects and those just viewed. Following this, a reviewing process operates so as to retrieve the characteristics of prior objects from the object files. Finally an implementation process operates as a comparator between old and new information so as to produce a percept of change or motion between the two views. Kahneman et al. were initially concerned with the reviewing process and attempted to examine the sorts of information said to be contained in the object files. In their experiments a number of letters was presented in a brief preview field. Then, following a short interval, a single coloured letter (the target) was presented and subjects had rapidly to name it. Kahneman et al., reported many variants on this theme, but for present purposes it is best to consider a subset of their preliminary results. In their first study, letter positions were marked by a random array of small rectangular frames. A given letter could then occur inside one of these frames. In the relevant conditions, two preview letters were presented for 250 ms. Their positions will be labelled P1 and P2. The letters then disappeared and the frames remained empty for 300 ms. This presentation of empty frames was known as the linking display. Finally a target letter was presented at either P1 or P2. The three conditions of interest were as follows; (i) Same Object – a preview letter occurred at its previous position, (ii) Different Object – the preview letter shown at P1 now occurred at P2 (or vice versa), and, (iii) No Match – a new letter was presented at P1 or P2.

Kahneman et al., discuss the results in terms of three kinds of effects – (i) a Same Object preview effect indexed by the difference in mean reaction times (RTs) to the No Match and Same Object conditions, (ii) a non-specific preview effect indexed by the difference in RTs to the No Match and Different Object conditions, and, (iii) an object-specific preview effect indexed by the difference in RTs to the Different Object and Same Object conditions. Each set of results from their experiments was discussed in terms of these three kinds of effects. Special significance was attributed to those cases where an object-specific preview effect arose that could not be accounted for solely in terms of some form of generalised priming. That is, their primary interest was in those cases where the object-specific preview effect was significantly larger than the corresponding non-specific preview effect. These cases were taken as being indicators of processes concerned with object constancy, and, were also taken to support the theoretical notions of object files and the putative processes such as object file “reviewing”.

According to Kahneman et al. (1992) the term reviewing “refers to the process in which a current target item evokes an item previewed in an earlier visual field” (p. 176). Although they at no stage discuss a detailed account of exactly how such processes interact or influence the processes underlying naming, they do state that reviewing facilitates recognition when the current and previous states of the object match and ‘hampers’ it otherwise. The tacit assumption, nevertheless, is that naming directly reflects the costs and benefits of reviewing. Whether such an assumption is warranted, is, in part, the focus of the current research.

Put simply, in their first experiment, and in several others, Kahneman et al. found significant object-specific preview effects: only occasionally, for instance, with long exposure preview fields, did they obtain significant non-specific benefits in addition to the object-specific effects.