ABSOLUTE PRODUCTION AS A ± POSSIBLE ± METHOD TO EXTERNALIZE THE PROPERTIES OF CONTEXT DEPENDENT INTERNAL REPRESENTATIONS

Absolute identification requires a participant to identify which stimulus has been presented, from a relatively small pre-specified set ± the stimuli might be a set of 9 lines, with the shortest labelled µ#1¶, the longest µ#9¶. Absolute identification yields many well-established behavioral phenomena- for example, more extreme stimuli (such as lines #1 and #9 in the above example) are identified more accurately, and responded to faster, than more central stimuli (such as lines #4 and #5). Contemporary models propose that the internal representations of the stimuli, and/or of the responses, are context dependent in that they depend on factors such as the number, range, and presentation frequency of the stimuli. We present data on absolute production that we interpret as µexternalizing ¶ the properties of such internal representations and that show effects paralleling those in absolute identification. Absolute identification involves identifying which stimulus has been shown out of a set of stimuli that vary on only one physical dimension. For example, a participant might be given a set of n lines varying in length, or tones varying in intensity, labelled from 1 through to n. The participant is then presented with one of these stimuli and asked to recall its label. There are standard results from such experiments (see Stewart et al, 2005) such as a bow effect in mean proportion correct (Figure 1a) and a W-effect in mean proportion correct as a function of the difference in magnitude between the stimulus presented on the current trial and that presented on the previous trial (Figure 2); these data are averaged over participants. Practice increases the magnitude of bow (Rouder et al., 2004). The SAMBA (selective attention mapping ballistic accumulators) model (Brown et al, 2008) successfully models various such data as illustrated by the fits in Figures 1 and 2. Figure 1b) plots the mean standard deviation (SD) of the numerical responses (10 in the present case) and complements the probability correct plot of Figure 1a). We include the SD plot as it ha