The discrepancy between visual information being delayed and behavior being accurate is unquestionable. It therefore might seem odd to take up the functionality of visual compensation. However, why should there be sensory compensation when later end effectors can compensate adequately? Of course, here one might ponder whether perceiving at all has functional significance, given that actions can be undertaken in the absence of accurate vision in terms of position information (Goodale et al. Reference Goodale, Pelisson and Prablanc1986). It is at this juncture in the present exposition that re-branding spatial extrapolation as visual prediction makes a tangible and productive difference.
In its current conception visual prediction does not treat vision as separate and independent of action. In fact, Nijhawan now argues for extrapolation being a two-stage process in that compensation first occurs in the dorsal action pathway and then is communicated to the ventral perception pathway. In this way, any conflict between action and perception systems is obviated. While the workings of the action system are grounded in accurate behavior, how does one go about proving the reality of extrapolated percepts?
The tack that I favor is to look for perception that varies lawfully as a function of spatial offset. If certain spatial percepts without recourse to extrapolation result in already established patterns of performance, and these very percepts can be created via extrapolation to render equivalent patterns of performance, then extrapolated percepts are no different than non-extrapolated ones. Furthermore, if the perceptual reality of the non-extrapolated percepts is unquestionable, then logically the same must hold for extrapolated percepts.
A spatially modulated phenomenon in face processing provides the perfect candidate for assessing the perceived reality of extrapolated vision. When two face halves belonging to two different individuals are spatially aligned, the resulting “facial chimera” is difficult to read for the identity of either face half (Young et al. Reference Young, Hellawell and Hay1987). However, the introduction of a spatial offset or misalignment greatly aids identification of the constituent face halves. Observers viewed facial chimeras that consisted of one moving face half and one flashed face half (the flash-lag effect [FLE] facial chimera) in a flash-initiated display with instructions to report either the offset or identify of the flashed half. Observers both “perceived” the face halves to be spatially offset and also produced identification times that were faster, as is typical for constantly present spatially offset face halves. Now, if one does not question the account given for the chimeric face effect without extrapolated stimuli, then the same should apply to stimuli that are a result of extrapolation. The observed reaction time pattern with FLE facial chimeras is clearly indicative of extrapolation being part and parcel of the ventral system and perception (Khurana et al. Reference Khurana, Carter, Watanabe and Nijhawan2006). The analogous logic applies to a situation in which the extrapolated percept is one of alignment and observers both perceive stimuli to be aligned and produce responses that are concordant with the percept.
Although these findings with FLE facial chimeras provide clear evidence for extrapolation in the ventral stream by measuring not only perceived offset but also the knock-on effects on identification latencies, other findings do not sit so comfortably with the notion that the flow of extrapolated information is from the dorsal pathway to the ventral pathway. It was also shown that when the two face halves are of the same familiar individual, the FLE is lawfully smaller. In the present context, these findings serve as counter examples. If R D is the dorsal representation of FLE facial chimeras and R V is the ventral representation in higher cortical face areas, then the smaller perceived spatial offset between face halves that belong to the same individual versus those that belong to different individuals can be accounted for as follows. The extrapolated moving face half makes contact with an underlying complete representation of the known face, thus priming the registration of the flashed face half. A reduction in the perceived spatial offset of same face halves is indicative of influences of ventral face processing on spatial dorsal computations. It might be that Nijhawan's proposed two-stage account is in actual fact two parallel processes – one dorsal and one ventral, each equally capable of influencing the other.
Until recently, the face data could be thought of as an existence proof of ventral influences on FLE; however, other well-learned representations can also reduce the magnitude of the FLE (Noguchi & Kakigi Reference Noguchi and Kakigi2008). For Japanese observers, it was shown that perceived FLE for Kanji character segments was significantly reduced relative to non-meaningful geometrical segments. Importantly, this diminution was neither observed for pseudo-Kanji characters viewed by Japanese observers nor for Kanji characters viewed by non-Japanese English-speakers with no knowledge of Kanji. Thus, both faces and meaningful complex visual forms can affect spatial computations.
Although the flash-lag effect is often likened to other non-veridical percepts such as the Fröhlich effect and representation momentum, the functional significance of extrapolation lies in the visual system generating a percept that matches reality at the time of perceptual awareness. Both the examples above of cortical post-constancy late representations superseding retinal pre-constancy early representations in visual awareness are indicative of extrapolated percepts, having real consequences for behavior.
