Gist!

I

In 'Perception and Its Objects', P. F. Strawson puts the following description of visual experience into the mouth of a 'non-philosophical observer': 'I see the red light of the setting sun filtering through the black and thickly clustered branches of the elms; I see the dappled deer grazing in groups on the vivid green grass ' (2011, p. 127). The context in which this passage appears suggests that Strawson intends it to capture not only the objects of visual experience but also the ways in which those objects are presented in perception. The words that Strawson attributes to his naive observer may sound commonplace, but their implications are of course highly controversial (as Strawson was well aware). Perceptual experience can certainly represent the clustered shapes of the elms and the illumination falling on the grazing deer, but can it also represent the elms as elms and the deer as deer? That is not so clear.

The question of whether properties such as being an elm and being a deer can be given in perceptual experience is at the heart of the 'admissible contents' debate (Hawley and Macpherson 2011). On

II

'The Meaning of a Scene'. In his 1964 film, The Pawnbroker, Sydney Lumet inserted a brief scene representing the protagonist's distant memory. 1 Although the scene is presented for less than a third of a second and is unrelated to the flow of the narrative, the viewer has no difficulty grasping its meaning (Biederman et al. 1983). Lumet was exploiting a phenomenon that psychologists refer to as 'gist perception'. Although no precise definition of gist perception is available, the notion is often glossed as the amount of information that typical observers can absorb 'in a single glance' (Oliva and Torralba 2006).

It is common to distinguish two forms of gist: scene-based gist, in which the gisty property is attributed to the perceptual scene as a whole, and object-based gist, where the gisty property is attributed to particular objects in the scene. With respect to scene-based gist, typical observers are astonishingly adept at determining whether a presented scene is (say) natural or constructed, whether it is an indoor scene or an outdoor scene, and (for natural scenes) whether it represents (say) a forest, a beach, or a mountain (Greene and Oliva 2009;Oliva 2005;Oliva and Torralba 2006;Potter 1975). With respect to object-based gist, typical observers are astonishingly adept at identifying the high-level properties of presented objects, such as whether they are animals or vehicles (Li et al. 2002;Thorpe et al. 1996). I will focus here on scene-based gist.

Gist raises a number of interesting issues for philosophers of perception. One question concerns whether it is possible for perception to be purely gisty. This idea has sometimes been suggested in connection with the interpretation of Sperling's experiments on the V C 2016 The Aristotelian Society This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.

reportability of briefly-presented visual stimuli (for discussion see Block 2007;Grush 2007;Stazicker 2011). In these experiments, subjects are unable to reliably report the identities of particular symbols that are presented to them, but they are able to identify them as natural language symbols-or at least, as items that look like natural language symbols. The perceptual states here are sometimes described as involving 'generic perception', which can be thought of as a kind of object-based gist.

My primary interest here is not with the question of whether purely gisty perception is possible, but with the more fundamental question of how exactly we should understand gist in the first place. Is it a genuinely perceptual phenomenon, or is it a matter of postperceptual judgement? Do we perceive presented scenes as natural, or do we merely judge them to be natural on the basis of their visual appearance?

One reason for taking a perceptual view of gist seriously is that the relevant sciences appear to treat gist as perceptual. Research into gist is carried out by vision scientists, it is published in journals devoted to visual perception, and papers on gist bear such titles as 'High-Level Scene Perception' and 'The Time Course of Abstract Visual Representation' (Tatler et al. 2003). The opening sentence of a representative paper on gist reads, 'One remarkable aspect of human visual perception is that we are able to understand the meaning of a complex novel scene very quickly even when the image is blurred . . .' (Oliva and Torralba 2006, p. 23). Of course, the fact that psychologists and neuroscientists treat gist as perceptual hardly proves that it ought to be so regarded, but it does-it seems to me-give us good reason to take the view seriously.

Four considerations favour a perceptual treatment of gist, two of which are noted by Fish (2103). The first consideration is temporal: gist is detected extremely quickly. Whereas objects must usually be presented for about 150 ms in order to be identified (Evans and Treisman 2005;Fei-Fei et al. 2007;Rayner et al. 2009;Tatler et al. 2003), scene-gist can be accurately detected with exposure times that are as short as 20 ms (Thorpe et al. 1996;Fabre-Thorpe 2001; see also Joubert et al. 2007;Gordon 2004;Potter 1975;Schyns and Oliva 1994). Of even more direct relevance here is the time required to process gist information. Here too gist seems to This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.

have an advantage over certain representations that qualify as clearly perceptual, for there is evidence that gist can be recovered within 150 ms of stimulus onset (Thorpe et al. 1996). A study by Macé et al. (2009) found a processing advantage for superordinate categories (animal versus non-animal) over basic categories (dog versus bird) in naturalistic scenes, indicating that 'in the visual domain, the superordinate level may not constitute an abstraction from basic levels as previously proposed, but rather the rudimentary level at which some coarse object representations can be accessed with early crude processing of visual information' (Macé et al. 2009, p. e5927).

A second motivation for a perceptual treatment of gist detection is that it requires very little in the way of focal attention (Biederman 1972;Wolfe 1998). Li et al. (2002) asked subjects to report as quickly and accurately as possible whether images that were presented for only 27 ms contained an animal (or animals) whilst concurrently performing an attentionally demanding task. Performance on the attentionally demanding task was not significantly different from performance when subjects were presented with just the attentionally demanding task, indicating that object-based gist could be detected in the near absence of attention. Rousselet et al. (2002) provided further evidence that object-based gist is not attentionally demanding, by showing that subjects were as quick and as accurate in discriminating animals from non-animals when two natural images were presented together as they were when only one image was presented. On the basis of these studies, the detection of gist is sometimes said to involve no attention at all. That rather bold claim has been undermined by Cohen et al. (2011), but even they admit that 'natural-scene perception is so efficient and requires so little attention that the perceptual system must be properly taxed if this attentional cost is to be identified' (Cohen et al. 2011, pp. 6-7).

A third consideration for treating gist as perceptual concerns its role in directing and guiding perceptual processing. As Oliva and Toralba put it, identifying the gist of a scene serves to 'constrain local feature analysis and enhance object recognition in cluttered natural scenes' (Oliva and Toralba 2006, p. 23). Information about gist can guide the allocation of attention Neider and Zelinksy 2006;Zelinsky and Schmidt 2009) and the direction of eye-movements to provide a more efficient visual analysis of the scene, so that areas of importance are scanned first and in more detail (Castelhano and Henderson 2007). Objects that are inconsistent with scene gist are detected more slowly and less accurately than those that are consistent with scene gist (Biederman et al. 1982;Friedman 1979;Henderson, Weeks and Hollingworth 1999), while alterations to a scene that change its gist are more easily detected than those that leave its gist intact (Sampanes et al. 2008). These effects require the perception of the gist, and are not obtained when subjects are merely primed with a verbal description of the scene (Biederman et al. 1983).

A fourth reason for treating gist as perceptual concerns adaptation. Adaptation occurs when exposure to a certain property biases the perceptual system away from that property, thus producing characteristic after-effects. A well-known example of adaptation is provided by the waterfall illusion: a stationary object will appear to move upwards following the visual presentation of downwards motion. Although the issue is controversial (Storrs 2015), there is some reason to think that certain gisty properties exhibit adaptation. Greene and Oliva (2010) found evidence of after-effects for four global scene properties: mean depth ('the scene takes up kilometres of space' versus 'the scene takes up less than a few metres of space'), naturalness ('the scene is a natural environment' versus 'the scene is a man-made, urban environment'), openness ('the scene has a clear horizon line, with few obstacles' versus 'the scene is closed, with no discernible horizon line') and temperature ('the scene environment depicted is a hot place' versus 'the scene environment depicted is a cold place'). Importantly, these effects were also accompanied by changes in the gist-related judgements that subjects were disposed to make. For example, an image that was near the border between the categories of forest and field was more likely to be classified as a forest when subjects had previously viewed images that were high in openness (and were thus classified as fields), but that same image was more likely to be classified as a field after adaptation to closed images. 2 It would clearly be unwise to regard the perceptual status of gist as settled. There are multiple ways of drawing the contrast between perception and cognition, and the considerations appealed to here represent only a subset of those that might be regarded as relevant. As Masrour (2011) has noted, any attempt to draw a boundary around the perceptual system will depend in part on views about its function, and that is an issue about which there is debate. However, it seems to me that the case in favour of a perceptual treatment of gist is prima facie compelling, and that the burden of proof rests with those who take gist to be a purely post-perceptual phenomenon. With that in mind, let us turn now to consider the bearing of gist perception on the admissible contents debate.

III

Twin Earth, Goldilocks, and the Spatial Envelope. Although Fish eschews the notion of 'phenomenal content', he does take gist perception to show that there are 'no compelling reasons to think that the sensory, presentational component of visual experience must be limited to basic properties' (2013, p. 54). Is he right to do so? Is gist reflected in the phenomenal character of visual experience, and are gisty properties-that is, the properties that are represented in gist perception-genuinely high-level? Gist provides a case in favour of liberalism only if affirmative answers to both of these questions can be justified.

Gisty properties certainly sound high-level. Subjects report scene gist with such phrases as 'it's an urban scene', 'it's a beach', 'it's a forest', and they report object-based gist with such phrases as 'it's an animal', 'it's a vehicle', and so on. Indeed, even theoretical treatments of gist employ apparently high-level categories, such as 'naturalness' and 'navigability' (Greene and Wolf 2011). Although there is no algorithm for deciding when a property is 'high-level' as opposed to 'low-level', it is fairly clear that being a beach and being an animal ought to be grouped with canonically high-level properties (such as being a tomato) rather than those that are canonically low-level (such as being square). Showing that gisty properties can figure in phenomenal content would provide at least a partial vindication of liberalism, even if it failed to motivate the claim that being a tomato, a pine-tree or a watch can be perceptually represented.

IV

Phenomenal Reflection. What might the liberal say in response to the foregoing considerations? To make progress here we need to return to the notion of phenomenal content and the question of what it is for a property to be reflected by perceptual phenomenology.

The notion of phenomenal content is often understood in terms of metaphysical supervenience-that is, it holds across individuals and across possible worlds. On this account, a property is reflected by the phenomenology of a perceptual state just in case its representation is metaphysically necessitated by that phenomenal property. On this view, my phenomenal twins-that is, those individuals who instantiate exactly the same phenomenal properties that I do-will share all (and only) my phenomenal contents. Thus, if each of my phenomenal twins is representing its immediate environment as containing a red square of such-and-such a size, then I too must be representing my environment as containing a red square of such-andsuch a size. 4 Let us call phenomenal content thus defined strong content, and the notion of phenomenal reflection that it is intended to capture strong reflection.

At least some contributors to the admissible contents debate have been concerned with strong content (for example, Brogaard 2013),

VI

In his commentary on Siegel's The Contents of Visual Experience, Prinz remarks that although her view aligns with common sense 'insofar as we take ourselves to see lions and tigers and bears . . . it breaks from the kinds of information-processing stories that dominate in vision science' (Prinz 2013, p. 827). The burden of this paper has been to suggest that the story told by vision science is not quite as one-sided as Prinz's remarks suggest. Vision science might not support the claim that we perceptually experience lions, tigers and bears, but it does-I have argued-provide some support for the claim that we perceptually represent beaches, forests and cityscapes. The more general lesson to be drawn from the foregoing is that we need to distinguish different senses in which a property can be given in the contents of visual experience, and thus different versions of 'the' admissible contents debate. 7

Philosophy

122

TIM BAYNE