Athanassios Raftopoulos University of Cyprus, Dept. of Psychology [e-mail]

1
Athanassios RaftopoulosUniversity of Cyprus,
Dept. of Psychologyraftop_at_ucy.ac.cy

• THEORY LADENNESS OF PERCEPTION, CONCEPTS, AND
  COGNITIVE IMPENETRABILITY OF PERCEPTION

2

• In this paper I argue that since there is
  evidence that there exists a stage of visual
  perception, namely early vision, which is
  cognitively impenetrable and has non-conceptual
  content (NCC), and since I take cognitive
  penetrability, conceptual encapsulation, and
  theory ladenness as coextensive, the cognitive
  impenetrability of early vision entails that the
  content of early vision is theory neutral.

3

• Cognitive encompasses activities involving
  propositional attitudes. Since concepts are
  construed as constant, context independent, and
  freely repeatable elements that figure
  constitutively in propositions cognition
  necessarily involves concepts.
• Concepts are embedded in frameworks wherever
  there are concepts there are conceptual
  frameworks. Hence, cognitive states have
  conceptual contents and the issue of the
  cognitive impenetrability of perception is the
  problem of whether the contents of perceptual
  states are influenced by ones concepts.

4

• I have explained why cognitive penetrability of
  perception and conceptual effects on
  perception should be treated as
  coextensive.Conceptual frameworks constitute
  theories (in a broad sense of the term and not
  necessarily in a strict scientific sense, since
  there is no requirement that conceptual
  frameworks be free of contradictions), and thus,
  the cognitive penetrability of perception, by
  implicating the conceptual apparatus, also
  signifies the theory-ladenness of perception.

5

• I have proposed a definition of nonconceptual
  content (NCC) according to which X is in a
  representational state S with NCC P, if X has (or
  is being disposed to have) a content that is
  (directly) causally connected in a certain way to
  instantiated Phood independent of the cognitive
  states of X. The content P of state S is
  independent of the content of the cognitive
  states of X iff the contents of the cognitive
  states do not enter into P, in that there is not
  an epistemic relation between the conceptual
  content and the content of the perceptual state.
  In this case, the content of the perceptual state
  is cognitively impenetrable. Hence, NCC is the
  content of perceptual states that are cognitively
  impenetrable (or conceptually encapsulated).

6

• Early vision includes a feed forward sweep (FFS)
  of signal transmission in which signals are
  transmitted bottom-up and which lasts, for visual
  areas, for about 100 ms, and a stage at which
  there lateral and recurrent connections between
  neurons allow recurrent processing.
• This recurrent processing, which starts at about
  80 ms after stimulus onset, is restricted within
  visual areas and does not involve signals from
  cognitive areas. Lamme (2003) calls it local
  recurrent processing (LRP). LRP culminates at
  about 120-130 ms. After that period, signals from
  higher executive centers including mnemonic and
  executive. This is the stage of late vision.

7

• The thesis of cognitive impenetrability of early
  vision says that cognition only affects
  perception by determining where and to what
  attention is focused. It does not in any more
  direct way alter the contents of perceptions so
  that they be logically/epistemically connected to
  the content of beliefs, expectations, and so on.
  By not affecting directly the contents of early
  vision, ones cognitive states do not determine
  the way that person perceives the visual scene.
  Thus, cognition affects perception only when a
  perceptual systems content is altered in a way
  that makes it bear some logical relation to the
  contents of ones cognitive states.

8

• When cognitive states affect directly perceptual
  states, their respective contents bear some
  epistemic relation and the perceptual content is
  conceptual content. The cognitive influences that
  operate either pre-perceptually or
  post-perceptually I call indirect causal
  influences on perceptual processing and states.
  They are distinguished from direct causal
  influences that affect perceptual processing
  itself. Only in the latter case do perceptual
  contents bear an epistemic relation to cognitive
  contents and are cognitively penetrated.

9

• if the phenomenal content of experiences is
  causally penetrated by the content of cognitive
  states, then the contents are cognitively or
  conceptually penetrated.
• The contents of experiences are causally
  penetrated by the contents of cognitive states,
  if the causal explanation of how or why it is
  that the perceiver is in a state with this
  content as opposed to some other content, has to
  take into account not only the perceivers
  position and the environmental perceptual
  conditions, but also the cognitive abilities she
  possesses.

10

• It is also arguable that the phenomenal content
  of this experience is NCC in that it does not
  constitutively depend on conceptual content (Tye
  1995, 140 2000, 61). It follows that a state can
  have cognitively penetrated content and yet this
  content be NCC.
• Thus, cognitive penetrability is not a necessary
  condition for NCC and the debate regarding
  cognitive (im)penetrability is orthogonal to the
  debate regarding the conceptual vs. nonconceptual
  character of the content of perception.

11

• Let the ambiguous figure be the duck/rabbit
  ambiguous figure and suppose that some cognitive
  state makes one focus attention at some point on
  an ambiguous figure and this focus makes one see
  a duck-like figure (the term duck-like figure
  signifies the NCC of ones visual experience of a
  duck, that is, what one perceives even if one
  does not possess the concept duck) as opposed
  to a rabbit-like figure.
• In this case, the content of the cognitive state
  must be invoked in a causal explanation of what
  is the phenomenal content. According to the
  intuitive view, the content is cognitively
  penetrated.

12

• The thesis of cognitive impenetrability as I
  construe it explicitly denies this possibility.
  What one chooses to attend to may be determined
  by cognitive factors. However, this type of
  modulation of neuronal activity by spatial
  attention consists of two contributions.
• The first concerns the enhancement of P1 ERP
  waveform for stimuli at the attended location in
  extrastriate visual areas with a latency of
  7-0-100 ms after stimulus onset. This is clearly
  an exogenous attentional factor since it is
  data-driven and not theory-driven. As such, it
  does not entail any sort of top-down conceptual
  influence on early visual processing.

13

• Second, attending to a location may enhance the
  spontaneous firing rates of the neurons tuned to
  the attended location in extrastriate areas V2,
  V3/V3a, V4, in parietal regions, and in V1 before
  the presentation of the stimulus.
• However, this effect does not determine what
  subjects perceive at that location because by
  enhancing the responses of all neurons tuned to
  the attended location independent of the neurons
  preferred stimuli it keeps the differential
  responses of the neurons unaltered and thus does
  not affect what it is perceived at that location.

14

• Therefore, spatial attention does not constitute
  a direct causal effect on early vision and does
  not entail its cognitive penetrability. By
  determining focus, the content of a cognitive
  state indirectly determines the content of an
  early vision state and this is why concepts
  figure in a causal explanation of the percept.
• However, this indirect kind of determination does
  not entail the cognitive penetrability of the
  phenomenal content, that is, it does not entail
  that the content bears some epistemic relation to
  the content of some cognitive states.

15

• Even if one grants that early vision is not
  cognitively penetrable and is, thus, conceptually
  encapsulated, one could argue that since learning
  affects the way one sees the world, some of our
  experiences are learned and form memories that
  are stored in visual memory. These memories
  contain conceptual elements that are somewhat
  built in the perceptual system and affect
  processing from its inception.
• Thus, our experiences shape the way we see the
  world and this means that people with different
  experiences see the world differently. Moreover,
  concepts do affect early vision not in a top-down
  manner but by being built in it.

16

• Familiarity can affect visual processing in
  different ways. It may facilitate object
  identification and categorization, which are
  processes that take time since their final stage
  occurs between 300-600 ms as is evidenced by the
  P3 ERP wave form in the brain) and their earlier
  stage starts about 150 ms onset as is evidenced
  by wave forms that are elicited about 150 ms and
  are thought to be early manifestations of P3.
• Familiarity is in general considered to intervene
  during the latest stage of object identification
  (300-360 ms) and its effects are considered to be
  post-sensory since they involve the cognitive
  levels of the brain at which semantic information
  and processing, both required for object
  identification and categorization, occur.

17

• These sorts of effects seem to pose a threat to
  the cognitive impenetrability of perception since
  they cannot be considered post sensory.
• The threat would materialize should the
  classification processes either require semantic
  information to intervene or require the
  representations of objects in working memory to
  be activated since this would mean conceptual
  involvement too.
• However, researchers agree that the early
  classification effects in the brain result from
  the feed forward sweep and do not involve
  top-down semantic information, nor do they
  require the activation of object memories.

18

• The early effects of familiarity may be explained
  by invoking contextual associations (that is,
  target-context spatial relationships) that are
  stored in early sensory areas to form unconscious
  perceptual memories, which, when activated modify
  the feed-forward sweep of neural activity
  resulting in the facilitating effects mentioned
  above. This is a case of rigging-up the early
  visual processing it is not a case of top-down
  cognitive effects on early vision.

19

• The early effects may also be explained by
  appealing to the storage of configurations of
  properties of objects or scenes.
  Neurophysiological, psychological research, and
  computation modeling all suggest that what is
  stored in early visual areas are implicit
  associations representing fragments of objects
  and shapes, as opposed to whole objects and
  shapes.
• Since the implicit associations can affect
  figure-ground segmentation, in view of the fact
  that figure-ground segmentation occurs early
  (80-100 ms) these associations must be stored in
  early areas and cannot be representations stored
  in, say, anterior IT, an area which is involved
  in working memory function. The earlier visual
  areas store object and shape fragments and not
  holistic figures and shapes.

20

• The associations in early visual circuits reflect
  the statistical distribution of properties in
  scenes. The statistical differences in physical
  properties of different subsets of images are
  detected very early by the visual system before
  any top-down semantic involvement as is evidenced
  by the elicitation of an early deflection in the
  differential between animal-target and non-target
  ERPs at about 98 ms (in the occipital lobe). The
  low-cues could be retrieved very early in the
  visual system by analyzing the energy
  distribution across a set of orientation and
  spatial frequency tuned channels. This suggests
  that the rapid image classification may rely on
  low-level or intermediate-level cues that act
  diagnostically and allow the visual system to
  predict the classification of images very fast.

21

• Though this is possible, notice that these
  concepts do not play the role that concepts are
  usually thought to play in cognition.
• First, they are not personal contents in the
  sense that one is not aware of them when they are
  employed in early vision. They are subpersonal
  informational processing contents. (Bermudez
  1995)
• Second, they can be used only in the processes of
  early vision and they are not available for
  cognitive tasks.
• Third, they do not allow re-identification across
  times and contexts of the objects formed during
  early vision.
• Fourth, they do not satisfy Evans (1982)
  generality constraint.

22

• Pylyshyn (2007, 52) calls them subpersonal
  concepts since, being inside encapsulated modules
  of early vision, they do not enter into general
  reasoning. Such concepts may be codes for
  proximal properties involved in perception, such
  as edges, gradients, or the sorts of labels that
  appear in early computational vision.

23

• two scientists working in two different paradigms
  and practicing different experiments, using
  different instruments, and facing different
  puzzles, have their early visual systems shaped
  differently enough to make them see the world
  differently, each one through the lenses of her
  own paradigm, which has been in essence
  relativisms main point all along.
• In addition, they may also focus their attention
  to different locations or configurations and
  perceive different things (for example, they will
  disambiguate ambiguous figures differently).

24

• Surely one can control for attention one could
  ask one scientist to focus on a certain location
  even if she does not expect to see anything
  important there. Then she will see in the
  phenomenal sense (her early vision will retrieve)
  what the other scientist sees in the phenomenal
  sense, the latter driven by her own theoretical
  commitments that dictate that she should focus on
  that location, their theoretical differences
  notwithstanding. The two scientists may
  conceptualize the scene differently but the NCC
  is the same.

25

• Since the role of attention is restricted only to
  determining focus before the onset of processing
  and in no other way does it affect processing,
  the phenomenal content of early vision (the NCC)
  is the same as the phenomenal content of late
  vision, except that in the latter parts of the
  content are conceptualized. Thus, the
  phenomenology of their experiences is the same
  and only the conceptualization may differ.

26

• Moreover, even if two scientists, because of
  differences in their theoretical commitments, and
  work in different environments have stored
  different associations, when they switch
  environments they will form each others
  associations (these associations can form easily,
  require few experiences, and are formed solely on
  the basis of the incoming signals independent of
  any top-down effects). Thus, they will be able to
  see what the other sees. The reason is that
  learning through experience is data-driven, which
  means that the same training produces the same
  implicit memories, and, thus, the persons that
  undergo the same training learn to perceive the
  same patterns despite their differing theoretical
  commitments.

27

• Dilworth (2005) argues that two scientists with
  different theoretical frameworks could train
  their perceptual apparatus so as to produce an
  invariant lower-level perceptual content, no
  matter how theoretically different the
  higher-level perceptual contexts are. Thus, one
  may remove theoretical biases among
  experimenters. Scientists in different paradigms
  could train their perceptual system so as to
  produce an invariant, theory-neutral
  representation of the object perceived.

28

• For example, one cannot subtract the theoretical
  factors that have led one to see a duck when
  viewing the duck/rabbit ambiguous figure in a way
  that would lead to perceiving a theory neutral
  figure, that is, a figure perceived by both
  persons that live in the duck and rabbit paradigm
  respectively there is no theory-neutral image in
  this case, as one perceives either a duck-like or
  a rabbit-like figure and there is no in-between.

29

• I disagree with Dilworth that ones perceptual
  system could be trained to perceive a
  theory-neutral representation. The associations
  acquired through training rig-up the feed-forward
  perceptual processing and this precludes the
  possibility of perceiving an invariant
  representational content, although one could be
  trained to store the same associations and,
  consequently, perceive what the other person does.

30

• Despite its cognitive impenetrability, the
  perceptual system does not function independently
  of internal restrictions. Visual processing is
  constrained by certain principles or operational
  constraints that modulate information processing.
  
• Such constraints are needed because distal
  objects are underdetermined by the retinal image
  and because the percept is underdetrmined by the
  retinal image. Unless the processing of
  information is constrained by some assumptions
  about the world, perception is not feasible.
• Most computational accounts hold that these
  operational constraints substantiate some
  reliable generalities of the natural world and
  reflect the geometry of our world.

31

• The operational constraints reflect general
  physical regularities that govern the behavior of
  objects and the geometry of the environment. They
  have been incorporated in the perceptual system
  trough causal interaction with the environment
  over the evolution of our (and others) species.
• They allow locking onto medium size lumps of
  matter (Haugeland 1998), by providing the
  discriminatory capacities necessary for the
  individuation and tracking of objects (the
  capacity for perceiving objects as cohesive,
  bounded, and spatio-temporally continuous
  entities), in a bottom-up, nonconceptual way
  (Raftopoulos and Muller 2006).

32

• These constraints are hardwired in perceptual
  systems in that the physiological mechanisms
  subserving vision implement these constraints,
  from cells for edge detection to mechanisms
  implementing the epipolar constraint.
• As such, they are not available to introspection
  and function outside consciousness.
• These constraints are not perceptually salient
  but one must be sensitive to them if one is to
  perceive the world.

33

• The constraints constitute the modus operandi of
  the perceptual system and not a set of stored
  rules used by the perceptual system as premises
  in perceptual inferences.
• The constraints are implicit, and are available
  only for visual processing, whereas explicit
  theoretical constraints are available for a
  wide range of applications.
• Implicit constraints cannot be overridden one
  cannot decide to substitute them with another
  body of constraints, even if one knows that under
  certain conditions they lead to errors.

34

• Since the operational constraints are hardwired
  in perceptual systems, they cannot be states of
  the system. A state is formed through the
  spreading of activation and its transformation as
  it passes through the synapses. The hard-wired
  constraints specify the transformation from one
  state to another but they are not the result of
  this processing. They are computational
  principles that describe automatic transitions
  between states.
• Although the states that are produced by these
  mathematical transformations have contents, there
  is no reason to suppose that the principles that
  specify the mathematical transformation
  operations are states of the system or contents
  of states in the system.