<p align=center> Last Update: 12/16/2019 </p>

Despite a few radically deviated theories, representation is still a critical concept in cognitive science research. The nature of representation, however, is a big problem for cognitive psychology, neuroscience, and philosophy of mind. I still remember Alex said in his class that the topic of representation is something we all refer to, but few people really know when they are referring to it.

Inspired by Jake Quilty-Dunn’s (2018) recent paper “Perceptual Pluralism”, I set out to explore the theme of representational pluralism in both perception and cognition. This is by no means a simple essay to write, especially when I feel I still have so little reading in the related topics. However, I do want to try to put something under this title, and hope to refine it bit by bit. Different from ACVS, this project might need me several years to finish, but I nevertheless feel worth it. Right now, a good plan is to brainstorm some ideas I already have in mind.

Perceptual Representation

In “Perceptual Pluralism”, Quilty-Dunn wrote:

This evidence suggests that, while features are bound to objects in MOT, the binding is not holistic. The representation of the object can persist even though features are changing or lost entirely. If the representation of a feature is syntactically separate from the representation of the object, then we would predict that they could detach from one another in the way suggested by these results. This syntactic separation precludes holistic binding and instead requires a distinct constituent that stands for individual objects apart from various features. Thus PORs are not icons. (p. 11)

When I first read about this, I was wondering how can we reach the conclusion that the representation of a feature is syntactically separate from the representation of the object, by showing that loss or change of feature does not disturb object identity in MOT? Could it be the case that the representations for both the object and features are still iconic, and the loss of some feature representations does not affect the representation of the object, since the binding is not holistic?

Now I’m understanding holistic binding as follows: two features are holistically bound in some representation iff the same part of the representation encodes both features (i.e., there are not separate parts of the representation corresponding to each feature). Here’s a holistically bound representation of a pink heart:❤ (PS, why it looks black here??). There isn’t a constituent for the color and a separate one for the shape; instead, they’re encoded together. Understood this way, to say that color and shape are not bound holistically–as in the phrase ‘pink heart’–just is to say that there is a syntactic separation between the features, i.e., that there are separate constituents corresponding to each feature.

Now what about MOT evidence? MOT evidence is suggesting that there is a representation of the object, and representations of its features, and that these are not holistically bound. The evidence for this is that they seem to come apart from each other in tracking. If I took the shape and color out of the heart ❤, for example, there wouldn’t be a little index left over that picks out the individual. Since the binding in an icon like an emoji is holistic, there’s no constituent that stands for the object independently of its features. But perceptual object representations seem not to be like this. The visual index can survive even after deleting the features.

So it appears that, instead of the holistic binding characteristic of icons, there is a part of the perceptual object representation that picks out the object (i.e., the visual index), and this is separable from the parts that encode features. As I understand things, to say that just is to say that there is a “syntactic separation” between the index and the featural information in an object representation.

In sum, the argument about holistic binding is as followed:

1. If PORs are iconic, then features are bound holistically.
   • Icons represent holistically, because parts of icons represent various features at once. So a heart like this ❤ has its shape and color encoded together, no constituents of the heart representation correspond to only the shape or the color of the heart. In other words, as you mentioned, there is no syntactic separation between the features.
2. Features are not bound holistically in PORs.
   • If features are bound holistically, then loss of one feature will cause the loss of the object identity. This is not the case with MOT, because observers can easily track up to about four objects even if their features are changed or deleted during their movement. Therefore, features are not bound holistically.
3. With (1) and (2), we know that PORs are not simply icons.

Conceptual Representation

Quilty-Dunn (2018) in his paper “Perceptual Pluralism” proposes that perceptual object representations have an iconic format and a discursive format, both of which are well supported by empirical research. Compared to a large body of literature discussing the content of perceptual representation, the “vehicle” of the representation, or format, is less discussed. The study of concepts also indicates conceptual representation format is also an area of interesting focus

The cognitive revolution in the last century shifted people’s attention to the internal states of our mind. With the rapid movement of computer science, people were convinced that our brain is an information processing unit, which takes symbolic representations from perception as an input, and manipulates them to generate an output. This view implies that our brain represents everything in the world with arbitrary symbols, similar to what a computer does, in which different instances of objects are stored in randomized memory locations. The first wave of