Connectionist Representations

In most of this chapter we have concentrated on the traditional symbolic approach to mental representation (see also Chapter 1). The basic view of this approach is that human cognition is centrally dependent on the manipulation of symbolic representations by various rule-like processes. Kosslyn's imagery theory is a prime example of theorising from this viewpoint, in which rule-based processes—like IMAGE and PUT—

manipulate various symbols. Even though the symbolic approach has been the dominant one within information processing psychology, some have questioned whether it is ultimately the best way to understand human cognition. These critics have highlighted some of the difficulties in the symbolic approach.

First, as we have seen in this chapter, within a symbolic tradition one has to explicitly state how mental contents are represented (whether they be images or propositions). Moreover, one has to specify how these representations are manipulated by various rules. So, even for relatively simple tasks, symbolic theories can be very complicated. When one moves away from laboratory tasks and looks at everyday tasks (like driving a car) it is sometimes difficult to envisage how such a complicated scheme could work. People can operate quite efficiently by taking multiple sources of information into account at once. Although a symbolic account might be able to account for driving, many feel that this account would be too inelegant and cumbersome. A second worry about the symbolic approach is that it has tended to avoid the question of how cognitive processes are realised in the brain. Granted, it provides evidence for the gross localisation of cognitive processes in the brain, but we are left with no idea of how these symbols are represented and manipulated at the neural level.

In response to these and other issues, in the 1980s a parallel processing approach re-emerged called connectionism (see Chapter 1; Ballard, 1986; Feldman & Ballard, 1982; Hinton & Anderson, 1981; Rumelhart, McClelland, & the PDP Research Group, 1986). As we saw in Chapter 1, connectionists use computational models consisting of networks of neuron-like units that have several advantages over their symbolic competitors.

As we shall see, connectionist schemes can represent information without recourse to symbolic entities like propositions; they are said to represent information sub-symbolically in distributed representations (see Smolensky, 1988). Second, they have the potential to model complex behaviours without recourse to large sets of explicit, propositional rules (see e.g., Rumelhart et al., 1986c; Holyoak & Thagard, 1989). Third, in their use of neuron-like processing units they suggest a more direct link to the brain (but see Smolensky, 1988). Connectionism clearly provides significant answers to many questions about human cognition. However, it is unclear how much of human cognition can be characterised in this way.

Distributed representation: The sight and scent of a rose

The concept of a distributed representation can be illustrated by an example involving a simple network called a pattern associator. Within the symbolic tradition, the sight and the scent of a rose might be represented as some set of co-ordinates (for the image of the rose) or as a proposition, i.e., ROSE(x). A distributed representation does not have symbols that explicitly represent the rose but rather stores the connection strengths between units that will allow either the scent or vision of the rose to be re-created (see Hinton, McClelland, & Rumelhart, 1986). Consider how this is done in the simple network in Figure 9.16a.

The sight and scent of the rose can be viewed as being coded in terms of simple signals in certain input cells (i.e., as pluses and minuses, see Figure 9.16). The input cells that take signals from vision are called vision units and those that take signals from the smell senses are called olfaction units. Essentially, the network is capable of associating the pattern of activation that arrives at the vision units with that arriving at the olfaction units. The distributed representation of the sight and scent of the rose is thus represented by the "matrix" of activation in the network; without recourse to any explicit symbol for representing the rose. Consider how this coding of the representation is achieved in more detail.

Figure 9.16a shows the vision and olfaction units. The sight of the rose is represented by a particular pattern of activation on the vision units (characterised by +1, -1, -1, +1), while the pattern of olfactory

Stop Anxiety Attacks

Stop Anxiety Attacks

Here's How You Could End Anxiety and Panic Attacks For Good Prevent Anxiety in Your Golden Years Without Harmful Prescription Drugs. If You Give Me 15 minutes, I Will Show You a Breakthrough That Will Change The Way You Think About Anxiety and Panic Attacks Forever! If you are still suffering because your doctor can't help you, here's some great news...!

Get My Free Ebook


Post a comment