Mental Representations

MENTAL REPRESENTATIONS

Common sense tells us that we’re able to think about objects or events not currently in view. Thus, you can decide whether you want the chocolate ice cream or the strawberry before either is delivered. Likewise, you can draw conclusions about George by remem-bering how he acted at the party, even though the party was two weeks ago and a hun-dred miles away. In these (and many other cases), our thoughts must involve mentalrepresentations—contents in the mind that stand for some object or event or state ofaffairs, allowing us to think about those objects or events even in their absence.

Mental representations can also stand for objects or events that exist only in our minds—including fantasy objects, like unicorns or Hogwarts School, or impossible objects, like the precise value of pi. And even when we’re thinking about objects in plain view, mental representations still have a role to play. If you’re thinking about this page, for example, you might represent it for yourself in many ways: “a page from the Thinking,” or “a piece of paper,” or “something colored white,” and so on. These different ideas all refer to the same physical object—but it’s the ideas (the men-tal representations), not the physical object, that matter for thought. (Imagine, for example, that you were hunting for something to start a fire with; for that, it might be helpful to think of this page as a piece of paper rather than a carrier of information.)

Mental representations of all sorts provide the content for our thoughts. Said differ-ently, what we call “thinking” is just the set of operations we apply to our mental rep-resentations—analyzing them, contemplating them, and comparing them in order to draw conclusions, solve problems, and more. However, the nature of the operations applied to our mental operations varies—in part because mental representations come in different forms, and each form requires its own type of operations.

Distinguishing Images and Symbols

Some of our mental representations are analogical—they capture some of the actual characteristics of (and so are analogous to) what they represent. Analogical representa-tions usually take the form of mental images. In contrast, other representations are symbolic and don’t in any way resemble the item they stand for.

To illustrate the difference between images and symbols, consider a drawing of a cat (Figure 9.1). The picture consists of marks on paper, but the actual cat is flesh and blood. Clearly, therefore, the picture is not equivalent to a cat; it’s merely a representa-tion of one. Even so, the picture has many similarities to the creature it represents, so that, in general, the picture looks in some ways like a cat: The cat’s eyes are side by side in reality, and they’re side by side in the picture; the cat’s ears and tail are at opposite ends of the creature, and they’re at opposite ends in the picture. It’s properties like these that make the picture a type of analogical representation.

In contrast, consider the word cat. Unlike a picture, the word in no way resembles the cat. The letter c doesn’t represent the left-hand edge of the cat, nor does the overall shape of the word in any way indicate the shape of this feline. The word, therefore, is an entirely abstract representation, and the rela-tion between the three letters c-a-t and the animal they represent is essentially arbitrary.

For some thoughts, mental images seem crucial. (Try thinking about a particular shade of blue, or try to recall whether a horse’s ears are rounded at the top or slightly pointed. The odds are good that these thoughts will call mental images to mind.) For other thoughts, you probably need a symbolic representation. (Think about the causes of global warming; this thought may call images to mind—perhaps smoke pouring out of a car’s exhaust pipes—but it’s likely that your thoughtspecifies relationships and issues not captured in the images at all.)For still other thoughts, it’s largely up to you how to represent thethought, and the form of representation is often consequential. If you form a mental image of a cat, for example, you may be reminded of other animals that look like the cat—and so you may find yourself thinking about lions or tigers. If you think about cats without a mental image, this may call a different set of ideas to mind—perhaps thoughts about other types of pet. In this way, the type of representation can shape the flow of your thoughts—and thus can influence your judgments, your decisions, and more.

Mental Images

People often refer to their mental images as “mental pictures” and comment that they inspect these “pictures” with the “mind’s eye.” In fact, references to a mind’s eye have been part of our language at least since the days of Shakespeare, who used the phrase in Act 1 of Hamlet. But, of course, there is no (literal) mind’s eye—no tiny eye somewhere inside the brain. Likewise, mental pictures cannot be actual pictures: With no eye deep inside the brain, who or what would inspect such pictures?

Why, then, do people describe their images as mental pictures? This usage presum-ably reflects the fact that images resemble pictures in some ways, but that simply invites the next question: What is this resemblance? A key part of the answer involves spatial layout. In a classic study, research participants were first shown the map of a fictitious island containing various objects: a hut, a well, a tree, and so on (Kosslyn, Ball, & Reisser, 1978; Figure 9.2). After memorizing this map, the participants were asked to

form a mental image of the island. The experimenters then named two objects on the map (e.g., the hut and the tree), and participants had to imagine a black speck zipping from the first location to the second; when the speck “reached” the target, the partici-pant pressed a button, stopping a clock. Then the experimenters did the same for another pair of objects—say the tree and the well—and so on for all the various pairs of objects on the island.

The results showed that the time needed for the speck to “travel” across the image was directly proportional to the distance between the two points on the original map. Thus, participants needed little time to scan from the pond to the tree; scanning from the pond to the hut (roughly four times the distance) took roughly four times as long; scanning from the hut to the patch of grass took even longer. Apparently, then, the image accurately depicted the map’s geometric arrangement: Points close together on the map were somehow close to each other in the image; points farther apart on the map were more distant in the image. In this way, the image is unmistakably picture-like, even if it’s not literally a picture.

Related evidence indicates enormous overlap between the brain areas crucial for cre-ating and examining mental images and the brain areas crucial for visual perception. Specifically, neuroimaging studies show that many of the same brain structures (prima-rily in the occipital lobe) are active during both visual perception and visual imagery (Figure 9.3). In fact, the parallels between these two activities are quite precise: When people imagine movement patterns, high levels of activation are observed in brain areas that are sensitive to motion in ordinary perception. Likewise, for very detailed images, the brain areas that are especially activated tend to be those crucial for perceiving fine detail in a stimulus (Behrmann, 2000; Thompson & Kosslyn, 2000).

Further evidence comes from studies using transcranial magnetic stimulation. Using this technique, researchers have produced temporary disruptions in the visual cortex of healthy volunteers—and, as expected, this causes problems in seeing. What’s important here is that this procedure also causes parallel problems in visual imagery—consistent with the idea that this brain region is crucial both for the processing of visual inputs and for the creation and inspection of images (Kosslyn, Pascual-Leone, Felician, Camposano, Keenan et al., 1999).

All of these results powerfully confirm that visual images are indeed picture-like; and they lend credence to the often-heard report that people can “think in pictures.” Be aware, though, that visual images are picture-like, but not pictures. In one study, partic-ipants were shown the drawing in Figure 9.4 and asked to memorize it (Chambers & Reisberg, 1985; also Reisberg & Heuer, 2005). The figure was then removed, and partic-ipants were asked to form a mental image of this now absent figure and to describe their image. Some participants reported that they could vividly see a duck facing to the left; others reported seeing a rabbit facing to the right. The participants were then told there was another way to perceive the figure and asked if they could reinterpret the image, just as they had reinterpreted a series of practice figures a few moments earlier. Given this task, not one of the participants was able to reinterpret the form. Even with hints and considerable coaxing, none were able to find a duck in a “rabbit image” or a rabbit in a “duck image.” The participants were then given a piece of paper and asked to draw the figure they had just been imagining; every participant was now able to come up with the perceptual alternative.

These findings make it clear that a visual image is different from a picture. The pic-ture of the duck/rabbit is easily reinterpreted; the corresponding image is not. This isbecause the image is already organized and interpreted to some extent (e.g., facing “to the left” or “to the right”), and this interpretation shapes what the imaged form seems to resemble and what the imaged form will call to mind.

Propositions

As we’ve seen, mental images—and analogical representations in general—are essen- tial  for representing  some  types  of  information.  Other  information,  in  contrast, requires a symbolic representation. This type of mental representation is more flexible because  symbols  can  represent  any  content  we  choose,  thanks  to  the  fact  that  it’s entirely up to us what each symbol stands for. Thus, we can use the word mole to stand for an animal that digs in the ground, or we could use the word (as Spanish speakers do) to refer to a type of sauce used in cooking. Likewise, we can use the word cat to refer to  your  pet, Snowflake; but, if  we  wished, we  could  instead  use  the  Romanian  word pisica˘ as the symbol representing your pet, or we could use the arbitrary designation X2$. (Of course, for communicating with others, it’s important that we use the same terms they do. This is not an issue, however, when we’re representing thoughts in ou own minds.)

Crucially, symbols can also be combined with each other to represent more complex contents—such as “San Diego is in California,” or “cigarette smoking is bad for your health.” There is debate about the exact nature of these combinations, but many schol- ars propose that symbols can be assembled into propositions—statements that relate a subject (the item about which the statement is being made) and a predicate (what’s  being asserted about the subject). For example, “Solomon loves to blow glass,” “Jacob lived in Poland,” and “Squirrels eat burritos” are all propositions (although the first two are true, and the last is false). But just the word Susan or the phrase “is squeamish” aren’t propositions—the first is a subject without a predicate; the second is a predicate without a subject. (For more on how propositions are structured and the role they play in our thoughts, see J. Anderson, 1993, 1996.)It’s easy to express propositions as sentences, but this is just a convenience; many other formats are possible. In the mind, propositions are probably expressed via net- work structures, related to the network models. Individual  symbols  serve  as  nodes  within  the  network—meeting  places  for  various links—so if we were to draw a picture of the network, the nodes would look like knots in  a  fisherman’s  net, and  this  is  the  origin  of  the  term  node  (derived  from  the  Latinnodus, meaning “knot”). The individual nodes are connected to each other  by  associative  links  (Figure  9.5). Thus, in  this  system  there might be a node representing Abe  Lincoln  and another node repre- senting President,  and the link between them represents part of our knowledge about Lincoln—namely, that he was a president. Other links  have  labels  on  them,  as  shown  in  Figure  9.6;  these  labels allow  us  to  specify  other  relationships  among  nodes, and  in  this way we can use the network to express any proposition at all (after J. Anderson, 1993, 1996).

The various nodes representing a proposition are activated when- ever  a  person  is  thinking  about  that  proposition.  This  activation then  spreads  to  neighboring  nodes, through  the  associative  links, much  as  electric  current  spreads  through  a  network  of  wires.

However,  this  spread  of  activation  will  be  weaker  (and  will  occur more  slowly)  between  nodes  that  are  only  weakly  associated. The spreading  activation  will  also  dissipate  as  it  spreads  outward, so that little or no activation will reach the nodes more distant from the activation’s source.

In fact, we can follow the spread of activation directly. In a classic study, participants were presented with two strings of letters, like NARDE–DOCTOR, or GARDEN–DOCTOR, or NURSE–DOCTOR (Meyer & Schvaneveldt, 1971). The participants’ job was to press a “yes” button if both sequences were real words (as in the second and third examples here), and a “no” button if either was not a word (the first example). Our interest here is only in the two pairs that required a yes response. (In these tasks, the no items serve only as catch trials, ensuring that partici-pants really are doing the task as they were instructed.)

Let’s consider a trial in which participants see a related pair, like NURSE– DOCTOR. In choosing a response, they first need to confirm that, yes, NURSE is a real word in English. To do this, they presumably need to locate the word NURSE in their mental dictionary; once they find it, they can be sure that these letters do form a legitimate word. What this means, though, is that they will have searched for, and activated, the node in memory that represents this word—and this, we have hypothe-sized, will trigger a spread of activation outward from the node, bringing activation to other, nearby nodes. These nearby nodes will surely include the node for DOCTOR, since there’s a strong association between “nurse” and “doctor.” Therefore, once the node for NURSE is activated, some activation should also spread to the node for DOCTOR.

Once they’ve dealt with NURSE, the participants can turn their attention to the sec-ond word in the pair. To make a decision about DOCTOR (is this string a word or not?), the participants must locate the node for this word in memory. If they find the relevant node, then they know that this string, too, is a word and can hit the “yes” button. But of course the process of activating the node for DOCTOR has already begun, thanks to the activation this node just received from the node for NURSE. This should accelerate the process of bringing the DOCTOR node to threshold (since it’s already partway there), and so it will take less time to activate. Hence, we expect quicker responses to DOCTOR in this context, compared to a context in which it was preceded by some unrelated word and therefore not primed. This prediction is correct. Participants’ lexi-cal decision responses are faster by almost 100 milliseconds if the stimulus words are related, so that the first word can prime the second in the way we just described.

We’ve described this sequence of events within a relatively uninteresting task— participants merely deciding whether letter strings are words in English or not. But the

same dynamic—with one node priming other, nearby nodes—plays a role in, and can shape, the flow of our thoughts. For example, we mentioned that the sequence of ideas in a dream is shaped by which nodes are primed. Likewise, in problem solving, we sometimes have to hunt through memory, looking for ideas about how to tackle the problem we’re confronting. In this process, we’re plainly guided by the pattern of which nodes are activated (and so more available) and which nodes aren’t. This pat-tern of activation in turn depends on how the nodes are connected to each other—and so the arrangement of our knowledge within long-term memory can have a powerful impact on whether we’ll locate a problem’s solution.