Custom Search

Tuesday, January 25, 2011


In contrast to Galton and Spearman, Alfred Binet focused on the universalities of human intellect. He proposed that we all pass through certain developmental stages, and that to understand these stages we should consider the ‘higher faculties’ of the mind rather than ‘low-level’ neural processing: ‘It seems to us that in intelligence there is a fundamental faculty. . . . This faculty is judgement, otherwise called good sense, practical sense, initiative,the faculty of adapting oneself to one’s circumstance’ (Binet& Simon, 1916, pp. 42–3). An emphasis on reason and judgement is perhaps not surprising given Binet’s formal training as a lawyer.The first intelligence tests In 1904, Binet was charged by the Parisian authorities to develop tests that would identify children in need of special education, without relying on the subjective reports of parents or teachers.So he set about finding a way to construct tests with objectively verifiable scales of difficulty that could measure rates of development in ‘higher mental processes’.

Binet’s technique for constructing the first test was based on an important insight: whatever intelligence is, we can be sure that it changes (develops) with age. So the first intelligence test was based on the central idea that the age at which the ‘average child’ can succeed at a particular problem is an indication of the difficulty of that problem. Using this yardstick, children can be characterized as ‘average’, advanced or delayed in their rate of development compared to their peers. Binet and his associate Théodore Simon used a range of tasks in their first intelligence tests. These included around 30 items of increasing difficulty, beginning with simple items that even children with intellectual disabilities were able to complete (such as following a lighted match with your eyes and shaking hands with the examiner). More complex tasks included pointing to body parts and defining words such as ‘house’ or ‘wheel’, and tasks that were harder still, such as repeating back strings of digits and constructing sentences involving several specified words. Interestingly, vocabulary and digit recall tasks are still used in our most advanced intelligence tests today. Binet was also the first psychologist to specify that such tests must be: 1. administered and scored in a careful and standardized manner if comparisons between children’s performance are to be valid and reliable; 2. presented in the same order to all children and in order of increasing difficulty so that each child can pass as many tests as possible; and 3. administered in a one-to-one setting and only where the examiner has first established a friendly rapport with the child.Psychologists still adhere to these very important principles of testing today.
Alfred Binet (1857–1911), a French lawyer and self-trained psychologist, came to the field of intelligence via a study of psychopathology, free will and hypnosis. His interest in intelligence was prompted by observation of his two daughters, Madeleine and Alicia. While he was interested in how their different personalities affected their understandings of the world, he also noted that with age came the ability to reason about events in increasingly abstract ways. Binet observed their performance on various puzzles and asked them to explain how they had solved them. He was fascinated with their different approaches. This informal case study methodology led to the development of intelligence tests as we know them today.


Galton also introduced the idea of ‘co-relation’ (Galton,1888), or correlation, which is a measure of the extent to which two variables, such asweight and height, are related.A correlation of +1 would reflect a perfect positive relationship between the two variables –
as height increases, so weight increases in direct proportion. But we know from our own xperience that there is not necessarily a perfect relationship (there are some short, heavy-set people andsome tall, skinny people) so the correlation between weight and height is ikely to be less than one but still positive. A correlationof –1 would reveal a perfect negative relationship, where an increase in scores on one variable is directly related to decreasing scores on the other – for example, the number of cigarettes smoked is negatively correlated with life expectancy. Together, the notions of normal distribution and correlation allow us to consider how our abilities vary in relation to each other and in relation to the abilities of others in the population, and how well we can use scores on one variable to predict scores on another.Early attempts to measure intelligence In his Anthropometric Laboratory in London in the late nineteenth century, Galton attempted to measure a range of attributes that show individual variation. These included physical attributes such as head circumference, height and hand size, as well as intellectual characteristics (which, remember, he believed were a function of neural processes). These intellectual measures included basic sensory-motor tasks, such as speed of reaction to sounds and visual stimuli. Galton then compared these innovative measures of ‘intelligence’ to subjective estimates of the intellectual prowess of his participants based on their ‘reputation’ and eminence in the family tree (There were no such things as intelligence tests at the time!). Unfortunately, his empirical efforts were not successful.
Subsequently, Charles Spearman (1904) set out to estimate the intelligence of 24 children in his village school. He discovered a relationship between each child’s performance in a number of domains (including teachers’ ratings of ‘cleverness’ and ratings by other students of their ‘common sense out of school’) and measures of their ability to discriminate light, weight and pitch. In other studies, he found strong associations between scores on examinations in different subject areas such as classics and maths. Linking together these strands of evidence, Spearman concluded that there was a ‘general’ intelligence underlying performance on these very different tasks. He regarded general intelligence, or g, as a unitary, biological and inherited determinant of measurable
intellectual differences. In apparent contradiction, Spearman also noted that there were some ‘specific abilities’, such as musical aptitude, that contributed to differentially exceptional performance in specific areas and seemed less related to performance in other disciplines. But his finding of a general feature that underlies performance in many areas was so radical that it became the hallmark of his work. Spearman likened g to mental energy – a limited resource available to all intellectual tasks. So the idea was that individuals differ in general intelligence because they have different amounts of this mental energy.

Normal distribution

Normal distribution
Another of Galton’s contributions was to bring statistical understandings from the physical sciences to the study of psychology – particularly, the notion of normal distribution. Galton noted that for any of our ‘natural gifts’ (physical, temperamentalor intellectual) there will be an ‘average’ amount of that feature, to which most people approximate. Then, as we consider scores increasingly higher or increasingly lower than that ‘average score’, there will be fewer and fewer people registering those scores. Galton explains it as follows:Suppose a million of the men . . . stand in turns, with their backs against a vertical board of sufficient height, and their heights to be dotted off upon it . . . The line of average height is that which divides the dots into two equal parts . . . The dots will be found to be ranged so symmetrically on either side of the line of average,that the lower half of the diagram will be almost a precise reflection of the upper. (1892, 27–8)The idea here was that, in this group, there would be many men of about average height (say 160cm) and increasingly fewer men as we approach 190cm, and similarly fewer as we approach 130cm.Studying the normal distribution of psychological characteristicssuch as intelligence enables us to estimate attributes withina group and to have a point of comparison for an individual’s abilities. So, we expect that most people will approximate average intelligence, and there will be a small but predictable number of people of exceptionally high intelligence and an equally small and predictable number will be severely mentally disabled.


Francis Galton can be credited with the first systematic, scientific attempt to both understand and measure human intelligence . Galton’s essential idea was that there are stable,biological differences in intelligence between people.‘I have no patience with the hypothesis . . . that babies are born pretty much alike, and that the sole agencies in creating differences between boy and boy, and man and man, are steady application and moraleffort,’ he wrote. ‘The experiences of the nursery, the school, and of professional careers, are a chain of proofs to the contrary’ (1892, p. 12). Galton considered intelligence to be a low-level property of our nervous system that we inherit from our parents.He believed that individual differences in intelligence reflect differences in the efficiency of operation of simple neural processes.
Galton pursued his theory in several ways – first, by constructing extensive family trees of ‘persons of reputation’ in one domain or another to investigate patterns in eminence and achievement within families. His book Hereditary Genius, first published in 1869, presents family trees of ‘Commanders, men of Literature and of Science, Poets, Painters and Musicians of whom history speaks’ to support his hypothesis.



Parents and teachers will both tell you that they notice differences in the rate at which siblings or classmates complete their work and progress from one level of learning to another. At one extreme, some children apparently have pervasive difficulty in completing daily tasks, while at the other extreme are children who seem ‘gifted’, excelling at almost everything. Think back to your own schooldays, and you will probably recollect a growing awareness of where you ‘fitted in’ relative to your classmates – in other words, which classmates tended to do better than you on
maths and English tests and which would come to you for helpwith their homework.Parents want to know if their child is capable of learning morethan they appear to be. They want to know whether problemsexperienced by their child at school are due to a general inability to keep up with their classmates, or due to a specific area of skill deficit (such as a difficulty mastering reading), or perhaps a personality style or ‘motivational’ factor and nothing to do with intelligence at all. Teachers want to know the answers to a number of important questions; for example, (a) how to give each child the best learning environment, (b) whether lessons should be targeted to a child’s preferred learning style and (c) whether all children can learn the same things if given enough time. Businesses, too, spend large sums of money each year on training new staff, so they want to know which candidates are most likely to learn quickly and accurately the skills and knowledge required to complete their jobs. Some companies also want to know how flexible potential employees are likely to be in dealing with new problems. They want to know whether the person who will ‘act most intelligently’ in one position will also act most intelligently in another. Is the best person for the job the one with thecollege degree or the one with only a basic formal education butten years’ experience working her way up from the factory floor?Our concern with intelligence leads to endless questions. Forexample: Can intelligence be effectively measured? What do traditionalintelligence tests measure? Is intelligence one thing or made up of many different abilities? Was Einstein’s intelligence of the same kind as Mark Twain’s, Leonardo Da Vinci’s or Helen Keller’s? Are we born with a fixed amount of intelligence? Are the same people who were smartest at school still smartest as adults?Are they the most successful? Is intelligence changed dramaticallyby education and culture? (Who do you think is more intelligent – Aristotle or a current undergraduate physics student whose understanding of the physical world is clearly superior?)Is it possible to compare the intelligence of different racial groups? If you placed Anglo-Saxon Australian children from the city into a remote Aboriginal community in central Australia, would they perform as well on local tests of judgement and reasoning as children of the same age from that indigenous community?Would they know how to find water in a desert terrain orhow to find a goanna? Probably not – but does that mean theyhave become less intelligent all of a sudden? Which group would we expect to perform better on conventional tests of spatial ability or verbal reasoning? If we do compare groups, do any differenceshave a genetic or cultural root? Does intelligence ‘run in families’? This chapter will address the core issues in understanding intelligence that bear upon these questions beginning with the notion of individual differences in intelligence.

Sunday, January 23, 2011


This brief overview of the literature on thinking might lead us to wonder whether we are capable of being rational and logical, or whether we fall short of that ‘ideal’. Caution is needed with this question. Survival depends on being good at doing things that confront us in the real world. Rather than think of rationality as an absolute, Herbert Simon (1991) introduced the idea of satisficing – that is, performing optimally with the limited data and time available to us. This is known as bounded rationality – it is about as close to the idea of being rational as we are likely
to get, and is the best we
could expect from any system
with finite resources.
It has also been argued
that many of the tasks used
in laboratories are artificial,
and that they lack ecological
validity. In other words, they are not typical of the kinds of problem
humans have to solve. (For a discussion of this important
idea, see Cosmides & Tooby, 1997; Gigerenzer & Hoffrage,
1995.) Gigerenzer and Hoffrage show, for example, that when
information is presented in terms of frequencies ( like 95 out of
100) rather than probabilities ( like 0.95), people do better at a range of reasoning tasks, and ignore base-rates to a lesser degree. They argue that this is because we are naturally adapted to frequency information because we tend to collect instances one at a time. These authors are working on a program of investigation into evolutionary cognition, which attempts to establish whether we are good at certain ways of thinking because we have evolved that way to adapt to our evolutionary environment (see also Piatelli- Palmarini, 1992).

[Herbert Simon (1916–2001) was a true cognitive scientist, crossing disciplinary boundaries in his efforts to understand human problem solving and decision making. He was awarded the Nobel Prize in economics for his work on administrative behaviour, but is best known in psychology for his work on the representation of problems, and problem-solving heuristics (with the eminent cognitive scientist Alan Newell). In the early 1950s, Simon and Newell conceived the idea that the best way to study problemsolving was to simulate it with computer programs. Computer simulation of human cognition subsequently became Simon’s central research interest, which he pursued until his death in 2001.]


Heuristics provide a means of reasoning, but they are short cuts, using strategies that generally work but are not guaranteed to work. At the same time, they can induce quite high levels of confidence in us regarding our decisions, even when we are wrong. To a large extent, heuristic reasoning overlaps considerably with the everyday idea of intuition. Intuitive thought is automatic, often fast and not derived from detailed analysis. It involves a strong feeling of conviction but – like heuristic reasoning – tends to be hard to justify. Problem-to-model mapping The mappings from the description of a problem to an automatic conception of that problem can be very strong, and constitute the basis of some very strong feelings of the intuitive ‘correctness’ of our understanding. Try the following problem (schematically illustrated in figure 12.5): Suppose there are three cups in front of you and the experimenter puts a coin under one of the cups. You don’t know which one it is under. Next you try to choose the cup ou think the coin might be under. Rather than tell you whether you are right or wrong, the experimenter removes one of the cups, but not the one you pointed at, and not the one the coin was under (which may be different).
The question is, would you have a greater chance of getting the coin if you stuck to your original choice, or shifted? Participants usually believe that they have a 1:3 chance of being correct when they start, and then that they have a 1:2 chance of being right once there are just the two cups left. They usually declare that there is no point in changing because after the cup has been removed thay have a 50/50 chance of being correct (and if they changed their choice at this stage they would still only have a 50/50 chance of being correct). This behaviour fits a simple mental model: with N choices, the chance of being correct is 1:N. The situation is mapped onto this simple model, and the result is coherent and compelling. Despite this, the answer is that you should shift. In the first place, the chance of being correct was 1:3, and the chance of being incorrect was 2:3. But the important point is that the experimenter does not remove the cup at random, and – the key point – she never moves the cup that conta ns the coin. So the chance of being wrong by sticking to the original decision is still 2:3 (as per the original decision), even though there are only two cups now left. But since there is only one other cup now remaining, the chance of that being the wrong choice is in fact 1:3 (because there is only one other of the original three cups under which the coin could now be located), so it makes sense to change. In fact, the odds in favour of changing are 2:1. This is a very difficult puzzle to think about (e.g. see Granberg & Brown, 1995). The usual mental model people set up does not have the capacity to deal with the correct solution, and yet it is very compelling. There is an intuitive way of making the point about shifting, though. Suppose there are 100 cups (each numbered), and one has a coin under it. The chance of your being incorrect in your choice is 99:100. You choose a cup – say, number 15. Now the experimenter takes away all of the cups except the one you chose and one other (say number 78), but you now she never takes the one with the coin under it. Do you now think that there are even odds on your having selected the correct one, or would you prefer to shift? Most people think it appropriate to shift under those circumstances.

The ‘Three Cups Problem’ is a good illustration of a stron mapping between a state of affairs (two cups are left) and a preexisting mental model (if there are two cups, one with a coin under it, then the odds on choosing the correct one are 50:50). The intuitive belief that goes with these problem-to-model mappings is very strong. Try it on your friends. The hindsight bias Just as discourse makes sense if it portrays a series of connected events that match some plausible possible world, so facts about things make sense if they fit a coherent scenario. Also, once we know the facts, it is often easy to find a way of linking them. Nowhere is this clearer than with the hindsight bias, in which people believe that they had a prior insight (‘I knew it all along’) and that an event was therefore not surprising. Hindsight judgements are made ‘after the fact’. In a typical hindsight experiment (Fischhoff, 1977; Slovic & Fischhoff, 1977), participants first answer binary-choice general knowledge questions, such as: Was laddin (a) Chinese? ( b) Persian? Subsequently, they are presented with the questions again, this time with the correct alternative marked, and are asked to say whether they got each one right on the previous occasion. In general, participants tend to falsely remember getting more right than they actually did, as though the correct answer interferes with their memories. Even if participants are paid for remembering correctly, the effect still occurs, so strong are the intuitions the paradigm generates. A major consequence of the hindsight bias is that things appear to be more obvious than they should. Before new experiments are carried out, it is never clear what the outcome will be – otherwise they would not be original experiments. Yet in one interesting study, the same information was presented to two groups of participants concerning an experiment with rats. One group was told that one result occurred, while the other group was told that another occurred. Although the two sets of results were quite diffe ent, both groups of participants rated the outcome as obvious (Slovic & Fischhoff, 1977).


Thinking, understanding and decision-making take place in the real world, where there are usually time pressures and rarely a full range of information available to support a complete appraisal of the problem at hand.
For instance, suppose you are buying a new washing machine. A good basis for the decision might include comparative data on reliability, ease of servicing, servicing and repair costs, ease of use, even noise levels during operation. The list could go on and on. Although sometimes data of this sort might be available, and sometimes it might be published in magazines, it is more likely that you will have to cut corners. In other words, you might not be able to obtain a machine that fulfils all of your desirable features, but you will instead settle for the closest that is available. Kahneman, Slovic and Tversky (1982) popularized the term heuristic reasoning for thinking and decision making that involves these types of short cuts. They also suggested that these short cuts are so common that they should be considered part of the machinery of thought itself.

Perhaps the simplest kind of heuristic reasoning is availability. The availability heuristic is a method of estimating the likelihood of something based on how easily it comes to mind. For instance, we might assess the divorce rate by thinking of its prevalence amongst people we know personally. Or when buying a car, we might estimate reliability from comments made by acquaintances and colleagues. Because there will generally be a correspondence between what comes to mind easily and the likelihood of the underlying event, this heuristic can be useful. Kahneman et al. (1982) point to two mechanisms that come under the availability rubric: ease of recalling relevant instances and ease of constructing representations. For instance, someone’s estimate of how many flower names they know will directly depend on how many they can think of in a short time – say, two minutes (Tversky & Kahneman, 1973). In this case, there is generally a good correspondence between initial rate of retrieval and the total number known. But this is not always the case. For instance, it is easier to recall the names of famous people than ordinary people. So if participants hear lists of names containing equal numbers of famous and non-famous names, they ill typically believe that there are more famous people on the list than ordinary ones (Tversky & Kahneman, 1973). Here, the heuristic leads to a biased outcome. Another example of bias occurs through the construction of representations. Consider the following problem: A group of ten people want to form a committee with only two people in it. How many possible committees are there? Now try this: A group of ten people want to form a committee with eight people in it. How many possible committees are there? Most people produce a higher figure for the first question than for the second, even though they are actually equivalent questions (because 8 + 2 = 10, so for every committee of 2 that is formed there is an equivalent committee of 8 formed from among the same group of 10 people). Tversky and ahneman argue that this is because it is easier to imagine several committees of two than several committees of eight. (This seems reasonable if we suppose that it is easier to form and manipulate a mental model with two rather than eight tokens in it.) The availability heuristic has been used to explain many, many phenomena. In risk perception, for example, people tend to overestimate car accidents, tornadoes and homicide as causes of death, and underestimate death from complications due to diabetes, stroke and smallpox vaccination. Furthermore, studies show a good correlation between the prevalence of these events in news reports (availability) and estimated likelihood as a cause of personal death (Slovic, Fischhoff & Lichtenstein, 1979). Social psychology research has established that individuals tend to think that they initiated arguments with significant others more than 50 per cent of the time, and that they did more than 50 per cent of the work in domestic situations. This applies to both partners! It s argued that this is because we each have ready access to information about our own contributions in these situations, so we are more likely to register and remember these than our partner’s contributions (because of the higher availability of the former) (Ross, 1981; Ross & Sicoly, 1979).

[Daniel Kahneman (1934– ) has conducted highly influential work over the last several decades into human reasoning, specifically regarding the role of heuristics (i.e. reasoning short cuts, using strategies that generally work but are not guaranteed to work). To a large extent, heuristic reasoning overlaps considerably with the everyday idea of intuition. Kahneman and colleagues have suggested that these heuristic short cuts are so common that they should be considered part of the machinery of thought itself. For example, the availability heuristic is a method of estimating the likelihood of something based on how easily it comes to mind. The representativeness heuristic is based on the principle that we can estimate the likelihood of something by seeing how well it fits a prototype of which it may be an exemplar. For his body of work investigating human judgement and decision-making under conditions of uncertainty, Kahneman was awarded the Nobel Prize in 2002.]

This heuristic is based on the principle that we can estimate the likelihood of something by seeing how well it fits a prototype of which it may be an exemplar. For instance, if you are trying to decide whether a person is a Christian, the more properties they have that fit your model of how Christians behave, and the fewer they have that do not fit, the more confident you would be that the person is a Christian. Like availability, representativeness is a double-edged weapon – it can lead to fallacious reasoning. Many of the examples Kahneman and Tversky (1972) give are about reasoning with distributions, such as the ‘Exact Birth Order Problem’: All families of six children in a city were surveyed. In 72 families, the exact order of boys and girls was GBGBBG. What is your estimate of the number of families found in which the exact order was BGBBBB?
The majority of participants thought that the first sequence was much more likely. In fact, the two orders are almost equally likely because, on any occasion, either a boy or a girl could be born with approximately equal probability. Both of these orders fulfil this requirement. From an intuitive viewpoint, the first seems much more likely because there is an equal number of girls and boys. But the equal number gives the impression of being more likely seemingly because it is judged to be more representative.

The impact of representativeness on exact order judgements can be seen even more clearly with the following: Which is more likely to occur: GGGBBB or GBBGBG? Most people think it is the latter, because it is more ‘randomlooking’ than the former. Yet on a random draw basis, both examples are equally likely. To make this clearer, draw out all the possible sequences that could occur using three boys and three girls. Although the sequences are all equally likely, there are more ‘mixed up’ ones like the second one above, and only one other (BBBGGG) that looks more extreme (and therefore less representative). Yet these possibilities are all equally likely. Another example shows how representativeness can apparently obscure the use of what is termed base-rate information. Consider the following scenario: 100 people, comprising 70 lawyers and 30 engineers, apply for a job. One of the applicants, Dick, is a 30-year-old man, married with no children. A man of high ability and motivation, he is likely to be quite succes ful in his field. He is well liked by his colleagues.
Is Dick more likely to be an engineer, a lawyer or equally likely to be either? Kahneman and Tversky (1972) found that the predominant answer given was ‘equally likely’ because the information does not discriminate between the two. Yet the prior odds are 70:30 in favour of Dick being a lawyer, so this should be the answer in the case where there is insufficient extra evidence in the description. In such cases, it is as if the representativeness of the description dominates the thinking of participants – a typical illustration of what is widely known as the ‘fallacy of ignoring the base-rate’.


Logical reasoning
A special form of problem solving is logical reasoning. In these kinds of task, people are required to draw conclusions that necessarily follow from a given, but not to draw conclusions about what might possibly follow. For example, in this syllogistic reasoning task, two premises enable conclusions to be drawn:
If all men have blood, and John is a man, then, necessarily, John has blood.
But syllogisms are not always this easy, and some can lead to false conclusions. For example:
If some As are Bs, and some Bs are Cs, what can be said about the relation of As to Cs?
A common error is to say: Some As are Cs. But while this may be case, it is not necessarily true. Those Bs that are Cs might be the ones that are not As. Johnson-Laird (1983) suggested that when people get this wrong, it is not because they are not ‘logical’; it is because they have an inadequate representation of the problem – what he calls a mental model. Johnson-Laird was able to show that forming such models is harder with some premises than others, and that the harder it is (i.e. the more complex the mental models), the more likely it is that we will make an error.

Conditional reasoning
Another much studied type of logical reasoning is conditional reasoning, which deals with ‘if–then’ statements. For instance: If a green light comes on, then the ball has rolled left. Suppose the ball has rolled left. What can we conclude? A common error is to conclude that the green light must have come on (Rips & Marcus, 1977), but this is not a necessary conclusion. The ball could have rolled left for any number of other reasons.

[Phillip Johnson-Laird (1936– ) has been a major contributor to the nature of reasoning and also to the psychology of language, becoming particularly well known through his book Mental Models (1983). Much of this work was conducted in conjunction with Peter Wason, especially regarding his work on deduction (as evaluated, for example, using the Wason selection task). Johnson-Laird proposed and developed the theory of mental models, which seeks to explain how understanding works through mental representations of the situations depicted by a text or problem description. According to Johnson-Laird, humans are not always rational, but they are not intrinsically irrational either.]

This error is called ‘confirming the antecedent’. Does the fact that the error occurs mean that people are not logical? This is the wrong way of thinking about the issue. Like the logical error, what it means is that some people have the wrong representation of the problem, and this leads to false conclusions. For instance, the abstract form of the problem, ‘If A then B. B, so . . . ?’, suggests that there is only A and B to consider, in which case it is reasonable to suppose that if B, then A. But, in general, there can always be some other cause for B – it simply is not stated. So it is easy to confirm the antecedent. For instance, if you commit murder, you go to jail. But if you go to jail . . . this does not mean you committed murder!

Detecting cheats
A very important way of testing if–then statements is known as the Wason Selection – or four-card problem (Wason, 1966). In this task, the participant is given a rule, and four cards are laid out that have information written on both sides. For example: Rule: If a card has a vowel on one side, then it has an even number on the other side. Card 1: A Card 2: D Card 3: 4 Card 4: 7 The task is to verify (i.e. test) whether the rule holds by turning over the two cards that will enable this to be determined. Which cards would you turn over to verify the rule? Try it before you continue reading. The most frequent response is to check A and 4. Turning A will provide information that is consistent with the rule if there is an even number on the other side of the card, and will falsify the rule if there is an uneven number, so that is fine. But turning 4 will achieve nothing, because the rule does not say, ‘If a card has an even number on one side, it will have a vowel on the other.’ Turning this card is very much lik confirming the antecedent. In fact, the crucial second card to turn is the card with the 7, because if this has a vowel on it, then the rule is false.

This problem is hard to think about. But real-life versions can be much easier. For instance:
If a student is drinking beer, then they are over 18.
Card 1: Over 18
Card 2: Drinking beer
Card 3: Drinking Coke
Card 4: Under 18
How would you test the rule? Most people would now think the crucial card to turn was card 4, ‘Under 18’, because if that had ‘Drinking beer’ on the other side, there is a clear violation of the rule. This is because testing for under-age drinking is an example of detecting cheating, which is something we appear to be good at (Cosmides, 1989; Gigerenzer & Hug, 1992). The argument is that we have social rules to live by, and that we are naturally attuned to be able to test whether these rules are being broken. Clearly the representation of the problem is crucial to how reasoning takes place. When a concrete form of the problem is used, we can bring in specific procedures that we have access to for detecting cheats, which is something that is socially important. With an abstract version of the task, this is not possible.


If I ask you, ‘What is 6 + 6?’, unless you are a young schoolchild, you will be able to retrieve the answer 12 straight from memory. On the other hand, if I ask you, ‘What is 37 + 568?’, you have to do some problem solving. Being numerate means that you know how to solve this problem: it calls for a standard application of arithmetic procedures, and these procedures can be drawn from memory. This kind of problem-solving is called simply routine problem solving. In contrast, creative problem solving cannot be done according to a formula because there are no standard procedures in memory. As we experience the same problem type over and over again, what was at first creative may become routine, of course.

Search space
Consider this anagram problem:
What two-word phrase can be made out of these letters:

What strategies would you employ to solve it? The simplest is blind search, in which you just move the letters around blindly until a phrase appears. The possibilities here are enormous, so blind search is clearly not a very smart way to proceed. But how do we constrain the search? There are some sequences of letters in English that are legal and commonplace (like ‘pro’), some that are rare (like ‘goli’), and some that are downright impossible (like ‘blvm’). So a smarter strategy is to try constructing fragments from common grammatically legal combinations, then trying sequences that are more and more rare. Fragments will serve to cue word possibilities that you know, which will help speed up the search. With practice, people who like anagrams in crosswords develop a number of ways to constrain the search space. All problems can be construed in terms of search spaces, though this is more obvious with some problems than with others. In their classic book Human Problem Solving, Newell and Simon (1972) illustrat d the problems of search space more thoroughly than anyone had before. One problem they studied in some detail is the following (cover the solution and try the problem first):
For each letter, substitute one digit, such that the whole thing fits the laws of (base 10) arithmetic; in the example below D = 5:
You will notice that your perception of what is involved in the problem increases as you work on it. For instance, to begin with, you may not have noticed the problem of carrying. That is, you will need to add 1 to a column left of the one you are working on if the sum exceeds 9.

Newell and Simon (1972) collected speak-aloud protocols – they required people to say aloud what they were doing while they were attempting problems like this. This helped them to analyse in detail the steps people go through in problem solving. There were two main findings: 1. People set up initial representations of problems, which influence the search space. 2. They employ general purpose techniques, called heuristics, which help constrain the search space. So, with the problem above, Newell and Simon found several possible representations. For instance, some people saw it as being one based on word meaning. Suppose the puzzle was:
A person might reason that BILL = William the conqueror → 1066, therefore B = 1, I = 0, L = 6. This kind of reasoning turns out to be inappropriate for our particular problem given above. Other examples might be described as typographic – E looks a bit like 3, etc. – and cryptographic – using some sort of systematic code, like A = 1, B = 2, etc. Neither applies to our particular example, but the important point here is that our initial conception of the problem can alter the way in which we attempt to solve it. Understanding how people develop a problem space – the representation of a problem in the head of an individual – is a major aspect of work on problem solving. (The more general idea of a Mental Model is discussed later in this chapter.) For instance, when we learn how to problem solve, we must first recognize when seemingly different problems have a common logical structure.

Looking for a common structure
A classic study of how underlying common structure might be spotted was carried out by Gick and Holyoak (1980; 1983). They examined how experience with a puzzle called the ‘military problem’
(Holyoak, 1984) affected performance on a second problem,
called the ‘radiation problem’ (Duncker, 1945):
The military problem
A general wishes to capture a fortress located at the center of a country. There are many roads radiating out from the fortress. All have been mined, so that although small groups of men can pass over the roads safely, any large force will detonate the mines. A full-scale direct attack is therefore impossible. What should the general do? (Holyoak, 1984, p. 205) The radiation problem Imagine that you are a doctor treating a patient with a malignant stomach tumor. You cannot operate because of the severity of the cancer, but you must destroy the cancer. You could use highintensity X-rays. However, the intensity needed is such that the beam would destroy the healthy tissue that the rays must pass through. A lower intensity beam would not harm the healthy tissue, but would also not destroy the cancer. How can you use X-rays to destroy the tumor without destroying the healthy tissue? (adapted from Duncker, 1945) The solution to the two problems is very similar. In the case of the radiation problem, the solution is o direct weak X-rays from a number of different points outside of the body, and to set the sources up so that the beams converge at the site of the tumor. That way, no single beam is strong enough to cause damage to healthy tissue, but the combined effect on the tumour is enough to destroy it. The military problem has a solution based on the same principle: small groups of soldiers are sent along different roads at the same time, converging as one big army at the fortress. Gick and Holyoak had participants do the military problem first. One group of participants simply read the problem in the belief that they were just to recall the wording. Under those circumstances, only 30 per cent derived the correct solution to the radiation problem. However, if the participants were given two similar problems before the radiation problem, then there was more transfer. In general, though, the more superficially similar problems are, the better the transfer (Holyoak, 1990). So spotting the similarity of problems is far f om automatic.


The study of thinking concerns how we come to understand the world, what our intuitions are, and how we reason, solve problems and make judgements and decisions about things. The cognitive approach to thinking attempts to discover the processing activities that characterize these mental acts. As with language, a great deal of progress has been made as a result of adopting a procedural approach. But the most striking thing to emerge from the study of thinking is that, as a species, although we can solve some amazingly difficult problems, we can also fail on others that seem quite simple. Two main strands have coexisted in the study of thinking for many years: problem solving and reasoning. Problem solving has revolved around the study of puzzles and how people solve them, while reasoning has been more concerned with what conclusions people draw, logical or otherwise, on the basis of knowledge and evidence.

More recently, studies in both areas have stressed the nature of the representation that results from trying to understand what a problem is about. This has led to the suggestion that people form mental models of problems, which represent, as far as possible, the crucial aspects of the problems. In this way, mental model theory links thinking to language comprehension ( Johnson-Laird, 1983), placing great emphasis on how problems are both understood and represented.


There is an interesting theory that the natural metaphors we use to talk about things influence our descriptions and the way we think. Over the past 20 years or so, Lakoff and his colleagues (Lakoff, 1987; Lakoff & Johnson, 1980) have presented a remarkable set of observations about the role that metaphorical systems play in both our thinking and our language. In general, the Lakovian claim is that the conceptual system relies on metaphor because this is equivalent to setting up mental models, and that these then constrain the way we think and communicate. Metaphors are much more prevalent than you might think (e.g. Cacciari & Glucksberg, 1994). Far from being restricted to specialist literary uses, they permeate our language in such a way that they surely must reflect something about the way our conceptual structures support understanding in general. Lakoff suggests that there are certain fundamental ways in which we think about things. This kind of analogical thinking finds its way into our language in stri ing ways.

For example, Lakoff (1987) considers the conceptions we have about anger. There are many expressions relating to anger, which, if taken literally, make no sense at all: John was so angry, he hit the ceiling [roof ] Mary blew her stack when she heard the news. When John broke the news, Mary exploded. There was steam coming out of his ears. Lakoff claims that mental models of anger result from simple observations, like an increase in internal pressure (blood pressure, heart pounding), becoming hot and sweaty, etc. These observations can be understood in terms of familiar everyday experiences with the material world, such as heating things up in containers. So Lakoff suggests that one way in which we conceptualize anger is in terms of heat being applied to a container that may contain a liquid (e.g. ‘she was boiling/seething’). Once the model of heat being applied to a container is being used, it is generative – that is, it leads to outcomes, like steam. To keep the steam in, a lid is normally used. So we get ex ressions like ‘Contain your anger’ and ‘Put a lid on it’. A lid on a container generates other possibilities, too. For instance, increased pressure leads to the lid coming off – ‘He flipped his lid’ – and ultimate explosion – ‘John exploded with rage’, ‘Mary blew her top’. In his case study of anger, Lakoff suggests many more metaphors may be produced on this well known, simple basis. And we can understand statements like ‘I thought he was going to erupt’ because of these well worn conceptual connections. If you overheard this statement in a conversation, you would likely infer that it was about anger. Lakoff’s basic argument is that we have very simple but significant and repeated experiences of certain types. For instance, we all go on journeys, we all use pots and containers, and we all engage in some sort of competition. We are also familiar with conflict and war, albeit to different degrees. These common experiences are used as models for the interpretation of a wide range of phenomena. So, in the anger ase, containers play a central role. In the same way, the idea of a journey can form the basis of understanding relationships – ‘This relationship is at a dead end/ isn’t going anywhere’ – or arguments – ‘At least we are getting near the conclusion.’ These attractive ideas merit very careful consideration, not just because they have the potential to explain the wide variety of metaphorical features of language, but because of the influence they have on thought and communication. It was a very deliberate act of dehumanization when the Nazi propagandists portrayed Jews as rats in films, justifying the treatment of people in an inhuman way. More recently, in Rwanda, propaganda by one group, Hutu, described the other group, Tutsi, as ‘cockroaches’. In a similar vein, many things that require action are thought of in terms of war. For instance, Lakoff and Johnson (1980) cite Jimmy Carter, one-time president of the USA, as reacting to the energy crisis by declaring ‘the moral equivalent of war’. They point out that his opens up a set of analogues of war concepts. So there will be an ‘enemy’, a ‘target’ will be set, ‘sacrifices’ will be called for, and so on.


Loss of language function is called aphasia – strictly dysphasia when there is partial language loss, but the term ‘aphasia’ is commonly used for all types of language loss. Aphasia is diagnosed when there are language difficulties that occur in the absence of sensory impairments or thought disturbances – in other words, the symptoms are specific to language. The traumatic event of a stroke often results in an inability to use language to some degree, and is a sadly common occurrence. Strokes (cerebrovascular accidents) affecting those parts of the brain that support language processing account for 85 per cent of aphasia cases.

The left hemisphere has long been known to be associated with language function. Damage the left hemisphere, and language dysfunction is likely to result. In particular, two areas of the brain have long been associated with specific aphasic symptoms: Broca’s area, and Wernicke’s area.

Broca’s aphasia (or production aphasia) Broca’s area is found to be damaged in patients with Broca’s aphasia. These patients have difficulty in the production of language, some being unable speak at all, others only with difficulty. When language is produced, it lacks fluency and is slow. Speech may consist of just one or two words, with no grammar and often an absence of verbs necessary for the production of well-formed sentences. Broca’s aphasics can understand language, though. This is demonstrated by their capacity to follow instructions or to verify whether scenes match sentences.

Wernicke’s aphasia (or sensory aphasia) Patients with Wernicke’s aphasia have a problem in comprehending the speech of others, and although they can produce a fluent flow of speech, it is usually garbled, containing many pseudo-words (so-called jargon). Because they cannot understand the speech of others, they also may not be aware that they are not making sense. They suffer from word retrieval deficits and cannot properly pars sentences. These effects result from lesions to Wernicke’s area.

Other types of aphasia include the debilitating global aphasia, in which heard speech cannot be comprehended or even repeated, there is no capacity to produce speech, and even objects cannot be named. Another category is conduction aphasia, in which patients have an apparently normal capacity to understand and produce speech. But they have difficulty repeating word-strings and ‘nonsense’ words. This condition has been attributed to damage to fibres connecting Broca’s and Wernicke’s areas. Psychologists who study the changes that occur in aphasia will explore specifics, such as whether the patient has difficulty finding the right words in normal speech, repeating words and sentences, using grammar so that they can understand sentences, or producing grammatical outputs themselves. For further information on treatments of aphasia, see Zurif and Swinney (1994).
Dyslexia means impaired reading. There are two broad categories: acquired dyslexia and developmental dyslexia.

1. Acquired dyslexia Brain damage in people who could previously read well can lead to acquired dyslexia. There are four main classes of this disorder:
1. People with visual form dyslexia might not be able to recognize all the individual letters. So they might read ‘mat’ as ‘cat’.
2. Those with phonological dyslexia have difficulty reading pronounceable pseudo words, like ‘pleke’, but they are good at reading real words. This shows that their problem is caused by damage to the mechanism that connects how a word looks (its orthography) to how it sounds (its phonology). By contrast, when they read well known real words, these patients can use direct routes between the whole word pattern and its sound – these direct routes are established when we learn to read.
3. Surface dyslexia is the opposite way round to phonological dyslexia. People with this disorder are unable to use this direct route to recognize words on the basis of their overall appearance, but they can read words by using orthographic knowledge. This means that they make errors pronouncing words that are irregular in the mapping between the letters and the sound, like ‘pint’ or ‘yacht’.
4. Deep dyslexia forms a very interesting category. On being asked to repeat concrete nouns, such as ‘uncle’, the patient may say ‘aunt’ instead: i.e. they substitute a semantically related item. These patients cannot read abstract words and pronounceable pseudo words. Deep dyslexia is associated with widespread left hemisphere damage, and tends to co-occur with aphasia.

2 Developmental dyslexia This refers to a developmental difficulty with reading, despite adequate intelligence. Attempts to match the reading difficulties to the categories of acquired dyslexia have led the division of syndromes into two main types: those associated with difficulties in ‘sounding out’ (as in acquired phonological dyslexia) and those related to difficulties in recognizing word forms (as in surface dyslexia). But one prevalent problem for most developmental dyslexics is poor phonological awareness: so they perform badly on tests of rhyme awareness, segmenting words into individual sounds (spelling out) and producing rhymes. The detailed study of dyslexia entails the application of well developed psycholinguistic techniques. For a review of one hundred years of work in this area, see Miles and Miles (1999).


Reading is a complex process, which can be broken down into a variety of activities:
fixating words with our eyes;
processing words in accordance with syntax and semantics;
representing meaning; and
understanding the significance of what is read.
Until now, we have focused on the last three activities – how we come to understand language. Now we will take a look at the first point in the process.
Some of the oldest studies of the reading process were concerned with the pattern of eye movements that occurs when we read text. Even today, many of our insights come from studies of eye-tracking.
Using modern eye-tracking equipment, it is possible to establish where the most sensitive part of the eye (the fovea) is fixating within a piece of text. Although we have the impression of a smooth process when we read, in fact the eye moves in jumps, called saccades, and then fixates, or remains stationary, upon successive pieces of text.

The dots are fixation points, and the lines are saccades. When the line moves back towards an earlier part of the sentence, this is a regression. Word information is only encoded when the eye is stationary, and then only about 15 letters can be encoded within a single fixation. From the perspective of understanding, it is interesting to note that small words are not always fixated. So a word such as ‘he’ may only be fixated 30 per cent of the time. Content words, on the other hand, are nearly always fixated. At one time it was thought that where the eyes fixated was simply a mechanical process, but now it is clear that eye movements are under the control of some of the complex processes underlying language understanding (Rayner & Polletsek, 1989). For instance, when someone has difficulty comprehending a piece of text, regressive eye movements take place – in other words, their eyes move back to earlier parts of the text. These movements are quite common, even in reading straightforward text, as a means of ch cking earlier information to aid interpretation.


Language consists of more than disconnected utterances. When sentences are put together to make a sensible message, the result is discourse. A substantial part of the psychology of language deals with discourse processing, especially when it concerns text. Many theories of discourse processing have been developed, for example by Gernsbacher (1990), Kintsch (1988), and Sanford and Garrod (1981).

The primary feature of discourse is that it is coherent – in other words, the individual sentences fit together in a meaningful way and do not contain any contradictions. Sometimes sentences are connected by explicit devices, called cohesion markers. Consider the following:
John fell off the cliff because he was walking too near the edge.
There are two cohesion markers here:
1. the connective ‘because’ indicates that the sentence ‘John fell off the cliff ’ portrays the result of a cause – i.e. ‘he was walking too near the edge’; and
2. the pronoun ‘he’ signals that some individual who is singular and male has been mentioned. The only thing that fits the bill is ‘John’, so we take it that it was ‘John’ who ‘was walking near the edge’.
But the establishment of coherence does not always rely on cues such as these. For instance: John was hit by a train. He had been walking down the track. This is coherent because ‘walking down the track’ was the condition that enabled ‘John’ to be ‘hit by a train’. But there is no explicit connector (‘because’): the connection is inferred. Coherence establishment may sometimes make use of cues in the text, but always relies on some degree of inference. As a final example, consider the following single sentence: John lent Harry some money because he was hard up. What is the referent of ‘he’? Obviously it is ‘Harry’, not ‘John’. Why? Because money is lent to people who are ‘hard up’, and this inference is automatically drawn and used to solve the reference problem.

These few examples show the complexity of the computational operations that underlie even the most mundane language processing at the discourse level, and they represent just a small sample of the issues. Inferences vs. scenarios Experimental work shows that it takes time to make inferences. Haviland and Clark (1974) asked people to read short texts made up of two sentences, and then measured the reading times for the second sentences. Compare the following pairs:
Inference version: Herb took the picnic supplies from the car.
The beer was warm.
Explicit control: Herb took the beer from the car.
The beer was warm.
The reading time for the second sentence was longer in the inference version, because participants had to infer that ‘The beer’ is part of the ‘picnic supplies’. The text demands that an inference be made, which demands cognitive resources.

But sometimes knowledge may be automatically recovered and included in the mental representation of the sentence. For instance, given ‘Harry drove to London’, there may be a default representation of the fact that a car was used. Subsequent mention of a car would not be a problem, because its default is already in the representation resulting from the sentence. This is what Garrod and Sanford (1982; 1983) found to be the case. In a fuller theory, Sanford and Garrod (1981; 1998) argued that we automatically relate what is being said to background knowledge, and that background knowledge is organized in long-term memory about specific situations. They called these structures ‘scenarios’, and argued that the basic, most fundamental operation of understanding is to recognize the situation in which the message is set. So, because we are retrieving further situation information from memory, sentences can lead to representations that go beyond their content.

As one final example of a study that seems to support this view, Garnham (1979) required participants to try to remember sentences they had seen previously: e.g. ‘The housewife cooked the chips.’ He found that participants remembered this sentence better if they saw the cue ‘fried’ than if they saw the cue ‘cooked’, even though ‘cooked’ is actually part of the original sentence. According to the scenario theory, this is because cooking chips has been implicitly represented as a situation in which frying is taking place. Of course, another possibility is that the word ‘fried’ simply provided more information, in terms of a cue for remembering. The ultimate questions For discourse studies, the ultimate questions are just which inferences are made (i.e. what knowledge is recruited) and when during language processing. Some theorists believe that sometimes there might not be much inferential activity taking place during natural discourse (McKoon & Ratcliff, 1992), and that inferences and elaborations will only ta e place when the relevant background knowledge is highly available in memory. Sanford and Garrod (1998) take the view that it is the task of the writer or speaker to say things in such a way that a scenario can easily be found, because this is essential for good message-level interpretation. Whatever they think about component processes, there would be few scientists who would disagree that understanding is based on bringing language input into contact with world knowledge – the basic question being how this is done. Noam Chomsky has been at the forefront of international thought over the past several decades regarding the individual development and intergenerational heritability of language. The classic Chomskian sentence ‘Curious green ideas sleep furiously’ is not intelligible at the message level, simply because it is hard to relate to anything we know about. But ‘The housewife cooked the chips’ is intelligible because we can easily relate it to many things we know about.


Comprehension of language requires the processor to use knowledge of the language (syntax), meaning (semantics), and our knowledge of the world (scripts) and inferences about the intentions of speakers (pragmatics).
The central questions for the study of the processing system are:

How and when are these sources of information called upon?
How is the architecture of the system organized?
Is syntactic analysis carried out first, and then meaning and interpretations ascribed later? Or are they all used at any point they might be needed?
There are too many studies in this area to present a full overview here. Instead we present just two sample problems (wordsense retrieval and nonliteral meaning) to indicate how the issues may be addressed experimentally.

Word sense retrieval
When reading or listening, it is important to retrieve word meaning, and that means retrieving the right sense of a word. This is an area where the role of background knowledge is important. For instance, in understanding ‘John put his salary in the bank’, it is necessary to select the appropriate sense of ‘bank’ – i.e. a place where financial transactions take place, not the side of a river. Context usually provides the solution to this problem, but the question is when during the sequence of processing? Is just one meaning of ‘bank’ selected at the outset, or are both meanings initially recruited, and then the right one selected later? There are two main possibilities:
1. The modular view is that word meanings are stored in a way that is not context sensitive. When we encounter a string of letters that represents a word, we automatically look up and retrieve the meaning. If the string (such as ‘bank’) represents more than one word, then both meanings should be retrieved.

The modular view is attractive because it keeps the mechanisms of looking up word meaning separate from context, and so is computationally simpler (see Fodor, 1983, for a discussion of this position).

2. The interactive view suggests that word meaning information is connected to other processes of comprehension, so that which aspects of word meaning are active depends on context. This view is attractive because it implies a very adaptive organization of knowledge and word meaning, but at the cost of more computational complexity (e.g. see McClelland and Rumelhart, 1981; Morton, 1969).

An important technique for finding the answer is priming (see Meyer & Schvaneveldt, 1971). When a word is read, it becomes easier to recognize words that are associated with it. So if you read the word ‘nurse’, you will then read the word ‘doctor’ more quickly than if you had just read an unrelated word, such as ‘bread’. What will be primed after reading the word ‘bank’? If there is no biasing context, then target words relating to both senses should be primed, such as ‘river’ and ‘money’. Swinney (1979) presented participants with spoken passages like these:
(a) Mary needed to buy some presents, so she went to the bank.
(b) Mary found the river cold, so she swam to the bank.
Immediately after the presentation of the ambiguous word, he presented a single letter string on a screen. Participants had to decide whether the letter string was a word or not (a lexical decision). When the string was a word, it could either be related to the intended sense of the ambiguous word (e.g. ‘money’), related to the other sense (e.g. ‘mud’), or unrelated to either. The question was whether there would be a response time advantage for the intended-sense associate alone, or whether there would also be an advantage for the other-sense associate of the word too.

It turned out that there was equal advantage (priming) for both senses. So context did not appear to affect initial sense selection. But if there was a delay of only 300 ms between hearing the ambiguous word and reading the letter string, the priming effect remained only with the intended (contextually cued) sense. This work suggests that word meaning information is initially stored in a modular fashion, and its retrieval is uninfluenced by context. On the other hand, very shortly after a word has been processed, contextual cues inhibit the activation of word sense information that is inappropriate. This one example represents a sample of work on the problem of modularity; research in this area remains very active (see Sanford, 1999, for a fuller review).

Nonliteral meaning
How do we understand sentences? One explanation is that we assign a literal meaning to them and then integrate this into the meaning of the discourse. But the literal meaning may not make any sense, especially if the sentence conveys an indirect speech act or a metaphor. For instance, if I say ‘My job is a jail’, I mean it restricts my freedom in a way that parallels being in jail. One prevalent view is that metaphors are first interpreted literally, then, if this fails, they are interpreted as nonliteral, or figurative (Searle, 1975, 1979). As a series of processing operations, this may be formulated as follows (from Glucksberg & Keysar, 1990):
1. Derive a literal interpretation of the utterance.
2. Assess the interpretability of that interpretation against the context of that utterance.
3. If that literal meaning cannot be interpreted, then and only then derive an alternative nonliteral interpretation.
The sequence above suggests that in order to make an appropriate interpretation of a statement, we need to know whether it is meant to be literally true or not. But it also makes strong assumptions about the processes underlying comprehension that subsequent work has suggested may be incorrect. The account has been examined for both indirect speech acts and metaphor comprehension. Gibbs (1979) showed that people take no longer to process indirect requests such as ‘Must you open the window?’ – meaning ‘Don’t open the window’ – than to understand literal uses of the same expressions (in the present case, meaning ‘Need you open the window?’). These data suggest that people do not need to obtain a literal meaning of an expression first in order to comprehend an indirect speech act. This runs against the traditional model (Glucksberg & Keysar, 1990). Gibbs (1983) claimed, more strongly, that participants do not always derive a literal meaning at any point. To establish this would be another blow to the traditional odel, since this model specifies that literal meanings are necessarily established. Gibbs had participants read stories that ended with critical sentences such as ‘Can’t you be friendly?’ In different stories, the sentence was given a literal meaning (‘Are you unable to be friendly?’) or an indirect interpretation (‘Please be friendly’). After reading a passage, participants had to decide whether a string of words was a grammatically correct sentence. Some of the strings were either the literal or the nonliteral interpretation of the critical sentence. Gibbs predicted that the literal context would prime the literal interpretation, and the nonliteral context would prime the nonliteral interpretation. These results should be reflected in a priming effect on the subsequent sentence judgement task. In two experiments, the results confirmed these expectations. In particular, when the context biased the interpretation of the critical sentence towards a nonliteral interpretation, there was no priming of the literal interpretation.

These findings show that the applicability of the standard comprehension model (Glucksberg & Keysar, 1990) is at best limited, although it is worth noting that the comprehension of sentences in stories (such as have been used in most of the studies reported here) and actual interactions in dialogue are very different situations, so we must guard against simplistic conclusions. Nevertheless, work on indirect speech act comprehension reinforces the view that literal interpretation is not always necessary. Similar findings have been obtained for metaphor comprehension. For example, Glucksberg, Gildea and Bookin (1982) asked participants to decide whether simple statements were literally true or false. For example, consider the statement ‘Some desks are junkyards.’ This is literally false, and so (according to the conventional model) the obvious metaphorical interpretation should not interfere with processing and the production of a ‘no’ response. Yet it does. A statement with an obvious figurative interpretation akes longer to reject as literally false than does a sentence with no obvious figurative meaning, such as ‘Some desks are roads.’ So, in the case of ‘some desks are junkyards’ it seems that the metaphorical meaning is computed automatically even though it is not needed, which indicates that testing for literal meaning cannot represent the previous, modular processing stage that the classic position would claim (see also Glucksberg and Keysar, 1990). Our sample of work on the comprehension of metaphors shows how simple response time studies can be used to evaluate the sequence of language processing events. The conclusions suggest that the straightforward classical view that literal interpretation takes place first, and then nonliteral interpretation takes place later if needed, is wrong.


It is conventional to divide up issues of language under the headings syntax, semantics and pragmatics. Syntax is the set of rules or principles that govern word order, and which words can be combined with which. The rules and principles have been determined by scholars but, in a sense, they reflect the way the brain analyses language. An example of a syntax rule, in English, is that a sentence consists of a noun phrase plus a verb phrase. This can be written as: S →NP + VP
So with the sentence ‘John loves Mary’, ‘John’ is the noun phrase (NP) and ‘loves Mary’ is the verb phrase (VP). Other descriptive rules specify what is an NP and a VP. The details are quite complex, but a descriptive grammar is one that allows only those strings of words that people accept as sentences. Psycholinguistics has been especially concerned with how people parse sentences – that is, how they break them down into their correct grammatical structures. This has to be done because, otherwise, it would be impossible to interpret a sentence at all. Consider the following: The horse raced past the barn fell. Is this an acceptable English sentence? What does it mean? In fact, it is a classic illustration of the problem of parsing. People normally find this a hard sentence to understand, because the parsing mechanism treats ‘The horse’ as an NP and ‘raced’ as the main verb, so it then expects more information consistent with the noun phrase. But the sentence actually contains what is called a reduced relative clause. Here it is in its unreduced version:
The horse that was raced past the barn fell.
By missing out the words ‘that was’, the relative clause is reduced. So, in fact, the structure of the sentence is:
NP: The horse (that was) raced past the barn
VP: fell.
The difficulty in understanding such sentences is ascribed to an initial misinterpretation, and is called a ‘garden path’ (see Frazier, 1987).

A large amount of time and effort has gone into studying the human parsing mechanism because it is central to language comprehension and production. By misparsing the sentence above, there is a resultant failure in comprehension at all levels. Another well known example is the sentence, ‘The old man the boats.’ Most people find this sentence ultimately intelligible but find there is a disturbance of understanding, because the string ‘The old man’ is parsed as an NP, and not as an NP + V (‘The old’ being a shortened version of ‘The old people’, and ‘man’ being a verb). Unless the sentence is properly parsed, it is unintelligible. Semantics concerns aspects of meaning. For instance, while ‘Green rain sinks frail grannies’ has good syntax, it is meaningless. The meaning of a sentence is somehow assembled from the meanings of the individual words that make up the sentence. Meaning at the sentence level is vital for comprehension, just like syntax. Compare the following:
Harry cooked dinner with his wife last night.
Harry cooked dinner with a wok last night.
In the first, ‘his wife’ is a co-agent, accompanying Harry, whereas in the second, ‘a wok’ is an instrument for cooking. To assign the wrong role (meaning) to ‘his wife’ would make Harry look like a cannibal!
And pragmatics concerns what we do with language.
At the level of sentence meaning, ‘Can you pass the salt?’ is a simple question, and should be interpreted as a question about competence. But when a child is asked this at the table and replies ‘Yes’, everyone knows this is a game. This is because there is a distinction between semantics, or sentence meaning, and pragmatics, which is sometimes called speaker meaning, and concerns the meaning of an utterance, not just a sentence.

The fact that sentence meaning is not sufficient to guide an interpretation led to a theory of speech acts (Searle, 1969), which treated utterances as actions on the part of a speaker, with the actions requiring their own interpretation. The introduction of pragmatics is essential to any account of language processing, and is especially obvious in cases where semantics (or literal meaning) appear to fall short. There are two principal classes of phenomena that obviously require more than literal meaning. One is indirect speech acts, like the salt example above. The other is metaphor and related phenomena. For instance, if I say ‘Adolf Hitler was a butcher’, I do not mean it literally. Similarly, if I say ‘John is really blue (or low) today’, I do not mean that he is covered in blue dye, or has shrunk in height. I mean that he is depressed. We appear to process many metaphors so readily that it is difficult to see what the problem is, but the processing problem is huge: not only does the processor have to par e sentences, but she has to determine their significance too. The psychology of language understanding is about just these issues. Finally, interpretation proceeds by linking language to our knowledge about people and situations. Consider the following:
A. John was hungry. He went to a restaurant and ordered some nine-inch nails.
B. John was really hungry. At the restaurant, he ate some Crepe Suzette, and then ordered steak, followed by Moules.
C. Harry put the wallpaper on the wall. Then he sat his full coffee cup on that.
D. Harry put the wallpaper on the table. Then he put his coffee cup on that. n In case A, a problem is detected because nine-inch nails are not edible. This information has to be retrieved in order to make use of it. It implies access to an almost encyclopedic knowledge of what one can eat. n In case B, the procedure for determining the order in which things are eaten is accessed. In this case, one would not normally consume a sweet dish (Crepe Suzette) before a savoury dish (Moules). Schank and Abelson (1977) suggested that we have mental scripts for stereotyped sequences, which are accessed under the right conditions, and as a result we can spot anomalies when they occur. Without such stereotyped knowledge, we would not have any knowledge of social norms.

In case C, wallpaper being on a wall puts it in a vertical position, so you cannot put your cup of coffee on it.

Detecting the problem suggests that we set up a mental representation of what putting wallpaper on a wall entails. n Case D is not a problem at all. But it is almost identical in linguistic terms to C; it is just that ‘on the table’ is taken to mean flat on the table, so sentence D is judged not to be problematic.

Language and Thought

Language gives us the capacity to let others know things and do things that would otherwise be impossible. It enables us to share knowledge and ideas, and to extend our spheres of influence beyond the immediate. Spoken language is the most basic form, especially dialogue, but most cultures have also developed written language systems. Written language not only allows the ready dissemination of information within our own culture, but also enables us to keep in touch with cultures that are remote in both time and place. The psychology of language is concerned with the organization and processing of both written and spoken language. It is a complex field, at the interface of pure psychology, linguistics and communication studies. And as we examine how language is processed, it will soon become clear just how complex and mysterious the process is. For instance, a colleague of mine recently mentioned that he was feeling ‘low’ because he had just received some severe criticisms of a paper he had written.


What we remember depends, in part, on how we were thinking and acting at the time of the original experience. This knowledge can allow us to develop strategies that help us modify what we remember. The role of rehearsal An early strategy often adopted by children is to repeat material over and over again. The mere repetition of information, with no additional thought about meaning or associations, can help us to retain information for a few seconds, but it is a very poor method of learning for the longer term, as demonstrated by Craik and Watkins (1973). Their participants learned lists of words. In one condition, they were encouraged to repeat the last few words over and over again for some time before recall. These participants recalled the repeated words well in the immediate test, but at the end of the experiment all of the different lists that had been
presented were tested. In the final test, the words that had been rehearsed repeatedly (and remembered better in the immediate test) were recalled no better than other words. This rehearsal was described as maintenance rehearsal – maintaining the memory temporarily but doing nothing for longer-term memory. In contrast to maintenance rehearsal is elaborative rehearsal. Rather than simply repeating information in an effort to maintain its availability, in elaborative rehearsal the meaning of the information is considered and elaborated. Although both types of rehearsal can keep information available for a short time, recall after a delay is much better when the information has been rehearsed elaboratively than when it has merely been rehearsed in a maintenance fashion (Bjork & Jongeward, 1975). Expanding retrieval practice Regardless of the type of rehearsal, later recall of information benefits from spaced retrieval practice – a technique for maximizing learning with the minimum of effort applied at the optima moment. The underlying principle here is that memory is strengthened most when recall is attempted just before it becomes too difficult to accomplish (Bjork & Bjork, 1992). When we first encounter some information, it may be relatively fragile in terms of memorability. By successfully recalling the information correctly a short while after studying it, we are more likely to recall it again later, so we can allow a somewhat longer delay before our next successful retrieval effort. With each successful effort, the delay can increase and still lead to further successes. The effectiveness of this expanding schedule for retrieval practice was demonstrated by Landauer and Bjork (1978). They showed fictitious first and last names to their participants, who were then asked to recall the last names when the first names were shown again. The tests were scheduled to explore a range of possibilities, including testing after short, moderate and long intervals filled with intervening items, and a further condition, the expanding schedule, in which the tests were at first introduced after a short delay and then the interval was steadily increased. For the expanding schedule, the first test took place immediately, the second test after three intervening items and the third after ten further items. Landauer and Bjork found that any retrieval practice was beneficial (relative to the control unpractised condition), but that the greatest benefit was found for the expanding schedule, which produced recall at approximately twice the level of unpractised items. Expanding retrieval practice is an excellent strategy for students. It is relatively undemanding in terms of the effort and creativity required, and can be applied to virtually any material (Morris & Fritz, 2002).

The benefits of spaced study
It is natural to plunge intensively into trying to learn new information, but this strategy has been shown repeatedly to be misguided (Dempster, 1996). The benefits of spacing study trials were observed by Ebbinghaus (1885/1964), who found that spreading his study sessions over three days approximately halved the amount of time actually spent in actively studying the lists. In fact, two spaced presentations of material to be learned are often twice as effective as two massed presentations (Dempster, 1996). Bahrick and Phelps (1987) demonstrated the robustness of the spaced study effect. They compared the performance of participants who had originally learned and then relearned Spanish vocabulary by testing them eight years after the teaching session. One group had originally learned and relearned the vocabulary with an interval between learning and relearning of 30 days. Another group had learned and relearned on the same day. Eight years later, the participants who had learned and relearned with a 30-day interval performed at a level 250 per cent higher than the same-day learning/relearning group!

Many students are familiar with rhymes such as ‘30 days hath September . . .’, whose rhythm and rhymes provide structures that aid recall (Morris & Gruneberg, 1979) and with first-letter mnemonics such as ‘Richard of York Gave Battle in Vain’ that help recall order – in this case, the colours of the rainbow (Morris & Cook, 1978).
But the oldest mnemonic method is the method of loci, traditionally attributed to Simonides around 500 BC and taught from Classical times until the present day. The technique involves knowing a series of places or loci that are familiar yet distinct – students might use places around their campus. The first item to be remembered is imaged in the first of these places, the second item in the second place, and so on. Recall then involves mentally revisiting the places and re-experiencing the images. Research has shown the technique to be highly effective (Morris, 1979), but its use is obviously limited by the availability of suitable loci and material to image. The method of loci has since been elaborated into the more flexible pegword system, using the phonetic mnemonic in the construction of the pegwords (Higbee, 2001). Easily imagined pegwords that can be relatively easily learned replace the places of the method of loci. For example, we might learn words to represent each of the numbers from 1 to 100. The w rds are easily learned because they are constructed according to a few simple rules that underlie the phonetic mnemonic. Each digit is replaced by a specific consonant sound and then vowel sounds are inserted in between to create concrete, imageable words instead of number combinations (for which it is more difficult to create images). In the phonetic mnemonic the consonant sounds for the digits 1 and 2 are ‘t’ and ‘n’, respectively. So, the number 21 can be represented by ‘net’ or ‘nut’. The full phonetic mnemonic and pegwords for the numbers 1 to 100 can be found in Higbee (2001). Pegword mnemonics allow a much more flexible use of the imagery mnemonic than the method of loci and can be dramatically effective (Bellezza, 1996; Morris & Reid, 1970); they form the basis of most professional memory improvement techniques. The pegs provide easily accessed memory cues, while the use of imagery links the cue and the item to be remembered through visuospatial interaction (Morris & Stevens, 1974). Imagery mnemonics ave been developed to tackle a range of practical memory problems. For example, Morris, Jones and Hampson (1978) evaluated an imagery mnemonic that was recommended by several stage memory performers. To remember a name, it had to be converted into some easy-to-image pegword form. For example, the name Gordon could be converted into a ‘garden’. Then a garden would be imagined growing on some prominent feature of the person’s face to link the pegword cue and the item to be remembered. By deciphering the pegword cue ‘garden’ into ‘Gordon’, this mnemonic produced an 80 per cent improvement in the learning of names. Similar techniques have been extended to language learning, such as the Linkword system – extensively investigated and developed by Gruneberg (1987, 1992). The foreign words are converted to some similar-sounding English word that can be easily imaged. A mental image is then formed to link the image with the actual meaning of the foreign word. So, for example, the French for tablecloth is nappe, so Gruneberg recommends imagining having a nap on a tablecloth. Wilding and Valentine (1997) describe studies of memory champions and other memory experts, many of whom have discovered for themselves the value of mental imagery as a memory improvement technique. The use of imagery is not essential for memory improvement, of course. It is just one way in which material that is superficially meaningless and disconnected can be made more meaningful and connected and therefore easier to remember. A simple way of connecting words from a list is to compose a story. Bower and Clark (1969) showed that getting people to make up a story that linked together a list of 12 words made later recall of the words very much better.

Reflecting on our own learning
Metamemory refers to the understanding that people have of their own memory. When attempting to learn something, it seems reasonable to assume that we will monitor our own learning and schedule subsequent study activities to attempt to improve it. But how accurate are we at judging how well we have learned something? If the judgment is made soon after studying the material, we are comparatively poor at predicting our later performance. On the other hand, when the judgment is made after a delay, we are relatively better at making this judgment (Dunlosky & Nelson, 1992). If we can adequately judge how well (or poorly) we have learned material, we can apply this knowledge to inform our subsequent study plans, spending additional time on material that is less well learned. Laboratory studies suggest that people schedule their time appropriately, in just this way. But preliminary work by Metcalfe and Son (1999) suggests that, in some more natural learning situations, people are more likely to schedule their stu y time with emphasis on areas that they know well or find particularly interesting, neglecting areas that need work.


Even when we believe that we are ‘playing back’ some previous event or information in our mind, as if it were a videotape, we are actually constructing a memory from bits and pieces that we remember, along with general knowledge about how these bits should be assembled. This strategy is usually very adaptive, minimizing our need to remember new things that are very similar to things we already know. But sometimes there can be a blurring between what actually happened and what has been imagined or suggested.

Reality monitoring
The issue of reality monitoring – identifying which memories are of real events and which are of dreams or other imaginary sources – has been addressed by Johnson and Raye (1981). These researchers maintain that the qualitative differences between memories are important for distinguishing external memories from internally generated ones. They argue that external memories have stronger sensory attributes, are more detailed and complex and are set in a coherent context of time and place. By contrast, they argue that internally generated memories have more traces of the reasoning and imagining that generated them. Although Johnson (1988) found support for these differences, applying them as tests can lead to accepting memories as real, even when they are not. Morris (1992), for example, asked participants to recall details from a videotape and to report both their confidence and the presence or absence of clear mental imagery and detail. Although clear images and details were found to occur more often with correct reports, their presence led people to be overly confident: incorrect details accompanied by mental images were reported with greater confidence than correct details that lacked these images. So there does not seem to be any sure way of distinguishing between ‘real’ and ‘imagined’ memories. Related to the concept of reality monitoring is source monitoring – being able to successfully attribute the origin of our memories (e.g. being able to state that we heard a particular piece of information from a friend rather than hearing it on the radio news broadcast). Errors in attributing memories can have important consequences – for example, during eyewitness testimony (Mitchell & Johnson, 2000).

The misinformation effect
The distortion of memory through the incorporation of new information has been an important research topic for psychologists concerned both with the practical implications for eyewitness testimony, and with theoretical accounts of the nature of memory. Loftus and colleagues have explored in depth the misinformation effect (Fruzzetti et al., 1992; Loftus & Loftus, 1980). This arises when misleading information is introduced indirectly. For example, Loftus, Miller and Burns (1978) showed participants a series of slides along with the story of a road traffic accident. Later, the participants were questioned about the event. One of the questions was slightly different for half of the participants, in that it referred to a Stop sign instead of a Yield (Give Way) sign. Participants who were asked the question with the misleading information were more likely to identify falsely that particular slide in a later recognition memory test. These participants tended to choose the slide with the road sign that had been mentioned in the misleading question, rather than the one they had actually seen. Loftus and colleagues have repeatedly demonstrated similar distortions of memory reports after intervening, misleading questioning. The findings are robust and have implications for the sort of questions that eyewitnesses of crimes and accidents should be asked if their recall is to be as accurate as possible. However, the basis of the misinformation effect is disputed (see Chandler & Fisher, 1996, for a review). It is possible that the participants’ original memories are permanently distorted by the questioning, but it is also possible that the questions supply information that the participants would not otherwise be able to remember (see Saunders & MacLeod, 2002).

False memories
Related to the misinformation effect, but with more potentially serious consequences, are recovered and false memories (Ceci & Bruck, 1995; Loftus & Ketcham, 1994). Under therapy, some adults have recovered memories of alleged abuse in childhood that have led to criminal convictions (Loftus & Ketcham, 1994). But substantial research has shown that, under certain circumstances, false memories can also be created. Sometimes these are benign (Roediger & McDermott, 1995). However, it is also possible to create, using suggestions and misleading information, memories for ‘events’ that the individual believes very strongly happened in their past but which are, in fact, false (Ceci & Bruck, 1995; Loftus & Ketcham, 1994). So it remains at least plausible that some abusive events that people ‘remember’ are in fact false memories.

How knowledge leads to errors

How knowledge leads to errors
Our previous knowledge is a very valuable asset, but it can also lead to errors. Owens, Bower and Black (1979) illustrated this point well. They gave their university student participants a description of the activities performed by a character. For example, one of the sketches was about a student named Nancy. Here is the first part of that sketch: Nancy went to the doctor. She arrived at the office and checked in with the receptionist. She went to see the nurse who went through the usual procedures. Then Nancy stepped on the scale and the nurse recorded her weight. The doctor entered the room and examined the results. He smiled at Nancy and said, ‘Well, it seems my expectations have been confirmed.’ When the examination was finished, Nancy left the office. (p. 186) Half of the participants were told in advance that Nancy was worried that she was pregnant. These participants included between two and four times as many pieces of incorrect information when tested on their recall of the sketch. For example, some f them recalled ‘usual procedures’ as ‘pregnancy tests’. The errors were made in both recognition and recall tests. People have many expectations about how conventional activities (going to the doctor, a lecture, a restaurant) will proceed, and these provide schemas or scripts that can both aid and mislead. Bower, Black and Turner (1979), for example, studied the influence of such scripts on subsequent recall. In another part of their study, they also gave their participants stories based on normal expectations, but including variations from the norm. So, for example, a story about eating in a restaurant might refer to paying the bill at the beginning. When recalling these stories, participants tended to reorder them back to their schematic form or script. Other common errors involved including actions that would normally be expected in that context, but which had not been mentioned in the original story, such as looking at the menu. In general, the findings of these and similar studies indicate that people end to remember what is consistent with their schemas or scripts and to filter out what is inconsistent.

How knowledge promotes remembering

How knowledge promotes remembering
Experts in any area find it easier and quicker to learn new information within their expertise than do novices. This indicates that what we learn appears to depend heavily on our existing knowledge. For example, Morris, Tweedy and Gruneberg (1985) showed that there was a very strong relationship between how much their participants knew about soccer and the number of new soccer scores they could remember after hearing them just once. Participants were read a new set of soccer scores as they were being broadcast. One set of scores were the real scores, and another set was simulated by constructing plausible pairs of teams and assigning goals with the same frequency as had occurred in an earlier week. Participants in the study were told whether the scores they heard were real or simulated. Only the real scores seemed to activate the knowledge and interest of the soccer experts. For real scores (the darker bars in the figure), level of memory recall was clearly related to expertise – so more knowledgeable fans recalled more. For simulated scores (the pale bars), where the scores were highly plausible but not the genuine results, expertise had little effect on recall performance. These results illustrate the interaction of memory capacity with existing knowledge (and, presumably, also interest and motivation) in determining what is remembered.


Schemas – what we already know Bartlett (1932) asked English participants to read and then recall a Native American folk tale, The War of the Ghosts, which came from a culture that was very different from their own. When they attempted to recall the story, their reports were obviously based on the original tale, but they had added, dropped and changed information to produce stories that seemed more sensible to them – what Bartlett termed an ‘effort after meaning’. Bartlett proposed that we possess schemata (or schemas), which he described as active organizations of past experiences. These schemas help people to make sense of familiar situations, guiding expectations and providing a framework within which new information is processed. For example, we might possess a schema for a ‘typical’ day at work or at school. People seemingly have trouble understanding things if they cannot draw upon memory, or schemas, for previously acquired knowledge. This point was nicely illustrated in a study by Bransford and Johnso (1972). They gave participants a passage to remember, which began as follows: The procedure is actually quite simple. First you arrange items into different groups. Of course one pile may be sufficient depending on how much there is to do. If you have to go somewhere else due to lack of facilities that is the next step; otherwise you are pretty well set. It is important not to overdo things. That is, it is better to do too few things at once than too many. . . . (p. 722). Recalling the passage proved difficult, even if a title was given after the passage had been read. Bransford and Johnson (1972) found that it was only when the title (‘Washing Clothes’) was given in advance that recall was improved. The title explained what the passage was about, cued a familiar schema and helped people to make sense of the statements. With the title provided first, the passage became meaningful and recall performance doubled. So it seems that memory aids understanding; and understanding aids memory. It is possible to remem er without understanding, though – especially with extra aids, such as having the information presented for verification. Alba, Alexander, Hasher and Caniglia (1981) demonstrated that, although recall of the ‘Washing Clothes’ passage was much improved when the title was known in advance, recognition of sentences from the passage was equivalent, with or without the title. Alba and colleagues concluded that the title allowed the participants to integrate the sentences into a more cohesive unit, but that it affected only the associations among the sentences, not the encoding of the sentences themselves (which is why recognition performance was apparently preserved). The research with the ‘Washing Clothes’ passage illustrates how our previous knowledge helps us to remember. Bower, Clark, Lesgold and Winzenz (1969) provided another demonstration. They asked participants to learn sets of words that were presented either as a random filled hierarchical chart or in a wellorganized one.

Bower and his colleagues found that presenting the words in meaningful hierarchies reduced the learning time to a quarter of that required for the same words when they were randomly positioned in the hierarchy. The organization of the hierarchy apparently emphasized aspects of the words’ meanings, which appeared not only to simplify the learning of the lists but also to provide a framework within which the participants could structure their recall.

[Sir Frederick C. Bartlett (1886–1969) was one of Britain’s greatest psychologists. Although he began his research using Ebbinghaus’s methods and materials, he was dissatisfied with the limits of simple artificial materials and turned his attention to how people recall stories and pictures. His studies remained experimental and carefully controlled, but he began to use materials ‘of the type which every normal individual deals with’ (1932, p. v). Whereas Ebbinghaus tried to limit the effects of meaning and studied the effect of other variables on memory, Bartlett’s emphasis was on the role of meaning and social influences upon remembering. Even today, his work is often cited and is the basis of much contemporary research.]