1

The mind’s eye in blindfold chess

Guillermo Campitelli and Fernand Gobet

School of Psychology
University of Nottingham

Address of correspondence:

Dr. Fernand Gobet

School of Psychology
University of Nottingham
Nottingham NG7 2RD
United Kingdom

+ 44 (115) 951 5402 (phone)

+ 44 (115) 951 5324 (fax)
frg@psyc.nott.ac.uk

After September 30th, 2003:

Department of Human Sciences
Brunel University
Uxbridge
Middlesex, UB8 3PH

fernand.gobet@brunel.ac.uk

Running head: The mind’s eye in blindfold chess

 2

Abstract

Visual imagery plays an important role in problem solving, and research into

blindfold chess has provided a wealth of empirical data on this question. We show

how a recent theory of expert memory (the template theory, Gobet & Simon, 1996,

2000) accounts for most of these data. However, how the mind’s eye filters out

relevant from irrelevant information is still underspecified in the theory. We describe

two experiments addressing this question, in which chess games are presented

visually, move by move, on a board that contains irrelevant information (static

positions, semi-static positions, and positions changing every move). The results

show that irrelevant information affects chess masters only when it changes during

the presentation of the target game. This suggests that novelty information is used by

the mind’s eye to select incoming visual information and separate “figure” and

“ground.” Mechanisms already present in the template theory can be used to account

for this novelty effect.

 3

The mind’s eye in blindfold chess

Mental imagery, and in particular visual mental imagery, has been the subject

of intensive research in psychology. This research is perhaps best known for the

protracted debate of whether imagery is made possible by propositions or by

analogue mental images. Recently, experimental and theoretical research has been

seconded by brain-imaging techniques, which have helped develop hypotheses about

the brain tissues involved in manipulating mental images (for a review, see Kosslyn,

1994).

In spite of this extensive research, little is known about the role of visual

imagery in problem solving, and, in particular, in expert problem solving. While

some elements of answer have been provided in domains such as physics (Larkin &

Simon, 1987), mathematics (Paige & Simon, 1966) and engineering (Ferguson,

1992), many questions have yet to be answered. For example, to what extent do

mental images favor the creation of efficient internal representations and allow new

information to be integrated with prior knowledge? How do they enable selective

search in problem-solving situations? What is the role of the “figure” and “ground” in

mental images? How does expertise mediate mental images?

Few domains have been as instrumental as chess in providing preliminary

answers to these questions. The focus of this research has been to understand how

chess masters can use mental imagery to maintain a representation of the positions

generated during look-ahead search. Blindfold chess, where players play without

seeing the board, has been especially informative on the cognitive processes and

representations used by chess players to manipulate mental images (see Saariluoma,

1995, for a review).

This study aims to understand the relationship between visual perception and

imagery in chess. In particular, we are interested in the overlap of internal and

 4

external information within the mind’s eye. We designed two experiments where

chess players had to create and maintain images using relevant stimuli. At the same

time, they perceived irrelevant stimuli. In order to perform the task, participants had

to heed relevant stimuli continuously, and, in so doing, they could not avoid

perceiving the irrelevant stimuli. We were also interested in the role played by the

amount and novelty of irrelevant information presented. A final goal of this paper is

to link empirical data on mental images in chess to a current theory of expertise, the

template theory (TT, Gobet & Simon, 1996a, 2000). In particular, since the amount

of information stored in long-term memory (LTM) is crucial in TT, we were

interested in the role of previous knowledge and its interaction with the problem of

overlapping information in the mind’s eye.

After reviewing work on blindfold chess, we describe the template theory, and

show how it can account for most of the available data on blindfold chess. However,

one component of the theory, the mind’s eye, is still underspecified. The two

experiments of this paper are aimed at gathering information about this component.

Mental imagery in expert problem solving

The question of the nature of mental representations is probably as old as

philosophy. For example, Aristotle suggested that imagery is the main medium of

thought (Eysenck & Keane, 2000). In psychology, interest in mental imagery has

been unabated since the classical studies on rotation and scanning of mental images

by Cooper, Shepard, Kosslyn and others (see Kosslyn, 1994, for a detailed history).

Beyond standard experimental studies of mental images, research has investigated

their neurobiological substrate (e.g., Kosslyn, 1994), their relation with external

representations (Newell & Simon, 1972), and their role in the use of multiple
 5

representations (Lane, Cheng & Gobet, 2001; Tabachneck-Schijf, Leonardo &

Simon, 1997).

Little is known, however, about the links between expertise and mental images,

in particular in problem-solving situations. While there is substantial anecdotal

evidence that experts do use mental images in domains such as science (Miller,

1986), engineering (Ferguson, 1992), visual arts (Zeki, 1999), and sports (Jarvis,

1999), few empirical studies have been carried out to flesh out the mechanisms

involved. This is an unfortunate situation, as the use or misuse of mental images may

have important consequences in these domains, including for training and education.

The difficulty of a problem is in large part determined by how it is internally

represented (Newell & Simon, 1972). Mental images, while containing less detail

than external pictures and diagrams, share the computational advantages offered by

these (mainly, localization of information and inference operators; Larkin and Simon,

1987). A natural consequence of these assumptions is that experts should learn,

among other things, to use an efficient mode of representation, including the

operators necessary to take full advantage of it. Experimental support for the

effective use of mental images by experts has been gathered from algebra word

problems (Paige & Simon, 1966), computer programming (Petre & Blackwell, 1999),

industrial design engineering (Verstijnen et al., 1998), architectural design

(Chan, 1997), surgery (Hall, 2002), and sport (e.g., diving; Reed, 2002). However,

the strongest experimental evidence comes from the literature on blindfold chess,

which we will now review in detail.

Blindfold chess

In blindfold chess, a player carries out one or several games without seeing the

board, typically against opponents who have a full view of it; the moves are
 6

communicated aloud using standard chess notation. As it seems to require remarkable

cognitive capabilities, this style of play has attracted the interest of a number of

psychologists, starting with Alfred Binet (1893/1966), who asked well-known chess

players to fill in a questionnaire about the characteristics of their representations

while playing blindfold chess. He found that skilled players do not encode the

physical properties of the pieces and board, such as the color or style of pieces,

preferring an abstract type of representation. In an introspective account of the way

he played simultaneous blindfold chess, grandmaster and psychoanalyst Reuben Fine

(1965) emphasized the role of hierarchical, spatio-temporal Gestalt formations, which

allow the player to sort out the relevant from the irrelevant aspects of the position. He

also noted the possible interference between similar games, and the use of key

statements summarizing the positions as a whole. Finally, he stated that the use of a

blank chess board was more of a hindrance than a help for him, although other

players, such as George Koltanowski, who held the world record for the number of

simultaneous blindfold games, found this external help useful.

As suggested by this brief review, most evidence about blindfold chess has

been anecdotal. It was not until about ten years ago that Pertti Saariluoma, in a series

of ingenious experiments, systematically explored the psychology of blindfold chess.

In these experiments, one or several games were presented aurally or visually, with or

without the presence of interfering tasks. With auditory presentation, the games were

dictated using the algebraic chess notation, well-known to chess players (e.g., 1.e2-e4

c7-c5; 2. Ng1-f3 d7-d6; etc.). With visual presentation, only the current move was

presented on a computer screen. Saariluoma (1991) uncovered several important

issues. First, blindfold chess relies mainly on visuo-spatial working memory, and

makes little use of verbal working memory. Second, differences in LTM knowledge,

rather than differences in imagery ability per se, are responsible for skill differences.
 7

Third, in a task where games are dictated, masters show an almost perfect memory

when the moves are taken from an actual game or when the moves are random, but

legal, but performance drops drastically when the games consist of (possibly) illegal

moves. Saariluoma took this result as strong evidence for the role of chunking (Chase

& Simon, 1973) in blindfold chess. Fourth, visuo-spatial working memory is

essential in early stages of encoding, but not in later processing. According to

Saariluoma (1991), this is because the positions are later stored in LTM and thus

insensitive to tasks interfering with working memory.

Continuing this line of research, Saariluoma and Kalakoski (1997) uncovered

additional phenomena. First, replacing chess pieces with dots had little effect on the

memory performance for both masters and medium-class players—a result that

supports Binet’s (1893) conclusion of abstract representation in blindfold chess.

Thus, when following a game blindfold, the critical information is that related to the

location of the piece being moved, and not information about color or size. Second,

transposing the two halves of the board leads to a strong impairment, which,

according to Saariluoma and Kalakoski, is due to the time needed to build a mapping

between the perceived patterns and the chunks stored in LTM. Third, they found no

difference between an auditory and a visual presentation mode. Finally, given more

time, less skilled players increase their performance, although they still perform

worse than highly skilled chess players.

In a final set of experiments, Saariluoma and Kalakoski (1998) investigated

players’ problem solving ability after a position had been dictated blindfold. They

found that, in a recognition task, players show better memory with functionally

relevant pieces than with functionally irrelevant pieces; this effect disappears when

players’ attention is directed towards superficial features (counting the number of

white and black pieces) instead of semantically important features (searching for
 8

white’s best move). Moreover, in a problem-solving task, players obtained better

results when a tactical combination is possible in a game rather than in a random

position. Finally, although there was no performance difference between visuo-

spatial vs. auditory presentation of the moves, a visuo-spatial interfering task

(Brook’s letter task) negatively affected problem solving.

Saariluoma (1991) and Saariluoma and Kalakoski (1997, 1998) explained

their results utilizing a number of theoretical ideas: Chase and Simon’s chunking

theory, Ericsson and Kintsch’s (1995) long-term working memory theory, Baddeley

and Hitch (1974) theory of working memory, and Leibniz’ (1704) and Kant’s (1781)

concept of apperception—that is, second-order perception. In the following two

sections, we show that most of their results can be explained within a single

framework.

The Chunking and Template Theories

While Chase and Simon’s (1973) chunking theory is best known for its

characterization of the patterns stored in chess players’ long-term memory, it also

describes at some length the role of the mind’s eye, an internal store where visuo-

spatial operations are carried out. Basing their account upon previous research (e.g.,

Simon & Barenfeld, 1969), Chase and Simon proposed that the mind’s eye consists

of a system storing perceptual and relational structures, both from external inputs and

from memory stores. These structures can be subjected to visuo-spatial mental

operations, and new information can be abstracted from them.

With practice and study, players acquire a large number of perceptual chunks,

which are linked to useful information, such as possible moves or plans. Chunks can

also be linked to a set of instructions allowing patterns to be recreated as an internal

image in the mind’s eye. The mind’s-eye model acts as a production system (Newell
 9

& Simon, 1972): chunks are automatically activated by the constellations on the

external board or on the internally imagined chessboard, and trigger potential moves

or plans that will then be placed in short-term memory (STM) for further inspection

through look-ahead search and/or used to update the imagined chessboard. The

choice of a move, then, depends both on pattern recognition and on a selective search

in the space of the legal possibilities.

In spite of its good explanatory power overall, Chase and Simon’s chunking

theory has two main weaknesses: it underestimates encoding speed into LTM, and it

is mostly silent about the high-level knowledge structures that players use, such as

schemata (Holding, 1985). As a consequence, it has recently been revised by Gobet

and Simon (1996a; 2000) in their template theory (TT). As with the chunking theory,

chess expertise is mainly due to the storing of chunks in LTM, which relate to

familiar patterns of pieces. Patterns elicit chunks in LTM, which allow players to

recognize automatically (parts of) the positions and give access to relevant

information, such as information about strengths and weaknesses of the position,

tactical possibilities, and potential moves. Pattern-recognition processes—which are

assumed to be automatic and unconscious—are performed both from the internal

image that chess players are constructing and from the stimulus they are perceiving

(Gobet & Simon, 1998). Often-elicited chunks evolve into “templates,” which are

more complex data structures similar to the “schemata” commonly used in cognitive

psychology (see Lane, Gobet & Cheng, 2000, for an overview). Templates contain

both stable information (the core) and variable information (the slots), and it has been

estimated that it takes about 250 ms to add information to the slots of a template

(Gobet & Simon, 2000).

Aspects of TT have been implemented as computer programs. CHREST

(Chunk Hierarchy and REtrieval STructures; De Groot & Gobet, 1996a; Gobet &
 10

Simon, 2000) accounts for empirical data on chess memory and perception, such as

eye movements, the effect of distorting chess positions, and the role of presentation

time on memory recall. CHUMP (CHUnking of Moves and Patterns; Gobet &

Jansen, 1994) implements the idea that chunks elicit possible moves and sequences of

moves. Finally, the relation of mental imagery to problem solving has been

investigated in SEARCH (Gobet, 1997), an abstract model of chess playing, which

combines assumptions about the role of chunks and templates with assumptions about

move generation and properties of the mind’s eye, including decay and interference

mechanisms.

As in the chunking theory, the mind’s eye in TT is considered as a visuo-

spatial structure preserving the spatial layout of the perceived stimulus, where

information can be added and updated (De Groot & Gobet, 1996; Gobet & Simon,

2000). The theory includes time parameters for carrying out various types of

operations, such as moving a Bishop diagonally or a Rook horizontally. These

mechanisms, which have been recently applied to modeling the way students learn

multiple representations in physics (Lane, Cheng & Gobet, 2001), are closely related

to the mechanisms proposed by Tabachnek-Schijf, Leonardo and Simon (1997) in

their CaMeRa model, which simulates the behavior of an economist solving a supply-

and-demand problem. CaMeRa, in turn, incorporates some of the ideas proposed by

Kosslyn (1994) in his theory of mental imagery, which details how several areas of

the visual system, including the visual buffer, take part in the generation,

transformation, inspection, and maintenance of visual images.1

1
Incidentally, the presence of these and other computational models, which provide detailed
mechanisms for processing information in the mind’s eye, indicates that this concept does not suffer
from the joint problem of the homunculus and infinite regress (the contents in the mind’s eye has to
be ‘seen’ by another mind’s eye, the contents of which has to be seen by a third mind’s eye, and so
on).
 11

Search processes are carried out in the mind’s eye: when an anticipated move is

carried out, the changes are performed there (Chase & Simon, 1973; Gobet, 1997).

The mind’s eye is subject to decay and to interference, the latter both from

information coming from the mind’s eye and from external information. Several

predictions of TT and SEARCH about the role of the mind’s eye in problem solving

are supported by the empirical data, including: the high-levels at which chess players

organize their knowledge (De Groot, 1946/1978; De Groot & Gobet, 1996; Holding,

1985); the fact that players often revisit the same positions during search (De Groot,

1946/1978); and the increase of mean depth of search and of the rate of search as a

function of skill (Charness, 1981; Gobet, 1998b, Saariluoma, 1990).

Applying the Template Theory to Blindfold Chess

A substantial part of Saariluoma’s research into blindfold chess was carried out

to identify the possible role of chunking mechanisms when playing without seeing

the board. Given that these mechanisms have been vindicated by empirical data
 12

(Saariluoma, 1991; Saariluoma & Kalakoski, 1997), it is not surprising that the

template theory, which is based upon the chunking theory, accounts for data on

blindfold chess reasonably well. We now discuss how the main empirical results can

be explained by this theory. Note that these explanations have not been developed ad

hoc for blindfold chess, but have been used to explain similar phenomena with plain-

view chess (Gobet, 1998a).

Several mechanisms inherent in TT are of importance in the application of the

theory to blindfold chess. First, positions that recur often tend to lead to the

development of templates; as a consequence, the initial chess position, as well as the

positions arising from the first moves in the openings familiar to players, will elicit

templates, in particular with masters. Second, templates can be linked to each other

and can be linked to moves. For instance, the initial position is linked, among other

moves and templates, to the move 1.e2-e4 and to the template encoding the position

arising after this move; in turn, this template is linked to the move 1…c7-c5 and to

the template describing the position arising after 1.e2-e4 c7-c5 (see Figure 1).

-------------------------------

Insert Figure 1 about here

-------------------------------

We can now apply the theory to blindfold chess research, starting with

Saariluoma’s (1991) results. The role of LTM knowledge and of chunking are

obviously at the center of TT. For example, the fact that actual games are better

recalled than random legal games, which are in turn better recalled than random

illegal games, can be explained as follows (essentially Saariluoma's, 1991,

explanation): masters, who have more chunks with which they can associate

information about moves, are more likely to find such chunks even after random

moves. With random illegal games, however, chunks become harder and harder to
 13

find, and masters’ performance drops. Random legal games drift only slowly into

positions where few chunks can be recognized, and, therefore, allow for a relatively

good recall.

The fact that players are sensitive to visuo-spatial interfering tasks early on,

but not later on, is explained as follows: early on, these tasks would interfere with the

access of chunks and templates, and with their potential modification; once this has

been done, interfering tasks are less detrimental because information is already stored

in LTM. The predominant role of visuo-spatial memory over verbal memory is

captured by the mainly visuo-spatial encoding of chunks (Chase & Simon, 1973).

Similarly, the results of Saariluoma and Kalakoski (1997) are consistent with

TT. Information about color and size may be hidden to players, because it is easy for

them to derive them from location, as chunks are location-sensitive (Gobet & Simon,

1996b; Saariluoma, 1994). The effect of transposing the two halves of the board is

explained by the difficulty to access chunks (basically the same explanation as

Saariluoma and Kalakoski’s). Modality of presentation (visual or auditory) does not

matter, as long as the information can be used to update the position internally, and

therefore access chunks. Finally, the speed of presentation time strongly affects

performance, a direct prediction of TT, where cognitive processes, including decay of

information in the mind’s eye and LTM storage, directly depend upon the amount of

time available.

With respect to problem solving (Saariluoma & Kalakoski, 1998), the fact

that functionally-relevant pieces are better encoded than irrelevant pieces follow from

the idea that functionally-relevant pieces are more likely to attract attention, and

therefore to elicit chunks and templates (in the simulations with CHREST, this can

already be observed in the early seconds of the presentation of a position (De Groot

& Gobet, 1996)). Similarly, orienting tasks (e.g., counting the number of pieces)
 14

change the object of attention; as a consequence, they affect which chunks will be

retrieved, which in turn influences memory performance. Better problem-solving

performance with game than with random background is explained by the fact that

game background is more likely to elicit relevant chunks, because it offers more

context and therefore more opportunity for accessing knowledge. Finally, visuo-

spatial interference tasks offer a checkered pattern of results. They affect problem

solving because search mechanisms occurring in the mind’s eye are impaired.

However, these tasks do not affect memory if the position is presented before the

interfering task (Saariluoma & Kalakoski 1998, exp. 4). On the other hand, if the task

is performed at the same time as the presentation of a game, the performance in

memory is indeed impaired (Saariluoma, 1991, exp. 6). According to TT, this result

is explained by the fact that, as soon as a template has been accessed, information can

be stored there rapidly, which makes memory less sensitive to the operations of the

mind’s eye.

As we have just seen, varying the background affects problem-solving

performance, and, presumably, the cognitive operations carried out in the mind’s eye.

It is also likely that background affects cognition in a memory task as well. To test

this hypothesis, and to explore how a possible effect is modulated by skill level, we

presented a game blindfold with a background, which is totally irrelevant to the target

game. This manipulation is interesting theoretically, because it raises the question of

how perceptual processes can separate the (relevant) figure from the (irrelevant)

ground before the information is used in the mind’s eye.

Experiment 1

The purpose of this experiment was to understand the relationship between

visual perception and imagery within the mind’s eye. We used a method similar to
 15

that used by Saariluoma and Kalakoski (1997): chess games were presented on a

computer monitor “blindfold”only the current move was displayed on an empty

board. While these authors were interested in the type of information used by chess

players (type of piece, color and location), we were interested in the effect of

background interference in blindfold chess, and manipulated the context surrounding

the piece being moved. Hence, the moves were presented normally, but, in the

interference conditions, we placed pieces not related to the target game throughout

the board. Two games were presented simultaneously, and the ability to remember

positions was measured three times: after 10 ply,2 30 ply, and 50 ply; memory for the

entire game was tested at the end.

Participants

Sixteen Argentinian players volunteered for this experiment: 8 masters

(including 3 international masters and 3 FIDE masters) and 8 Class A players. The

masters had an average international rating (ELO)3 of 2,299 and an average national

rating (SNG) of 2,193. Class A players had an average national rating of 1,885.

(They did not have international rating.) The average age of the sample was 20.5

years (SD = 5.2) with a range from 14 to 31.

2
A ply (or half-move) corresponds to a piece movement by either White or Black. A move consists of
two ply, one by White and one by Black.
3
The Elo rating is an interval scale with a standard deviation of 200 points, which is used to delimit
skill classes. Players with more than 2200 points are called ‘masters’, from 2000 to 2200 ‘Experts’,
from 1800 to 2000 ‘Class A players’, from 1600 to 1800 ‘Class B players’, etc. The international
Chess Federation (FIDE) recognizes three titles: international grand master (usually players above
2500 Elo), international master (above 2400) and FIDE master (above 2300). There are also national
ratings that resemble the Elo system. In the case of Argentina, it is called SNG (National Grading
System).
 16

Material

Six grandmaster games were carefully chosen so that they were not played by

elite grandmasters and did not follow common opening lines. The mean number of

pieces for the three stages of reconstruction was 32 (SD = 0) after the 10th ply, 26.7

(SD = 1.5) after the 30th ply, and 20 (SD = 1.2) after the 50th ply.

Design and Procedure

We used a 2x3x3 design, where Skill (masters and Class A players) was a

between-participant variable, and where Interference (Empty Board, Initial Position

and Initial Position in the Middle) and Depth (10, 30 and 50 ply) were within-

participant variables. The orders of the conditions and of the games were

counterbalanced.

The six games were presented on a computer screen “blindfold,” that is, the

players could see only the moves but not the current position. Three experimental

conditions were used. In the control condition, the moves were played on an empty

board. In the first interference condition, the moves appeared on a board that

contained the initial position of a chess game. In the second interference condition,

the moves were carried out on a board where the initial position was transposed to the

middle of the board (the 32 pieces were placed on rows 3 to 6, rather than on rows 1,

2, 7, and 8, as in the normal initial position).

In the three conditions, the participants were told that they had to follow two

games mentally, starting with the initial position and updating the position with the

moves presented on the board. The target piece was first presented for one second in

its origin square and then for two seconds in its destination square. It was surrounded

by a green square to discriminate it from irrelevant pieces.

 17

The games were presented as follows. The first 10 ply of game 1 were played,

followed by the first 10 ply of game 2. At this point, an empty board appeared on the

screen, and participants had to reconstruct the last position in each game. Then, ply

11 to 20 of game 1 were played, followed by ply 11 to 20 of game 2, and then ply 21

to 30 of game 1, followed by ply 21 to 30 of game 2. At this point, participants had

to reconstruct the last position of each game. Finally, ply 31 to 40 of game 1, ply 31

to 40 of game 2, ply 41 to 50 of game 1, and ply 41 to 50 of game 2 were played. At

the end, participants had again to reconstruct the final position of each game.

When they had finished reconstructing the positions after 50 ply, participants

were presented with a board containing the initial position and had to reconstruct the

moves of game 1 and, then, the moves of game 2. Players were allowed a maximum

of 4 minutes to reconstruct the positions and a maximum of 10 minutes to reconstruct

the moves of a game. The time spent in reconstruction was recorded with a

stopwatch. At the end of this procedure, participants had a five-minute break, after

which they started the same cycle with games 3 and 4 in a different experimental

condition, followed by another five-minute break and the same cycle with games 5

and 6 with the third condition.

Before starting the experiment, all subjects went through a practice session in

order to familiarize themselves with the procedure and the use of the mouse in the

reconstruction of positions. The practice consisted in following the procedure

explained above, for 6 practice games (2 per condition) until the reconstruction of ply

10 in all the games.

 18

Results

Recall of Positions

Figure 2 shows the means for the percentage of pieces correctly replaced.

ANOVA indicated a main effect of Depth [ F(2,13) = 92.1, MSE = 20,320; p < .

001]. Post-hoc Scheffé tests showed that the differences were between Ply 10 and Ply

30, Ply 10 and Ply 50, as well as between Ply 30 and Ply 50. There was also a main

effect of Skill [F(1,14) = 122.2, MSE = 26,956; p < .001]. However, no main effect

was found for Interference [F(2, 13) < 1, MSE = 20.7]. We found only one

significant interaction: Depth x Skill [F(2,13) = 25.3, MSE = 5,581, p < .001], due to

the fact that Class A players were more affected by depth than masters.

-------------------------

Insert Figure 2 about here

-------------------------

We also analyzed the errors of omission (total number of pieces minus the

number of pieces replaced) and of commission (pieces incorrectly replaced). Errors

of commission consist of pieces placed on the board that were not present in the

actual position and pieces placed on an incorrect square. Since the positions did not

have the same number of pieces, we report the results as percentages.

With respect to the percentage of errors of commission, we found a pattern

similar to that found with the percentage of pieces correctly replaced. The mean

percentages for Skill were 6.0 (SD = 12.0) for masters, and 25.5 (SD = 26.2) for

Class A players. The mean percentages for Depth were: 1.7 (SD = 3.8) for Ply 10;

15.5 (SD = 16.3) for Ply 30; and 30.0 (SD = 29.1) for Ply 50. Finally, the mean

percentages for Interference were: 16.3 (SD = 23.3) for Empty Board; 14.9 (SD =

21.5) for Initial Position, and 16.0 (SD = 22.8) for Initial Position in the Middle.

There were main effects for Skill [F(1,14) = 109.7, MSE = 27,332; p < .001], Depth
 19

[F(2,13) = 76.8, MSE = 19,146; p < .001], but not for Interference [F(2,13) < 1, MSE

= 55.34]. Again, Depth x Skill was the only significant interaction [F(2,13) = 24.1,

MSE = 6,010; p < .001]. The same pattern of results was found for the percentage of

errors of omission: Skill [F(1,14) = 12.5, MSE = 1,842; p < .001]; Depth [F(2,13) =

12.6, MSE = 1,860; p < .001]; Interference [F(2,13) < 1, MSE = 13.36]; Interaction

Depth x Skill [F(2,13) = 4.0, MSE = 592.04; p < .02].

Finally, we found a similar pattern of results for reconstruction time, with

main effects of Skill [F(1,14) = 7.7, MSE = 36,450; p < .01] and Depth [F(2, 13) =

98.9, MSE = 469,204; p < .001], but not of Interference [F(2,13) < 1, MSE =

555.96]. No interaction was significant.

Reconstruction of games

For this variable we utilized a 2 x 3 ANOVA model (Skill x Interference).

The mean percentages of moves correctly reported were 86.4 (SD = 13.9) for masters

and 41.1 (SD = 24.6) for Class A players. Regarding Interference, the means were

60.7 (SD = 32.5) for Empty Board, 67.0 (SD = 28.1) for Initial Position and 63.6 (SD

= 30.5) for Initial Position in the Middle. Again, we found a Skill effect [F(1, 14) =

122.8, MSE = 49,232; p < .001], but no Interference effect [F(2,13) < 1, MSE =

322.4]. The interaction term was not significant [F(2,13) < 1, MSE = 392.6].

Discussion

In this experiment, we repeatedly found the following pattern of results: (a) a

main effect of Skill; (b) a main effect of Depth; (c) an interaction between Skill and

Depth, and (d) no main effect of Interference. The first result naturally flows from

TT, as we have seen above. As a consequence of their experience with the game and

their study of chess literature, chess players have acquired a considerable knowledge

base of both chunks and templates, which allows them to recognize familiar patterns
 20

automatically. Since masters’ knowledge base is much larger than that of Class A

players, the main effect of Skill arises.

The second and third findings can also be explained easily by TT. The

difference in performance between masters and Class A players arises at Ply 30 and

increases at Ply 50, but it does not exist at Ply 10. The positions and the moves close

to the starting position are well known both to masters and to Class A players; hence,

we do not expect big differences in the corresponding recall performances. However,

as soon as the game progresses, masters can recognize more chunks and templates

than Class A players. Interestingly, some of the masters (but not all of them)

experienced impairment in their performances at Ply 50. We suggest that this is due

to the lack of familiarity with positions corresponding to Ply 50, which makes pattern

recognition harder.

The fourth finding is more challenging, and led us to design the second

experiment. Surprisingly, performance was not impaired in the conditions displaying

the initial position, either on its normal location or in the middle of the board. In

addition, there was no sign of interaction for this variable. We suggest two

explanations. First, it is not necessary for chess players to use the perceived

representation of the external chessboard as an aid to update the internal

representation of the position in the mind’s eye. Therefore, they just process the

move that is being presented, and ignore the board and the other pieces. Thus, the

interference position never gets processed. Second, chess players use the external

board’s percept as a help to refresh the image of the current position, but, early on,

they can avoid processing the other (irrelevant) pieces on the board in depth, because

they are not unexpected (the interference position remained unchanged during the

whole task).
 21

In order to tease apart these hypotheses (no processing of the board, or

processing-plus-early-filtering), we designed a second experiment where the

interference positions changed during the task. We speculated that the novelty of the

position would cause its automatic processing, therefore impairing performance for

all skill levels. This assumption flows naturally from TT and, and in particular from

its computer implementation, CHREST, where novelty is one of the heuristics used

to direct eye movements and is at the heart of its discrimination learning mechanism.

A further goal of this experiment was to gain additional information on the presence

of templates by an analysis of the types of errors made during reconstruction.

Experiment 2

The purpose of this experiment is to test our hypothesis that the lack of main

effect of interference in the first experiment was due to the lack of novelty in the

interference positions. We used two interference conditions, different from those used

in experiment 1. In the first condition (move-by-move condition), fifty positions (one

for each move in the target game) were used as interference. These positions

belonged to an unrelated game, which started with the same opening as the target

game. In the second condition (semi-static condition), five different interference

positions were displayed, which corresponded to five positions in a game, ordered

chronologically.

In a pilot study, we found that a Class A player could not do the task at all.

Therefore, we decided to increase the skill level of the two groups in comparison to

the first experiment, recruiting stronger masters and having Experts instead of class A

players. In addition, a FIDE master tested in the pilot study reported that the task had

been demanding and that he felt extremely tired in the second part of the experiment.
 22

Hence, we decided to eliminate some components of the experiment, leaving intact

the elements that we considered essential for our research questions.

Participants

There were 16 volunteers: 8 masters (4 international masters and 4 FIDE

masters) and 8 Experts (all of them with international rating but without any FIDE

title). The mean international rating was 2,351 (SD = 45.8) for the masters and 2,113

(SD = 51.9) for the Experts. The mean national rating was 2,256 (SD = 67.85) for

the masters and 1,996 (SD = 70.14 ) for the Experts. The average age of the sample

was 25.9 years (SD = 8.1; range 15 to 39).

Material

From a pool of grandmaster games with relatively uncommon openings, we

selected 6 games for the stimuli to memorize, and 6 games for constructing the

interference positions. For the stimuli positions, the mean number of pieces at the two

moments of reconstruction were 26.8 (SD = .75) at Ply 30, and 21.0 (SD = .89) at Ply

50. For the Interference Positions, the means were 27.8 (SD= .98) at Ply 30 and 21.5

(SD = 1.0) at Ply 50. The interference positions were similar to the experimental

games on the first moves. In the move-by-move interference condition, we used 50

different positions; in the semi-static condition, we used 5 different positions.

Design and Procedure

We used a 2x3x2 ANOVA design. Skill (masters and Experts) was a between-

participant variable; Interference (Empty board, move-by-move and semi-static) and

Depth (30 and 50 ply) were within-participant variables. The procedure was similar

to that in Experiment 1, with changes mainly with the interference conditions. In

addition, we did not ask the participants to reconstruct the position after 10 ply, and
 23

we did not ask them to reconstruct the games at the end. We took these decisions in

order to keep the length of the experiment within bearable bounds.

Every interference position corresponded to a game starting with similar

moves in the first ply. In the move-by-move condition, we presented a different

interference position at each ply. (In this condition, participants reported a sensation

of motion.) In the semi-static condition, the interference positions lasted 10 ply.

Hence we used 5 interference positions, taken after 10, 20, 30, 40 and 50 ply. The

presentation of the game in the control condition was similar to that in Experiment 1.

Before starting the experiment, all subjects went through a practice session.

The practice consisted in following the procedure explained in Experiment 1, for 6

practice games (2 per condition) until the reconstruction after 10 ply in all games.

(Note again that during Experiment 2 itself, the players did not reconstruct the

positions after 10 ply.)

Results

Figure 3 shows the means for the percentage of pieces correctly replaced.

There was a main effect of Skill [F(1, 14) = 42.6, MSE = 16,894; p < .001), Depth

[F(2, 3) = 20.3, MSE = 8,066; p < .001) and Interference [F(2, 13) = 13.7, MSE =

5,432; p < .001). No significant interaction was found. Post-hoc Scheffé tests

showed a significant difference between the means in the empty-board and semi-

static conditions (17.3%; p < .001), and between the empty-board and move-by-move

conditions (14.5%; p < .001). However the difference between the two interference

conditions (3.1%) was not statistically significant.

--------------------------------

Insert Figure 3 about here

--------------------------------
 24

A similar pattern of results was found for the percentage of errors of

omission: main effects of Skill [F(1,14) = 16.1, MSE = 6,958; p < .001], Depth

[F(2,13) = 16.7, MSE = 7,226; p < .001], and Interference [F(2,13) = 4.4, MSE =

1,927; p < .02]. No interaction was significant. The only discrepancy was that the

post-hoc Scheffé tests showed that only the difference between the empty-board and

semi-static conditions (10.2%; p < .03) was significant. The results for the errors of

commission were the same in all respects except for Depth: main effects of Skill

[F(1, 14) = 13.4, MSE = 4,134; p < .001) and of Interference [F(2,13) = 4.4, MSE =

1,346; p < .02], but not of Depth [F(2,13) = 2.7, MSE = 837.46; ns]. There was no

statistically significant interaction. Post-hoc Scheffé tests showed reliable differences

only between the empty-board and semi-static conditions (8.9%; p <. 02).

We further classified errors of commission into the following subtypes: Same

(the piece is placed on a correct previous [but not actual] location of the same game),

Interference (the piece belongs to the interference position), Pair (the piece belongs to

a previous position of the game presented in the same pair), and Inference (the

location of the piece is inferred incorrectly). In order to help visualize the relative

influence of errors of commission and omission, we plotted them together in Figure 4

, which shows the average percentages of errors after 30 and 50 ply, all players

included.

The use of schemata for each game, as proposed by TT, predicts that there

should be few errors where information is confused between the two games (pair

errors) and where the information is confused with the interference position

(interference errors). On the other hand, TT predicts that there should be more errors

within the same game (same errors), since players may maintain a template for the

game but fail to update it correctly. If, however, games are mainly coded as chunks,

interference and pair errors should be more frequent, as an entire chunk may be
 25

incorrectly assigned to a game. Errors of omission would reflect more the difficulty

of the task, and would not differentiate between these two hypotheses. Figure 4

shows that errors of omission and errors of the same game are the more frequent, thus

supporting the presence of templates.

--------------------------------
Insert Figure 4 about here
--------------------------------

Discussion

Five main results were found in the second experiment: (a) even though the

skill difference between groups was reduced in comparison to the first experiment

(from 308 to 260 in national rating), there was a significant Skill effect (expectedly,

the effect was smaller); (b) in the control condition, masters performed worse than in

the same condition of Experiment 1; (c) in comparison to Experiment 1, the

interaction Depth x Skill has disappeared; (d), there was an Interference effect: both

Interference conditions differed from the control condition, but there was not any

difference between each other; and (e) the presence of templates was supported by an

analysis of the types of errors made during reconstruction.

The first finding has been already explained in the discussion of Experiment

1. The second finding is due to the fact that we eliminated the reconstruction phase

after 10 ply. It is likely that the reconstruction at that stage helped participants to

memorize the game, an opportunity that was not present in Experiment 2. The lack of

interaction can also be explained easily with the elimination of the reconstruction task

at ply 10. Indeed, the pattern of results in Experiment 2 for the reconstruction of the

positions after 30 and 50 ply is similar to that in Experiment 1.

The most important finding is that both interference conditions had a reliable

effect, unlike in Experiment 1. Somewhat surprisingly, the condition with only 5

 26

different interference positions impaired performance as much as the condition with

50 different interference positions. These results support our second hypothesis —

chess players use the board’s percept as a perceptual help to refresh the positions of

the game they are following mentally. In the second experiment, unlike in the first,

interference positions were varied enough to impair players’ performance. We

hypothesized that the novelty of the irrelevant pieces would make it hard to avoid

their processing. However, although novelty did cause a decrement in performance,

larger amounts of novelty did not cause larger impairment. We will take up this result

in the general discussion. Finally, the analysis of errors provided direct support for

the construct of templates.

Subjects’ Strategies and the Role of Study Time

Given the complexity of the task, it is likely that players developed strategies

to memorize games. Informal comments indicate that there are several moments

during which players engage in visual rehearsal. One such moment is after the

presentation of each move, when, according to TT, the new position is updated in the

mind’s eye. Another such moment is before the shifting between the games. Reports

also indicate that, as soon as the game goes further away from the initial position, it

becomes harder to update. This is in line with TT’s prediction that middlegame

positions are harder to categorize, and hence are less likely to activate a template. The

difficulty in updating creates a time pressure: participants have to choose between

either continuing rehearsing while the next move is presented, or quitting rehearsing

and focusing in the next move. As a consequence, according to TT, the image in the

mind’s eye decays and becomes subject to errors. This general mechanism is also

supported by the fact that, in the control condition of the second experiment, there

was an impairment in performance as compared with the control condition in the first
 27

experiment. The difference was that players had to reconstruct the position after 10

ply in the first experiment, but not in the second. After having reconstructed the

position, participants spent additional time studying the position before moving to the

reconstruction of the second game or to the continuation of the presentation of

moves. We believe that this study time, not available in the second experiment,

allowed players to consolidate the LTM encoding of the game, and hence to obtain a

better recall later. This explanation is consistent with that given by Gobet and Simon

(2000) to account for the role of presentation time in a recall task.

General Discussion

In the introduction of this article, we have shown how TT, a general theory of

expertise, explains most of the data on blindfold chess available in the literature,

without amendment or ad hoc assumptions. We also noted that the mind’s eye is still

underspecified in the theory. In order to shed light on this issue, two experiments

were carried out, where chess games were presented visually, move by move, on a

board that contained irrelevant information; this background information was either

static, semi-static positions, or updated after every move. These experiments had the

novelty of presenting interference patterns embedded in the context of the stimuli and

opened questions about the convergence of images generated by external input and

images generated by internal processes.

While the two experiments emphasized memory and involved only a small

amount of problem solving, they required players to engage mechanisms that lie at

the core of their expertise: the processing of (sequences of) moves. The results were

consistent with TT. It was found that additional time to process moves led to better

recall, a direct (but not surprising) prediction of the model. Another less obvious

prediction was that the number of errors of the “same game” type should be higher
 28

than the other errors of commission. Finally, as expected, depth had a reliable impact.

We also acquired some evidence that changes in the background may affect

performance, and proposed that novelty processing may be at the core of this

phenomenon. Specifically, results showed that irrelevant information affects chess

masters only when it changes during the presentation of the target game. This

suggests that novel information is used by the mind’s eye to select incoming visual

information and separate “figure” and “ground”. In the conditions used in

Experiment 1, the lack of novelty of the interference stimuli may have led to their

inhibition. However, in the interference conditions of Experiment 2, the positions

were changed, leading novelty-detection mechanisms to process them automatically.

Within the framework of TT, two mechanisms are possible to explain these

results. First, novelty detection essentially happens at the perceptual level. Eye

movements are directed, perhaps by cues in peripheral vision, to squares where novel

information is provided (e.g., the move in the target game), and are inhibited from

returning to squares where information is static. Second, novelty detection occurs at

the stage of memory encoding: for example, when discrimination learning

mechanisms are engaged (see also Tulving et al., 1994). In the second, but not in the

first case, chess players can be said to perceive a stimulus at the same time as they

hold a different image in memory. In this case, it is likely that the time necessary for

processing the information after presentation of a new move (update in STM, update

in the mind’s eye) impairs image rehearsal, leading to a lessening in information.

This is because there is potentially overlapping and competing information in the

mind’s eye: the position presented on the screen and the image corresponding to the

actual game. The results of Experiment 2 supported the second mechanism.

A question that is not resolved by the data is whether we are dealing with

selective attention (the attention is directed towards the last move in the target game) or
 29

whether we are dealing with selective inattention (the attention towards the interfering

position is somehow inhibited). Laeng and Teodorescu (2002) have demonstrated that

eye movements during mental imagery are instrumental in image

generation. Supplementing the blindfold experiments with eye-movement studies

may offer a means of answering whether attention, or inattention, is selective in

blindfold chess.

The mechanisms we have proposed are consistent with Kosslyn's (1994) theory of

imagery. In this theory, which we can only outline here, the visual buffer roughly plays

the same as the mind's eye in TT. According to Kosslyn, it is located in the occipital

lobe (primary visual area) and is topologically arranged. Associative memory, thought

to be located in the superior, posterior temporal lobes and in the area at the junction of

the temporal, parietal, and occipital lobes, is the place where outputs from the ventral

system (encoding object properties) and from the dorsal system (encoding spatial

properties) come together. Associative memory stores associations between perceptual

representations as well as conceptual information. What we call “chunks” and

“templates” seems to correspond to what Kosslyn calls “structural descriptions”: rich

associative and conceptual representations encoding both parts of scenes or objects and

their spatial relations. Unlike the visual buffer, associative memory is not

topographically organized. Kosslyn (1994) suggests that visual patterns might be stored

in arrays in the ventral system (inferior temporal lobe); in order to generate a visual

image, these arrays should activate cells in the visual buffer. Once accessed, these

memory structures are free of interference; by contrast, information in the visual buffer

fades quickly. As a consequence, processing a large amount of information in the visual

buffer leads to interference and loss of performance.

Applied to our experiments, Kosslyn’s theory predicts that, at the beginning of a

game, all players are able to access a structural description of the initial position. When
 30

the initial moves are dictated, players can access other structural descriptions linked to

the current one, without the need to generate a visual image in the visual buffer.

However, when no structural descriptions are available, which happens earlier in

experts than in masters, it becomes necessary to maintain an image in the visual buffer.

How does this affect performance? In the first experiment, the irrelevant perceptual

information, not being novel, does not attract attention and does not cause interference.

However, in the second experiment, the irrelevant incoming perceptual information is

indeed novel. This causes interference in the visual buffer between information coming

from the retina and the image of the position coming from the object properties

encoding subsystem. As far as we now, Kosslyn’s (1994) attention-shifting system,

which disengages, moves, and engages attention, does not seem to be specified to the

point to make clear-cut predictions about the role of novelty in our experiments.

Beyond its theoretical relevance, this study has implications for applied

research. In education, there is currently a renewed interest in the potential benefit of

multiple representations, both internal and external (Ainsworth, 1999). In many

technical and scientific domains, being able to combine multiple representations is

one of the required skills. However, using multiple representations may be time and

resource consuming (Gobet & Wood, 1999). For example, diagrammatic and other

external representations need to be translated into mental images in order to be fully

integrated into other types of knowledge. Learning (how to generate) such images

and how to link them to other information has a cost, that must be weighted against

the benefits of using them. In particular, as suggested in our study, depth of

anticipation, as well as changes of information in the background environment, may

lead to a decrease in performance, perhaps to the point where mental images cannot

be relied upon any more.

There is still much to learn about how expertise mediates mental images. In
 31

this paper, we have used an existent computational model to account for previous

data on blindfold chess, and have described two experiments aimed at understanding

the role of background in the manipulation of mental images. We found that

irrelevant information affects masters’ performance only when it changes during the

presentation of the target game. We have argued that background novelty has

important repercussions for theories of expertise and imagery, in that it plays a key

role in what is attended to and/or learned.

Most research on experts’ mental images has been carried out with blindfold

chess, and this article is no exception. We urge researchers to replicate studies pursued

with blindfold chess in other domains of expertise—both ‘classic’ domains (such as

physics, medicine, or sports) and new domains—in order to test the validity of the

theoretical ideas discussed in this paper.

 32

References

Ainsworth, S.E., (1999) A functional taxonomy of multiple representations.

Computers and Education, 33, 131-152.
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), The
Psychology of Learning and Motivation. New York: Academic Press, pp. 47-89.
Binet, A. (1966). Mnemonic virtuosity: A study of chess players. Genetic Psychology
Monographs, 74, 127-164. (Original work published in 1893).
Chan, C. S. (1997). Mental image and internal representation. Journal of
Architectural and Planning Research, 14, 52-77.
Charness, N. (1981). Search in chess: Age and skill differences. Journal of
Experimental Psychology: Human Perception and Performance, 2, 467-476.
Chase, W. G., & Simon, H. A. (1973). The mind's eye in chess. In W. G. Chase
(Ed.), Visual information processing (pp. 215-281). New York: Academic Press.
De Groot, A. D. (1978). Thought and choice in chess. The Hague: Mouton. (Original
edition in Dutch: 1946)
De Groot, A. D. & Gobet, F. (1996). Perception and memory in chess. Assen: Van
Gorcum.
Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological
Review, 102, 211-245.
Eysenck, M. W., & Keane, M. T. (2000). Cognitive psychology: A student's
handbook. Hove, UK: Psychology Press.
Ferguson, E. S. (1992). Engineering and the mind's eye. Cambridge, MA: MIT Press.
Fine, R. (1965). The psychology of blindfold chess. An introspective account. Acta
Psychologica, 24, 352-370.
Gobet, F. (1997). A pattern-recognition theory of search in expert problem solving.
Thinking and Reasoning, 3, 291-313.
Gobet, F. (1998a). Expert memory: A comparison of four theories. Cognition, 66,
115-152.
Gobet, F. (1998b). Chess players' thinking revisited. Swiss Journal of Psychology, 57,
18-32.
Gobet, F. & Jansen, (1994). Towards a chess program based on a model of human
memory. In H. J. van den Herik, I. S. Herschberg, & J. W. Uiterwijk (Eds.),
Advances in computer chess 7. Maastrich: University of Limburg Press.
 33

Gobet, F., & Simon, H. A. (1996a). Templates in chess memory: A mechanism for
recalling several boards. Cognitive Psychology, 31, 1-40.
Gobet, F., & Simon, H. A. (1996b). Recall of random and distorted positions.
Implications for the theory of expertise. Memory & Cognition, 24, 493-503.
Gobet, F. & Simon, H. A. (1998). Pattern recognition makes search possible:
Comments on Holding (1992). Psychological Research, 61, 204-208
Gobet, F. & Simon, H. A. (2000). Five seconds or sixty? Presentation time in expert
memory. Cognitive Science, 24, 651-682.
Gobet, F. & Wood, D. J. (1999). Expertise, models of learning and computer-based
tutoring. Computers and Education, 33, 189-207.
Hall, J. C. (2002). Imagery practice and the development of surgical skills. American
Journal of Surgery, 184, 465-470.
Holding, D. H. (1985). The psychology of chess skill. Hillsdale, NJ: Erlbaum.
Jarvis, M. (1999). Sport psychology. London: Routledge.

Kant, I. (1781). Kritik der Reinen Vernunft (The critique of pure reason). Stuttgart:
Reclam.
Kosslyn, S. M. (1994). Image and brain: The resolution of the imagery debate.
Cambridge, MA: MIT Press.
Laeng, B., & Teodorescu, D. S. (2002). Eye scanpaths during visual imagery reenact
those of perception of the same visual scene. Cognitive Science, 26, 207-231.
Lane, P. C. R., Gobet, F., & Cheng, P. C. H. (2000). Learning-based constraints on
schemata. Proceedings of the Twenty Second Annual Meeting of the Cognitive
Science Society (pp. 776-781). Mahwah, NJ: Erlbaum.
Lane, P. C. R., Cheng, P. C. H., & Gobet, F. (2001). Learning perceptual chunks for
problem decomposition. Proceedings of the 23rd Meeting of the Cognitive Science
Society (pp. 528-533). Mahwah, NJ: Erlbaum.
Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes) worth 10,000
words. Cognitive Science, 11, 65-99.
Leibniz, G. (1704). New essays on human understanding. Cambridge: Cambridge
University Press.
Miller, A. I. (1986). Imagery in scientific thought. Cambridge, MA: The MIT Press.
Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ:
Prentice-Hall.
 34

Paige, J. M., & Simon, H. A. (1966). Cognitive processes in solving algebra word
problems. In B. Kleinmuntz (Ed.), Problem solving: Research, method and theory.
New York: Krieger.
Petre, M., & Blackwell, A. F. (1999) Mental imagery in program design and visual
programming. Int. J. Human-Computer Studies, 51, 7-30.
Reed, C. L. (2002). Chronometric comparisons of imagery to action: Visualizing vs.
physically performing springboard dives. Memory & Cognition, 30, 1169-1178.
Saariluoma, P. (1990). Apperception and restructuring in chess players’ problem
solving. In K. J. Gilhooly, M. T. G. Keane, R. H. Logie & G. Erdos (Eds.), Lines
of thinking (Vol. 2). New York: John Wiley.
Saariluoma, P. (1991). Aspects of skilled imagery in blindfold chess. Acta
Psychologica, 77, 65-89.
Saariluoma, P. (1994). Location coding in chess. The Quarterly Journal of
Experimental Psychology, 47A, 607-630.
Saariluoma, P. (1995). Chess players' thinking: A cognitive psychological approach.
London: Routlege.
Saariluoma, P. & Kalakoski, V. (1997). Skilled imagery and long-term working
memory. American Journal of Psychology, 110, 177-201.
Saariluoma, P. & Kalakoski, V. (1998). Apperception and imagery in blindfold
chess. Memory, 6, 67-90.
Simon, H. A., & Barenfeld, M. (1969). Information processing analysis of perceptual
processes in problem solving. Psychological Review, 7, 473-483.
Tabachneck-Schijf, H.J.M., Leonardo, A.M., & Simon, H.A. (1997). CaMeRa: A
computational model of multiple representations. Cognitive Science, 21, 305-350.
Tulving, E., Markowitsch, H., Kapur, S., Habib, R., & Houle, S. (1994). Novelty
encoding networks in the human brain: Positron emission tomography data.
NeuroReport, 5, 2525-2528.
Verstijnen, I. M., van Leeuwen, C., Goldschmidt, G., Hamel, R., & Hennessey, J. M.
(1998). Creative discovery in imagery and perception: Combining is relatively
easy, restructuring takes a sketch. Acta Psychologica, 99, 177-200.
Zeki, S. (1999). Inner vision: An exploration of art and the brain. Oxford: Oxford
University Press.
 35

Figure Captions

Figure 1. The template theory proposes that positions that recur often in a player’s

practice (such as the position after 1.e2-e4, diagram on the left, and the position after

1.e2-e4 c7-c5, diagram on the right) lead to the creation of templates. Templates may

be linked in LTM, for example by the move or sequence of moves that lead from one

to another (in our example, the move 1… c7-c5).

Figure 2. Experiment 1: percentage correct as a function of the experimental

condition (empty board, initial condition, and initial condition in the middle) and of

the number of ply (top, masters; bottom, Class A players).

Figure 3. Experiment 2: percentage correct as a function of the experimental

condition (empty board, semi-static, and move-by-move) and of the number of ply

(top, masters; bottom, Experts).

Figure 4. Experiment 2: Detail of errors of omission and commission.

 36

Figure 1
 37

Masters
100
80 Empty board

60 Initial position
40
Initial position in
20 the middle
Percentage
0 correct
10 30 50
Ply

Class A players
100
80 Empty board
60 Initial position
40
Initial position in
20 the middle
Percentage correct
0
10 30 50
Ply

Figure 2
 38

Masters
100
80
60 Empty board
Semi-static
40 Move-by-move
20
Percentage correct
0
30 50
Ply

Experts
100

80

60 Empty board
Semi-static
40 Move-by-move

20
Percentage correct
0
30 50
Ply

Figure 3
 39

25
Omission
Same
20
Inference
Interference
15 Pair

10

Percentage of Errors
5

0
Empty Board Semi-static Move-by-move

Figure 4