DeFT: A conceptual framework for considering learning with

multiple representations
Shaaron Ainsworth*

Reviewed by:
JANICE G. PAGTALUNAN
MAFIL-1
ABSTRACT
Multiple (external) representations can provide unique benefits when people are learning complex
new ideas. Unfortunately, many studies have shown this promise is not always achieved. The DeFT
(Design, Functions, Tasks) framework for learning with multiple representations integrates research on
learning, the cognitive science of representation and constructivist theories of education. It proposes that
the effectiveness of multiple representations can best be understood by considering three fundamental
aspects of learning: the design parameters that are unique to learning with multiple representations; the
functions that multiple representations serve in supporting learning and the cognitive tasks that must be
undertaken by a learner interacting with multiple representations. The utility of this framework is
proposed to be in identifying a broad range of factors that influence learning, reconciling inconsistent
experimental findings, revealing under-explored areas of multi-representational research and pointing
forward to potential design heuristics for learning with multiple representations.

Keywords: Multiple representations; Diagrams; Pictures; Multimedia

1. INTRODUCTION
Research on learning with representations has shown that when learners can interact with an
appropriate representation their performance is enhanced. Recently, attention has been focused on
learning with more than one representation, seemingly predicated on the notion ‘that two representations
are better than one’. Yet, as research on learning with multiple external representations (MERs) has
matured, it is increasingly recognized that the issue is not whether MERs are effective but rather concerns
the circumstances that influence the effectiveness of MERs (see Goldman, 2003). The most common
approach to considering the effectiveness of representations emphasizes the sensory channel and/or the
modality of the representations (i.e. either auditory/visual, or textual/pictorial). Two theories that are
particularly associated with this approach are the Cognitive Theory of Multimedia Learning (e.g.,
Mayer,1997) and Cognitive Load theory (e.g., Sweller, van Merrienboer, & Paas, 1998). They share a
focus on the nature of working memory (and its relation to long term memory) with its multiple,
modality-specific limited capacity subsystems. Presenting information in multiple modalities is
advantageous to learners who actively process such information. Schnotz (2002; Schnotz & Bannert,
2003) focuses not on pictures and text per se, but on depictive (iconic) and descriptive (symbolic)
representations. In this approach, mapping happens at the level of mental model construction and what
results is not an integrated representation but complementary representations that can communicate with
one another.
The purpose of this paper is to present an alternative approach addressing different aspects of
learning with representations. Instead of focusing on the form of the representational system, it suggests
that there are a number of additional design factors that should be considered. Given its wider scope it is
premature to advance design principles. So instead of proposing predictive guidance, it aims to suggest a
complementary set of factors that should guide research into the design of effective multi-representational
software. Thus, this paper serves as a review of research on MERs, an argument about the importance of
acknowledging a wide range of factors that influence learning with MERs and some proposed
applications of this approach.
 The DeFT (Design, Functions, Tasks) framework suggests that many dimensions combine to
influence whether someone will be able to benefit from learning with a particular combination of
representations. The dimensions considered in DeFT are the design parameters that are unique to learning
with more than one representation, the different pedagogical functions that MERs can play, and the
cognitive tasks that must be undertaken by a learner when interacting with MERs. This framework has
been developed by reviewing a broad range of current research from a variety of perspectives (e.g.,
cognitive psychology/science, education, artificial intelligence, and curriculum studies), addressing
methodologies such as case studies, experiments and computational modelling, and from empirical work
conducted in domains such as mathematics, physics, biology, and alchemy.
Understanding the role played by MERs first requires understanding how external representations
influence learning. Consequently, this paper begins by briefly describing the design factors that DeFT
addresses before turning to the advantages of employing the right representation (the functions aspect of
the framework) and cognitive tasks associated with learning with a single external representation (tasks).
Then these same issues are reconsidered addressing learning with multiple representations. The final
section reviews how DeFT might be applied to increase understanding of the impact of different designs
on learning with MERs.

2. DESIGN PARAMETERS IN DeFT

There are a number of ways to design multi-representational systems that influence the processes
and outcomes of learning. Systems differ in their content, in the target users of the system and in the
teaching strategies they employ. Often there are specific reasons to use a particular representation. An
external representation consists of (1) the represented world, (2) the representing world, (3) what aspects
of the represented world are being represented, (4) what aspects of the representing world are doing the
modelling and (5) the correspondence between the two worlds (Palmer, 1977). So when considering the
effectiveness of a representation both the information provided in the representation (represented world)
and the way it is presented (representing world) must be considered. Consequently, designing effective
representations is substantial endeavor in its own right. However, there are a set of design dimensions that
uniquely apply to multi-representational systems and it is these that are reviewed here: (a) number of
representations; (b) the way that information is distributed; (c) form of the representational system; (d)
sequence of representations; and (e) support for translation between representations.
Number: Multi-representational systems employ at least two representations, but commonly many
more are available, either simultaneously or at some point during a learner’s interaction with a system.
Information: Multi-representational systems allow flexibility in the way that information is
distributed over representations, impacting both upon the complexity of each representation and the
redundancy of information between representations. At one extreme, each representation can convey
completely different content (refer to different represented worlds). In this case, there is no redundancy
across representations. Distributing information in this way simplifies each representation but at the cost
of requiring additional representations, which learners may then be required to integrate. Systems can also
be partially redundant, so that some of the information is constant across (some of) the representations.
Finally, each representation can express the same information and so the only difference between
representations is in their computational properties (representing worlds).
Form: A typical multi-media system can display pictures, text, animations, sound, equations, and
graphs e a key question is whether it should. Much research has focussed on heterogeneous systems e
ones that combine text and graphics (Mayer, 1997; Schnotz, 2002) or multi-sensory systems such as
written text or pictures presented with spoken narration (Kalyuga, 2000). However, form can refer to
many different aspects of representations such as dimensionality, abstraction or dynamism. Consequently,
much is still unknown about how the form of a representational system influences learning.
Unfortunately, it is not sufficient to consider each type of representation in isolation e representations
interact with one another in a form of ‘‘representational chemistry’’. As a result, there is a potentially vast
space to explore.
 Sequence: If not all representations are drawn upon simultaneously, a number of further issues
arise. The first issue is the sequence in which the representations should be presented or constructed and
even if this is known, the learner or the system still needs decide at what point to add a new representation
or switch between the representations.
Translation: Computerized environments have a wide variety of ways to indicate the relation
between representations. Two dimensions have received some attention. Firstly, how active a role the
environment plays in supporting learners. Secondly, whether support is provided at the syntactic level or
the semantic level (also called surface and deep levels, or representation and domain levels (Seufert &
Bru¨nken, 2004).
In the next sections, the pedagogical functions and cognitive tasks associated with learning with
one representation and then with multiple representations are considered. Subsequently, these dimensions
are reconsidered to examine if considering learning in this way helps designers with these complex
decisions.

3. LEARNING WITH AN EXTERNAL REPRESENTATION

3.1. The functions of an appropriate representation

There is abundant evidence showing the advantages that external representations play in
supporting learning. Much research has shown that matching the type of representation to the learning
demands of the situation can significantly improve performance and understanding. Scaife and Rogers
(1996) proposed that external representations differ in their advantages for learning by varying the extent
to which they support computational offloading, re-representation or graphical constraining.
Consequently, combinations of representations can play a number of functions in supporting learning
(discussed in detail in Section 5).
Computational offloading is the extent to which different external representations reduce the
amount of cognitive effort required to solve equivalent problems. Larkin and Simon (1987) argue that
representations that are informationally equivalent still differ in their computational properties. For
example, diagrams can exploit perceptual processes by grouping together relevant information so making
search and recognition easier.
Re-representation refers to the way that external representations that have the same abstract
structure differentially influence problem solving. Zhang and Norman (1994) showed that problem
solving with isomorphic versions of the Towers of Hanoi was enhanced when representations
externalized more information. By utilizing external perceptual processes rather than cognitive
operations, graphical representations can often be more effective.
Graphical constraining describes the limits on the range of inferences that can be made about the
represented concept. Stenning and Oberlander (1995) argue that text permits expression of ambiguity in a
way that graphics cannot easily accommodate. It is this lack of expressiveness that makes diagrams more
effective for solving determinate problems. Schnotz (2002) makes a similar point when emphasizing the
distinction between descriptive (symbolic) and depictive (iconic) representations. Thus, depictive
representations are most useful to provide concrete information and are often efficient as specific
information can just be read off. Alternatively, descriptive representations can more easily express
abstract information as well as more general negations and disjunctions.

3.2. Cognitive tasks involved in learning with an external representation

Unfortunately, the benefits of an appropriate representation do not come for free. Learners are
faced with complex learning tasks when they are first presented with a novel representation. They must
understand how it encodes information and how it relates to the domain it represents. In addition, learners
may need to select an appropriate representation or to construct one for themselves, which can provide
advantages but also new cognitive tasks. The following sections describe these cognitive tasks and the
problems learners can face in mastering them. It should be noted that although these cognitive tasks are
presented in sequence, it is not meant to imply that learners would approach the task of understanding a
new representation in this same order.

Learners should understand the form of representation

Learners must know how a representation encodes and presents information (the ‘format’). In the
case of a graph, the format would be attributes such as lines, labels, and axes. They must also learn what
the ‘operators’ are for a given representation. For a graph, operators to be learnt include how to find the
gradients of lines, maxima and minima, and intercepts.
A number of studies have shown how complex this is (e.g., Friel, Curcio, & Bright, 2001).
Children have difficulty in applying and understanding the format and operators of graphs. They may
have trouble with reading and plotting points, interpret intervals as points, or confuse gradients with
maxima and minima. Petre and Green (1993) describe some similar effects with adults using a visual
interface. Novices lacked proficiency in secondary notation (i.e., perceptual cues that are not described by
the formal semantics of a representation) and found navigation of graphical representations difficult as
they don’t have the required reading and search strategies.
Additionally, the operators of one representation are often used inappropriately to interpret a
different representation e for example, when graphs are interpreted ironically, learners use the operators
for pictures (e.g., Leinhardt, Zaslavsky, & Stein, 1990). When learners are given a velocity time graph of
a cyclist travelling over a hill, they should select a U-shaped graph, yet many show a preference for
graphs with a hill shaped curve. Elby (2000) proposes that in many of these cases learners tend to rely on
an intuitive knowledge element, what-you-see-is-what-you-get, and that this is cued by the most
compelling visual attribute of a representation (e.g., straight lines mean constancy, hill shape means hill).
Learning to interpret a representation can involve learning to ignore this intuition.

Learners should understand the relation between the representation and the domain
Interpretation of representations is an inherently contextualized activity (Roth & Bowen, 2001) as
learners must also come to understand the relation between the representation and the domain that it
represents. This task will be particularly difficult for learning, as opposed to problem solving or
professional practice, as this understanding must be forged upon incomplete domain knowledge. Learners
need to determine which operators to apply to a representation to retrieve the relevant domain
information. For example, when attempting to read the velocity of an object from a distance time graph,
children often examine the height of line, rather than its gradient (Leinhardt et al., 1990). These problems
do not only arise with abstract representations. Boulton-Lewis and Halford (1990) point out that even
concrete representation such as Dienes blocks, and fingers still need to be mapped to domain knowledge.
Processing loads may be too high for children to obtain the anticipated benefits of such apparently simple
representations.

Learners may need to understand how to select an appropriate representation

In some situations, learners select a representation that they find most appropriate, and so they
may have to consider factors such as the representation and task characteristics as well as individual
preferences. There is evidence that learners can select effective representations. Zacks and Tversky
(1999) found that people were successful at choosing bar graphs to represent discrete comparisons
between data points and line graphs to depict trends. Novick, Hurley, and Francis (1999) found that
students were able to choose which of hierarchical, matrix or network representations was most
appropriate to represent the structure of a story problem. diSessa (2004) argues that students often have a
deep meta-representational competence which includes their abilities to judge the value of representations
along such dimensions as epistemic fidelity, compactness, parsimony, systematicity and conventionality.
However, there may well be individual differences in insight into the effectiveness of representations
(Roberts, Gilmore, & Wood, 1997). Selecting appropriate representations will be more difficult for
novices than experts as they can lack understanding of the deep nature of the tasks they are trying to solve
(Chi, Feltovich, & Glaser, 1981). Indeed one characteristic of expertise is the knowledge of what
representations are appropriate for what tasks (Kozma & Russell, 1997). One key unsolved issue is how
explicitly these skills should be taught; with some researchers arguing that teaching is crucial
(McKendree, Small, Stenning, & Conlon, 2002).

Learners may need to understand how to construct an appropriate representation

In many situations learners may construct or even invent a representation rather than interpret a
presented representation. Indeed, diSessa (2004) argues that learners are often strikingly good at
designing their own representations. Furthermore, even if learners construct their representations
inaccurately, they sometimes still draw the correct inference even so (Cox & Brna, 1995). There is
evidence that creating representations can lead to a better understanding of the situation. Grossen and
Carnine (1990) found that children learned to solve logic problems more effectively if they drew their
responses to problems rather than selected a pre-drawn diagram. This may in part be due to the support
they were provided with during the construction process. Van Meter (2001) found that drawing was most
helpful in learning from science texts when students were prompted with guidance questions whilst
creating diagrams. However, there is unlikely to be a simple relationship between interpretation and
construction of a representation. Cox and Brna gave students reasoning problems which could be solved
with Euler diagrams. They found six students made errors in constructing representations but not in
interpreting them, and four students made errors in interpreting diagrams but not in constructing them.

LEARNING WITH MULTIPLE REPRESENTATIONS

Early research on learning with MERs concentrated on the ways that presenting pictures
alongside text could improve readers’ memory for text comprehension (e.g., Levin, Anglin, & Carney,
1987). In the last two decades, the explosive increase in multi-media learning environments have widened
the debate to include combinations of representations such as diagrams, equations, tables, text, graphs,
animations, sound, video, and dynamic simulations. Furthermore, a number of influential educational
theories discuss the importance of MERs. For example, Dienes (1973) argues that perceptual variability
(the same concepts represented in varying ways) provides learners with the opportunity to build
abstractions about mathematical concepts. In cognitive flexibility theory, the ability to construct and
switch between multiple perspectives of a domain is fundamental to successful learning (Spiro & Jehng,
1990). Research on analogical reasoning shows how comparison processes help people make new
inferences (Gentner & Markman, 1997). However, research on the benefits of providing learners with
more than one representation has produced mixed results. For example, a number of studies have found
that learners benefit from MERs (e.g., Cox & Brna, 1995; Mayer & Sims, 1994; Tabachneck, Koedinger,
& Nathan, 1994), but unfortunately, just as many studies fail to find these benefits (e.g., Chandler &
Sweller, 1992; Van Someren, Reimann, Boshuizen, & de Jong, 1998).
In the next sections, the benefits that MERs can bring to learning situations are reviewed. The
functions aspect of the DeFT framework proposes that there are many advantages of MERs and that these
should be clearly identified as they often have different design implications. These functions of MERs are
only possible if learners master the cognitive tasks associated with their use. Learning to use MERs
requires learners to understand each individual representation. This is a complex process in its own right
(see Section 3.2). But, in addition, when interacting with MERs, learners must often understand the
relationship between representations and many studies have shown that learners tend to treat
representations in isolation and find it difficult to integrate information from more than one source.
FUNCTIONS

Complementary functions
When MERs complement each other, they do so because they differ either in the processes each
supports or in the information each contains. By combining representations that differ in these ways, it is
hoped that learners will benefit from the advantages of each of the individual representations.

Complementary processes
Representations that theoretically contain the same information still differ in their advantages for
learning in certain situations due to the extent to which they support computational offloading, re-
representation, or graphical constraining. Consequently, by providing MERs complementary processes
can be supported. This can be advantageous for a number of reasons.
Individual differences: Theorists following a learning styles approach argue that if learners are
presented with a choice of representations, they can choose to work with the representation that best suits
their needs (e.g., Dunn & Dunn, 1993). There is limited evidence that this can improve learning. For
example, Plass, Chun, Mayer, and Leutner (1998) found that students comprehended a story in a second
language better when they had the opportunity to receive their preferred mode of annotation
(visual/verbal/both). However, the assumption that ‘visualizers’ will necessarily do better with visual
representations is not always warranted (Coffield, Moseley, Hall, & Ecclestone, 2004; Klein, 2003;
Roberts et al., 1997). Alternative accounts of the individual differences effect emphasize differing
expertise with the subject studied or with the representations. For example, Stenning, Cox, and
Oberlander (1995) found
that high performing students benefited from graphical instruction of logic but lower performing students
performed better when given traditional textual instructions. ChanLin (2001) found novices learning
physics benefited from the use of still graphics rather than text but found no differences between formats
for experienced students.
Task: Performance is most likely to be facilitated when the structure of information required by
the problem matches the form provided by the representational notation (Gilmore & Green, 1984).
Learners given MERs can benefit from choosing the best representation for the current task. For example,
Tapiero (2001) found that when subjects were given textual descriptions of a city, they performed spatial
judgement tasks more accurately than those given a map of the city. However, when presented with a
transfer task of a map of a modified city, the reverse was true e map subjects performed better than text
subjects.
Strategy: Different forms of representation can encourage learners to use more or less effective
strategies (Ainsworth & Loizou, 2003). MERs encourage learners to try more than one strategy to solve a
problem. Tabachneck et al. (1994) examined the representations that learners created to solve algebra
word problems and found that each representation was associated with a different strategy. The use of
MERs and hence multiple strategies was about twice as effective as any strategy used alone. As each
strategy had inherent weaknesses, switching between strategies made problem solving more successful by
compensating for this.

Complementary information
Multiple representations are used to provide complementary information when each
representation in the system contains (some) different information. This may occur if a single
representation would be very complicated if it presented all the information or if the information is on
radically different scales.

Constraining functions
A second advantage of using MERs is that certain combinations of representations can help
learning when one representation constrains interpretation of a second representation. This can be
achieved in two ways. Firstly, learners’ familiarity with one representation can constrain interpretation of
a less familiar one. For example, concrete animations are often employed in simulations alongside
complex and unfamiliar representations such as graphs. Secondly, these constraints can be achieved by
taking advantages of inherent properties of representations. Graphical representations are generally more
specific than textual representations (Stenning & Oberlander, 1995). The phrase ‘the cat is by the dog’, is
ambiguous about which side of the dog the cat is sitting, but in a picture, the cat must be either on the left
or the right of the dog. So, when these two representations are presented together, interpretation of the
first (ambiguous) representation may be constrained by the second (specific) representation. Depictions
can perform this same role for descriptions (Schnotz, 2002).

Constructing functions
Multiple representations support the construction of deeper understanding when learners integrate
information from MERs to achieve insight that would be difficult to achieve with only a single
representation. Furthermore, insight achieved in this way increases the likelihood that it will be
transferred to new situations (Bransford &Schwartz, 1999).
Abstraction is the process by which learners create mental entities that serve as the basis for new
procedures and concepts at a higher level of organization. Learners can construct references across MERs
that then expose the underlying structure of the domain represented. Schwartz (1995) showed that the
representations that emerge with collaborating peers are more abstract than those created by individuals e
abstracted representation may emerge as a consequence of requiring a single representation that could
bridge both individuals’ representations.
Extension can be considered as a way of extending knowledge that a learner has from a known to
an unknown to representation, but without fundamentally reorganizing the nature of that knowledge. For
example, learners may know how to interpret a velocity time graph in order to determine whether a body
is accelerating. They can subsequently extend their knowledge to such representations as tables or
acceleration time graphs.
Relational understanding is the process by which two representations are associated again without
reorganization of knowledge. The goal of teaching relation between representations can sometimes be an
end in itself. For example, much mathematical education concerns how to construct a graph given an
equation. On other occasions, it may serve as the basis for abstraction.
It should also be noted that the functions that representations serve often depend upon learners’
knowledge and goals not system designer’s intent. For example, one learner may be familiar with tables
and extend his or her knowledge to graphs (extension), another may already be familiar with both but not
have considered their relationship (relation). The differences between these functions of MERs are subtle
and all may be present at some stage in the life cycle of encouraging deeper understanding with a multi-
representational environment.

Functions summary
MERs can play many advantageous roles in learning complex material and these different roles
fall into three distinct categories. However, the picture is complicated by the need to acknowledge that
MERs can support more than one of these roles simultaneously. For example, Ainsworth, Wood, and
O’Malley (1998) found that a combination of table and place-value representations in a primary math’s
environment provided different information and supported complementary processes, constrained
interpretation in two alternative ways and might also have supported abstraction.
COGNITIVE TASK: LEARNERS MAY NEED TO UNDERSTAND HOW TO RELATE
REPRESENTATIONS
Many of the proposed benefits of MERs result from the integration and coordination of more than
one source of information and a characteristic of expertise is the ability to integrate different
representational formats. Unfortunately, a very large number of studies have observed that learners find
translating between representations difficult (e.g., Anzai, 1991; Schoenfeld, Smith, & Arcavi, 1993).
Learners can fail to notice regularities and discrepancies between representations (Dufour-Janvier,
Bednarz, & Belanger, 1987) and in the worst cases this can even completely inhibit learning. Ainsworth,
Bibby, and Wood (2002) compared children learning estimation with two representations, either
mathematical, pictorial or a mixed system of one pictorial and mathematical representation. They showed
that whilst pictorial and mathematical representations helped learning, the combination of pictorial with
mathematical representations inhibited learning of the task. It was possible to isolate the problem as
resulting from relating representations. Each representation in the mixed system was present in either
mathematical or pictorial systems where it was used successfully, so it is known that learners could
understand the form of representations and their relation to domain, just not how they relate to each other.
Teaching learners to coordinate MERs has also been found to be a far from trivial activity.
Yerushalmy (1991) provided students with an intensive three-month course with multi-representational
software that taught functions. In total, only 12% of students gave answers that involved both visual and
numerical considerations and those who used two representations were just as error prone as those who
used a single representation. Resnick and Omanson (1987) gave children instructions about the
correspondence between Dienes blocks and written numerals to help them master the symbolic
subtraction procedures. They were disappointed by how little children referred to the blocks and found,
for the most part, they were not helpful.
Consequently, it is important to understand the factors that influence the difficulty of relating
representations. Here the characteristics of the representations and characteristics of the learner are
considered.

Representation characteristics
A reasonable heuristic for considering how the representational system will impact upon how
learners integrate information from multiple sources is to suggest that the more the formats of the
representations and the operators that act upon them differ, the more difficult it will be for learners to
translate between them. There are a number of dimensions that are likely to maximize these differences
between representations.
The sensory channel of the representation: A common combination of representations is one that
combines an auditory with a visual representation. A number of researchers working in the cognitive
tradition (e.g., Kalyuga, 2000; Mayer, 1997) propose that referential connections between these types of
representations are facilitated because combining both auditory and visual stimuli take maximum
advantage of short-term memory and so facilitate translation between representations. However,
Gyselinck, Ehrlich, Cornoldi, de Beni, and Dubois (2000) show that visual spatial working memory can
still be heavily loaded in such situations.
The modality of the representations: A heterogeneous system is one that contains both a text
based representation and a graphical/diagrammatic representation. These representations are known to
have very different computational proprieties (Larkin & Simon, 1987) and some researchers also consider
them to be processed separately in the brain (e.g., Mayer, 1997; Tabachneck-Schijf, Leonardo, & Simon,
1997). Learners may find it difficult to see the relationship
between such different forms of representation.
The level of abstraction: Peirce (1906) distinguishes between a symbol, which has an arbitrary
structure (a description in Schnotz’s terminology), and an icon (depiction) that does not. Bruner (1966)
adds an extra mode, enactive, to represent events through motor responses. Purchase (1998) further adapts
Bruner’s scheme by dividing the iconic category in two: concrete iconic, which has a direct perceptual
relationship to the object, and abstract iconic, which has a related but non-direct relation. Blackwell and
Engelhardt (1998) identify eight schemes that consider level of abstraction. There seems little agreement
about the granularity of the dimension, but its importance is widely recognized.
The specificity of representations: Specificity determines the extent to which a representation
permits expression of abstraction (Stenning & Oberlander, 1995). Learners will interpret and act upon
representations of different levels of specificity in unlike ways, so they may find integrating information
across representations that differ in specificity more difficult.
The type of representation (e.g., histogram, equation, table, line graph, narrative text, picture):
There are many schemes proposed for categorizing representations into different types (e.g., Cox & Brna,
1995; Lohse, Biolsi,Walker, & Rueler, 1994). For example, Lohse et al. identified 11 major clusters:
graphs, numerical and graphical tables, time charts, cartograms, icons, pictures, networks, structure
diagrams, process diagrams and map clusters. These taxonomies have been created by a variety of
methods (e.g., intuition, analysis of domain properties and card sorts) and although there is some overlap
between the taxonomies, no one classification is universally accepted. They differ in the domains
addressed, the granularity with which representations were described and the task for which they were
created.
Integrated presentations of representations: When presenting textual and graphical
representations learners find it easy to understand physically integrated material rather than separately
presented material (Chandler & Sweller,1992).
Whether representations are static or dynamic: Dynamic representations such as animations,
dynamic graphs and spoken text require different operators to interpret them and have different formats
than their static equivalents and consequently learners draw different inferences from pictures and
animations (Jones, 1998; Lowe, 2003). Thus, representational systems that combine static and dynamic
representations may be particularly complex. Furthermore, different types of dynamic representations also
have different format and operators (Ainsworth & Van Labeke, 2004).
Dimensionality: With the mounting availability of virtual reality and other visualization tools,
learners are increasingly being placed in situations where they must integrate information from both two-
dimensional and three-dimensional representations. There is evidence that learners can fail to build such
links easily (Moher, Johnson, Ohlsson, & Gillingham, 1999).

Individual characteristics
How well individuals cope with the relating different representations is likely to depend upon a
number of learner characteristics. Probable candidates include familiarity with the representations and
domain, age and cognitive style.
Representational familiarity: If learners are already familiar with the representations, then they
should understand (to some degree) the format and operators of representation and the relation between
the representations and the domain. The lower the learning demands are on other parts of the task, the
more resource for translating between representations should be available. Furthermore, if learners are
less likely to misinterpret the representations, this should enhance the possibility of recognizing the
similarity between them.
Domain familiarity: Generally, novices tend to characterize problem representations by their
surface features, not their deep structure (Chi et al., 1981). Therefore, as learners generally lack expertise
either in the domain or in the representations they are using, they are likely to be hampered in recognizing
deep structural relations between representations due to their surface dissimilarity. This lack of domain
knowledge interferes with their ability to transfer knowledge across representations appropriately (Stern,
Aprea, & Ebner, 2003).
Age: A learner’s age may also affect his or her abilities to translate between representations.
Often children’s performance can be seen as characteristic of novices in a domain. Nevertheless, there are
likely to be developmental factors that affect integration of MERs. Moore and Scevak (1997) found
developmental differences in children’s use of text and accompanying visual aids with explicit linking of
text and visual aid by older students that was not as evident in the younger students. A number of
researchers have proposed that information-processing capacity such as short-term memory span or
processing speed increases with age (e.g., Case, 1985). For example, Halford (1993) defines
dimensionality as the number of independent items of information that must be processed in parallel. He
proposed that it is not until children reach 11 years of age that they can process four-dimensional
structures. If MERs exceed this capacity then children would need to re-represent the problem, for
example, by chunking. This suggests that younger children would require considerable experience with
the representations in order to relate them successfully.
Individual differences: There has been much research relating both personality and cognitive
factors to learning with external representations (see Section 5.1.1). There is less research into aptitude
treatment interactions and MERs. An exception is that of Oberlander, Cox, Monaghan, Stenning, and
Tobin (1996). They suggest that a distinguishing characteristic of people who were classified as
diagrammatic reasoners was their ability to translate information across representations more
successfully.

Representation and individual characteristics

These two levels come together in the way that individual factors and representation factors
influence the strategies that learners use. Using MERs can encourage learners to try different strategies
(e.g., Tabachneck et al., 1994;Watson, Campbell, & Collis, 1993). This is often advantageous as by
switching between representations learners can compensate for weaknesses in their strategy. However, if
learners are attempting to relate different representations, then this may provide a source of difficulty.
Ainsworth et al. (2002) hypothesized that one of the reasons why learners did not integrate information
across pictorial and mathematical representations is that the pictorial representation encouraged the
development of a perceptual strategy and the mathematical one encouraged the generation of a rule based
upon symbol manipulation.

APPLYING DeFT TO EXPLORE DIFFERENT DESIGNS

In Section 2, the (implicit or explicit) design parameters of multi-representational software were
identified. In this section, I return to these parameters to explore if considering the functions and
cognitive tasks of learning with MERs can help someone make these design decisions. It is fair to say that
at this stage DeFT raises far more questions than it answers. However, it may still be useful if it helps to
identify those aspects of multi-representational design that are under-explored. DeFT may also encourage
researchers to describe in more detail the design parameters of their systems and the pedagogical
functions they intend to their systems to play. Although, research might initially be considered as
addressing one question (e.g., simultaneous versus sequential presentation), other researchers interested in
alternative questions (e.g., redundancy) are often unable to draw conclusions about their interests if this
aspect of the system is not described.

Number
Given the research reviewed on the cognitive tasks associated with adding representations to a
system, it seems wise to use the minimum number of representations consistent with the pedagogical
function of the system. In many cases it may not be appropriate to use MERs at all, since one
representation may be sufficient and will minimize the split attention affect. However, there are many
circumstances where MERs are appropriate. The decision about the number of representations often
depends upon the informational (Section 7.2) and computational (Section 7.3) properties of the desired
representational system.

Information
Information can be distributed in multiple ways over MERs which may simplify individual
representations and impact upon the redundancy of the representational system. Consequently, there may
be a way of distributing information that best supports learning (for a particular task and a particular type
of learner). One possibility is that it is easier to learn complex ideas when each part is represented
separately in a simpler representation. Alternatively, it may be easier to learn from complex
representation(s) as all the information is presented together.
Kalyuga, Chandler, and Sweller (1998) report a number of studies that showed that less
experienced learners benefit from redundant text but for those with more experience adding text
interfered with performance with a diagram. Learners who can benefit from a diagram in isolation do not
need text and so eliminating it reduces cognitive load. This suggests that redundancy should be reduced as
expertise grows. However, Ainsworth, Bibby, and Wood (1997) gave students two representations to
describe their performance on computational estimation tasks e these were either non-redundant where
each representation provides one dimension of information each or completely redundant with both
representations displaying two dimensions of information. They found that learners given nonredundant
representations understood aspects of estimation accuracy faster than those given fully redundant
representations. Examining the apparent contradiction between these experiments, it seems likely that the
conflicting results are due to different functions of the MERs. The text in Kalyuga et al.’s studies seems
to have been used by novices to constrain interpretation of an unfamiliar diagram, whereas in Ainsworth
et al.’s study, the representations were used to complement each other.
Furthermore, the impact of the way that information is distributed may be modified by the form
of the representation. Ainsworth and Peevers (2003) examined the interaction between the form (tables,
diagrams, or text) and number (four simple or one complex one) of representations. Participants were
provided with instructions about how to operate a complex device using these representations. Problem-
solving based on tables or diagrams was equally effective with one complex or four simple
representations. However, those participants given a single text spent much longer studying
representations than those who saw four texts but were also more likely to find the ideal solution to the
task.
It is apparent that whilst one of the advantages of using MERs is that information can be distributed to
simplify individual representations, little research has directly addressed that question. Most of the
experimentation holds informational equivalence across representations constant in order to explore
computational nonequivalence. Furthermore, describing the information in a representation is problematic
as information differs in how explicitly it is represented (Kirsh, 1991). One possibility is to base it on a
theoretical description assuming a perfect information processor with unlimited resources and knowledge.
However, people are not perfect information processors, and they differ in their background knowledge,
skills, and cognitive capacities. More research is needed to explore these issues.

Form
This aspect of designing for effective learning has received the most attention and there is
consistent evidence that differences in the form of representational systems strongly impact upon learning
processes and outcomes. However, the majority of research has concentrated on modality and sensory
aspects of representations. This focus has meant that other forms of representational system remained
significantly under-explored (Reimann, 2003) and one role that DeFT can play is to draw attention to that
fact. Furthermore, a number of studies have found contradictions in the apparent usefulness of specific
combinations of representations. For example, whether simultaneous presentation of written and spoken
text is beneficial (e.g., Kalyuga, 2000; Mayer & Sims, 1994). From a DeFT perspective, one key reason
for these differences is that MERs play different pedagogical functions. If MERs are used to support
different
computational properties and one of the representations is not needed by the learner, then simultaneous
presentation of both representations it is not likely to aid learning. However, this is not the case if the first
representation is needed to constrain understanding of the second representation or if both representations
are needed to encourage deeper understanding. One strong prediction of DeFT, which requires empirical
validation, is that different design principles will apply for different pedagogical functions (see Section
8).
Sequence
The approaches to deciding upon a sequence of representations can be placed on a continuum
ranging from domain-specific to domain-general. Some researchers start with an analysis of the properties
of the domain to be taught in order to identify any representational consequences. Only in the absence of
any particular constraints arising from this domain analysis are more general representational factors
considered. Alternatively, a representational perspective can be taken that favours a domain-general
approach.
An example of a domain-specific approach can be seen in MathsCar, a multi-representational
system to teach introductory calculus (Kaput, 1994). Kaput argues that when teaching calculus,
introducing integration before differentiation best supports understanding and so velocityetime graphs
should be introduced before position time graphs. At a mid-point on the continuum lies the approach of
Plo¨tzner (1995), who analyses one-dimensional motion in classical physics problems to argue that
qualitative knowledge should be taught before quantitative knowledge and consequently qualitative
representations should be introduced before quantitative ones. Evidence for this proposal is provided by a
cognitive model and by the performance of collaborating pairs taught with different sequences of
qualitative and quantitative representations (Plo¨tzner, Fehse, Kneser, & Spada, 1999).
At the other end of the continuum, a more domain-general approach can be seen in Kulhavy’s
model of text learning with organized spatial displays, which suggests that graphical representations
should precede text. Verdi, Johnson, Stock, Kulhavy, and Whitman (1997) showed that learners presented
with a visual display before related text recalled significantly more information than when presented with
text and then the visual display. Another domain-general approach is to introduce representations in such
a way as to increase their abstraction. For example, COPPERS (Ainsworth et al., 1998) presents coin
problems to children first as pictures but then through increasingly abstract representations such as mixed
text and pictures and then text only and finally as algebra. This approach can be seen in many systems,
but its validity has rarely been evaluated. It does seem reasonable to start by offering learners the least
complex available representations. This may be the most concrete/least expressive representation that the
increasing abstraction route suggests. However, in some situations, concrete and realistic representations
can actually be more complex for learners (Lowe, 2003).
The question of whether learners make strategic decisions about when to change a representation or
introduce a new one has been addressed by Cox and Brna (1995) who allow users to move at will
between their self-created representations. They found mixed success with some learners switching
effectively but others choosing new representations which did not help them move nearer the goal.
Another possibility is that learners should switch when they have exhausted all the information available
in the representation they are currently using. For example, Graphs and Tracks (Trowbridge, 1989)
suggest that users should switch from a velocityetime to a distanceetime graph in order to gain
information about the represented object’s starting position. Alternatively, the system may take
responsibility for determining when to change the representation. In this case, the task for the system is to
determine when users have learnt all they can about the domain with the given representations, but not
switch so soon (or so often) that the learning demands of the new representations overburden the user.
Unfortunately, as Resnick and Omanson (1987) observe, it is possible to introduce new representations
too late. In their study of children learning to subtract using the standard written symbols, Dienes blocks
were introduced to help children understand this task in a more conceptual way. The researchers were
disappointed by how little children referred to the blocks and suggest that once children had reached
automated performance with symbolic manipulation, it does not easily allow for application of principled
knowledge. This suggests that a new representation should be introduced before learners have achieved
automated performance with an existing representation. This raises the question of what aspects of
learners’ behavior would need to be captured by a system and interpreted in a student model to be able to
switch representations appropriately. It is apparent that there are many questions about sequencing
representations, but that few of them have comprehensive answers. Deciding on a specific order of
representations will almost always require a domain analysis, which could be supplemented with general
representational principles. However, decisions about when to switch representations have received little
attention and await systematic experimentation.

Translation
Computers can automatically support translation between representations and have variety of
ways to indicate the connection between representations. Learning environments differ in how actively
they support learning and whether this support is provided at the syntactic level or the semantic level.
The least active way that environments (computational or not) provide support for relating
representations is in the use of implicit cues. For example, the relation between representations is easier to
identify if they have consistent labels. Dufour-Janvier et al. (1987) suggest that children have a tendency
to recognize that two representations concern the same problem only when they contain the same
numbers. Other cues include using the same colors to represent the same objects over different
representations.
More active support is seen when learners can select part of a first representation and see how this
corresponds to a second representation. For example, in Bru¨nken, Plass, and Leutner (2003) learners can
click on a hyperlink and arrows point to equivalent part of an accompanying picture.
Sometimes, learners act on one representation and see the results of those actions in another. This
is commonly referred to as dyna-linking. For example, graphical calculators present dyna-linked algebraic
expressions and graphs. Dynamic linking of representations is assumed to reduce the cognitive load upon
the student e as the computer performs translation activities, students are freed to concentrate upon their
actions on representations and their consequences in other representations. Kaput (1992) argues that this
is particularly beneficial when the representations involved are expressing actions sequences rather than
just final outcomes as previous research has shown just how difficult this task is for learners. However,
direct empirical support for the benefits of dyna-linking is not easy to find. For example, van der Meij and
de Jong (2003) found no difference in learning between separate and dyna-linked representations.
Finally, some systems require learners to actively integrate representations and monitor students’
success in so doing. Bodemer, Ploetzner, Feuerlein, and Spada (2004) gave learners spatially separated
pictorial and symbolic representations of statistics concepts and asked them to drag the symbols and
dropping them within the pictures. Compared to split source or integrated representations, learners did
better when required to integrate representations.
Only a few systems can vary the amount of help for relating representations that they offer to
learners. However, this is probably the ideal. For example, there are reasons to hesitate about the
invariable dynamically linking of representations. If we aim to encourage users to understand the
mapping between representations, then we may be in danger of over-automating the process. This over-
automation may not encourage users to reflect upon the nature of the connection and could in turn lead
learners to fail to construct the required deep understanding. Alternatively, dyna-linking may encourage
learners to attempt to relate concepts that are beyond their level of understanding. Seufert (2003) provides
support for varying the amount of support according to a learner’s expertise. She found that high prior
knowledge learners did not benefit from help relating representations, as presumably they could make
these links for themselves. Low prior knowledge students also did not benefit because they became
overwhelmed. It was only learners with an intermediate level of prior knowledge who benefited from this
help.
The level at which help is provided has received less direct research. Seufert provided help at a
deeper level whereas most other research (particular with computers) have used more surface strategies.
Van Labeke and Ainsworth (2003) report a case study of three learners working with complex multi-
representational software for up to 8 h and found that they differed in their strategy for relating
representations. One learner without relevant background knowledge used dyna-linking to support his
surface level strategy for relating representations, whereas the two learners with more background
knowledge used dyna-linking less and attempted to relate representations using deeper structural features.
Consequently, it may be the case that both the degree of support and the level at which this help is
provided should vary depending on learners’ expertise.
DESIGNS HEURISTICS
One key purpose of DeFT is to help delimitate the complex demands faced by learners when
interacting with MERs. It adds to the substantial literature on cognitive load accounts of learning with
MERs (e.g., Kalyuga, Chandler, & Sweller, 1999; Kirschner, 2002; Mayer & Moreno, 2002) by
identifying factors that add to cognitive load in these situations. Thus, it can help explain the intermittent
success of learning with MERs by examining which cognitive tasks learners mastered and which they did
not. In addition, by emphasising the pedagogical functions of MERs and their implications for the
cognitive tasks associated with MERs, DeFT provides an alternative way to consider how multi-
representational systems might be designed to support learning. These proposals should be read as
heuristics for guiding experimentation not as cast-in-stone principles.
When MERs are used to support complementary functions, learners need to understand each
representation in isolation, how to select appropriate representations but need not understand the relation
between them. The main design consideration therefore becomes one of selecting appropriate
representations for the situation and the learners, rather than supporting learners in mastering the complex
task of relating representations. Consequently, systems could provide
dyna-linked representations and/or minimize co-presence of representations as learners often attempt to
translate between co-present representations even if they do not need to do so to achieve the task.
When MERs are used to constrain interpretation, it is imperative that the learner understands the
constraining representation. Thus, using concrete representations with simple format and operators is
ideal. But, in addition and in contrast to the first use of MERs, designers need also to ensure that learners
understand how the constraining representation relates to the constrained representation. Consequently, a
way must be found to signal the mapping between representations without over-burdening learners by
making translation complex. DeFT predicts that these representations should be co-present. Factors that
increase the perceived similarity between representations should be applied, as could dyna-linking where
appropriate.
The third function is when MERs are designed to allow learners to construct a deeper
understanding of a domain. This goal provides designers with hard choices. If users fail to translate across
representations, then abstraction and extension cannot occur. Learners find it difficult to translate over-
representations that are superficially dissimilar, but if made -too easy, for example, by providing
representations that do not provide sufficiently different views on a domain, then abstraction of
invariances does not occur. However, if the system performs all the translation activities for students, then
students are not afforded the opportunity to actively construct this knowledge for themselves. Approaches
such as those of Bodemer et al. (2004) may provide one solution to this issue. Although little research has
addressed the way that that information distribution influences translation, Ainsworth et al. (1997) found
tentative evidence that increasing the redundancy of information between (simple) representations
increased learners’ abilities to reconcile representations that differ in format. In addition, to maximize
opportunities for learners to build cognitive links over representations, then representations should be co-
present. Although, much is still left to uncover, researchers are beginning to specify the factors that
encourage or discourage deeper understanding with multiple representations.

CONCLUSION
This paper has illustrated the DeFT framework that describes some of the important aspects of
learning with MERs. It clarifies the pedagogical functions that MERs serve, the often-complex learning
demands that are associated with their use and in so doing aims to consider the ways that different designs
of multi-representational systems impact upon the process of learning. It is hoped that DeFT will prove to
be helpful to other researchers analyzing learning with MERs by highlighting areas of study that are
relatively under-investigated, providing an explanation for apparently opposing findings, offering a
common language for describing aspects of system design allowing generalizations across studies to be
more easily achieved and ultimately aiding in the development of design heuristics and principles for
learning with more than one representation.
References
Ainsworth, S. (1999). The functions of multiple representations. Computers and Education, 33(2e3),
131e152.
Ainsworth, S., Bibby, P., & Wood, D. (2002). Examining the effects of different multiple representational
systems in learning primary mathematics.
Journal of the Learning Sciences, 11(1), 25e61.
Ainsworth, S., Bibby, P.A., & Wood, D.J. (1997). Evaluating principles for multi-representational
learning environments. Paper presented at the
7th European conference for Research on Learning and Instruction, 1997, August, Athens.
Ainsworth, S. E., & Loizou, A. T. (2003). The effects of self-explaining when learning with text or
diagrams. Cognitive Science, 27(4), 669e681.
Ainsworth, S. E., & Peevers, G. J. (2003). The interaction between informational and computational
properties of external representations on
problem-solving and learning. In R. Altmann, & D. Kirsch (Eds.), Proceedings of 25th annual conference
of the Cognitive Science Society, 2003, August, Mahwah, New Jersey: LEA.
Ainsworth, S. E., & Van Labeke, N. (2004). Multiple forms of dynamic representation. Learning and
Instruction, 14(3), 241e255.
Ainsworth, S. E., Wood, D., & O’Malley, C. (1998). There is more than one way to solve a problem:
Evaluating a learning environment that supports the development of children’s multiplication skills.
Learning and Instruction, 8(2), 141e157.
Anzai, Y. (1991). Learning and use of representations for physics expertise. In K. Anders-Ericsson, & J.
Smith (Eds.), Towards a general theory of expertise: Prospects and limits. Cambridge: Cambridge
University Press.
Blackwell, A.F., & Engelhardt, Y. (1998). A taxonomy of diagram taxonomies. Paper presented at the
Thinking with diagrams, 1998, August, Aberystwyth.
Bodemer, D., Ploetzner, R., Feuerlein, I., & Spada, H. (2004). The active integration of information
during learning with dynamic and interactive visualisations. Learning and Instruction, 14(3), 325e341.
Boulton-Lewis, G.M., & Halford, G.S. (1990). An analysis of the value and limitation of mathematical
representations used by teachers and young children. Paper presented at the 14th international conference
for the Psychology of Mathematics Education, 1990, July.
Bransford, J. D., & Schwartz, D. L. (1999). Rethinking transfer: A simple proposal with multiple
implications. Review of Research in Education, 24, 61e100.
Bruner, J. (1966). Towards a theory of instruction. New York: W.W. Norwood & Company.
Bru¨nken, R., Plass, J., & Leutner, D. (2003). How instruction guides attention in multimedia learning. In
H. Niegemann, R. Bru¨nken, & D. Leutner (Eds.), Instructional design for multimedia learning.
Proceedings of the EARLI SIG 6 Biannual Workshop 2002, Erfurt. Case, R. (1985). Intellectual
development. Orlando, FL: Academic Press.
Chandler, P., & Sweller, J. (1992). The split-attention effect as a factor in the design of instruction.
British Journal of Educational Psychology, 62(2), 233e246.
ChanLin, L. (2001). Effects of formats and prior knowledge on learning in a computer-based lesson.
Journal of Computer Assisted Learning, 17(4), 409e419.
Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation in physics
problems by experts and novices. Cognitive Science, 5(2), 121e152.
Coffield, F., Moseley, D., Hall, E., & Ecclestone, K. (2004). Learning styles and pedagogy in post-16
learning: A systematic and critical review. Learning and Skills Research Centre. Retrieved May 8, 2004.
<http://www.lsda.org.uk/pubs/dbaseout/download.asp?code¼1543>.
Cox, R., & Brna, P. (1995). Supporting the use of external representations in problem solving: The need
for flexible learning environments. Journal of Artificial Intelligence in Education, 6(2/3), 239e302.
Dienes, Z. (1973). The six stages in the process of learning mathematics. Slough, UK: NFEReNelson.
Dufour-Janvier, B., Bednarz, N., & Belanger, M. (1987). Pedagogical considerations concerning the
problem of representation. In C. Janvier (Ed.), Problems of representation in the teaching and learning of
mathematics. Hillsdale, NJ: LEA. Dunn, R., & Dunn, K. (1993). Teaching secondary students through
their individual learning styles: Practical approaches for grades 7e12. Boston: Allyn & Bacon.
Elby, A. (2000). What students’ learning of representations tells us about constructivism. Journal of
Mathematical Behavior, 19(4), 481e502. Friel, S. N., Curcio, F. R., & Bright, G. W. (2001). Making
sense of graphs: Critical factors influencing comprehension and instructional implications.
Journal for Research in Mathematics Education, 32(2), 124e158.
Gentner, D., & Markman, A. B. (1997). Structure mapping in analogy and similarity. American
Psychologist, 52(1), 45e56. Gilmore, D. J., & Green, T. R. G. (1984). Comprehension and recall of
miniature programs. International Journal of Mane Machine Studies, 21(1), 31e48.
Goldman, S. R. (2003). Learning in complex domains: When and why do multiple representations help?
Learning and Instruction, 13(2),239e244. Grossen, B., & Carnine, D. (1990). Diagramming a logic
strategy e Effects on difficult problem types and transfer. Learning Disability Quarterly, 13(3), 168e182.
Gyselinck, V., Ehrlich, M. F., Cornoldi, C., de Beni, R., & Dubois, V. (2000). Visuospatial working
memory in learning from multimedia systems.
Journal of Computer Assisted Learning, 16(2), 166e176.
Halford, G. S. (1993). Children’s understanding: The development of mental models. Hillsdale, NJ: LEA.
Jones, S. (1998). Diagram representation: a comparison of animated and static formats. Unpublished PhD
thesis, School of Computing and Cognitive Sciences, University of Sussex, UK.
Kalyuga, S. (2000). When using sound with a text or picture is not beneficial for learning. Australian
Journal of Educational Technology, 16(2), 161e172.
Kalyuga, S., Chandler, P., & Sweller, J. (1998). Levels of expertise and instructional design. Human
Factors, 40(1), 1e17.
Kalyuga, S., Chandler, P., & Sweller, J. (1999). Managing split-attention and redundancy in multimedia
instruction. Applied Cognitive Psychology, 13(4), 351e371.
Kaput, J. (1994). Democratizing access to calculus: New routes using old roots. In A. Schoenfeld (Ed.),
Mathematical thinking and problem solving (pp. 77e156). Hillsdale, NJ: LEA.
Kaput, J. J. (1992). Technology and mathematics education. In D. A. Grouws (Ed.), Handbook of
teaching and learning mathematics. New York: Macmillan.
Kirschner, P.A. (2002). Cognitive load theory: Implications of cognitive load theory on the design of
learning. Learning and Instruction, 12(1), 1e10.
Kirsh, D. (1991). When is information explicitly represented. In P. P. Hanson (Ed.), Information,
language and cognition (pp. 340e365). New York: OUP.
Klein, P. D. (2003). Rethinking the multiplicity of cognitive resources and curricular representations:
Alternatives to ‘learning styles’ and ‘multiple intelligences’. Journal of Curriculum Studies, 35(1), 45e81.
Kozma, R. B., & Russell, J. (1997). Multimedia and understanding: Expert and novice responses to
different representations of chemical phenomena. Journal of Research in Science Teaching, 34(9),
949e968.
Larkin, J. H., & Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand words.
Cognitive Science, 11(1), 65e99.
Leinhardt, G., Zaslavsky, O., & Stein, M. M. (1990). Functions, graphs, and graphing: Tasks, learning
and teaching. Review of Educational Research, 60, 1e64.
Levin, J. R., Anglin, G. J., & Carney, R. N. (1987). On empirically validating functions of pictures in
prose. In D. M. Willows, & H. A. Houghton
(Eds.), The psychology of illustration: I. Basic research (pp. 51e85). New York: Springer.
Lohse, G. L., Biolsi, K., Walker, N., & Rueler, H. (1994). A classification of visual representations.
Communications of the A.C.M., 37(12), 36e49.
Lowe, R. K. (2003). Animation and learning: Selective processing of information in dynamic graphics.
Learning and Instruction, 13(2), 157e176.
Mayer, R. E. (1997). Multimedia learning: Are we asking the right questions? Educational Psychologist,
32(1), 1e19.
Mayer, R. E., & Moreno, R. (2002). Aids to computer-based multimedia learning. Learning and
Instruction, 12(1), 107e119.
Mayer, R. E., & Sims, V. K. (1994). For whom is a picture worth 1000 words e Extensions of a dual-
coding theory of multimedia learning. Journal of Educational Psychology, 86(3), 389e401.
McKendree, J., Small, C., Stenning, K., & Conlon, T. (2002). The role of representation in teaching and
learning critical thinking. Educational Review, 54(1), 57e67.
van der Meij, J., & de Jong, T., (2003). Supporting students’ translation between representations in a
simulation-based learning environment. Paper presented at the 10th EARLI conference, 2003, August,
Padova, Italy. Moher, T., Johnson, A., Ohlsson, S., & Gillingham, M. (1999). Bridging strategies for VR-
based learning. Paper presented at CHI ‘99, 1999, May, Pittsburgh, PA.
Moore, P. J., & Scevak, J. J. (1997). Learning from texts and visual aids: A developmental perspective.
Journal of Research in Reading, 20(3), 205e223.
Novick, L. R., Hurley, S. M., & Francis, M. (1999). Evidence for abstract, schematic knowledge of three
spatial diagram representations. Memory & Cognition, 27(2), 288e308.
Oberlander, J., Cox, R., Monaghan, P., Stenning, K., & Tobin, R. (1996). Individual differences in proof
structures following multimodal logic teaching. Proceedings of the eighteenth annual meeting of the
Cognitive Science Society (pp. 201e206). Palmer, S. E. (1977). Fundamental aspects of cognitive
representation. In E. Rosch, & B. B. Lloyd (Eds.), Cognition and categorization. Hillsdale, NJ: LEA.
Peirce, C. S. (1906). Prolegomena to an apology for pragmaticism. The Monist, 16, 492e546.
Petre, M., & Green, T. R. G. (1993). Learning to read graphics: Some evidence that ‘seeing’ an
information display is an acquired skill. Journal of Visual Languages and Computing, 4(1), 55e70.
Plass, J. L., Chun, D. M., Mayer, R. E., & Leutner, D. (1998). Supporting visual and verbal learning
preferences in a second-language multimedia learning environment. Journal of Educational Psychology,
90(1), 25e36. Plo¨tzner, R. (1995). The construction and coordination of complementary problem
representations in physics. Journal of Artificial Intelligence in Education, 6(2/3), 203e238.
S. Ainsworth / Learning and Instruction 16 (2006) 183e198 197.