Psychology Of Transfer Of Learning Research Paper

2. The Quest For A General Perspective Of Learning And Transfer

In order to give an overview, a general perspective for learning and transfer is outlined here. Learning, on the one hand, is thought (a) to occur within an immense body of prior knowledge, and (b) to be a process of continual integration of knowledge by constructions or reconstructions. Such (re-)constructions include:

(a) relating meaningful units of different conceptual levels that represent objects and their features,

(b) organizing these related elements into larger units according to the constraints imposed by the content structure of its concepts (e.g., money and goods have to be related first in order to become organized further to explain the higher-order concept of ‘inflation’), and

(c) chunking: the process of condensing built-up structures for both memory capacity reasons and ease of subsequent use. The chunks may be related further to other units of different conceptual levels, organized, and chunked anew. Chunks are indexed for retrieval and further processing, highlighting the top-level or goal concept: ‘and all this is called inflation’; or ‘and this is how acceleration is calculated in mechanics’; or ‘this protection by the bird’s color almost guarantees survival.’

Transfer, on the other hand, is thought of as a ubiquitous, continuous, systematic use of selected parts of the immense body of prior knowledge. It starts from a cognitive need for information that has to be (a) retrieved from memory, (b) selected for tentative application, and (c) mapped on a corresponding subprocess of learning as described above.

3. Transfer Of Skills: The ‘Common Element’ Theory Of Transfer—An Old And A Modern Version

Experimenters in cognitive psychology look for learning or, preferably, problem-solving tasks that can be constructed such that identities or similarities between two tasks (a source and target task) can be controlled. Some of the favorite research domains include learning programming languages, use of text processing programs, solving algebra word problems mainly from physics, and motor ability training.

3.1 Playing The Violin—An Example

One of the most obvious transfer situations can be illustrated by the use of common elements in playing a musical instrument, e.g., the violin. When practicing, a sequence of several notes and accordingly a sequence of finger movements of the left hand are learned. Once mastery has been achieved, the required skill can be applied to the whole range of the violin’s touch board. This is the violinist’s economy to use the practiced motor sequences to generate the very same melody in different pitches. This ‘very same melody’ was the kind of holistic unit or ‘totality’ Christian von Ehrenfels (Ehrenfels 1890) was working on, calling it a ‘Gestalt,’ a term used for a whole movement early in this century: the Gestalt psychology. The underlying common elements in violin playing that are transferred from string to string or from one hand shift to another are complete knowledge structures (chunks) containing the condensed motor program or action plan for performing that invariant melody. The specific parameters of the respective program or plan may be adapted if necessary, e.g., the higher the shift position of the left hand the shorter the absolute distances between the set fingers will be. However, the proportions of the distances between the fingers and the single movements within the sequence remain the same.

Motor programs do transfer if the relative timing of a motor sequence is not violated. The music pattern can be speeded up without too high a risk of a loss of transfer.

3.2 Common Elements—An Early Approach

A first version of a ‘common element’ theory of transfer was developed early in the twentieth century by Thorndike. He showed how difficult it is to transfer general abilities from one academic domain to another (Thorndike and Woodworth 1901), except in cases where both share common elements: stimulus–response associations, typical for the explanation of behavior at that time. Transfer increased proportionally with the number of such overlapping associations in the learning and the test tasks (Thorndike 1913).

3.3 Production Systems—A Modern Approach

The common element theory of transfer has made considerable progress during the twentieth century. One of the well-known contemporary theoretical approaches for explaining what kinds of common elements are transferred in mastered skills comes from Anderson’s ACT theory in its several versions (Singley and Anderson 1989) (the acronym ACT stands for adaptive control of thought). In the beginning of the violin play, each of the single knowledge elements (the printed notes to be transformed into movements) are elements of declarative knowledge (facts), each having both a certain single and a relational musical meaning (regarding rhythm and harmony). Repeating the sequence leads to what Anderson calls knowledge compilation: Some tones get closer to each other, are ‘heard together’ as a subpart within the sequence until, finally, larger and larger subparts are experienced as forming a totality. With practice, procedural knowledge is gradually built up. It is composed of specific conditions followed by actions (IF … THEN associations, so-called productions). In our example the condition (the IF part of the production) can be defined by the set of notes; it is followed by the corresponding action (the THEN part), namely the finger movements. If learned and practiced in a violin sonata by Handel, the production would transfer to a Mozart sonata if the same class of conditions were met again. Productions are, thus, the common elements that can explain transfer of skills. The progress in transfer theory in the twentieth century consists of (a) the elaboration of the former stimulus and the response side of knowledge use by introducing the production concept and (b) the integration of productions into compound production systems for knowledge acquisition and use.

4. Analogical Transfer

The most extensive transfer research since the mid- 1980s has focused on so-called analogical transfer whereby true analogies are found in those problems that share a similar deep structure but not necessarily specific content (e.g., the analogy of the atom and the solar system). According to current research, an adequate theory of analogical transfer would have to account for how analogical transfer develops over the processing time span, and for the role that surface and schematic knowledge play. The process would have to follow several stages as the following ‘green walnut’ example illustrates (Fig. 1).

4.1 The Green Walnut—Another Example

Assume that a young girl has acquired the skill to handle several kinds of fruit properly: small apples, bananas, and even a coconut with the help of her father. She is now holding a green walnut in her hand (some time before harvest, as it is still covered by its soft green skin that later dries out), an object she has never seen before.

If we look at the process of analogical transfer the following four stages can be distinguished:

(a) Stage 1 refers to encoding some elements of the target problem. Our young girl encodes her object as something small, round, and green (top right in Fig. 1) which reminds her of some other object with which she is familiar: small, green apples (top left). The corresponding REMINDING 1 subprocess is based on surface similarities between a visible object and an imagined apple (as an exemplar of the class of small apples).

(b) Stage 2: retrieval of one or more source analogues (i.e., elements from specific prior knowledge) when presented with the target problem. The retrieval process can result in a variety of working memory activities: spontaneous spreading-out, systematic decomposing of chunks, or inferencing, leading to a list of features such as … is juicy, … has skin, … is biteable, that are candidate analogs to some corresponding characteristics of the target task.

(c) Stage 3: selection and mapping of the source analogue to the target task. The crucial question is which one of the retrieved items will be adequate for use. There are constraints determined by the target task limiting the selection of candidates to be used for attaining the goal of the target task. To identify the object, the selection will be an action schema or an assimilatory schema (Piaget 1952). Thus, the ‘bite’ schema is selected (subprocess SELECTION 1) for mapping (subprocess MAPPING 1). The corresponding schema use is an application of abstract schematic knowledge, the schema being a generalized unit of the prior knowledge repertoire of the child. Mapping is the subprocess of establishing a contact with the main learning process where still further processing will take place: the schema application as an operation on that green object to be identified.

(d) Stage 4: knowledge integration. The green object is now assimilated by the ‘bite’ schema: The object is indeed biteable. This knowledge is easily integrated into the child’s knowledge, but this is not necessarily the case, as will be shown below. Although the taste while biting is terribly bitter, there is the important additional recognition that the object has a removable skin. (Status and ‘locus’ of stage 4 subprocesses will be discussed in detail.)

From here it becomes obvious that transfer is an oscillating process with recurrent subprocesses: The removable skin is encoded (a recurring stage 1 subprocess) and reminds (REMINDING 2) the child of another object with removable skin, a banana. It is certainly a very fast retrieval process followed by a similarly fast selection (since the task’s constraints do not leave a great choice) of the ‘peel’ schema (recurring stage 3 subprocesses SELECTION 2 and MAPPING 2). These are, again, followed by knowledge integration in the form of the application of the ‘peel’ schema. This ‘peel’ schema is successful with the recognition of a hard shell covering the object. In a next step the hard shell is encoded (another stage 1 subprocess), leading to REMINDING 3, which, in turn, lets the child recall another object with a hard shell, a coconut. Here, stage 2 subprocesses will recur: memories about opening that coconut when the saw from her father’s toolbox had to be used. The list of features of the retrieved coconut is relatively long (… is brown, … has fibers around, … is shakable, etc.) but the ‘open-it’ schema prevails according to the constraints of the task at hand (SELECTION 3 as another recurrent stage 3 subprocess). The ‘open-it’ schema will be mapped (MAPPING 3) for tentative use. But the assimilation by the ‘open-it’ schema will not be realized. Before looking for the saw that was adequate for opening the coconut, her father will explain the inadequacy of the old ‘open-it’ schema and suggest the use of the nutcracker. This is, in Piaget’s terms, an accommodation of the assimilatory schema that would otherwise not satisfy the constraints of the task. So, knowledge integration adopts a specific form in the last phase of the learning process: The ‘open-it’ schema is accommodated and adapted into a ‘crack-open’ schema which, then, serves for assimilation. This is, of course, a specific stage, not of transfer subprocesses but of the main learning process: the accommodation of a schema enabling it for final assimilation in a whole series of accumulating assimilations (notice the embedded rectangles labeled L through L in Fig. 1), leading finally to the identification of the once unknown object as a ‘walnut.’ The crucial point of this final knowledge integration stage is that important processes take place: triggering further constructions after the arrival of a mapped element, partially reorganizing knowledge leading to new structures, chunks or insights. These processes certainly do not belong to the set of transfer subprocesses. However, several authors plead explicitly for a fourth stage in analogical transfer focusing on schema induction and structural abstraction (Gentner 1989, Holyoak 1985, Gruber et al. 1995).

4.2 Theoretical Conclusions From The Green Walnut Example

The following conclusions can be drawn.

(a) The interesting part of transfer research is to be found in the subprocesses and not so much in the outcome or the problem solution.

(b) Transfer (and mapping as part of it) has a definite serial time structure. It should be kept in mind that the need for knowledge stemming from a target task depends on the actual stage of the ongoing learning or problem solving process. This fact dictates the time sequence of mapping activities.

(c) Mapping is in itself not a unitary subprocess. Its components are selection for mapping, tentative matching (i.e., looking for fit), evaluation of fit (including feedback), and either return to a next selection or else finishing the mapping process and giving the information over to the ongoing learning process (knowledge integration).

(d) Mapping is an ancillary process in favor of the ongoing learning or problem-solving process. All subprocesses after a fit are part of knowledge integration, which is part of the main learning process, in our example the accommodation of the inadequate assimilatory schema.

(e) The transfer subprocesses are oscillating processes. They follow the constraints of the ongoing learning or problem solving process, including possible feedbacks from nonfit situations of some mappings (not exemplified in Fig. 1).

4.3 Analogical Transfer Research

There is a huge research literature on specific details while transferring source knowledge on to target knowledge: (a) the structure-mapping approach stresses the fact that ‘analogy is a way of focusing on relational commonalties’ (Gentner 1989, p. 201), (b) the pragmatic schemata approach emphasizes the construction of pragmatic schemata by comparing and abstracting the analogues of the two tasks (Holyoak 1985), and (c) the exemplar approach holds that concrete exemplars instead of abstract structures play the dominant role in analogical transfer. All three approaches agree on the above-mentioned stages of transfer but differ in attributing relative importance to content-specific surface analogues on the one hand and abstract schemata on the other across the stages of transfer. No single theory can explain all available data concerning the effects of source analogues on retrieval and application; there is a plea for a hybrid of the structural, pragmatic, and exemplar views for that purpose.

5. Enhancing Transfer Effects: Knowledge Constructions Need Consolidation

Lack of transfer is the rule rather than the exception—at least in every-day learning. It seems that the primary reason for this is the fact that basic learning processes have not been led up to a level of sufficient understanding, the students often remaining on a level of surface processing or mere symbol manipulation (especially in mathematics and physics). Conceptual and procedural constructions (including problem solutions) in the main learning process need consolidation, as follows.

5.1 Strengthening Knowledge Structures And Making Them Transparent

Consolidation for strengthening cognitive structures involves modified reconstructions of previous constructions: Basically the same conceptual elements are connected again with each other

(a) in a different sequence (sometimes a backward reconstruction),

(b) under different conditions (e.g., time),

(c) from different perspectives, or

(d) with a different goal (e.g., for a different audience),

providing transparency and coherence to the structure. Much of this kind of consolidation can be implemented by means of self-explanations (Chi et al. 1994, Renkl et al. 1998).

5.2 Flexibilizing Knowledge Structures

Several approaches focus on flexibilizing knowledge structures, one stemming from Spiro et al.’s cognitive flexibility theory (Spiro et al. 1991). Another is the progressive transformation procedure (Steiner and Stoecklin 1997), aiming at structural flexibility by systematically transforming the original construction (mainly mathematical problem solving) forcing the learner to define similarities as well as differences by intensive comparisons. To secure flexibility of structures, it is important that knowledge is abstracted by the learners themselves (e.g., from examples) during the learning process instead of being taught as abstract units by some teacher or tutor.

One specific point refers to retrieval flexibility. In order to optimize stage 1 transfer subprocesses (encoding elements of the target task or problem), it is necessary that chunks formed during the previous construction process are decomposed in due time to keep the elements accessible for further processing (relating). Such decomposing activities can be seen as a specific form of retrieval practice that, at times, come close to self-explanations as mentioned above. In order to speed up retrieval of subroutines for learning and problem solving (e.g., conceptual knowledge in addition to algorithms for mathematical transformations), ‘proceduralized’ knowledge (after knowledge compilation; Anderson 1983) may yield an important contribution to transfer. However, the automatic use of proceduralized knowledge may include task-inappropriate effects if disconnected from the original declarative knowledge from which it is built up.

5.3 Transforming The Representational Format—Multiple Knowledge Representations

A specific kind of flexibilizing knowledge structures is transforming one representational format into another: a symbolic representation (e.g., an algebraic equation) into an iconic (pictorial) or an enactive (i.e., action-related) format and vice versa (Bruner 1966). The ability to transform representational formats and to keep knowledge structures accessible in several different formats renders parts of knowledge highly flexible and apt to play a role in several subprocesses in transfer.

5.4 Decontextualizing The Structures

Transfer often fails because knowledge is strongly caught in the training context, so that the common elements of source and target tasks cannot be found. Therefore, an important goal for learning for transfer is decontextualization. This means changing the content of the chosen examples, and thereby replacing contextual facts such as the agents of a problem situation (soldiers attacking a fortress from different directions instead of X-rays attacking a tumor in the same way), the concrete material (sand instead of water) or cultural circumstances (rural instead of urban) in order to have the learners recognize the invariant structural aspects or the abstracted schemata, respectively (Renkl et al. 1998).

5.5 Reviewing—Focusing On One’s Own Processes, Not On Outcomes

There is one severe lack in today’s teaching of learning for transfer, namely reviewing one’s own learning processes: ‘How did I acquire this wonderful problemsolving procedure, this marvelous proof in geometry, or this striking insight into a network of economic relations?’ Reviewing one’s own learning processes means mentally unpacking some of the chunks formed, recalling the relations established during listening to the teacher’s explanations or the studentcentered discussion; in short, reactivating the how of that multistep process of relating and condensing (chunking).

To sum up: the emphasis on consolidation is not an emphasis on automating (and thus not on knowledge compilation; Anderson 1983), but rather on making structures transparent, enhancing the coherence of relations and chunks, but also flexibilizing the elements, relations and larger units of a total knowledge representation including decomposing chunks. Consolidation contributes to the comprehension or understanding of the built-up structures (Kintsch 1998, Steiner 1999).

The crucial point for transfer is not whether or not chunks, productions, schemata, scripts, or mental models are constructed, nor is it their very nature that counts. It is how all these knowledge representations were dealt with in the original learning process to fight from the outset against structural deficits. There is a long-term aspect involved in transfer with accumulating learning effects that does not seem to be reflected sufficiently in current research.

6. Social, Emotional, And Motivational Aspects Of Transfer

In the ‘situated’ learning approach (Gruber et al. 1995), the social context plays a significant role; it is encoded and stored as part of episodic memory. Indeed, actual features of target task elements or requirements may remind someone of episodic in- formation that, in turn, may activate some conceptual source task information either to trigger a transfer process or to keep it going. Furthermore, learning, whether individual or social, is permanently accompanied by individual evaluations of both the potentially emotional aspects of the content and some individual aspects of learning such as goal attainment, success, or failure. It is therefore not surprising that emotional effects are encoded along with cognitive contents and activities that may be retrieved later in transfer situations when the target task itself evokes emotional or mood memories (Bower 1981). More-over, there are motivational and personality-related aspects that contribute to the overall quality of the learning and transfer process and outcome: volitional self-regulations (Kuhl 1985) along with high personal involvement and mental effort.

6.1 External Support Or The ‘Proximal Zone Of Development’

Experimental work and educational field studies suggest that transfer is usually not performed without external support, be it by hints with regard to similarities or analogues in two tasks, tutoring, or other sociocognitive guidance. All these measures correspond to the ‘proximal zone of development’, a concept introduced by Vygotsky (1978): the psychological space where individuals are able to develop cognitive achievements not by themselves but through the support of a tutor.

6.2 About Future Transfer Research

A lot of admirable experimental work has been done within transfer research but there is a strong bias toward artificial microworlds in the tasks to be learned that diminishes the usefulness of the results with regard to broader, less experimental settings. As far as the ‘analogues’ are concerned, the nonexperimental reality often looks different: at the beginning of many problem-solving processes there are no analogues whatsoever in the learner’s mind. It would be promising for future transfer research to focus on the subprocesses at the outset of transfer when individuals try to find and define the mental gaps to be filled in their learning or problem-solving tasks or when the retrieval subprocesses are set in motion. There is, furthermore, a definite need for more research focusing on both metacognitive and noncognitive aspects for explaining the lack of transfer so often observed.

Bibliography:

1. Anderson J R 1983 The Architecture of Cognition. Harvard University Press, Cambridge, MA
2. Bower G H 1981 Mood and memory. American Psychology 36: 129–48
3. Bruner J S 1966 Toward a Theory of Instruction. Belknap Press of Harvard University Press, Cambridge, MA
4. Chi M T H, DeLeeuw N, Chiu M H, La Vancher C 1994 Eliciting self-explanations improves understanding. Cognitive Science 18: 439–77
5. Cormier S M, Hagman J D (eds.) 1987 Transfer of Learning. Contemporary Research and Applications. Academic Press, San Diego, CA
6. Detterman D K, Sternberg R J (eds.) 1993 Transfer on Trial: Intelligence, Cognition, and Instruction. Ablex, Norwood, NJ
7. Ehrenfels von C M 1890 Uber Gestaltqualitaten. (On the qualities of ‘Gestalt’). Vierteljahreszeitschrift fur Wissenschaftliche Philosophie. 14
8. Gentner D 1989 The mechanisms of analogical learning. In: Vosniadou S, Ortony A (eds.) Similarity and Analogical Reasoning. Cambridge University Press, New York, pp. 199–241
9. Gruber H, Law L C, Mandl H, Renkl A 1995 Situated learning and transfer. In: Reimann P, Spada H (eds.) Learning in Humans and Machines: Towards an Interdisciplinary Learning Science. Pergamon, Oxford, UK, pp. 168–88
10. Holyoak K J 1985 The pragmatics of analogical transfer. In: Bower G H (ed.) The Psychology of Learning and Motivation: Advances in Research and Theory. Academic Press, New York, Vol. 19, pp. 59–87
11. Kintsch W 1998 Comprehension. A Paradigm for Cognition. Cambridge University Press, Cambridge, UK
12. Kuhl J 1985 Volitional mediators of cognition–behavior consistency: self-regulatory processes and action versus state orientation. In: Kuhl J, Beckmann J (eds.) Action Control: From Cognition to Behavior. Springer-Verlag, Berlin, pp. 101–28
13. Piaget J 1952 The Origins of Intelligence in Children. International Universities Press, New York
14. Renkl A, Stark R, Gruber H, Mandl H 1998 Learning from worked-out examples: the effects of example variability and elicited self-explanations. Contemporary Educational Psychology 23: 90–108
15. Singley M K, Anderson J A 1989 The Transfer of Cognitive Skill. Harvard University Press, Cambridge, MA
16. Spiro R J, Feltovich P J, Jacobson M J, Coulson R L 1991 Cognitive flexibility, constructivism, and hypertext: random access instruction for advanced knowledge acquisition in ill-structured domains. Education and Technology 31(5): 24–33
17. Steiner G 1999 Learning. 19 Scenarios from Everyday. Cambridge University Press, Cambridge, MA
18. Steiner G F, Stoecklin M 1997 Fraction calculation—a didactic approach to constructing mathematical networks. Learning and Instruction 7(3): 211–33
19. Thorndike E L, Woodworth R S 1901 The influence of improvement in one mental function upon the efficiency of other functions. Psychological Review 8: 247–61
20. Thorndike E L 1913 Educational Psychology, Vol. 2: The Psychology of Learning. Teachers College, New York
21. Vygotsky L 1978 Mind and Society. Harvard University Press, Cambridge, MA