Summary:
 

Models of learning described by Shuell and by Bruner are compared in relation to classifications of types of learning. Three instructional programs for Special Needs learners are described and their learning models discussed. In the process, the adequacy of the models is evaluated. With Special Needs software for users with profound learning difficulties, cognitive models seem less appropriate and behaviourist terminology could be used. However, this learning can be interpreted in terms of hypothesis testing in a discovery environment. The importance of affective factors is noted.

Introduction

Shuell (1992) claims that every teacher or developer of instructional software has an implicit model of learning, which influences the decisions of instructional design and therefore the interaction with the learner. Shuell describes four such models, by which he means not a formal information processing model but a characterization of the activity required in the learning situation. Before attempting to identify which models are implied by some software for Special Needs learners, I discuss his classification and that of Bruner (1986) in the context of types of learning.

Models of learning

We can distinguish between models of learning and taxonomies of learning types. Models such as those described by Shuell (1992) and Bruner (1986) (Table 1) are caricatures of common assumptions about learners which may or may not be based on real learning processes, as described in taxonomies of learning such as those of Gagne (1977, 1985). Before using models to describe software, we should evaluate them by comparison with other models and with the broader range of types of learning in the taxonomies of learning.

A very broad taxonomy is behaviourism versus cognitive approaches, which while having different approaches also describe largely different learning phenomena. While behaviourism describes learning in terms of associations between stimulus and response (Skinner 1971, Greene and Hicks 1984) and explicitly avoids modelling internal learning processes, the latter are the purpose of cognitive psychology, which represents learning as a progressive construction of meaningful personal knowledge (Gage and Berliner 1988, ch 12). Hannafin and Rieber (1989) describe the insights into learning of both behavioural and cognitive schools (p 99, Table 1).

Another taxonomy is described by Lovell (1982) in his summary of the psychology of learning: classical conditioning and operant conditioning (behavioural tradition); Gestalt theory where perceptions are organised by the mind into patterns which most simply structure understanding; and insightful learning where a flash of inspirational understanding can occasionally result from a reorganisation of cognitive structures. Gagne's (1977) taxonomy of eight learning types is well known: signal processing (classical conditioning); stimulus-response learning (operant conditioning); motor chain and verbal chain reasoning (learning sequences of actions or words through practice); multiple discrimination (of individual items from a set); concept learning (the identity of classes); the acquisition of rules (allowing learners to organise information); and problem solving (learning new rules or applying them to new situations). All but signal processing are on a ladder in which each level requires the underlying levels to have been learnt. Signal processing is independent and is important in affective learning of emotions and attitudes. (His 1985 edition retains these elements in a different framework.) (See table 1.)

These taxonomies include a wide range of types of learning of which only some are the focus of mainstream school, college and adult education: the cognitive rather than the behavioural, Gestalt and insightful learning, and in Gagnes hierarchy, from the level of concepts upwards. However, in Special Needs education the full range may be involved.

We can now relate these learning types to Shuell's classification of models:

The learner models of Bruner (1986) cover a wider historical and philosophical landscape:

The affinities have been noted between the learning models of Shuell and Bruner. We can accept Shuell's classification for practical purposes, except that its models are concerned with conceptual learning, or the top three of Gagne's types. However, in Special Needs education, some learners may be operating at lower levels described by behavioural approaches, consistent with Gagne's first five types. Furthermore, Special Needs software might be aimed at affective responses which Shuell's models do not include, but are described by Bruner's Carrot and Stick or Gagne's signal learning.

Table 1

Summary of Models

Bruner 1986

Shuell 1992

Summary of Taxonomies

Hannafin & Rieber 1989

Lovell 1982

Gagne 1977

A hierarchy with each level (but the bottom one) a prerequisite for the next above.

Some Special Needs software

The use of software in Special Needs Education is particularly interesting for the range of learning abilities and types with which it is concerned. Broadly, software can be classed as being appropriate for those with mild, moderate or profound learning difficulties. The first and second examples are for profound learning difficulties. The third example covers a range of learning difficulties, from moderate through mild.

This is not an evaluation of the software; the applicability of the models is instead discussed. They all use an Acorn computer.

1. Keyboard fun

Typical of simple but useful software for learners with profound learning difficulties, this was produced by a student on the Special Needs Computing Graduate Diploma course at Keele, for use with a particular adult client with profound learning difficulties at the Llwynypia Day Centre. The opening screen is coloured baloons and the title. The following screens are intended for two types of users: the tutor and the learner. For the tutor, a following screen asks for the learner's first name, a choice of two graphics screens (baloons or kaleidascope), a choice of two tunes, a time period in minutes (1-30) for this run, and a choice of input device (single switch, concept keyboard or standard keyboard). Having set up the run, the tutor then hands over to the learner.

The program waits for any press of the input device, when it builds up a colour graphics screen of balloons or a kaleidoscope for 15 seconds, while playing the tune in time with the pattern. The screen then goes blank until a further input is made, when the pattern and tune repeat. This continues for the period specified. The data on keypresses, display and tune are appended to a disk file under the learner's name. A tutor screen then appears asking if the stored data on keypresses should be displayed or printed, and if another run is required. This collection of management data implies the possibility of progress in the learner, and would be used to alter the options and judge when a learner could gain no further benefit from the software.

Only Shuell's Discovery model seems relevant here, and this seems a rather grand description. The program's purpose is to encourage any positive input by rewarding this with interesting images and sound. The computer output reinforces an interaction, no doubt as the first step in a programme of increasingly demanding computer use, in a way that could be described as operant conditioning. As well as training a particular behaviour, the hopefully pleasing output may be useful in generating a general positive motivation towards computers, as would any reward in conditioning, and a curiosity about the randomness element of the kaleidoscope in particular. Gagne's signal learning and the Carrot and Stick model seem to apply although there is no stick, only a carrot.

2. I can do it!

This is a suite of 14 programs recently published by Barnados, intended for children with profound learning difficulties. Programs are grouped in three levels of user ability and one program from each level is described briefly below. They all produce good quality sound and graphics. The learner is prompted to look or listen at appropriate moments by displays of symbols for eyes or ears, and to touch the input device by a hand symbol. As in Keyboard Fun, text instructions to the tutor offer simple choices for setup, sound volume and pitch and the learner's name.

Level one programs require only the activation of whatever input device is being used. The program 'Swans' starts by showing swans swimming while playing a related piece of ballet music. After the 'touch' instruction is given, the dispay and tune stop periodically until the input device is touched. Level two programs require the use of two switches or input devices. For example, 'Happy Birthday' first draws a birthday cake with the appropriate number of candles and the learner's name on it, and plays the appropriate tune. After the touch instruction, the cake is drawn again. One switch lights the candles one at a time and the other then puts them out. Level three programs require choices between 6 or 8 switches. A concept keyboard or touch screen is needed for input. In 'Susie' a picture of a girl's face appears with hairbow, earings and necklace. On overlays for the concept keyboard are either pictures of some of the screen objects (e.g. earrings), or tools for producing screen effects (e.g. lipstick for colouring the lips). The correct object must be touched on the keyboard to achieve the screen effect and build up the picture.

Swans is similar to Keyboard Fun. It is simultaneously motivational of computer use, with rewards of pictures and tunes. Levels two and three require the use of simple but fundamental concepts of choice and then matching. These concepts are probably not gained as declarative knowledge but as procedural skills, through practice. The software gives non-verbal prompts but human guidance would also be needed, so the most appropriate model is guided discovery. The types of learning encouraged range from Gagne's stimulus-response, motor chaining and multiple discrimination, by the selection of correct objects. While the software manual refers to the learning of the 'concept' of choice, this is obviously not learnt in a declarative way; it could not be articulated by the learner, who actually gains procedural skills in matching and selection. Declarative and procedural knowledge are separate (as proposed by the ACT* theory of learning, see MSc Course Team 1991). While use of the program might be thought of as operant conditioning to use an input device (in behaviourist terms), to the extent that the learner has a choice s/he is exercising intention and then conditioning is an inadequate description of what is happening. On the other hand, if the learner is pressing any randon key and the software responds to this, then operant conditioning is an adequate description.

3. Mr Ugh

Mr Ugh was produced by the Computer Applications to Special Education unit of the Department of Psychology at Keele. Described as an introduction to arcade games, it offers 30 levels of difficulty and so spans severe to mild learning difficulties. The aim of the game is to direct a small screen sprite (space invader) to hit Mr Ugh (a sad face) placed at a random position on the graphics screen, whereupon he expodes. Keyboard, joystick or concept keyboard can be used to guide the sprite. Sound feedback accompanies the sp[rite movement but offers no information on position.

In contrast to most arcade games the program has two features significant for Special Needs environments. Firstly, the game adjusts its level of difficulty automatically to the skill of the user. It starts with a run at level 1 and if the user is successful in under 15 seconds it proceeds to level 2 for the second run, and so on. If the user takes between 15 and 30 seconds to hit the target, the level of difficulty stays the same for the next run. If the user fails after 30 seconds, the game continues but drops to a lower level for the next run. The first level is extremely easy with the face filling some quarter of the screen. The final level is quite difficult for an average person (it took me 20 minutes to reach it), involving Mr Ugh moving away from the sprite, the appearance of mazes of increasing complexity in which the target hides, and the need to use a 'hyperspace' feature to move the sprite instantly to new but random positions. For the last stages it is possible to design one's own mazes in which Mr Ugh will hide.

The second feature is that the input activity and level are monitored and stored, and can be displayed as a graph on the screen. This gives a feedback on performance which would be useful for the user or a tutor in later (metacognitive) reflection on the activity.

The aim of the software is not to convey conceptual knowledge but to develop skills. At easy levels this can be seen as stimulus-response learning, rewarding correct input. Motor chaining learning is needed to link keypresses to control the sprite's movement. As obstacles appear, multiple descrimination is necessary to distinguish obstacles of varying size and shape from the clear space in which the sprite can move. At higher skill levels, some concept learning is involved, for example, the concept of a maze to be run, or an enclosure in which the sprite is trapped unless 'hyperspace' is used.

Mr Ugh has a predictable behaviour which is partly dependent on the user, moving away from the current sprite position and breaking through maze walls which the sprite cannot. At the highest difficulty levels, the game resembles instructionless learning about a device (Shrager and Klahn, 1986) and corresponds with Shuell's Discovery Learning model. If the learner is told (or reads in the manual) that the Spacebar creates a hyperspace jump for the sprite, which is necessary to escape some mazes, it is Guided Discovery learning. But generally the game is a microworld in which the learner is a problem solver, overcoming obstacles by discovering new strategies for achieving the goal of hitting Mr Ugh, discovery with or without some degree of guidance from outside the software. The learner will need to embark on hypothesis formulation and testing, reflective thinking and self regulation. 'This active, constructive attitude of the learner encourages meaningful incorporation of information into the learner's cognitive structure' (Goodyear et al. 1991, p.269). While the specific concepts learnt are limited in this arbitrary microworld, general problem solving skills may be transferable to other domains.

The automatic adjustment of level of difficulty means that the learner should generally be successful within 30 seconds. This illustrates the notion of the 'zone of proximal development' (Vygotsky, page 7 in DeCorte 1990). While the game does not explicitly help the learner to find methods of achieving the next level of difficulty, it does stimulate learning by continuously presenting new problems slightly harder than the last one, though similar to it. The storage and graphical presentation of feedback information on the learner's performance is available to a tutor and the learner. It may thus encourage the user to reflect on their learning activities and thus gain metacognitive skills as well as domain knowledge (DeCorte 1990, page 3).

Discussion

The validity of Shuell's classification is reinforced by being broadly cognate with that of Bruner. Generally, the programs described assume Discovery learning, but it is not clear if Shuell's models accommodate all the programs for profound learning difficulties, where learning is often not at a conceptual level. The learning types at the bottom of Gagne's (1977) hierarchy often seem appropriate, and it is an attractive idea that the sequence his learning types span the range from simple association to conceptual learning encouraged by these programs. However, Gagne's description of association learning is a re-working of Skinnerian behaviourism. The simpler types of learning are those most easily described in behaviourist terms, and in the preceding descriptions of software it has been convenient to use behaviourist descriptions of simple associations which the software encourages the learner to gain. In fact, though generally discredited, behaviourist methods continue to be useful with learners with Special Needs (Gallagher 1994). Nonetheless, the apparent continuum within Gagne's hierarchy is misleading; it conveniently includes the very different approaches of behaviourism and cognitive psychology. The continuum of learning types encouraged by this software therefore exposes a difficulty if it is to be described in terms of two fundamentally different approaches

However, it may not be necessary to accept a behaviourist philosophy, based on the interpretation of association between stimulus and response as conditioning (Skinner 1971, p32). Tenant (1988, p. 113) describes the work of Martin who reinterprets operant conditioning as an organism guiding its own behaviour according to the patterns it discovers in the world. 'The behaviour of animals ... can be explained without recourse to the term conditioning ... this is also true of human behaviour in natural settings'. This seems similar to simple hypothesis testing in the learner. We can therefore avoid behaviourist concepts and describe the simpler forms of learning (the lower part of Gagne's hierarchy) in a way more consistent with concept learning.

For example, Mr Ugh provides a smooth progression of learning types ending in discriminations, concepts and rule learning. As an instructionless learning environment, discovery learning can be described here as the process of hypothesis creation and testing (Shrager and Klahr 1986) The easiest levels of the software can also be described as hypothesis testing where the hypotheses are simple, for example: 'pressing an arrow key moves the sprite', 'the up-arrow moves the sprite up', 'repeated arrow keypresses repeat the sprites movement', and 'Mr Ugh moves away from the sprite'. While the simplest forms of learning can be described in the terms of behaviourism, they need not be, and can equally well be described in terms more consistent with concept learning, allowing a consistent approach across the whole range of learning types. We should ask 'how useful is this explanation' rather than 'is this the most accurate description'. Cognitive explanations may provide us with more insight into the learning, although behaiourist theory may provide us with simple practical guidance as to how to plan future learner activity.

If this hypothesis testing interpretation of association learning is accepted then the association learning prompted by the software fits Shuell's model of discovery learning, which in practice would be guided by a carer. In effect the software provides a simple simulation and the learning processes are those of hypothesis formulation and testing, 'encouraging active experimetation and exploration' (Goodyear et al. 1991). Learners with profound disability are not going to be able to articulate such hypotheses, but that does not refute this interpretation. Procedural knowledge gained by learners from using the Special Needs software need not be accompanied by declarative knowledge of the programs' behaviour. For example, Shrager and Klahr (1986) describe the cognitive 'device model' gained by normal adult learners in the instructionless (discovery) learning environment they provided, but the learners themselves reported their knowledge in the procedural terms of how to make the machine do specific things.

A second general issue is motivation. All the software described here attempts to reward the learner and create a positive affective response. Bruner's classification of learner models includes this aspect as the Carrot and Stick model. Shuell explicitly recognises the absence of affective and motivational factors in cognitive models (p 27) although acknowledging that motivational components of tutoring are equally important as cognitive considerations. Shuell treats motivational and affective states as aspects of mental activity outside the scope of the cognitive information processing model but yet capable of affecting the efficiency of that processing.

The relationship between behaviourism and emotions is unclear. While a behaviourist description of learning does not include emotional states, classical conditioning has been used to explain the adoption development of emotional responses (Lovell 1982, p.34). For example, if math classes are repeatedly associated with embarrassment or fear due to a teacher's criticism of wrong answers, then the situation of a maths problem can become a conditioned stimulus and induce the same emotions in the absence of a teacher in future. Lovell suggests that emotional components of attitudes are acquired in this way. Presumably, the sensory rewards of the computer programs described will create a positive attitude to computers generally (or help overcome a negative attitude) as well as reinforcing the specific learner activities. In the absence of an extension to the information processing model that accommodates emotions, we can use the conditioning model or Bruner's Carrot and Stick to explain the development of affective responses. However, there may be other theories of motivation which are as compatible with cognitive explanations of learning, or more so.

References