Jutta Treviranus, University of Toronto


Contents

Abstract
Introduction
Prerequisites to Skill Acquisition Skilled Behavior
Considerations in Promoting Skilled Use Case Example Conclusions and Future Directions
Acknowledgments
References

Abstract

Many individuals use computer-based assistive devices as alternatives for the fluent, automatic skills of speech, writing and touch-typing. This paper proposes that controlling an alternative access technique should become as automatic as touch-typing or speech. The process of acquiring a cognitive-motor skill is discussed. Prerequisites to skill acquisition and impediments to motor automaticity are outlined. Factors which promote or impede skilled control of alternative computer access systems at each stage of the design, assessment, prescription, training and evaluation process are reviewed. Several case examples are used to illustrate relevant issues.

"...we must make automatic and habitual, as early as possible, as many useful actions as we can..The more of the details we can hand over to the effortless custody of automatism, the more our higher powers of mind will be set free for their own proper work" ...James, 1913(1)



Introduction

Controlling a computer through an alternative access technique is usually a slow, physically and cognitively demanding process. Many individuals who cannot control a standard computer keyboard use computer systems to speak and write (2). Alternative computer access techniques are therefore used to replace the fluent, automatic skills of speech, writing and touch typing. As controlling alternative access systems is not the end goal, but merely a means to an end, it is imperative that cognitive and perceptual energies be preserved for the end goal of communicating. In speech, writing and typing this is done by developing sufficient skill in the motor tasks, which then become automatic, leaving cognitive resources available for formulating and communicating the message(3). Alternative access techniques can also become largely automatic skills thereby making the process of alternative and/or augmentative communication (AAC) faster and less cognitively demanding.

Human factors researchers and cognitive scientists have long been concerned with designing systems and training strategies that produce skilled computer users (4-6). The benefits of skilled use for even occasional users of specific software programs are the focus of much discussion. For users of computer based AAC systems the computer is not a tool they occasionally use. It is a tool they depend on for important if not vital human functions. For many users it becomes part of their identity. As a result it plays a much more intimate and personal role in the user's life than the average computer. It is therefore of great importance that users of computer based AAC systems be assisted in acquiring skill and ultimately automaticity in controlling their alternative access systems. To this end, insights and knowledge can be borrowed from the vast body of literature on skill acquisition in the fields of psychology, education, cognitive science and human factors, and adapted to users of AAC systems. Insights and guidelines specific to the field of assistive technology are also required.

This paper will address skill acquisition as it applies to control of an alternative access technique. Discussion has been limited to this one aspect of a communication system because alternative access techniques replace processes that are usually skilled and automatic. This paper is not a literature review on skill acquisition but an attempt to glean general principles and practical implications from the literature and apply them to the area of alternative computer access. The recommendations made in this paper have not been experimentally verified in the alternative computer access field. It is hoped that this paper will assist in providing the theoretical background for such investigation. The second half of the paper will illustrate concepts discussed using an example alternative access technique.



Prerequisites to Skill Acquisition

Two important prerequisites to skill acquisition are understanding and trust of the tool, in this case the alternative access system. Without an adequate conceptualization of the system to be able to predict the actions of the system and plan appropriate motor strategies to achieve desired outcomes, the user cannot begin to gain skill. Without the feeling that the system is dependable and predictable the user is not free to master the use of the system.

 

Understanding

The mental processes whereby human beings understand or conceptualize the world are the topic of much theoretical debate (7). Although the terms and the specific processes vary, the majority of theorists in cognitive science and human factors agree that people gain an understanding of their world and of phenomena in the world by actively constructing mental models (also referred to as schema (8), scripts (9), productions(10) or frames(29)). These models allow us to make sense of the world, to predict what will happen next, and to determine how to respond. Initially these models may be very sketchy pictures of a phenomenon. We draw upon previous experiences and existing mental models to construct these new models. These constructs evolve, being constantly revised and updated as new information is gained about the phenomena or system being modeled. As experience with the system accumulates the model is fleshed out and may include "motor programs", or the sequence of motor commands required to achieve specific goals, and "feedback templates" or familiarized patterns of how the feedback should appear if motor actions are performed correctly(4).

How well we understand an access system can be seen as the completeness and accuracy of our mental models of the system. Misunderstanding, lack of understanding or confusion about the system can also be seen as the presence of chaotic, disorganized, incomplete, inconsistent or over-simplistic models(7).

Human factors literature discusses both the mental models formed by the learner and those prescribed by the designer/developer or trainer. Because it appears easier to reuse or recycle existing models than to construct a model without reference to existing knowledge, some designers prescribe metaphors they feel are familiar to prospective users, in their interface design(7). One well-known example of this is the desktop metaphor. A limitation of prescribed metaphors is that they function best if the user is familiar with the metaphor. Metaphors or models prescribed by the trainer or system design should be relevant to the user's world view. The desktop/filing metaphor is not optimal for individuals who have never worked in an office or handled files. In choosing metaphors the user's age and experiences need to be considered. Another limitation is that the metaphor should adequately explain the structure and behavior of the device. It has been found that functions inconsistent with the metaphor are learned more slowly than those consistent with the metaphor. When the typewriter metaphor was used to assist students in learning to use a word processor, the delete and return functions were last to be mastered (7).

Studies have shown that it is beneficial to suggest metaphors or prescriptive models in system design and training. Foss, Rosson and Smith found that users learn more in less time when a metaphor has been suggested (11). Researchers found that users perform better on novel tasks (7, 30). There is also evidence to suggest that prescribing a model benefits users even when the model does not hold. It is postulated that the model although inaccurate provides a referent for the exceptions(12). Thus, in introducing a metaphor, similarities and dissimilarities to the task to be learned should be pointed out to the learner.

Prior to introducing a new access method to the user it is imperative that the practitioner have a complete and accurate mental model of the system. In introducing the system the practitioner should remain cognizant of the user's conceptualization of the system and the task. It is not safe to assume that the user has remembered everything that has been said, and recognized the significance and function of all that has been demonstrated. Accurate rote repetition of the steps required to complete the task does not imply understanding or retention. The practitioner should guide the user in building a robust and sufficiently complete and accurate model of the system and the task. (Light and Lindsay (23) provide an instructive example; describing the consequences of incomplete models and suggesting strategies for facilitating the formulation of accurate models.) The goal in introducing an access system should therefore be to encourage the user to construct a mental model that is appropriate to the task, helps the user deal with novel situations, and allows the user to accurately predict the actions of the system.

 

Trust

Another prerequisite to skilled tool use is trust of the tool or system (13). This is dependent to a large degree on understanding of the system and on the trustworthiness of the system. Trust depends on how well the system meets the expectations the user has of it. If the understanding of the device is incorrect or incomplete the device may be perceived as unpredictable because the user is unable to predict the devices' actions or responses.

A trustworthy device is both dependable and predictable. It is consistent in its performance. It lives up to its expectations. Technical breakdown, or technical errors are not conducive to trust, or new learning.

If trust is broken or mistakenly assigned the process of regaining trust is slow and difficult, the whole learning process may be jeopardized and the user may abandon the use of the tool. It is important therefore not to instill unrealistic expectations regarding the performance of the access system. Another risk associated with unrealistic expectations is that the user will blame his/her own performance if the device does not meet these expectations.

The more the system is trusted, the less it will be monitored and observed, the less wary and vigilant the user will be when using the system. When trust is established the user will be more willing to experiment and take risks.

 

Superstitions, Bad Habits and Old Habits

Conceptualizations of a system are not necessarily accurate. Automatic behavior does not necessarily aid in efficient completion of the task. Erroneous beliefs about a system have occasionally been termed superstitions (7). This analogy can be expanded to include maladaptive mental models, motor programs and automatic behaviors.

Superstitions are usually seen as strategies for avoiding bad luck or encouraging good luck. Bad luck in the case of an access system is the malfunctioning of the device or the entry of errors. Good luck is the successful completion of the goal. Although superstitions are not based on completely rational knowledge of the world, many superstitions are gained through experiences of events whose occurrence is incorrectly attributed. An example is a user who assumed that globally changing the font of a text file before printing a file caused the file to crash. This belief was based upon an actual experience, but the blame was incorrectly attributed to the act of globally changing the font, rather than a power surge which happened simultaneously.

Human beings constantly seek meaning in their world. In the absence of accurate information we construct models with whatever information we have; however erroneous or poorly founded this information may be (12). We attempt to incorporate superstitions or erroneous beliefs into our interpretation of the world by creating misleading mental models. The motor programs or automatic behaviors arising from these maladaptive models could be termed bad habits. A user of a relative headpointer was seen to occasionally shake his head rapidly from side to side. When asked why, he responded that this brought the cursor back to the center of the display. In fact simply turning his head to one side of the screen until the cursor is appropriately positioned would have been a more efficient and accurate method of reaching his goal. His model of the pointing system described the cursor as stuck to one side of the screen, shaking his head would loosen it.

Superstitions when ingrained into the model of the system are very difficult to eradicate and persist even after the user has been given concrete evidence to disprove them. Once these superstitions have led to bad habits, remediation becomes even more difficult. A user of multiple head switches persisted in hitting one switch much more gently and slowly than the remaining switches. Following inquiry it was discovered that the user had used a switch with a loose connection for many months. Hitting the malfunctioning switch with average speed and force resulted in double hits. The habit, developed to cope with the malfunctioning switch, persisted even after the switch was replaced.

Old habits, which were functional at one point, but do not apply to the task to be learned, can interfere with task mastery. The greater the similarity between the two tasks, the greater the negative transfer is likely to be. A common example is moving from operation of a car with a standard transmission to one with an automatic transmission. The habit of pressing the clutch, frequently interferes with appropriate use of the brake and gas pedal.

Superstitions or erroneous beliefs about the system need to be detected and eliminated early in the learning process before they lead to bad habits. This can be done by watching the user perform the target task and noting extra steps, inefficient use, avoidance of certain steps, or errors. The practitioner should assist the user in constructing mental models of the access system that allow the user to correctly diagnose errors and assign blame or responsibility appropriately.



Skilled Behavior

Once the user has gained understanding and trust of the access system s/he can begin to develop skill in controlling the access system. As anyone who has achieved skill in playing a musical instrument, touch-typing or playing a sport can attest to, becoming a skilled user is a lengthy process. Many hours of practice are required even after consistent, error-free performance has been reached.

Descriptors used to characterize skilled use include (4, 14-16):

  1. economy of effort (far less energy and attention is required largely due to the automaticity of skill components, this also implies that the mechanisms for performing the task tend not to be readily available to conscious awareness),
  2. consistency of performance,
  3. adaptability (performance mechanisms are automatically adjusted to compensate for a wide variety of task conditions),
  4. speed,
  5. dependence on internalized (or intrinsic) prompts and feedback rather than prompts and feedback from the environment or computer system (extrinsic prompts),
  6. ability to anticipate the actions of the system; and
  7. the ability to multitask, to carry on more than one activity at a time.


Most authors see skilled behavior as multi-component, including cognitive, motor and perceptual sub skills (3, 4, 5, 15). Automatic processing is felt to be an important component of skilled behavior. As argued by Fisk, automaticity, however, is necessary but not sufficient for complex skilled behavior to occur (16).

 

Stages of Skill Acquisition

Most theorists agree that there are three distinct stages in learning a skill. The specific labels for these stages vary from author to author (4, 17, 18). In the first phase (labeled as the cognitive phase by Fitts or the declarative phase by Anderson) the learner collects facts, background information, and general rules related to the skill. The learner attempts to approximate the desired behavior experimenting with and testing out various strategies. The learner usually concentrates on superficial features of a task, ignoring deeper structures or organizing principles (19). Performance is slow and effortful, requiring the learner's full attention. Outside sources of stress or distraction significantly impair or disrupt the learner's performance (20). In this stage learners who speak will frequently verbally rehearse information needed to guide their behavior. During this phase the learner is very dependent on external cues, prompts, feedback and information about the system (e.g., cue sheets). The learner leaves this phase with a basic understanding of the task requirements and a set of strategies. These are not yet fully elaborated or integrated.

In the second stage (Fitts associative phase and Anderson's knowledge compilation phase) the skill is refined or smoothed out (4, 17, 18). Deficiencies in the conceptualization of the task are systematically eliminated and strategies are elaborated. Task components are integrated and whole-task practice begins. Although frequently repeated task components become automatic, task performance remains largely under conscious control continuing to require the user's full attention. The learner continues to require detailed, accurate and immediate feedback regarding his/her performance. It is during this stage that significant improvements in performance are noted, not just in speed and accuracy but also in the ability to anticipate system actions and to generalize strategies used to other tasks. Toward the end of this stage performance gains begin to plateau.

Only in the third stage does the learner achieve skilled or automatic performance. This phase is characterized by gradual improvements during many repetitions of the task. It is the longest stage in achieving skilled behavior. At this stage the conceptual model of the task has been integrated. The user is able to concentrate on the higher level principles of the task. Performance strategies are fine-tuned for the most efficient execution. The learner is able to both generalize the process to similar tasks and to specialize the process for specific task situations. The skill is over learned to the point of automaticity. The skill no longer requires the user's full attention. The skilled learner replaces or supplements external feedback with internalized patterns of expected events which act as referents for successful task completion.



Considerations in Promoting Skilled Use

All team members have a role to play in assisting users in becoming skilled device users, those who:

  • design the system,
  • provide long term support of an access method,
  • assist the user in choosing the most appropriate access system,
  • train the user,
  • evaluate the user's performance, and
  • help the user in making decisions about updating the access system. Important research questions remain at each stage of this process. The role to be played by each of these team members will be discussed. Throughout the following sections it as assumed that the end user is an active team member and primary decision maker at each stage of the process.


 

Design

The most important question for designers of access systems is what properties of an access system make it most conducive to skill acquisition. Human factors literature contains abundant and frequently conflicting design guidelines (6, 21, 22). However, it is clear in the literature that a system that is optimally configured for the novice user is usually not optimally configured for the skilled user (4, 14). The access system therefore needs to be sufficiently flexible to accommodate the learner at all stages of skill acquisition.

Systems that make the required steps to task completion self evident (or operationally obvious as termed by Light and Lindsay, 23), lend themselves well to early acquisition. The early learner requires much of the knowledge needed to operate the system to be externally available (e.g., . prompts or help files displayed in text, graphic or auditory form). The novice requires easy access to help functions and extensive support for error diagnosis and correction. The novice user may need many of the advanced features of the system to be turned off or hidden until the basic system has been mastered. Learners at all stages of skill acquisition benefit from a system that is consistent, predictable and coherent (i.e., having a high degree of internal organization) (24, 25).

Automatic performance is more easily achieved when a system has the following properties:

  1. the user can perform the task without reference to or dependence on external prompts, cues or timing (4)
  2. the system is predictable and relatively stable
  3. the system does not require visual or auditory vigilance (15)
  4. the number and variety of steps required to complete the task are kept to a minimum, and
  5. decisions to be made are kept to a minimum or the decisions to be made are routine, repeated decisions.


Another design consideration which is critical at all stages is error strategies. The possibility of making errors should be minimized as much as possible. Error correction should be simple and direct. There should be no opportunity to make additional errors in the process of correcting an error. It is estimated that even skilled users spend a quarter of their time making and correcting errors. Errors are "disorienting," frequently disrupting automatic performance and therefore requiring conscious intervention before the "rhythm" of automatic operation can be regained.

Unlike novice users, skilled users rely on internalized, or memorized feedback, knowledge and cues (5, 14, 15). Extraneous, external cues are distracting and cumbersome. Help which is not needed is frequently seen as irritating or even insulting to the user. The skilled user should have the option to trim down the system for more efficient use.

 

System Upgrading or Updating

Designers, developers, distributors and manufacturers have one additional responsibility in developing systems that promote skilled users. Once a user has developed skill in using an access system, that skill is jeopardized if the access method is not supported by the manufacturer, if upgrades to the technology are not compatible with the access system that has been learned, or if consistency from one version of an interface to the next version is not maintained. Mainstream computer manufacturers are well aware of this responsibility. The QWERTY keyboard layout, despite its inefficiencies and ergonomic flaws, has survived the evolution of the typewriter and the computer simply because of the large number of skilled users in the market. Alternative access designers do not have the same market forces to encourage them to support skilled use. Many skilled users of directed scanning, quadrant scanning or reed switch keyboards must abandon years of skill development and relearn a new access technique because those techniques are no longer supported by commercially available alternative access systems.

 

Assessment

The team that assists the user in making the best match between his/her skills and an access system should seriously consider all stages of skill acquisition. The process of determining which access system best utilizes the user's physical, cognitive and sensory abilities has been dealt with extensively elsewhere and will not be discussed in this paper (31). If the goal is skill acquisition and ultimate automaticity in controlling the access technique the team must choose an access technique which meets the needs of the user for an extended period of time and not only for the lifetime of the present device. A system that can grow and evolve with the user without abandoning skills mastered in earlier stages must be chosen. The team is frequently faced with the dilemma of choosing an access technique that affords the user immediate functional use without a large learning load but also affords fast and efficient skilled performance. These criteria are frequently in conflict.

It is commonly agreed by researchers that initial performance on a task is not a predictor of skilled performance (4). Thus to base a decision on whether a user would ultimately perform better using Morse code or scanning on initial performance would be greatly misleading. The assessment team which advises the user, must look ahead, considering what properties of an access system are conducive to skilled use (as listed earlier in the design section of this paper). As the time and energy investment of mastering more than one access technique would be prohibitive, the assessment team must extrapolate their knowledge of the user's skills and needs to encompass likely changes which would affect the choice of access systems. Prescription guidelines and modeling techniques that help to predict skilled performance can assist the team in making long term recommendations (32,33). This field also needs to continue to amass a body of data regarding access control at various stages of the skill acquisition continuum. This data would help clinicians predict potential for skill-based use and provide general guidelines for facilitating skilled alternative computer access.

 

Training

If skill acquisition is the goal it becomes immediately apparent that training must be much more systematic and long term than is presently possible in most clinics or assessment centers. The relevant question for the trainer and user is what type of training is most conducive to skill acquisition. There is considerable controversy in the literature on the optimal training strategy (26, 15, 4). Researchers disagree on whether it is better to train in isolation or in context, whether to practice the whole task or its components, whether uninterrupted consistent repetitions of the same task or separated variations of a task are more beneficial, and whether a more structured guidance approach to learning or an unstructured discovery approach is better. Synthesis of the data suggests that the optimal training strategy depends on the user, the task. A mixture of training and practice strategies and schedules is probably required. User preference should play a large part in designing a training paradigm.

Lane (4) lists three methods of decomposing a task for training and practice, these are:

a) Task simplification - here the task is redesigned into gradations of complexity. The user learns and practices a functional but less complex whole task which incorporates all or nearly all the component skills and informational concepts of the complete task. Once the simpler version of the task is learned the next level of complexity is introduced. This approach is also called the training wheels approach, in that the user immediately attains the goal of the task but through a simpler method (27). This approach would likely be beneficial for very young users who require the incentive of acquiring the goal and for tasks where the relationship between task components is an important skill component in itself. In the field of alternative access, Koester and Levine have found that task simplification can promote skill acquisition (28). Two groups of users were introduced to a scanning access technique with word completion prediction. The first group initially used the scanning technique without the word-completion-prediction feature. The word-prediction feature was added after several training sessions had been completed. The second group initially used the scanning keyboard with the prediction feature active. The first group performed better and improved their performance at a faster rate.

b) Skill decomposition - in this method the task is broken down into separate skills or concepts and these skills are practiced one at a time (e.g., hitting a switch, visually searching through a list of words), before being recombined and practiced as a complete task. This approach allows the trainer to determine which task components have been mastered and which haven't. As stated by Lane this approach is also well suited for tasks in which one skill is more critical or problematic than others.

c) Segment decomposition- here an intact sequential segment of the task is isolated and practiced separate from the whole task. This method is well suited for tasks where one segment of the task is more critical or problematic.

There is an increased body of evidence to support the contention that training which promotes optimal initial performance is not necessarily conducive to long term retention and transfer of learning. Variations in the training task, training within context, interference or interruptions in the training schedule, and a larger emphasis on discovery learning all result in poorer initial performance, but lead to greater retention and transfer of learning (15, 26,4).

Just as there are no formulas for an optimal training paradigm, the time taken to become skilled is dependent on the learner (e.g., the skills brought to the learning situation) and the appropriateness of the training methods (4). Most authors agree that the more opportunities the learner has for successful repetitions of a task the greater the confidence in the behavior, and the greater the resistance to disruptions by changes in the task condition (3,4, 14, 21). Automaticity is achieved long after most existing training programs have been terminated.

 

Evaluation

The evaluator should consider more than the objective measurements of speed and accuracy when evaluating performance with a device. Possible impediments to automaticity should also be investigated. These could include inconsistencies or inefficiencies in the functioning of the device, incomplete understanding of the device, or inefficient strategies for completing a task. Possible superstitions and bad habits should also be identified. The user may need assistance in dispelling superstitions and replacing bad habits with more functional approaches.

Mental models should evolve with use. As well as evaluating successful task completion, the development of the user's conceptualization of the system and the task should also be evaluated. Is the conceptualization adequate to handle variations in the task? Can it accommodate new knowledge? Does it enable the user to accurately diagnose problems? Does it allow the user to feel secure about his/her ability to perform the task?

 

Making Changes to the User's System

If the goal is skilled, automatic computer access, the decision to upgrade or change the system cannot be made lightly. Questions the team must answer include: if the system is to be changed, which skills developed in using the old system can be transferred to the new system, how much will the system changes disrupt the performance of the skilled behavior, and, are the reasons for changing the system adequate to justify the learning time wasted and the learning time which must be invested.

It is surprising how minor changes to a tool disrupt automatic task completion, forcing the user to resort to conscious control. Many of us experience this phenomenon when we drive a rented or borrowed car, attempt to speak when our mouth is anaethesized following a dental appointment, or switch from one word processor to another.

Not only the justification for changing the system and the nature of the change need to be considered, but also the timing of the change. For example, if it is recognized that the access system will not meet the user's long term needs, it may be better to change the system early on than to wait until the user has become skilled in controlling the present access system. The user must be given adequate time to master the new system before s/he can be expected to perform at his/her previous level. Also, the user should not be expected to adequately perform secondary tasks with the system (e.g., writing essays in school) before achieving basic mastery.



Case Example

To illustrate the concepts discussed in the previous section, the design, development and use of one access technique will be discussed. The author has chosen to discuss a particular access technique because the technique is intended to meet the needs of a group of users for whom the choice of access techniques is frequently unclear. Commercially available access alternatives designed for this user group have been very unstable in their design and availability. As a result, these users have experienced many changes to their access technique and have frequently mastered a technique only to find it is no longer supported. Case examples will be presented of three users, each in a different stage of skill acquisition.

 

Background

Presently, there are few commercially available computer access systems that effectively exploit the abilities of users who have limited pointing abilities or who are unable to time their actions. These users must accept less than optimal speed or accuracy due to present technical limitations. Users must frequently choose between direct selection or scanning. Many individuals who are presently using direct selection techniques do so with great effort and frequent errors. Accuracy can be improved at the cost of increased physical effort and decreased speed by strategies such as a key delay which forces the user to hold down the desired key longer, key guards, or larger spacing between keys. The most common alternative, scanning, is extremely slow and indirect. Another limitation of scanning is that young children frequently require an access technique before they can conceptualize the operational demands of scanning input (34).

Alternative access techniques that bridge the gap between direct selection and scanning can be grouped into three approaches. One approach is to combine direct selection and scanning in one device. Thus, users would directly select to the extent of their abilities, and then scan to choose items which they are unable to select, or select from a larger group and then scan items within that group (hybrid selection)(35). This approach is not offered in any commercially available device at present but is frequently practiced by graphic communication display users and their communication partners.

Another approach is to use a form of coding where two or more direct selections are used to choose each item. The sequence of actions required to choose an item could be either cued or memorized. In the third approach, termed disambiguation, the user chooses a group of items and relies upon the computer to guess (or disambiguate) which item within the group is intended. Thus the user makes only one direct selection to choose each item, unless the computer has "guessed" wrong. It was felt that the coding approach was most conducive to motor habituation and automatic use.

From the perspective of developing automaticity the coding approach seemed the most promising. An approach using hybrid selection techniques could involve more than one set of operational rules for controlling the access system. It may be more difficult to combine the rules from different selection techniques into a cohesive mental model. By combining two access techniques additional error types could also be introduced. Disambiguation approaches are less predictable and require the user to maintain visual vigilance to possible system errors or wrong guesses. The user is also more dependent on external cues to plan the next action. Disambiguation techniques also depend on linguistic information to guess which targets are intended, ruling out the use of icons or graphically represented objects. While each of these techniques may be most appropriate for specific clients, it was felt that a coding approach lent itself well to motor habituation.

 

Present Applications of Coded Access

The most common code is Morse code. Although Morse code is a potentially fast and efficient means of computer input, several factors restrict its use by children with congenital neurological impairments. Some users have difficulty meeting the motor demands of Morse code. Users must reliably control the timing of one or two actions, to distinguish dits from dahs and to enter all elements of the code within the allotted time. The training requirements of Morse frequently restrict its use because it depends on the user‰s ability to memorize the code (2). Systems less dependent on literacy are frequently chosen when users cannot read (31). Even when the codes are reassigned to symbols, children frequently have difficulty in understanding the operational demands. Morse code in its traditional form is designed for the skilled literate user. The initial learning load is very high and must be mastered before the access technique can be used functionally. Knowledge required to control the system, cues and prompts are not represented externally but must be internalized by the user.

Various modifications to traditional Morse code have been suggested or developed to make it more suitable for AAC users. Vanderheiden (36) proposed a selection technique that uses three reliable actions without the timing requirements usually associated with Morse code (three-switch Morse code). This technique requires three reliable switch sites and does not overcome the training demands of Morse code. The Handicode™ software program uses this strategy.

Providing the Morse code user with a visual display as implemented in the RealVoice™ reduces the need to memorize the code. The choices are displayed in a binary tree structure. The user guides the cursor to the selection by activating the appropriate switch at each branching. Although the necessary knowledge and prompts are externally presented, the perceptual and visual-motor demands of this approach make it difficult for many users to make use of this information.

In another form of coding, users point with their eyes to a group of items on an eye-gaze frame (Etran), subsequently indicating which item is intended within the group by gazing to another group which indicates the location, color or number of the item (2). Clinicians report that this is an efficient selection method whose primary limitation is the listener. As listeners must decode the message, it is reported that even trained listeners are unable to keep up with skilled users. The Etran is infrequently used because the augmentative communication user is dependent upon trained listeners and because of the physical barrier an Etran frame creates between the individual and the listener (37). An electronic system modeled on the Etran is the EyeTyper Model 200.™ This unit displays eight groups of eight items. The user must gaze at the appropriate group and then at the group that corresponds to the position of the desired item. The display remains static. The gaze is detected remotely by a built-in camera system (38).

Several researchers have explored a coding system for users who can accurately target to a telephone keypad. Three letters of the alphabet are assigned to each of nine keys. The simplest application of the keypad requires 2 keystrokes to select each letter (39). The user selects the key to which the desired letter is assigned and then one of the remaining three keys to indicate which of the three letters is intended.

Researchers have demonstrated that selection strategies such as those used with Etran systems are potentially far more efficient than scanning (40, 36). Use of present coding systems is restricted due to one or more of the following reasons:

  • the initial training demands are high because the user must memorize a code,
  • the system is based on traditional orthography and therefore requires literacy,
  • the user must be able to time his/her response, or
  • the possible controlling actions monitored by the device are limited (e.g., eyegaze or pressing keys on a telephone keypad).


 

Statement of the Problem and Design Criteria

A selection technique which:

  • could be effectively controlled through inaccurate or imprecise pointing,
  • supported the learner in all three stages of skill acquisition,
  • could be used with any representational set (e.g., traditional orthography, symbols or pictures),
  • did not require response timing, and
  • could be controlled through a variety of controlling actions,
    was required.


The designers wanted to approximate the conceptual simplicity of pointing, where the user points directly to the choice and not also to another target that forms a code for the desired item. An additional criterion was that error correction should be as simple as possible: there should be no opportunity to make additional errors in the process of correcting an error.

 

Design and Development

An iterative design procedure was used with numerous cycles of prototyping, user testing and design adjustments. Several on-screen keyboards for imprecise pointing were developed. These keyboards were modeled on the ETRAN but made use of the dynamic nature of the computer screen. The keyboards were implemented in an on-screen keyboard program called WiViK. The program runs on any IBM or compatible computer running Windows 3.0 or 3.1. The on screen keyboard can be addressed with any mouse compatible pointing device (including head pointers, touch screens, joysticks, and trackballs) or switches (41).

The first keyboard developed was a straightforward quartering keyboard. In this keyboard all the selectable items are divided into four quarters. The user points to the quarter containing the item they want. That quarter expands to fill the whole keyboard and is again divided into four quarters. The user then chooses the quarter containing the desired item as before and that quarter expands to display one item per quarter. The user then points directly to the desired item. Thus a user who can only point to four targets can accurately pick one of 64 items with three selections. Each selection act involves locating the desired choice and pointing to it. (See Figure 1). The user corrects errors or backs up through the process by activating an external switch. The item entered is echoed by a voice synthesizer.

Quartering keyboard
FIG. 1. Sequence of displays showing expanding quarters when choosing the letter "a" on a quartering keyboard

This keyboard was tested by several users who pointed using an absolute headpointer. It was found that vertical and horizontal head movements were easier to control than diagonal head movements. To allow control through vertical and horizontal movements the keyboard was redesigned. The keyboard was rotated to form a diamond, thereby maintaining the effect of expanding quarters (See Figure 2).

Diamond Quartering keyboard
FIG. 2. Diamond Quartering keyboard. Quarters can be selected using hrizontal and vertical rather than diagonal head movements

When using a headpointer or other pointing device, users indicated a selection by pressing a switch, vocalizing (to activate a sound switch) or pausing over a square for a predetermined time (dwell time). Switch activation frequently caused the pointer to stray from the target. When using dwell time, users were more prone to make unwanted selections (often when the user was resting between selections). Some users had difficulty in maintaining the cursor on the target. To account for this an averaging feature was incorporated into WiViK such that the dwell criterion could be met by passing over the target several times.

Another solution exploited both the form and location of head or hand gestures. In this keyboard named the Threshold Keyboard each quarter of the diamond is bisected with a vertical or horizontal line. To select the quarter the user passes over the line and back into the center of the keyboard. The desired item is therefore selected by a series of three horizontal and/or vertical head movements. The size of the movements required can be adjusted by moving the lines further from the middle of the keyboard.

Some users found the lack of tactile feedback when using a headpointer disconcerting. These users found it difficult to maintain visual vigilance of the position of the cursor on the screen. They preferred to use four mechanical switches to select the quarters. Direct selection of each quarter using 4 mechanical switches was added as an option.

Throughout the iterative process keyboards evolved whose selection technique became less and less dependent on visual cues and visual feedback. The skilled user, selecting through thresholds or mechanical switches, can access the keyboard through fluid habitual gestures whose successful completion is signaled by auditory and tactile feedback.

The experiences of three AAC users will be discussed. Each client is at a different stage of skill acquisition and is using a different permutation of a quartering keyboard. All three clients have a diagnosis of cerebral palsy and do not use speech as their primary mode of communication.

 

User Example 1

The first user example illustrates the incorporation of a familiar metaphor into the introduction of a new computer access system. Benefits of drawing upon the child's world knowledge to assist her in conceptualizing the task are discussed.

The user was a 4 year old girl with no previous computer experience. At the time of the assessment she was enrolled in an integrated junior kindergarten program. Her modes of communication included vocalizations, facial expression, and a communication display that she accessed by pointing with a head-mounted lightpointer and listener-assisted scanning. She gained attention by pointing with the head-mounted lightpointer and vocalizing. Appropriate use of the lightpointer was one of the educational goals identified by her teacher. Her favorite activity was playing with a dollhouse. This dollhouse had a removable roof which gave her an aerial view of the rooms. She enjoyed directing the rearrangement of the dolls and furniture in the dollhouse using her lightpointer.

She was introduced to the original (square) version of the quartering keyboard and selection using a remote headpointer and dwell time. Due to experiences with a lightpointer she had overcome the first cognitive hurdle of using a headpointer, namely associating movement of the cursor on the screen with her head movement.

To link the task to her world view and maintain motivation, the selectable objects on the keyboard were presented as animate beings. The task goal was presented as making the desired item aware that it was wanted or chosen (gaining its attention). The client's experiences with the dollhouse were used as a metaphor. The relatively abstract concepts of groupings were presented as rooms. The effect of expanding quarters was presented as the chosen items stepping forward.

A sample dialogue between the practitioner and the client would go as follows:
"Can you find the one you want? (Client vocalizes yes) Which room is he in? Can you point to his room? (client points to appropriate quarter but not long enough to fulfill dwell time) You know what, they're a little slow maybe you should point a little longer. (Client points long enough to fulfill the dwell time) Good, there, now they are stepping forward, which room is he in now, can you point to that room?....."

Once the basic operation was grasped, the task of predicting which quarter the item would appear in, following expansion, was turned into a guessing game. Thus, this subgoal was separated from the complete task, giving the client more immediate gratification (a successful guess). The client was assisted in "discovering" the principle. "There seems to be a pattern, look, if she's in this corner of the square where does she go?..."

The client responded well to this approach. After eight half hour sessions over a period of two weeks, she was able to anticipate which quarter she would need to point to next, occasionally positioning the cursor without waiting for the screen to completely redraw. However, the activity continued to require her full attention, disruptions made her lose her place or forget which item she wished to select.

The task to be learned was made immediately more accessible, by using a metaphor which: the client was familiar with, had associated task mastery with, and found motivating. The metaphor provided the trainer with a vocabulary and task breakdown appropriate to the child's cognitive level. The metaphor also assisted the trainer in making abstract concepts concrete and within the realm of the user's understanding. Introduction of the new skill was simplified by drawing upon previously mastered subskills (e.g., use of headpointer, gaining attention by pointing and vocalizing).

 

User Example 2

The second example discusses a user who moved from mastery of one system to skilled control of a new system. Changes which disrupt automaticity, positive and negative transfer of skills between tasks, and characteristics of the second stage of skill acquisition, are touched upon.

The user was a 15 year old high school student who had used a directed scanning technique for more than 7 years. The access technique was used to control a word processor (for schoolwork and personal writing), and computer games. He controlled his directed scanning system using three head switches. A switch behind his head toggled, with each activation, between causing the cursor to move up or down. A switch to the right of his head controlled left and right cursor movement, also toggling with each activation. A switch to the left of his head selected the item the cursor was hi-liting. Following selection the cursor remained in the same location. As a result, moving the cursor to any one letter or item did not involve a consistent motor pattern but depended on the location of the previous selection and the state of the two toggle switches (e.g., for a keyboard with 100 items this could mean up to 99 x 2 x 2 different motor paths for each item). The client also made use of diagonal movements of the cursor by pressing the two switches, which controlled movement, at the same time. He frequently chose to wrap around the screen, rather than traversing over the screen, if this was a more efficient path (e.g., taking the cursor beyond the bottom of the screen to cause it to appear at the top of the screen). Despite the complex operation of this selection technique, and the obvious impediments to automaticity, the client had developed considerable skill in controlling the access system. Although there were up to 12 possible paths from one item to the next (when both diagonal movement and wrap-around are factored in) he consistently chose the most efficient path. The client appeared to have internalized the layout of the keyboard. When the screen was adjusted so that the keys could not be identified the client was still able to select items without errors.

Unfortunately the client's access technique was no longer commercially available and his alternative access device was plagued with technical breakdown. Given his considerable skill in controlling directed scanning and given his academic demands the team was very reluctant to change his access technique. The team questioned whether the access technique made most efficient use of his motor and cognitive skills but did not wish to abandon the many years of practice invested in the technique. An alternative keyboard was setup using WiVik which mimicked as accurately as possible the directed scanning system. The client was also exposed to a quartering keyboard that was controlled using 4 mechanical switches.

Once the client understood the operational demands of the two systems, he was given an opportunity to practice using each keyboard for one hour. Comparing his speed and accuracy, when using a) his familiar directed scanning system, b) the system which mimicked his familiar system, and c) the quartering keyboard, revealed interesting and unexpected results. Although he was twice as fast using his familiar system than when using the quartering technique, copy typing using the mimic of his familiar system was slower than the other two systems (see Figure 3). The WiVik-based directed scanning system had an identical key layout, supported toggling of the two switches, and allowed the same diagonal movement and wrap around as his original system. The only differences noted between his familiar system and the WiVik system were: the visual display features (e.g., higher resolution, a larger color palette), the presentation of the cursor movement (e.g., jumping from one item to the next rather than moving smoothly between items), and the sound of the auditory tone used to denote movement from one item to the next. It appeared that these differences were sufficient to prevent transfer of the automatic behavior.

Client's input rate in correct selections per minute using a) his familiar access technique, b) a keyboard mimicking his familiar technique, and c) a quartering keyboard.
Fig. 3. Client's input rate in correct selections per minute using a) his familiar access technique, b) a keyboard mimicking his familiar technique, and c) a quartering keyboard.

The client chose to practice using the quartering technique for a trial period of 4 weeks, while continuing to use his familiar system for necessary writing tasks. During this period the client made rapid gains in speed and accuracy, surpassing performance using his familiar system by the third week (See Figure 4). During the fourth week the client began completing his homework using the quartering keyboard. The task of operating the new access technique was not yet automatic. The client was still heavily reliant on visual prompts to guide his movements, but could use the position of the desired item on the first keyboard display to anticipate which movements were needed next. The client could enter several commonly used functions and letters without visually referencing the keyboard layout. Negative transfer of the habituated motor acts from his original access system did not appear to interfere with mastering the new system. This may be because the system was sufficiently different to prevent interference.

Changes in the client's input rate in correct selections per minute using the quartering keyboard over a three week period of use.
Fig. 4. Changes in the client's input rate in correct selections per minute using the quartering keyboard over a three week period of use.

The weakest component of the client's new system was the mechanical switches. Three types of problems occurred with the switches: one switch did not reliably close on activation; another switch required greater travel than the rest; and the switch signal from two of the switches was noticeably delayed. Unfortunately, the client adapted to consistent or predictable switch problems (greater travel or delay). When the problem was inconsistent or unpredictable (i.e., unreliable closure) learning was disrupted until the switch was replaced.

In this example abandoning a mastered access system was unavoidable. Characteristics of the old access system (the large number of possible steps to complete the task, and the dependence on visual tracking) made it difficult to develop automaticity. Contrary to expectations, the skills did not appear to be transferred to a nearly identical simulation of the mastered access system. This may imply that the skill was not very stable or well integrated. Developing skill in the new access system was aided by the relative simplicity and predictability of the new technique. Required motor acts (holding a switch to move a cursor and accurately timing the release of the switch vs. hitting the appropriate switch in sequence), were sufficiently different to prevent negative transfer. The client's tendency to accomodate consistant problems with the access system hi-lites the importance of diagnosing and eliminating problems or inefficiencies in the technology before they are integrated into the skilled behavior of the user.

 

User Example 3

The third example describes a user who has reached the third stage of skill development. Design changes required to accomodate his skill level are hi-lited. Factors which supported or disrupted his automatic performance of the task are discussed.

The third client was a 36 year old man employed in a supported work environment. He used a diamond quartering keyboard for six months. He controlled the keyboard using a headpointer which detected relative headmotion. He indicated selection using the threshold technique. The main page of his keyboard contained the alphabet, numbers, punctuation and necessary commands, with frequently used words, phrases and system macros on subsequent pages. Subsequent pages were accessed by selecting a menu key which displayed available page keys. The vocabulary and keyboard layout of his main page did not change over the six month period.

Prior to using the quartering keyboard he had used four different access techniques over a ten year period. The client reported that his most successful access means was a reed switch keyboard (the Autocom) which was discontinued by the manufacturer. Arthritis in his thoracic spine prevented him from continuing to use arm movement to access a computer.

The client made rapid gains in his performance with the quartering keyboard during the first three months. During the third month his speed and accuracy plateaued (see figure 5). By the fourth month the user could enter items from the main keyboard page without looking at the display. Frequently the user allowed his application window to occlude the on-screen keyboard and continued working. The client had no difficulty in learning a complex application program while using the quartering keyboard.

Changes in client's input rate in correct selections per minute using a diamond quartering keyboard over a six month period.
Fig. 5. Changes in client's input rate in correct selections per minute using a diamond quartering keyboard over a six month period.

At six months the client's input rate was restricted by the limitations of the access system. He was using a computer with a relatively slow processor. Redrawing of the screen following each selection of a quarter was not instantaneous. As a result he had to wait for the computer before he could make his next move. The client had learned to pace his actions to the speed of the computer.

The keyboard must be redesigned to better accommodate the skilled user who no longer needs the visual cues of expanding quarters. This keyboard could remain static, hi-liting the quarters chosen. If the user has completely internalized the code, only the threshold boundaries, the section hi-lighting and the auditory feedback could be maintained (with an option to display the visual cues when necessary or when accessing additional keyboard pages).

After the sixth month he felt that he no longer required the key labels or visual cues of expanding quarters. The client's performance was measured with several changes to the system. In completing a copy-type task with the key labels removed the client's input rate was 24.3 net selections per minute (the client was asked to correct all errors during the task). This was very close to his performance with key labels (26.2 selections per minute). Interestingly, the client had difficulty choosing the actions required to select a particular letter from a list of possible movements. The client was presented with three choices, in text form and read aloud by the author (e.g. "how do you enter the letter 'a': 1) up, down, up 2) down, down, down or, 3) down, up, up). When the client was asked to make the movements required to retrieve the letter, he had no difficulty. It appears that the sequence was learned and habituated as a motor pattern but not as a verbal pattern.

The client agreed to experiment with other changes to his system to discover what changes would affect his skilled performance. Several changes caused marked degradation in his input rate. Turning off the auditory cues, which signaled successful completion of required movements (with key labels on), lowered the client's input rate to 18.2 net selections per minute. The client appeared to depend on the auditory cues to signal successful completion of each component task. Without the auditory cues the client was forced to reorient his attention to the visual cues. Changing the keyboard layout lowered his input rate to 13.5 net selections per minute. After one hour of practice with the new layout the client's rate went up to 15.8 selections per minute.

The vocabulary, or keyboard layout is frequently the least stable component of a client's system, however, this client's skilled behavior was dependent on a consistent layout.

prompts and cues which make it easier for novice user can slow down and impede skilled performance systems should provide these as options which can be turned off when they are no longer needed

 

User Example Conclusions

Thus each client at each stage of skill acquisition required a different configuration of the quartering keyboard. While the effect of expanding quarters initially made the task more operationally obvious and reduced the number of keys to be visually scanned, this effect became cumbersome in the later stages of mastering the system. The clients moved from reliance on visual cues to reliance on auditory and internalized motor cues. Careful attention to the client's conceptualization of the system appeared to promote skill learning. Consistent system weaknesses were accommodated in the user's behavior leading to inefficient habits, and reducing the impetus to fix the system. Certain kinds of "minor" changes in the access tool caused disruption of skilled behavior.



Conclusions and Future Directions

Although much can be borrowed from skill acquisition theory and research in other fields, we need to build up a body of knowledge regarding skilled behavior in the field of alternative computer access and augmentative communication. The common sense dicta, presented in this paper and by other authors, need to be experimentally verified. Pertinent questions include:

  • How much difference in skill acquisition time and difficulty do the practices recommended her actually make?
  • Which alternative access systems are most conducive to skill achievement?
  • What are optimal times to make changes to the access system?
  • What changes to the access system cause degradation of skilled performance?
  • How can changes in the client's system requirements (caused by maturation or changes in life situation) be accommodated without disrupting the client's transition to skilled alternative access?
  • What skills transfer from one task to the next, and how can this transfer be facilitated?
  • How do predictive systems and disambiguation systems affect the achievement of automaticity?
As well as empirical research specific to this field, practitioners need to share case studies and strategies for promoting skilled users.

Choosing the right access system for a client is only the first step in providing a client with assistive device control. We must insure that the user understands, places appropriate trust and develops skill in its use. Access systems must be designed to meet the needs of the user through all stages of skill acquisition. To maintain skilled use the access system must remain relatively stable. As a result the access system prescribed must meet the client's long term needs. If we expect users to be able to participate fully in social relationships, integrated classrooms and employment, they cannot devote large amounts of cognitive energy to controlling their assistive devices.



Acknowledgments

The author wishes to acknowledge: Fraser Shein, Gil Hamann and Russell Galvin for their technical innovations in programming WiViK; Debra Fels for her assistance in organizing this paper; and Morris Milner and Penny Parnes for their leadership.
The Ontario Ministry of Colleges and Universities University Research Incentive Fund, IBM Canada Ltd., and IBM Corporation are thanked for their generous support of this work.



References

  1. James, W. Talks to teachers on psychology; And to students on some of life‰s ideals. London: Longmans, Green, 1913.
  2. Beukelman D , Yorkston K, & Dowden P. Communication augmentation: A casebook of clinical management . San Diego, CA: College Hill Press 1985.
  3. Colley AM. Cognitive motor skills. In Human Skills 2nd edition Dennis H. Holding (ed.). New York: John Wiley & Sons, 1989: 229-248.
  4. Lane NE. Skill Acquisition Rates and Patterns: Issues and Training Implications. New York: Springer-Verlag, 1987.
  5. Newell KM. Skill Learning. In Human Skills D. Holding (ed.). New York: John Wiley & Sons, 1981; 9: 203-226.
  6. Kantowitz BH, Sorkin RD. Human Factors: Understanding People-System Relationships. New York: John Wiley & Sons. 1983.
  7. Carroll JM, Olson JR. Mental models in human-computer interaction. In: Handbook of Human-Computer Interaction M. Helander (ed.). North-Holland: Elsevier Science Publishers B.V., 1988;2:45-65.
  8. Schmidt RA. The schema concept. In Human motor behavior: An introduction J.A. Kelso (ed). Hillsdale, NJ; Erlbaum.
  9. Schank RC, Abelson RP. Scripts, plans, goals & understanding. Hillsdale, NJ: Lawrence Elbaum Associates, 1977.
  10. Anderson JR. The architecture of cognition. Cambridge, MA: Harvard University Press 1983.
  11. Foss DJ, Rosson MB, Smith PL. Reducing manual labor: An experimental analysis of learning aids for a texteditor. Human Factors in Computer Systems Proceedings. Washington, DC: National Bureau of Standards 1982.
  12. Carroll JM, Mack RL. Metaphor, computing systems, and active learning. International Journal of Man-Machine Studies. 1985; 22:39-57.
  13. Moray N, Muir B. Intelligent Aids, the Theory of Machines, and Trust between Humans and the Systems they Control. Unpublished Manuscript, University of Toronto, 1987.
  14. Salmoni AW. Motor skill learning. In Human Skills 2nd edition Dennis H. Holding (ed.). New York: John Wiley & Sons, 1989: 197-226.
  15. Logan GD. Automaticity, resources, and memory: Theoretical controversies and practical implications. Human Factors 1988; 30(5): 583-598.
  16. Fisk AD, Oransky NA, Skedsvold PR. Examination of the role of "higher-order" consistency in skill development. Human Factors 1988; 30(5): 567-581.
  17. Fitts PM. Perceptual-motor skill learning. In: Categories of human learning A.W. Melton (ed). New York, NY: Academic Press. 1964.
  18. Anderson JR. Acquisition of cognitive skill. Psychological Review, 89(4): 369-406.
  19. Rabinowitz M, Glaser R. Cognitive structure and process in highly competent performance. In The gifted and talented: Developmental perspectives F.D. Horowitz & M. O‰Brien (eds). Washington, DC: American Psychology Association. 1985.
  20. Norman D. Learning and memory. San Francisco, CA: W.H. Freeman & Co. 1982.
  21. Holding DH. Skills Research. In Human Skills D. Holding (ed.). New York: John Wiley & Sons, 1981; 1: 1-26.
  22. Colley AM, Beech JR. Discussion: Changing representations of skill. In: Cognition and Action in Skilled Behavior A.M.Colley and J.R. Beech (eds). North-Holland: Elsevier Science Publishers B.V., 1988; 397-403.
  23. Light J, Lindsay P. Cognitive science and augmentative and alternative communication. AAC 1991; 7(3): 186-203.
  24. Briggs GE, Naylor JC. Relative efficiency of several training methods as a function of transfer task complexity. Journal of Experimental Psychology 1962; 64: 505-512.
  25. Gardlin GR, Sitterley TE. Degradation of learned skills- A review and annotated bibliography. Seattle, WA: Boeing. 1972.
  26. Lee TD, Magill RA. Can forgetting facilitate skill aquisition. InDiffering Perspectives in Motor Learning, Memory, and Control D.Goodman, R.B. Wilberg, and I.M. Franks (eds). North-Holland: Elsevier Science Publishers B.V., 1985; 3-21.
  27. Carroll JM, Carrithers C. Training wheels in a user interface. Communications of the ACM. 1984; 27: 800-806.
  28. See Koester and Levine, this issue.
  29. Minsky ML. Frame-system theory. In Thinking: Readings in cognitive science P.N.Johnson-Laird & P.C. Watson (eds). Cambridge: Cambridge University Press 1977; 355-376.
  30. Mayer RE. Elaboration techniques for technical text: An experimental test of the learning strategy hypothesis. Journal of Educational Psychology. 1980; 72: 770-784.
  31. Lee K S , & Thomas DJ Control of computer-based technology for people with physical disabilities: An assessment manual . Toronto, ONT: University of Toronto Press 1989.
  32. Rosen MJ, Goodenough-Trepagnier C. The Tufts-MIT Prescription Guide: Assessment of user to predict the suitability of augmentative communication devices. Assistive Technology. 1989; 1(3): 51-61.
  33. Horstmann HM, Levine SP. Modeling of user performance with computer access and augmentative communication systems for handicapped people. AAC 1990; 6(4): 231-241.
  34. Light J. A vision of simplicity for augmentative communication systems. Proceedings of the Visions Conference, AI duPont Institute, University of Delaware, DE, 1991; 36-39.
  35. Treviranus J, Shein F, Haataja S, Parnes P, & Milner M. Speech recognition to enhance computer access for children and young adults who are functionally nonspeaking. Proceedings of the 14th Annual Conference of RESNA , Kansas, MO. 1991;
  36. Vanderheiden G. A unified quantitative modeling approach for selection based augmentative communication systems. Ph.D. Thesis, Department of Technology in Communication Rehabilitation and Child Development, University of Wisconsin, Madison, WI, 1984.
  37. Trefler E, & Crislip D. No aid, an Etran, a Minspeak: A comparison of efficiency and effectiveness during structured use. Augmentative and Alternative Communication, 1, 151-155, 1985.
  38. Brandenburg S , Vanderheiden G. Communication, control, and computer access for disabled and elderly individuals. Resource Book 1: Communication Aids. Boston, MA: College Hill Press, 1987.
  39. Witten, I. H. Principles of computer speech. New York, NY: Academic Press, 1982.
  40. Goodenough-Trepagnier C. Rate of language production with a SPEEC nonvocal communication system. Proceedings of the First International Conference on Rehabilitation Engineering, Toronto, ONT, 1980; 92š93.
  41. Shein F, Hamann G, Brownlow N, Treviranus J, Parnes P, & Milner M. WiViK: A visual keyboard for Windows 3.0. Proceedings of the 14th Annual Conference of RESNA, Kansas, MO. 1991; 160-162.