Introduction

Augmentative and Alternative Communications consists of two components: the selection set and the selection method.  The selection set is the message components (words, phrases, concepts) that the user of an AAC device can choose to include in a message (utterance).  It is generally accepted that a user requires a large vocabulary to express nuanced ideas in a range of topics and environments (Weitz, Dexter, & Moore, 1997; Zangari, 2013). A number of studies have been conducted to identify the core vocabulary that would allow communication parity for the user of an AAC system (Balandin & Iacono, 1999; Banajee, Dicarlo, & Stricklin, 2003; Hill, 2001; Marvin, Beukelman, & Bilyeu, 1994; Stuart, Beukelman, & King, 1997). Various strategies have been developed to manage large vocabularies for a person with limited movement skills, including semantic compaction, word and phrase prediction, and communication macros.

The selection method of an AAC device or system encompasses the behaviors of the user must produce to generate the message to be communicated to the communication partner.  The selection method includes both the motor demands of making selections and the cognitive and perceptual processes of identifying the appropriate selection.  This area appears to have been less well studied. For example, the Dynavox Dynamic AAC Goal Grid offers only a single strategy for improving communication speed: increasing the number of buttons. It does not address the relative advantages of scanning, switch encoding, or direct selection with finger or eye-gaze that might be available to the user.  Even within any of the selection methods, physical layout of the choices may have a significant impact of performance.

With the ubiquitous availability of dynamic displays, one unexplored issue of AAC device layout is the relative effects of paging of the selection set (displaying a subset of the choices at a time) versus displaying all of the choices on a single page. Displaying the entire core vocabulary at once, if the user is able to perceive the display, reduces the demand of remembering the location of undisplayed choices, but increases the cognitive and perceptual load of the single display.  On the other hand, using multiple pages of choices requires the user to remember the location of desired targets, but provides, for a given display, larger targets.  Fitt’s Law tells us that larger targets are easier to select than smaller ones. However, it is also known, though not as well quantified, that complex visual fields are more difficult to process than simple ones.

This leads to the following research questions:

1. How do the selection rates differ between a single display of the entire selection set and a paged display of the same selection set when display size is constant?
2. How does the accuracy of selection differ between a single display of the entire selection set and a paged display of the same selection set when display size is constant?

In order to examine the effect of layout on selection speed and accuracy, this study strove to remove the effects of language interpretation from the selection process.

Methods

Subjects

The participants in this study were 25 able-bodied adults age 18 or older. None of the participants had significant experience with using AAC systems or scanning selection.

Instrumentation

Selection Set

All communication messages were generated using Proloquo from Assisted Technology.  Custom communication layouts were created by selecting two symbols from each of the first 20 communication categories in the Layout Kitchen component of Proloquo.  These symbols were randomly assigned to a row and column of the single display layout, which was divided into quarters for the multipage layout (see below).

To minimize the impact of language, each item of the selection set printed only its name, followed by a space.  Hence, the symbol of a dog caused “Dog “ to be inserted into the message.

Selection Layout

We created two layouts using the selection set: The first was a single page display arranged in six rows of 10 cells.  The second layout was a set of four displays with four rows of cells.  The first row contained four cells that allowed the user to jump between pages.  The next three rows of the layout contained five cells each, and replicated a single quadrant of the 60-cell layout.  The displayed size of the layouts on the screen was the same.

Target Message

A set of four target messages was generated by creating a random string of symbols from the selection set.  Each of these messages was approximately 300 symbols long, so that our participants would not be able to complete the message in the allotted time.

Computers

All message input was conducted using Apple iMac computers with 24 inch displays and the standard keyboard as the input switch.  These computers each had Core 2 Duo processors and 8 GB of memory.  The software configuration of the computers, including background utility programs, was identical.

Control Software

The messages were entered using Assistiveware’s SwitchXS program.  The program was configured with a step speed of .5 seconds, and to repeat step when the switch was held down, autoselect when the switch was released for the step period.  This setup was selected so that waiting for SwitchXS to offer the desired row or symbol would not be the limiting factor in message generation.  The .5 second step interval was faster than the typical user was able to process the display, so cognitive/perceptual and motor issues became the limiting factors. 

Procedures

Each participant completed three, 20-minute message sessions with each of the screen layouts. The order of presentation of the layouts was balanced across users to control for learning effects.

At the beginning of each session, the participant was seated at the computer with the mouse placed out of reach, the appropriate keyboard layout open on the screen in the lower right corner, and a blank Microsoft Word document open in the upper right of the screen. A printed copy of the target message was placed on a copy stand to the left of the screen. A digital countdown timer was set for 20 minutes, and placed so that its display was not visible to the participant.

The participant was instructed, “When I say go, I’d like you to copy this message using this keyboard as quickly and as accurately as you can.  Are you ready?  Go!”  At the word Go, the participant began copying the message.  When the timer rang, the subject was instructed to stop.

At each session, the participant was allowed to complete no more than three trials, to control for possible fatigue effects.

Data Analysis

After all of the data was collected, each participant message was compared with the source material for length using Word’s Word Count feature, and for accuracy by comparing the content with the text version of the source message using Word’s Compare Documents feature.  Each block of difference between the original and participant message counted as a single error, whether this was a single wrong symbol or a skipped line.  The speed and accuracy were compared between the two keyboard layouts.

Results

How do the selection rates differ between a single display of the entire selection set and a paged display of the same selection set when display size is constant?

Using the single-page layout, participants were able to enter an average of 193 symbols in 20 minutes on the first trial, and 216 symbols on the third.  This difference was statistically significant (p<.05).  Similarly, with the multipage layout, participants were able to enter 133 symbols on the first trial, and 180 symbols on the last trial.  This difference was also significant (p<.05) indicating that subjects were faster as they learned the layouts.  However, participants started faster, and ended faster on the single page layout than they were on the multipage layout (p<.05), with a difference in of communication rate of approximately 20%.

This would appear to be a clear win for the single page input.

How does the accuracy of selection differ between a single display of the entire selection set and a paged display of the same selection set when display size is constant?

Using the single-page layout, participants averaged, on their first trial, 23 errors, and 20 errors on their final trial.  This difference was  not statistically significant (p>.05), indicating that accuracy was not improved over time.  However, on the paged layout, participants showed an average of 18 errors on their first trial, and only 13 errors on their final trial.  This difference was significant (p<.05), indicating that with practice, participants improved on the paged layout.  Further, the difference in errors between the two layouts, on the final trial, was also significant (p<.05), with the subjects averaging 35% fewer errors on the paged layout.

This would appear to be a clear win for the paged layout.

Discussion

The results of this study would appear to be ambiguous, with the single page layout being preferred for speed (a major limitation of AAC), and the paged layout being preferred for accuracy (a major desire for communication).  However, the magnitude of the difference would seem to indicate a preferred approach.  The gain in accuracy obtained by using paged layout was nearly double the cost in speed, indicating that paged layouts with larger targets are generally superior for communication.

The current study was crafted to remove the effects of language from the results.  The icons had no labels, so that the participant could mentally name them freely, but no series of icons conveyed any meaning, so it was not possible to predict the next icon from the current selection.  In AAC, the inclusion of language may affect the outcome in any number of ways. 

Because language has a high level of redundancy, the inclusion of a small number of errors might not hinder the conveyance of meaning to the communication partner.  This may be particularly true when the communication partner is familiar with the speech patterns of the user and can recognize an utterance that is unlikely to be what the individual was meaning to say. (“I want more fish,” when it is known that the speaker abhors fish.) In this case, the difference in accuracy would be less important.

Because language has internal structure, communication systems can be created that incorporate that structure.  One page might contain people and favorite things; another might contain actions; while a third might include modifiers. With such a structure, the user would be able to predict the likely location of a symbol to convey specific meanings.  This might allow for faster message generation.

Because language might provide improved performance for both types of keyboards, future research should be conducted using meaningful messages, to evaluate how language changes the performance of single-page versus multipage communication layouts.

Conclusions

This study was designed to assess the impact of physical targeting and cognitive/perceptual processing on performance using single and multi-paged communication layouts.  It was designed to, as far as possible, remove the impact of language from the performance.

The participants in this study were all typically developing individuals.  The benefits of larger targets would likely be greater for individuals with motor control limitations or perceptual difficulties, while the benefits of a single screen would likely be greater for a person with learning or memory deficits.

Because different individuals bring different skill levels to the use of AAC devices, there is no ideal solution.  This study does not identify a magical solution for everyone.  It does, however, provide some evidence basis to guide the selection of the layout of communication systems.