[UW Augcomm Home]

[AAC Glossary]

[AAC Resources]

[AAC References]

[Contact Info]



Missing something?

This site has been completely revised, but the old modules are still available. Follow these links to the old modules:
Old vocabulary module,
Old features module
Old funding module


Home > AAC Features > Message Composition > Rate Enhancement

C. Rate Enhancement

Communicating through AAC is always slower than speech. In some situations the difference in rate can be a tremendous obstacle to AAC users. For example, speakers commonly produce bursts of speech at 150 - 200 words per minute (wpm). AAC is always slower, sometimes as slow as 2 to 5 wpm for an individual using scanning access. For this reason, many devices include rate enhancement features.

There are two primary features for increasing the rate of selection and delivery of messages in AAC:

1. Encoding:
2. Prediction

Terminology Note:

The word "encoding" describes one device feature: the use of a code to retrieve a word, phrase or sentence.

However, this single feature serves TWO purposes in AAC:
1) message retrieval (described under Message Composition).
2) rate enhancement.(here)

1. Encoding

If an entire word, phrase or sentence can be retrieved with a one or two symbol code, then communication can happen much faster. There are three different types of encoding, depending on what type of symbol is used for the codes.

a. Icon or Symbol Encoding
Icon encoding uses one or more iconic symbols, for example line-drawings, to stand for a concept or an entire message. The most well-known type of icon encoding in AAC is Minspeak encoding. This is discussed elsewhere in this course as a method of retrieving messages, but it is also considered a means of rate enhancement.

More on Encoding

You can download demo software for the Pathfinder at the Prentke Romich website on their AAC product demo page. I highly recommend it.

All Minspeak devices permit icon encoding:

The Pathfinder has numerous small keys and a bright display with symbols in it.
(click picture to enlarge)
Prentke Romich, Inc.

b. Letter Encoding
Letter encoding is a method used primarily in spelling-based systems. The user can quickly call up a word or a phrase, such as "please help me," with letter codes, such as "PHM" -- the first letters of the words. Another technique uses letters that correspond to categories. For example, "I need suctioning, now!" might be stored as MS for "Medical-Suctioning."

One device that permits letter encoding:

LightWriter is a small device with a keyboard and dual displays
(click to enlarge)
Zygo Industries, Inc.

c. Alphanumeric or Numeric Encoding
Messages can also be stored under combinations of letters and numbers. So "I need suctioning, now!" might be stored as M1 for "Medical-#1" or simply under 1-1 for "1st category-1st item."

One device that permits alphanumberic or numeric encoding:

DynaWrite communication device
(click to enlarge)
DynaVox Systems, Inc.

2. Prediction

All prediction techniques speed up the rate of message composition by offering possible choices or guesses to the user. If the guess is correct, the user doesn't need to continue typing and can go on to the next word. In a device, if the guess is incorrect, there may be additional guesses offered, which then may or may not be accepted. New words are stored in the computer for prediction the next time the user needs them. The choices change as the user continues to type or select items; the better predictors offer increasingly closer choices as the user continues.

Although there is considerable disagreement within the field, it seems reasonable to some practitioners to divide prediction techniques into five different types:

a. Word Completion
(This is sometimes called "current word prediction" or simply "word prediction")
Some devices predict the end of a word, based on the first letters typed by the user. If you use Quicken™ financial software, you've seen this and it is the most common type of prediction in AAC systems. Note that "Word Completion" is a better term than "word prediction" because the software does not predict the entire word, only the ending of the word.

Here is one device that has this type of prediction, in addition to other prediction features.


(click to enlarge)
Prentke Romich Company

(click to enlarge)
Pathfinder screen with word prediction
Prentke Romich Company

b. Next Word Prediction
Some devices with Word Completion also predict words that follow in the text, which I call Next Word prediction. This method predicts words based on word pairs and/or the grammatical rules of the words that have preceded it. For example, the word “kitchen” is likely to be followed by “sink.” The prepositions (e.g. in, to, by) are likely to be followed by articles (e.g., a, an, or the). The word "computer" is often followed by the word "screen" or "keyboard." The best software will learn from your typing over time. So, if you tend to talk about your computer "crashing" then that is what will appear in the list of choices.

There are many devices that use both “Word Completion” and “Next Word” prediction. Here are two systems that have both types of prediction:

(click image to enlarge)
Co:Writer SmartApplet
Don Johnston, Inc.

The Handheld Impact is a small computer that can rest in your hand, shown here with an onscreen keyboard of the alphabet.
Handheld Impact
Enkidu Research, Inc.

c. Linguistic Prediction
Next Word prediction could be taken a step further with software that has information about both the grammar and the meaning of word combinations. Systems with "Linguistic Prediction" offer words that agree in case and number based on the preceding words in the text. For example, the sentence "Yesterday, I..." will be followed by past tense verbs (e.g. went, had, saw, etc.) and not words for the future (e.g. will, won't, shall, hope to, etc.). "She has two new..." will be followed by plurals (e.g. dresses, ideas, friends, etc.) only. Similarly, the sentence "The wrench was ..." will be followed only by words that can describe inanimate objects (e.g. cold, hard, new, etc.) and not by descriptors for living things (e.g. angry, proud, etc.). Presumably, this feature could be turned off by an AAC user who wishes to write science fiction or just use colorful language (e.g. think about what Alice encounters in Through the Looking Glass).

To my knowledge the only software that provides prediction on this level is Co:Writer Software, which is (to some degree) installed on the AlphaSmart 3000 communication system:

AlphaSmart 3000
AlphaSmart 3000
AlphaSmart, Inc.

d. Whole Message Prediction
This is a technique that devices don't use, but communication partners use all of the time. Whether it is with AAC users, or speakers who are not quite as fast as other, there are many partners who guess what the speaker is trying to say and finish sentences. Some AAC users like it when knowledgeable partners do this because it speeds up the communicative exchange. Other AAC users absolutely hate it when partners predict messages.. It is very important to ask each individual whether he or she minds.

e. Icon Prediction
In this technique, the device predicts the next icon or symbol to be used in a symbol sequence used for encoding. The first icon prediction system that we know of was used in Minspeak systems by Prentke Romich. To use Minspeak, the individual must select two or three symbols for each utterance or each word in an utterance. After selecting the first one, there is only a limited set of possibilities for the second symbol and even fewer possibilities for the third symbol. Developers at Prentke Romich felt they could speed up communication by highlighting the symbols that are possible, based upon the previous selection. Users need only look at the highlighted symbols for many of their selections.

(click image to enlarge)
Prentke Romich, Inc.

Go back to Clinical Considerations of Message Composition