Does AAC always use AI?

No. Many effective AAC approaches use no AI or no electronics.

Can someone combine speech and AAC?

Yes. The approach can supplement speech and vary across situations.

Should a language model automatically speak its top suggestion?

A suggestion is not established user intent. Interfaces should preserve the user's control over message selection and output.

AI Glossary / Natural language and speech / Intermediate

AAC

Definition

AAC stands for augmentative and alternative communication: ways of communicating that supplement or replace speech when needed. AAC can include gestures, writing, communication boards, and speech-generating devices. In AI-related speech and language work, language models can support AAC by suggesting words or phrases, while the person using the system retains control over what they communicate.

What it means

“Augmentative” means adding to existing speech; “alternative” means providing another way to communicate. A person may use different approaches in different situations, and their system may combine several methods. AAC is therefore broader than one device, app, or artificial voice.

The intended meaning here comes from the speech-and-language source: assistive communication, not the audio-compression format also abbreviated AAC. Technology can help reduce the effort of composing a message, but successful communication also depends on access, vocabulary, interaction with other people, and the user's preferences.

How it works

An aided AAC system provides a way to select symbols, letters, words, or phrases and a way to present the resulting message. Selection might use touch, switches, eye gaze, or another suitable access method. Output might be displayed text, symbols, recorded speech, or synthesized speech.

A language model can rank possible continuations using what has already been entered. A person can accept, reject, edit, or ignore a suggestion. The prediction is a convenience offered by the interface, not evidence of what the person intended to say.

Design and selection should involve the person using AAC and, where relevant, communication partners and qualified professionals. Individual requirements differ. Reliable access to familiar vocabulary and a way to express unexpected messages can matter more than impressive automatic text generation.

Three practical examples

1. Workplace discussion: a person composes a question using a speech-generating system, with optional word completion to reduce repetitive entry.

2. Everyday interaction: a communication board provides a usable way to express choices or ideas without depending on a powered device.

3. Personalized vocabulary: a system makes the user's preferred names, terminology, and frequently used phrases easier to select while retaining a route to new messages.

Think of it this way

Think of an assistant arranging possible words within easier reach, while the author decides which words belong in the message. This illustrates the supporting role of prediction. The analogy has limits: an AAC interface is part of a person's communication access, so unwanted substitutions or inaccessible controls can directly interfere with their voice and participation.

Worked example

Consider a simplified text-entry example for the word “agenda,” which has six letters. Without completion, entering one letter per selection requires six selections. Suppose the system suggests the intended word after the user enters “ag.” Accepting that suggestion takes one additional selection, for three selections in total.

The selection count falls from six to three, a reduction of 50% for this particular word. That is not automatically a 50% reduction in communication time. The person must locate, read, and decide whether to accept the suggestion.

For an invented timing illustration, assume each selection takes 1.5 seconds. Six direct selections take 9 seconds. Two letter selections take 3 seconds; reviewing the suggestion takes 2 seconds; accepting it takes another 1.5 seconds. The assisted route totals 6.5 seconds, saving 2.5 seconds under these assumptions.

If reviewing a changing suggestion list instead takes 5 seconds, the assisted route totals 9.5 seconds and is slower than direct entry. If the suggestion is wrong and requires correction, the cost can increase further.

This arithmetic shows why evaluating prediction quality alone is insufficient. Selection effort, visual attention, correction burden, message accuracy, and user preference all matter. These are hypothetical interface calculations, not clinical results or estimates of any individual's communication speed.

Illustrative case study

Fictional case: Rowan Design develops an optional prediction panel with AAC users participating in testing. A fluent phrase generator impresses the developers, but participants report that suggestions sometimes change tone or introduce details they did not intend.

The team separates suggestions from the composed message and makes review and editing explicit before speech output. It preserves direct entry and familiar stored phrases, then evaluates whether the panel helps participants express their own messages. The case describes a participatory design approach; it claims no measured therapeutic or communication outcome.

Common misconceptions and mistakes

AAC is not restricted to people who never speak, and it is not necessarily electronic. Using a communication aid also does not justify assumptions about a person's intelligence, understanding, or preferences.

AI-generated text must not be treated as the user's statement merely because the system predicts it confidently. A plausible sentence may be factually wrong, socially inappropriate, or simply not what the person wants to say.

Do not evaluate an AAC feature only through keystrokes saved or words per minute. An interface that produces text quickly but makes correction difficult can reduce useful control. Different users and contexts can favor different arrangements.

Business use

For businesses, AAC is relevant both to inclusive workplaces and to accessible customer interactions. Allow time for message composition, address the person directly, and support the communication methods they choose. Do not require a particular AI tool as the sole route to participation.

Product teams should involve AAC users in defining success and examine privacy, vocabulary ownership, offline availability, and recovery when technology fails. Personalized clinical selection belongs with the individual and appropriate professionals; a glossary or generic software demonstration cannot substitute for that assessment.

Check your understanding

A prediction feature halves the number of selections for one word. Has it proved that communication is twice as fast?

Answer: No. Review time, corrections, access method, and the full interaction must also be considered.

Related terms

Text-to-speech
Human-computer interaction
assistive technology
speech-generating device
word prediction
eye-gaze access
accessibility

Learn next

compare direct entry with optional word prediction from the user's perspective, then study how communication autonomy and correction design shape an AAC interface.

Sources

Primary sources: Jurafsky and Martin, Speech and Language Processing, Chapter 3, AAC application: https://web.stanford.edu/~jurafsky/slp3/3.pdf ; ASHA's AAC practice overview: https://www.asha.org/practice-portal/professional-issues/augmentative-and-alternative-communication/ . Relevant passages were read. Scope is augmentative and alternative communication, not the audio codec. The numerical scenario is original and is not clinical guidance.

Previous: A-optimal design Next: Abductive reasoning

AAC