To Model or Not to Model: A Critical Look at Aided Language Modeling

A familiar refrain in the field of augmentative and alternative communication (AAC) is “model, model, model.”

The reasoning makes sense. Children who speak hear other people speaking throughout the day. People who use AAC, however, may rarely see anyone else communicating with their AAC system. Aided language modeling (ALM) attempts to address that mismatch. A communication partner selects symbols on an AAC system while also speaking.

There are good reasons to model AAC. But how much do we really know about how well aided language modeling works, particularly for autistic individuals?

What do we mean by “modeling”?

In traditional behavioural instruction, modeling typically comes with an expectation of a response. An instructor demonstrates something and then gives the learner an opportunity to do it: essentially, “I do, now you do.” Modeling may also be combined with prompting, reinforcement, practice, and feedback.

Aided language modeling is often recommended differently, sometimes described as “modeling without expectation.” Communication partners use the AAC system themselves during everyday interactions. The AAC user is not required to imitate the model or produce a response.

There are good reasons for this approach. We don’t want communication to feel like a constant test. AAC users also shouldn’t have to demonstrate something every time a communication partner uses their device. From a learning perspective, however, there is an important difference between expecting a learner to respond to a model and providing a model simply for the learner to observe.

Modeling without expectation relies heavily on learning through exposure and observation. The reality is that this may work better for some learners than others.

What does the research actually support?

When I completed my doctoral dissertation on aided language modeling in 2016, research involving people with complex communication needs was generally encouraging. The evidence specific to autism, however, was much thinner. Since then, the research base has grown, including more studies involving autistic participants.

One challenge remains when interpreting this research. Many successful interventions that include aided language modeling also use other teaching strategies. These may include prompting, reinforcement, structured communication opportunities, or systematic instruction. If communication improves when several strategies are used together, we know the intervention package was effective. It is harder to know how much of the learning resulted from modeling itself.

My own dissertation looked at this question with two adolescents with autism. I used an aided modeling intervention without explicitly prompting them to produce the modeled responses. I also did not reinforce or shape correct responses. One participant acquired only one of three targeted responses, while the other acquired none.

Of course, a study with two participants can’t settle the question of whether aided language modeling works. It does, however, fit with a broader pattern in the research: modeling without expectation, when used alone, may not always be an effective way to teach new skills.

Why might this matter for autism?

Learning from a model requires more than simply being exposed to one. A learner has to attend to important parts of another person’s behaviour. They also have to tell symbols apart, connect those symbols with meaning, and eventually use what they observed in another situation.

Skills related to observational learning include imitation, joint attention, receptive language, and discrimination. These are not necessarily areas of strength for every autistic learner. This may be especially relevant for autistic people with more significant communication and cognitive support needs.

Autistic people can certainly learn through modeling, but we shouldn’t assume that exposure alone will be enough for everyone.

Using someone’s AAC system while talking can provide valuable language input without placing demands on the AAC user. There is value in that even if the person doesn’t immediately imitate what was modeled. But if our goal is to teach a particular communication skill, we also need to pay attention to whether learning is actually occurring. If it isn’t, the answer shouldn’t automatically be to model more.

So, to model or not to model?

To deflect from my own question for a moment, I think the most important responsibility is to effectively teach people to use AAC to communicate.

For some people, aided language modeling may be a key part of that process. They may learn vocabulary, symbol meanings, and ways to communicate through repeated exposure to other people using their system. For others, modeling without an expectation to respond may produce much less learning. More explicit strategies may be needed, including prompting, reinforcement, systematic instruction, or planned opportunities to communicate.

Simply put, the evidence is stronger for interventions that include aided language modeling as part of a treatment package than it is for modeling alone. As a result, it remains unclear how much modeling without expectation, by itself, contributes to learning, particularly for autistic individuals.

Rather than assuming modeling without expectation should be the foundation of AAC instruction, we should start with the individual and ask:

• How does this person learn most effectively? Do they learn readily through observation, or do they benefit from more direct instruction and practice?
• What are they actually learning? Are we seeing meaningful, independent communication or simply familiarity with a particular routine?
• Is progress occurring at a meaningful pace? Could another approach help the person learn more effectively or efficiently?
• What should we change when progress is limited? Rather than simply modeling more, what additional teaching strategies might help?

Ultimately, what matters is whether the person is becoming a more effective and independent communicator.

*The takeaway: Aided language modeling may* be an important part of AAC instruction, but the best teaching strategy is the one that helps the individual learn effectively and efficiently.**