Speech Therapy That Doesn’t Tell Autistic People ‘Your Brain is All Wrong’

The popular view of speech therapy looks something like this: A person, often a child, has problems talking, so they attend repeated sessions with a professional who works to “fix” them so they speak more like the rest of us.

A multidisciplinary University of Maryland team is working to replace that idea of speech therapy with a more inclusive approach—one that leverages artificial intelligence (AI) to help speech therapists and their clients set goals based on the client’s specific daily challenges and strengths. It’s supported by a $86,036 Cross-College Collaborative Award from the Artificial Intelligence Interdisciplinary Institute at Maryland.

“We know that as a population, not a lot of autistic adults are getting speech therapy,” said research team member Sorrah Edwards-Thro M.A. ’26, who’s studying speech-language pathology. “At the same time, they’re experiencing high rates of loneliness and mental health challenges and a struggle with connecting and communicating in a variety of settings, like the workplace, romantic and family relationships.”

One barrier, said Edwards-Thro–who is autistic–is the unintentional message from the field that the goal of speech therapy for neurodivergent people is to essentially teach them to be neurotypical. It’s an idea that’s both insulting and impossible.

“It’s telling people ‘Your brain is all wrong, and you need to change everything about how you talk,’” he said. “How can we do it in a way that’s more affirming and more helpful?”

The research team’s current project—which has roots in an earlier UMD Grand Challenges Team Project Grant—aims to accomplish that by giving autistic clients an active role in analyzing communication difficulties and shaping the design of their own therapy. The key to this is using a variety of AI approaches to transcribe and objectively analyze conversations between autistic people and friends or family members.

“We’ll use the AI to identify important moments in the conversation for more analysis about issues that are arising with communication,” said Associate Research Professor Shevaun Lewis Ph.D. ‘13, a language science researcher who directs the University of Maryland Autism Research Consortium. “We want the client to feel a sort of agency in the process, and providing these AI-annotated conversations lets them work with a speech therapist to make a decision about what to work on.”

For example, the collaborative focus might center on how to manage different topics in a conversation: Autistic people may not pick up on non-autistic people’s social cues about when they’re interested in a topic or weary of it, while non-autistic people sometimes don’t understand how autistic people share information in an attempt to connect. By analyzing a conversation in more detail, speech therapists and their clients could identify what each person contributes to the conversation, and where they might be missing the other person’s intentions. 

Although AI has the potential to provide analysis and insight, its flashy recent forms present challenges as well, Lewis said. The large language models (LLMs) that power the new wave of advanced chatbots frequently exhibit biases that arise from the internet data that AI companies built the models on. “LLMs can be pretty judgy about what they think is the norm,” Lewis said. “We want to make sure our tool is not biased toward non-autistic ways of conversation.” 

The team also includes Clinical Professor Kathy Dow-Burger ’86, M.A. ’92, Friedman Family Director of Neurodiversity and Autism Transition Services in the Department of Hearing and Speech Sciences; and Rida Cuddy M.S. ’26, a master’s student in applied machine learning. 

Because a significant number of autistic people also have intellectual disabilities, the team will also be recruiting people who have both diagnoses, Dow-Burger said.

“Often, therapy done for those with intellectual disabilities is very rudimentary, focusing on things like taking turns in conversation, which is not necessarily something that is going to benefit relations and quality of life like we’re talking about here,” she said. “So I appreciate Sorrah’s dedication to being very inclusive in planning this tool. Because that’s why we’re really here—to be inclusive of everyone’s situation.”