Quiet. Sing-song-y. Robotic. Too fast. Too slow. Despite the contradictions among some of these terms, researchers and clinicians have noticed that these various atypical qualities of spoken language are more common than average among individuals on the autism spectrum. Explanations and practical uses for that observation are harder to come by. But it is clear that there is enormous potential to learn in the area of overlap between autism spectrum disorder (ASD) and speech-language variation.
For example, measurements of voice characteristics and speech patterns could be a valuable tool for research because they are a naturally occurring indicator of what is happening in the brain, noted Julia Parish-Morris, PhD, a postdoctoral fellow in the Center for Autism Research at The Children’s Hospital of Philadelphia, during her research presentation at the inaugural Workshop on Linguistic Approaches to Autism and Neurodevelopmental Disorders (LAAND) May 9.
Dr. Parish-Morris spearheaded organizing the workshop to bring together a small, diverse group of researchers and clinicians who share an interest in this emerging area of research. The workshop was held at the Linguistic Data Consortium (LDC) at the University of Pennsylvania in the week preceding the International Meeting for Autism Research (IMFAR) in Baltimore.
“I hope that bringing together researchers from multiple disciplines looking at autism through the lens of language will lead to new and stronger collaborations,” Dr. Parish-Morris said. “Different people bring different backgrounds and perspectives to the same questions.”
Using Voice as a Measurable Indicator of ASD
Efforts to use voice or speech as a systematic and measurable indicator of ASD or its clinical symptoms are still in their early days. Published empirical research to date on voice in ASD has been “scarce, unsystematic, and contradictory,” according to Riccardo Fusaroli, PhD, an assistant professor at Aarhus University in Denmark who led a meta-analysis of this work. He presented findings from the analysis at the LAAND workshop and has a manuscript under review.
Sustained and systematic research into using voice as an indicator of ASD could yield dividends, however. Dr. Parish-Morris noted that language metrics may be especially useful for measuring ASD because speech is fundamentally social — while many ASD-associated deficits are also social — and most children on the autism spectrum acquire spoken language, or at least vocalize in some way that could be measured even if spoken language is not their primary form of communication.
One way she and colleagues are beginning to explore the use of voice as an indicator of ASD is through analysis of existing recordings of ASD behavioral assessment interviews. Worldwide, thousands of video and audio recordings of Autism Diagnostic Observation Schedule (ADOS) assessments fill the closets and hard drives of autism research centers and behavioral health clinics. This widely used behavioral assessment method for ASD involves play-based interactive interviews between a clinician and patient or research participant. These interview sessions are commonly recorded for reliability and clinical supervision purposes, then stored away.
Seeking to Describe and Explain Patterns of Speech Disfluency in ASD
At the LAAND workshop, Joseph Donaher, PhD, of the Center for Childhood Communication at CHOP, shared one project focused on ASD and lapses in speech fluency that uses recorded speech from ADOS interviews as a data source.
He pointed out that in his specialty of speech-language pathology, differentiating specific types of speech disfluency gives clues to the underlying cognitive or neurological process that a speaker is experiencing. For example, repeating a word commonly reflects that the speaker is struggling to think of the next word or phrase, whereas repeating syllables is a stuttering-like disfluency, which in many cases has a neurological cause.
Working together for the past year or so, Dr. Donaher, Dr. Parish-Morris, and colleagues, have used ADOS recordings to better understand the relationship between ASD and speech disfluency patterns. In their preliminary results discussed at the workshop and presented at IMFAR, children with ASD had an overall similar level of disfluency compared to other children, but the pattern of disfluency type was unexpected. Children with ASD were more likely than the comparison groups to have stuttering-like disfluencies and less-studied atypical disfluencies, such as prolongation of syllables.
Linking Linguistic Characteristics With Clinical Characteristics in ASD
In another project shared at the LAAND workshop, Dr. Parish-Morris and CAR’s director, Robert Schultz, PhD, have teamed up with the LDC at Penn, with collaborators including LDC Director Mark Liberman, PhD and Executive Director Christopher Cieri, PhD.
This CAR/LDC collaboration is processing and analyzing ADOS recording samples in an effort to correlate linguistic factors with children’s clinical characteristics. The project is currently in a pilot phase to assess feasibility and to identify which linguistic factors can be most useful, and it has included ADOS evaluations of 100 children so far. The team has evaluated differences between children with and without ASD in subject matter (such as frequency of using words about friends vs. words about family), rate of speech, and overall word choice.
This type of analysis is potentially extremely powerful because these ADOS recordings come with metadata from the original CAR research studies for which families enrolled — ranging from behavioral data, to genetics and brain scans. (CHOP’s Institutional Review Board approved the reuse of the selected ADOS recordings for linguistic study.)
If the team is successful in its goal of applying machine learning tools to classify and learn from these language samples, then such research could potentially grow into a large-scale enterprise, including larger volumes of speech samples from sources other than ADOS recordings. The LDC’s experience with scalable speech-sampling methods via phone banks could provide the groundwork for such future collections.
Future Aspirations: Continued Collaboration and Shared Data Collection on Language and ASD
Dr. Liberman, who is also a professor of Linguistics and Computer and Information Science at Penn, closed the workshop with a discussion of how a large-scale, multi-site collaboration and data source for speech-language studies of ASD could proceed in the future.
The LDC has a repository of more than 250,000 speech samples collected via phone calls from 25,000 individual speakers over decades of linguistic studies. Open sharing of data about these samples has transformed speech/language research, and similar shared-data repository efforts could be the future of biomedical research, including the linguistic analysis of ASD, Dr. Liberman said.
Dr. Parish-Morris is hopeful that the inaugural workshop on language and ASD will become the first in a series of annual meetings to bring together more researchers as interest in this subject continues to grow.
The McMorris Autism Early Intervention Initiative Fund at CHOP and Penn funded the workshop. Additional speakers were Lisa Blaskey, PhD, a pediatric neuropsychologist in the department of Radiology at CHOP who brought a clinical perspective on language disorders and ASD; Ani Nenkova, PhD, an associate professor of Computer and Information Science at Penn, who shared ongoing research on associations between ASD-like characteristics and perceptions of written text; and Ethan Weed, PhD, of Aarhus University, who shared research on how interactive alignment between speakers differs among speakers on the autism spectrum.