Researchers at A-State are Taking on Reading Comprehension, One Font at a Time
JONESBORO – A group of researchers at Arkansas State University have been working together for the past three years to provide better learning opportunities for those with dyslexia.
“We wanted to know if we can use artificial intelligence (AI) to create a brand-new font that could make reading easier,” said Dr. Beverly Boals Gilbert, professor of teacher education.
Nearly a dozen faculty members and students have been working as a team, examining existing fonts to know the path they would need to take to develop a new font.
Researchers include a speech-language pathologist, educators, computer scientists, a data analyst, a web designer, a psychologist, an engineer and a graphic designer.
Gilbert said this project started after a discussion during an open forum hosted by the Arkansas Biosciences Institute (ABI). She said they received a $75,000 grant from ABI for a two-year-period with a no-cost extension for a final third year.
Researchers then began to explore the educational connections between AI and other disciplines. Following an open forum where the topic was discussed, Gilbert said she pondered ways research could be conducted to expand existing summer camps at A-State.
“Whenever I heard the presentations about ways to use AI, I thought, ‘We should use AI for dyslexia.’ We have done some camps for dyslexia, and we have received some grants to do summer programs for individuals who are struggling readers, so we knew what a great need there was,” she continued.
Gilbert worked alongside Allen Hays, director of the A-State Delta STEM Education Center, to better formulate how to approach this topic.
"There was little conclusive data concerning whether fonts worked or not as an intervention for dyslexia," said Hays, an instructor in teacher education. "Frankly, there was not a large body of data on the subject. I did find fonts such as Open Dyslexic were no better than Times New Roman, Helvetica or Arial."
"Dr. Gilbert and I knew we wanted to do something to help students, but we also knew we did not have the expertise in computer science or AI outside of what we had researched."
Arianne Pait, director of clinical services at the A-State Speech and Hearing Center, joined the team with her vast knowledge on the topic paired with her desire to do more for those with dyslexia.
“Dyslexia is the most common neurogenic disorder. It is very underdiagnosed. There are estimates that 20% of the entire population has dyslexia or their brains think about reading differently than the other 80%,” said Pait, assistant professor of communication disorders.
Dyslexia is a word recognition disorder of written languages and can cause impairments in all five domains of language. She said while about two million people in the U.S. have been diagnosed with dyslexia, they think that number could be closer to 40%.
Pait said the project was initially funded with an Arkansas Biosciences Institute seed grant. With that funding they were able to hire student researchers to help with the project.
Dr. Jake Qualls, interim chair of the Department of Computer Science, said they built a regressor model for readability and used the model to predict the characteristics of a font with maximum readability. They worked with the team of researchers to develop a website that would be used for all of the participants.
“We created a website with reading games on it and that is how we collected our data,” said Pait.
With parental permission, students in kindergarten through sixth grade from four Delta schools participated. Students with fewer reading skills, including those in kindergarten and first grade, would see an image and select the word they thought matched. The words used in the game were from Dynamic Indicators of Basic Early Literacy Skills (DIBELS).
Pait said they worked with Leslie Parker, assistant professor of graphic design, along with some of her students, to create images using words provided by DIBELS.
“We spent a lot of time working with her senior classes to make sure the images developed represented the DIBELS words, to give every participant a fair chance at getting it right,” Pait added.
Once they had all of the images to go with the words, Pait said it was time to code the website.
She said a lot of work was put in by Tamra Washington, a graduate communications disorders student and Eli Richmond, a senior undergraduate computer science major, both of Jonesboro, in organizing the content and coding the website.
“The website is a place where all our participants interacted with the systems and all the data we collected came from those interactions. We collected the type of font that was presented to the individual and then whether or not they were able to correctly answer the question,” explained Qualls.
Students in second to sixth grades were shown different sentences in different fonts.
“By going through this in multiple settings with several school districts we have been able to generate a large amount of data from each one of those assessments and from each one of the questions that are presented within that assessment,” added Qualls, associate professor of bioinformatics.
He said they used a gradient-boosted tree model to predict readability based on the font attributes.
“Not only are we looking at which font was answered correctly the most, we are also looking at the attributes that exist within those fonts and then using those attributes to start generating new, unseen fonts based on those attributes,” he continued.
“We had over 300 students take part from four surrounding schools and the A-State Speech and Hearing Center reading boot camp that is hosted each summer,” said Charlotte native Grace Carter Mitchell, a graduate student in the Department of Communication Disorders.
“Of the 134 fonts, I was interested in seeing which were the highest performing and which were the lowest performing. We looked at fonts of which students were able to answer the most questions correctly and which fonts were answered the most incorrectly,” said Qualls.
They found through their research, students had less trouble with thinner fonts and more trouble with thicker fonts.
“One of the nice things about these fonts we have gotten from Google is that they have a list of attributes that are associated with them. They have been judged and measured on a scale of how artistic or dramatic those fonts are,” noted Dr. Jonathan Stubblefield, assistant research professor at the Arkansas Biosciences Institute.
Stubblefield said it was his job to go beyond the fonts and go into the attributes that each font was assigned and discover the most correct ones for those with dyslexia.
“Our investigation is promising in that we can make easy-to-implement recommendations to those who serve those who struggle to read. Our results may also have impact to improve literacy accessibility for all,” said Pait.
“It looks like there really are some tips and some recommendations that we hope to make for people who are interacting and trying to teach those who are struggling to read,” she added.
Throughout the project, the kindergarten and first-grade students answered 2,313 questions, while the second through sixth-grade students answered 6,183 questions.
Gilbert served as the principal investigator, alongside project co-principal investigators Qualls, Stubblefield, Hays, Pait, and Dr. Karen Yanowitz, professor of psychology.
Pait said the project is ongoing and they will continue to recruit school districts to participate in the fall.
“We have a spin-off project where we will be looking at how people with dyslexia write their letters. We are currently looking for grants to help hire more student researchers to continue to expand our reach for both projects,” said Pait.
She said there is no end date in sight for the project. They will continue to gather data and use what they discover to help those with dyslexia. The team is actively pursuing external funding in order to continue this vital research.
Dr. Arianne Pait (far left) stands alongside students who have participated in dyslexia intervention and research.
