IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

AI, Young Voices Build Literacy Tools With Regional Accents

Ireland-based SoapBox Labs built a speech-recognition engine based on specific accents in 193 countries, using a “for kids, by kids” approach to reading and writing instruction to make it inclusive.

language processing, AI,Artificial,Intelligence,Entity,Using,Voice,To,Communicate,As,Represented,By
Across the English-speaking world, a speech recognition engine embedded in reading and writing tools for younger students is fostering educational equity by identifying “age-specific” regional and cultural dialects.

SoapBox Labs, an Irish company based in Dublin, spent a decade building an AI-enabled database from voice recordings of unidentified young learners across the United States and 192 other nations where English is either the main language or a common second language. This includes samples of accents to the most local levels — regions, states and individual city neighborhoods. In turn, that information is applied when teachers assess a student’s functional literacy progress, the company’s CEO Martyn Farrows said in an interview Tuesday with Government Technology.

SoapBox’s engine has powered learning tools by McGraw Hill, Scholastic, EarlyBird, Imagine Learning and various other companies for almost four years now, Farrows said. It’s also licensed in some entertainment technologies by PBS Kids and Mattel.

A typical learning scenario, noted by Farrows and illustrated on the company website, involved a first-grade classroom where students use tablets while the teacher views a dashboard on her laptop that displays activity reports for every child in the room. A student sees the words “she is the president” on his screen and reads it back as “she is the pree-sident.” A voice on the device tells him to read it again using the “other ‘e’ sound.” He then reads it correctly on his second try.

“It’s a very personalized and in-the-moment feedback loop,” Farrows said. “We can identify how a child sounds it out.”

By location, the pronunciation of words would vary, and the engine adjusts accordingly.

In the greater New York City and Boston areas, for example, the young learners might not be asked to correct their pronunciation if they read the words “fire truck” as “fi-ah truck.” Likewise, in certain parts of Pennsylvania, Maryland and Delaware, “po the handew” would be recognized when reading the words “pull the handle.” And on the contrary, while Americans generally do not use the soft “t” sound in words like “butter,” “better” or “later” (common U.S. pronunciations are “bud-der,” “bed-der” and “lay-der”), the engine would address such mispronunciations by a student in the United Kingdom.

Farrows acknowledged that speech recognition technology is not new to education but said he’s unaware of other tools that are based on the voices of children, not adults. He said his engine is also in tune with speech patterns and idiosyncrasies of younger learners, which differ from those of students in the higher grades, let alone their teachers.

“In some cases, you are more interested in what they didn’t say instead of what they did say,” Farrows said, adding that the tool can also help screen for dyslexia.

Regarding cultural dialects, the Soapbox engine considers how children communicate differently with their parents and siblings, or even if English is a child’s second language. That context is important to understand the use of double negatives. Example: “you don’t need no money” instead of “you don’t need any money.” Grammar instruction for the written language can be provided without mistaking the student’s intent or miscalculating their literacy comprehension ability, Niamh Bushnell, SoapBox Labs chief marketing officer, said in an interview with Government Technology Tuesday.

“The bias piece is massive. There are too many kids (put) in remedial classes without the appropriate assessments,” she said, adding that the engine assesses the user’s spoken and written language on the teacher’s dashboard, but it does not make judgements or recommendations. “And the teachers can always override the technology.”

Farrows said SoapBox is still working on a Spanish language engine, and the technology will be expanded to teach literacy to older students and adults. He also sees potential for foreign language instruction.

Editor's note: An earlier version of this story misstated the SoapBox Labs chief marketing officer's title.

Editor's note: An earlier version of this story said the SoapBox Labs' speech-recognition engine could diagnose dyslexia. In fact it can flag a student struggling with reading or pronouncing words, but diagnosis requires a credentialed professional.
Aaron Gifford has several years of professional writing experience, primarily with daily newspapers and specialty publications in upstate New York. He attended the University at Buffalo and is based in Cazenovia, NY.