�鶹Ӱ��

“Alexa, how can I tell if I sprained my ankle?”

“Siri, where’s the nearest all-night pharmacy?”

“OK, Google, how much sleep do I need to be alert at work?”

In today’s rapidly expanding world of AI-powered virtual assistants, producing content that is easy to find and understand by voice command is more important than ever. Nowhere is that more vital than in health information searches, which constitute 15 to 20 percent of audio queries, says David Kauchak, chair of �鶹Ӱ��’s Department of Computer Science.

Kauchak, whose research focuses on natural language processing, uses the example of a voice query to Amazon’s Alexa virtual assistant. Alexa simply “searches the web, finds a source and reads the source back,” he says. The listener may or may not understand the response, because it was likely written to be read and not heard.

Kauchak and colleagues from the University of Arizona are currently researching ways to improve health and medical literacy in the virtual-assistant world. They have been awarded a grant from the National Library of Medicine to build a tool that will help health-information creators prepare audio-friendly content. Funding for the four-year project continues through 2025.

The group used a first grant, which ran through 2021, to build an online text editor. It gave content creators concrete, scientifically validated suggestions to make written health and medical literature easier for patients and consumers to read. The new grant supports similar work for search results returned for listeners.

“We’re trying to figure out how to modify the written text itself to make it easier to understand in audio form,” says Kauchak. “People listen and relate better to people who sound like them. How might different speakers affect comprehension? How might bias affect speaker choices? Should speech be sped up or slowed down to improve understanding? Should certain words be replaced? The tool will identify parts of text that should be changed and make concrete suggestions for improvements.”

A prototype tool housed on a publicly accessible website allows Kauchak and his colleagues to conduct real-world user testing. “We’re researchers,” Kauchak says. “We do proof of concept. My hope is that between the prototype tool and what we publish, more permanent tools will be created down the road.”