For democratizing machine learning—bringing speech and language technology to underserved communities
As someone who lives and breathes data science and machine learning, Kathleen Siminyu understands better than most the awe-inspiring potential of speech and language technologies. From voice assistants to text-to-speech apps, they have the potential to reshape how people interact with their homes, cars, jobs and even one another. But these innovations typically only work across certain languages—leaving them out of reach for millions of Africans.
Siminyu set out to change that. “The goal of my efforts is to ensure that our African languages are not left behind,” Siminyu says. First up: leading a project at the Mozilla Foundation to help computers understand Kiswahili as it’s spoken and written. There are more than 200 million Kiswahili speakers, and by annotating the language for a variety of machine learning tasks, Siminyu’s team is building speech transcription models that help level the playing field. That means Kiswahili speakers increasingly will have access to tools like videoconference apps and talk-to-text services.
Before their models could be built, though, the team had to crowdsource text and speech from hundreds of Kiswahili speakers—a feat that required translating the program vision to a wide variety of participants. Siminyu then collaborates with developers to incorporate new, more expansive models into their services.
“We’re not experts in any of these areas, so we don’t want to say we own these resources,” Siminyu says. “Instead, we’d rather give funding to organizations that already have networks of farmers, maybe networks of table banking groups on the continent—or in a Kiswahili speaking environment—and see if voice can be injected into any part of their process or a workflow.”
Emphasizing collaboration and stakeholder engagement not only ensures accurate feedback from users but helps guide how the team pivots. Another way to build agility into the team? She established bimonthly status updates, to amp up response time.
“The biggest benefit of working this way is realizing that each project will constantly evolve, and we need to be flexible and move with it,” Siminyu says. “If we stay on top of things, we’re still able to complete the project within a reasonable time frame.”
Her work at Mozilla Foundation is just one way this tech influencer is honing her leadership skills. At Nairobi Business Angel Network, she helps empower tech startups, and she’s also collaborating with the African Union Development Agency to establish a blueprint for scaling AI across Africa. Siminyu is a board member of Masakhane, a grassroots org that seeks to spark natural language processing research for multiple African languages. And she also co-founded a community for women in machine learning and data science.
Yet even with that schedule, Siminyu recognizes the power of unplugging to maintain a big-picture perspective amid so much disruption. “Young leaders need to have the ability to disconnect from all the noise, from all the information around them—and just get a sense for their own peace,” she says. “Being able to step away and find your center is a definite must-have skill to keep you focused and quick on your feet.”
Q&A: Kathleen Siminyu on how automation will change project management and why she’d recruit Beyoncé
Fast forward: What’s one way managing projects will have changed over the next decade?
Given a high-level plan, an automated system will be able to create granular tasks and do the day-to-day follow-up work.
What famous person would you want to recruit for your team?
Beyoncé. I admire her work ethic and her ability to outdo herself with each subsequent work.
What podcast are you recommending right now?
Unlocking Us With Brené Brown. I enjoy expanding my vocabulary of feelings, emotions and human experiences while better understanding myself as an individual.
What moonshot project would you most like to work on?
Intelligent tutoring systems that are entirely voice- and audio-based and available in African languages. I’d like to build these for early childhood education to enhance literacy learning and for adult vocational training to tackle the unemployment crisis.