For a brief period, I worked as a Smart Cities Fellow in Providence, RI. I was recruited from the neuroscience department here at MIT, by a newly appointed Chief Economic Officer hoping to incorporate behavioral science into his policy decisions. One of the first programs I was introduced to was Providence Talks, an early life literacy intervention program. Basically, UK put out a widely cited study showing children from lower-income families across the US hear ~30,000,000 fewer words than their peers by the time they reach kindergarten. This word exposure gap leads to an early education performance gap, as measured by literacy. 2/3 of children in Providence start kindergarten already falling short of national literary test benchmarks, and the city struggles with how best to include a largely bilingual child population in public school curriculums which are largely English-only.

The intervention relies on a new tech called a LENA device, essentially a conversation tracker–how many words are heard and said by a child, with a person, with a television, with a radio, alone…these numbers are then given to social workers and educational coaches, who use the data to speak to caretakers, who then use this advice to change the way they speak to children in subtle ways and increase words exchanged. The data, gathered from hundreds of low income homes across the city, shows the program is remarkably effective: 2/3 of participating children have had word exposure increase by >50%.

I worked next door to Providence Talks, from the Economic Development Department. There were early questions that came up in our discussions, different from the purely ed intervention side. Firstly, the system was setting up an efficient sort of audio surveillance, of the most sensitive members (children) of a vulnerable group (largely poor latino families), directly to a government body. The data was parsed purely algorithmically (this was meant to mean with no human influence), we were assured, with no audio recordings saved. Secondly, the system was built on a few assumptions that deserve questioning: that the results from the UK study generalize, that increasing word exposure at home via intervention is the same as living in a home which organically involved more word exposure, that this relationship is causal and not correlative, and that English language learning is necessarily what young Spanish-speaking children need.

The educators running this program were well aware–they gathered data not only on word exposure, but have begun gathering on relationship to longitudinal ed performance; they have begun a LENA based program for ESL, which aims to strengthen Spanish skills as well as English skills. They were largely bilingual themselves, and training coaches who came from the communities they were serving. The PVD Talks group have trouble addressing the LENA device privacy concerns, because none have a background in computation or speech processing, such that the input-output of the device from audio to word count is largely mysterious. But they have executed the largest educational behavioral intervention in the US, in hundreds of homes with thousands of children, on minimal funding from city gov (~500k), utilizing a new technology that is hard to understand, working alongside communities. And as a model, translating research to action at scale–which happened remarkably rarely with educational neuroscience–I think there’s a lot to learn from PVD Talks.

Adam