Work Design: Translating the web a few words at a time

Luis von Ahn presents the Duolingo project. Like Tom Sawyer persuading the neighborhood kids to paint his fence for him, Duolingo exchanges language lessons for help translating the web. Duolingo builds on predecessors like reCAPTCHA, which uses the free labor from proving you are human to translate old books, Duolingo addresses a massive task, breaks it down into small fragments, and aligns the task with a strong motivator. Big project, tiny tasks, strong motivation.

This design looks obvious in hindsight. I think it is harder than it looks.

Selecting the right problem from the universe of problems is hard. You need a problem that lets people feel good about participating, aligning with common values. You need a problem that is already understood by the public or that can be explained in a few words. Your problem should have measurable value in time saved or in money. And the problem should have a major component that can be solved by humans.

Breaking down knowledge work to 5 to 30 second snippets is a fairly new skill, although industrial engineers have designed simple repetitive snippets of work for factories for more than a century. Now we must learn to decompose to exploit what the human mind does well, leaving the rest to algorithms.

Motivation is the last leg of this triangle. We don’t have prior art or proven models to discover contexts that marry the human task to incentives. We have a few examples, like getting access to something valuable, learning a language, or running a cool screensaver. In each example the contributor benefits are immediate, in your face, and powerful enough for high completion rates.

The last project I saw that worked like this was 1-800-GOOG-411. GOOG411 was a directory assistance robot; you’d call for the phone number or directions of US businesses. It used caller voices and behavior to build a vast corpus of speech and search data. Google used the data to engineer its transcription service and to learn about mobile-local search. I loved it; it usually had better answers than expensive phone company operators.

Big project, tiny tasks, strong motivation.

YouTube video: TEDxCMU — Luis von Ahn — Duolingo: The Next Chapter in Human Computation.