Community-Trained Transcription
A capacity-building model for documentation work. Instead of an outside team building an automatic transcription system for a community and handing it over, a five-day workshop taught language workers to build their own automatic speech-to-text systems to take recorded audio and produce text in the community's own orthography. It was reported at the Sunaŵi workshop as having involved O'odham and Piipaash language workers. Every team succeeded in training a working model and, more importantly, came away understanding how the technology works. The Piipaash effort, in particular, has continued robustly since.
Importantly, the deliverable was not a tool but an ability.
Why it works
- The capacity stays local. When the people who will use and maintain a system are the ones who built it, it's much less likely to stall when an outside developer moves on.
- Community Ownership. Building the thing yourself gives a kind of understanding and control that receiving a finished product does not. It positions community members as agents who can adapt, extend, or rebuild, rather than users dependent on someone else.
- Speech-to-text keeps a human in the loop. The system produces a draft transcription in the community's orthography that a person reviews. This is a good example of technology as an aid to human work, rather than a replacement for it.
What to watch
- It asks more of everyone up front. Teaching people to build systems is slower and harder than delivering a finished tool, and it depends on participants having the time and support to learn. This won't fit every community or every timeline.
- "Build your own" isn't always the goal. Whether a community wants the capacity or just wants a working product depends on the community. This case shows the capacity-building path can work and can last, not that it's always the right one.