Crisis management organisation taps into language skills
When a disaster hits, the first thing that breaks is not the power grid. It is communication. Survivors know exactly what they need and where they are. The people flying in to help cannot read a word of it.
Crisis Commons exists in that gap. The organisation works to provide manpower for the technologies that prove invaluable in emergencies, and it taps into language skills as deliberately as it taps into software skills. The two turn out to be the same problem.
Haiti, and a five-digit number
After the 2010 Haiti earthquake, the requirement was blunt: they needed people who could speak Creole. Crisis Commons helped the Mission 4636 project and found an entire global network of Creole speakers willing to work.
The system they supported was as simple as emergency technology gets. Anyone on the ground could send a text message to the number 4636, describing what they needed and where they were. A remote team then translated those requests so that aid workers could get the right help to the right survivors.
Consider the chain that makes that work. A trapped person types a message in Haitian Creole, using local place names and the shorthand of someone in a hurry. Within minutes it reaches a volunteer in Montreal or Miami or Paris, who renders it into English, adds a location a mapper can use, and hands it to a responder in Port-au-Prince who now knows there are four people under a collapsed school on a street he cannot find on any official map.
The diaspora made that possible. Haiti's Creole speakers are scattered across the Americas and Europe, and for once that scattering was an asset. The translation capacity was global, the crisis was local, and a text message closed the distance.
Pakistan, and the same call going out again
The scale of the disaster in Pakistan generated the same appeal. Crisis Commons and OpenStreetMap both put out requests for people to translate for Pakistan, and volunteers arrived to do work that has become alarmingly routine.
The organisation also runs short-term events called Crisis Camps, gatherings that assemble aid and expertise quickly in the days after a disaster. The Pakistan camp produced a very specific list of outputs:
- Developers for the Sahana disaster management system
- Drupal development for the Disaster Accountability Project
- People locating information and translating SMS messages for Pakreport.org, the Ushahidi project collecting reports inside the country
That last line is the one worth reading twice. Pakreport was gathering raw reports from people living through the floods, in Urdu and in regional languages, and those reports were useless to an international responder until somebody moved them into English. Urdu translation became a form of disaster response, performed by volunteers at laptops thousands of miles from the water.
Aid arrives speaking the wrong language
The uncomfortable truth underneath all of this is that humanitarian aid is delivered overwhelmingly in English and French, and received overwhelmingly in neither. Assessment forms, needs surveys, public health messaging and the notices telling people where to queue for water are drafted by international staff and then, if there is time, translated. Often there is no time, and the information simply does not reach the people it was written for.
Studies of post-disaster response keep finding the same thing: survivors rank information about what is happening and where to get help alongside food and shelter as an immediate need, and it is the need least likely to be met. A relief operation can be well funded, well staffed and still functionally mute.
Where the machines come in
Nobody in this world imagines that machine translation alone can be trusted with an emergency message. A mistranslated location or a garbled description of an injury has consequences that a mistranslated marketing slogan does not.
What machines do well is triage. Combining human language skills with automated translation, there is a real opportunity to take a flood of incoming data and make it more useful: sort it, cluster it, strip the duplicates, flag the messages that contain a location or a medical term, and put the ones that matter in front of a human translator first. The machine handles volume. The person handles meaning.
The bottleneck in that arrangement is data. Systems trained on formal documents perform badly on a panicked text message written in a language with limited resources online. Creole and many Pakistani regional languages sit exactly there. Better machine translation for a crisis depends on collecting real speech and real messages from the people who use those languages, which is slow, unglamorous work that nobody funds until the week after it is needed.
The volunteer model, and its limits
What Crisis Commons demonstrated is that crowdsourcing a translation team in 48 hours is possible. Thousands of people will give their evenings to it. The Ushahidi platform, the OpenStreetMap mappers who drew Port-au-Prince from satellite imagery, and the 4636 translators all proved that a distributed volunteer network can outrun an institutional response.
It also has obvious weaknesses. Volunteers burn out. Quality control is difficult when nobody is checking the work. Attention drains away as the news cycle moves on, usually while the emergency is still in progress. And the whole thing has to be rebuilt from scratch each time, because the network assembled for Haiti does not persist until Pakistan needs it.
The fix is not more goodwill. It is standing infrastructure: rosters of vetted translators in the languages most likely to be needed, agreed protocols for handling emergency text, and funding that exists before the earthquake rather than after it. Crisis Commons showed what the volunteers can do. The gap it exposed is that a formal system for emergency communication across languages still does not exist, and every disaster starts by improvising one.