Realtime crowdsourced translation for emergency response and beyond

2,830 views
2,709 views

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
2,830
On SlideShare
0
From Embeds
0
Number of Embeds
4
Actions
Shares
0
Downloads
33
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Realtime crowdsourced translation for emergency response and beyond

  1. 1. Robert Munro<br />Stanford University & Energy for Opportunity<br />Workshop on Crowdsourcing and Translation<br />10 June 2010<br />Realtime crowdsourced translation for emergency response and beyond<br />
  2. 2. Where I’ve sourced crowds <br />Industry<br />Research<br />Esp Psycholinguistics Munro et al 2010)<br />Social development<br />
  3. 3. Daily potential language exposure<br />On a given day, what is the average number of languages that someone could potentially hear?<br />How has this changed?<br /># of languages<br />Year<br />
  4. 4. Daily potential language exposure<br /># of languages<br />Year<br />
  5. 5. Daily potential language exposure<br /># of languages<br />Year<br />
  6. 6. Daily potential language exposure<br /># of languages<br />Year<br />
  7. 7. Daily potential language exposure<br />You can make a realtime difference in the life of anyone you can communicate with<br />We will never be so under-resourced to do this as we are right now<br />But we can leverage many of the same technologies to crowdsource<br />
  8. 8.
  9. 9. Haiti, January 12<br />Text messaging (SMS) is the dominant form of remote communication in Haiti<br />The existing emergency response systems failed<br />Most text messages were getting through<br />A team of people came together quickly, all connected by Josh Nesbit of FrontlineSMS:Medic, to put together an SMS-based emergency response system<br />We launched in under 48 hours<br />
  10. 10. The responders don’t speak Kreyol<br />FanmimwennanKafou, 24 Cote Plage, 41A bezwenmanjeakdlo<br />MounkwensenanSakreKènanPòtoprens<br />Ti ekipmanLopital General genyenyopakaminmfè 24 è<br />Fanm gen tranche poufè yon pititnanDelmas 31<br />
  11. 11. The responders don’t speak Kreyol<br />FanmimwennanKafou, 24 Cote Plage, 41A bezwenmanjeakdlo<br />MounkwensenanSakreKènanPòtoprens<br />Ti ekipmanLopital General genyenyopakaminmfè 24 è<br />Fanm gen tranche poufè yon pititnanDelmas 31<br /><ul><li>My family in Carrefour, 24 Cote Plage,41A needs food and water
  12. 12. People trapped in Sacred Heart Church, PauP
  13. 13. General Hospital has less than 24 hrs. supplies
  14. 14. Undergoing children delivery Delmas 31</li></li></ul><li>Making a difference, from anywhere<br />Translate 1 message = help 1 person<br />
  15. 15. Messages: ~1000 per day<br />
  16. 16. The process<br />
  17. 17. The process<br />
  18. 18. The process<br />Average turnaround = 10 mins<br />
  19. 19. At launch – utilizing volunteers<br />Coordinate response<br />Response<br />Message to ‘4636’<br />Translation and geo-location<br />@Tufts<br />Refine coordinates and identify actionable items<br />Volunteer translators<br />
  20. 20. Today – creating jobs<br />Coordinate response<br />Response<br />Message to ‘4636 / 177’<br />Translation and geo-location<br />@Tufts<br />Refine coordinates and identify actionable items<br />Paid workers in Haiti<br />
  21. 21.
  22. 22. Collaborating organizations<br />
  23. 23. Local knowledge counts<br />Apo<br />Dalila<br />Haiti responders<br />(18.4957, -72.3185)<br />Workers collaborating to find locations:<br />Dalila: I need Thomassin Apo please<br />Apo: Kenscoff Route: Lat: 18.495746829274168, Long:-72.31849193572998<br />Apo: This Area after Petion-Ville and Pelerin 5 is not on Google Map. We have no streets name<br />Feedback from responders:<br />"just got emergency SMS, child delivery,USCG are <br />acting, and, the GPS coordinates of the location we <br />got from someone of your team were 100% accurate!"<br />The ability for someone to make a real-time difference at any other place in the world:<br />Apo: I know this place like my pocket<br />Dalila: thank God u was here<br />‘here’ = anywhere<br />
  24. 24. Local knowledge counts<br />“Rue Casseus no 9 gen yon santkap bay swenakmounkiblese e mounkibrile”<br />Street Casseus no 9, there is a center that helps people that are wounded or burnt<br />Marc, Union Haiti:<br />“I will pass it on that is my cousins hospital.”<br />
  25. 25. A person alone is not in a crowd<br />
  26. 26. Ushahidi – Jan 14<br />
  27. 27. Ushahidi – Jan 17<br />
  28. 28. Ushahidi – Jan 17<br />
  29. 29. Feedback - responders<br /><ul><li>Clark Craig of the Marine Corps:
  30. 30. “I cannot overemphasize to you what the work of the Ushahidi/Haiti has provided. It is saving lives every day.”
  31. 31. Secretary of State Hillary Clinton:
  32. 32. “The technology community has set up interactive maps to help us identify needs and target resources. And on Monday, a seven-year-old girl and two women were pulled from the rubble of a collapsed supermarket by an American search-and-rescue team after they sent a text message calling for help.”
  33. 33. Craig Fulgate, FEMA Task Force:
  34. 34. “[The] Crisis Map of Haiti represents the most comprehensive and up-to-date map available to the humanitarian community.”
  35. 35. Ushahidi@Tufts
  36. 36. “The World Food Program delivered food to an informal camp of 2500 people, having yet to receive food or water, in Diquini to a location that 4636 had identified for them.”</li></li></ul><li>>1000 Volunteers<br />Apo:<br />“When I found out about the project I thought it was a really good thing because I could help people, instead of just donating to charity”<br />Sarah, Union Haiti:<br />“You kept us going”<br />
  37. 37. Crowdsourced mapping – Jan 12<br />
  38. 38. Crowdsourced mapping – Jan 23<br />
  39. 39. Preparing for the future<br />Microsoft Research have built new MT models with the SMS (Lewis 2010)<br />Issues for MT<br />Needed to re-translated (too much paraphrasing)<br />Needed to normalize (too many abbreviations)<br />
  40. 40. NLP for social development<br />Bantu<br />Beyond Haiti and emergency response<br />Those with the least resources have the most to gain<br />Chichewa<br />
  41. 41. Professional communications<br />Partnering with FrontlineSMS:Medic and a clinic in rural Malawi<br />Classifying medical text messages (Munro and Manning 2010)<br />
  42. 42. Why apply NLP to medical SMS?<br />Identify outbreaks early<br />Malaria Prevalence Model, MARA<br />Route important data quickly<br />Save time<br />2 doctors for 250,000 patients<br />= 5 seconds per patient per year<br />
  43. 43. Variation is the norm<br />Keyword-based models are suboptimal<br />
  44. 44. Example<br />
  45. 45. The potential<br />medicine, banking, market information, literacy …<br /># of languages<br />Year<br />
  46. 46. Thank you<br />References<br />Lewis, Will. 2010. Haitian Creole: How to Build and Ship an MT Engine from Scratch in 4 days, 17 hours, & 30 minutes. Annual Meeting of the European Association for Machine Translation<br />Munro, Robert. 2010. Haiti emergency response: the power of crowdsourcing and SMS. Relief 2.0 in Haiti<br />Munro, Robert, Steven Bethard, Victor Kuperman, Vicky Tzuyin Lai, Robin Melnick, Christopher Potts, Tyler Schnoebelen and Harry Tily. 2010. Crowdsourcing and language studies: the new generation of linguistic data. Workshop on Creating Speech and Language Data with Amazon's Mechanical Turk.<br />Munro, Robert and Christopher Manning. 2010. Subword Variation in Text Message Classification. Annual Conference of the North American Chapter of the Association for Computational Linguistics<br />

×