Researchers Submit Jibbigo Translation System to CrisisCamp Haiti-Silicon Valley Campus - Carnegie Mellon University

Researchers Submit Jibbigo Translation System to CrisisCamp Haiti

Carnegie Mellon researchers Joy Zhang, Matthias Eck and Ian Lane participated in CrisisCamp Haiti last weekend, held at Microsoft’s Silicon Valley campus, in Mountain View.

In response to the crisis unfolding in the wake of Haiti’s earthquake’s devastation, the Crisis Commons team mobilized. CrisisCamps were announced for six locations in the United States and around the world – all on Saturday, January 16, 2010: Washington, DC, Silicon Valley, Los Angeles, Brooklyn, London, and Boulder/Denver.

As part of Mobile Technologies, a start-up founded by Carnegie Mellon professor Dr. Alex Waibel, Eck and Lane demonstrated the Jibbigo speech-to-speech translation app designed by Mobile Technologies. Based on their previous experiences from a 2009 exercise in Thailand where their system was tested by the US military for medical dialogs in Thai-to-English and English-to-Thai, they are able to streamline the development of a Haitian Creole translation tool.

“Our goal now is to rapidly produce a speech-to-speech translation system for English-Haitian Creole. The intended use would be simple medical dialogs, but also civil engineering,” said Eck, a Carnegie Mellon research technician who has extensive background in language portability to rapidly support new language pairs and machine translation on small, mobile devices.

Eck and Lane used the event to get in contact with Haitian Creole speakers who can help to build a translation system. Professor Zhang and Dr. Eck have setup a Google Docs document at http://translation4haiti.org to collect translations from volunteers as well as professional translators.

In order to be able to build such a translation system they will need:

  • translations from English to Haitian Creole text (at least 10,000 sentences)
  • speech recordings in Haitian Creole (at least 10 hours)

“All we need is some seed translation data so that our statistical model can learn from the examples and develop a voice-translation system in a very short period of time. In a crisis moments like these, time is critical, and at Carnegie Mellon we have been working on this technology for years. The rapid, deployable voice translation system could really help people in Haiti and the rescue effort,” said Zhang. “With more help from volunteers to build up the data, we could deliver the system earlier.”

If any Haitian Creole speakers would like to participate in the development of this system, please contact haitian.translation@jibbigo.com.

Once a significant number of text translations is available they will contact the volunteers to come in and do Haitian Creole speech recordings.

For more information, visit http://www.crisiscommons.org.