Most of you know that we released the first publicly available Haitian Creole statistical machine translation engine last week and have been hard at work making it even better. I am pleased to announce since last night we rolled out a new update to the system and our site which bring several improvements:
1) More training data = better translations. We trained the system on even more training data (including data that we hand translated) which should reflect in better translations. We are nowhere near done yet, and we will continue to work on this.
2) Please don’t forget the broad set of APIs and webmaster resources that are available for those that are building applications and websites to help with the relief efforts. There are several efforts underway to develop mobile apps (using the SOAP or HTTP API) and websites (using the AJAX API). If you are working on something along those lines, leave a link to your app/site in the comments and I will make sure to surface them up here so people can find them more easily.
We will continue to work on improving the system and we wish to thank everyone in the community that has been instrumental in helping us get this much-requested translation engine out of the door. Stay tuned for more announcements!
Also, let me once again point to a resource where you can help with the broader Haiti relief efforts. Please help in any way you can!
Update (1/31): The DIPLOMAT project at CMU in the 1990s was an earlier project to create a Haitian Creole system for DOD/DARPA. As I mentioned in our earlier blog post, our system makes use of CMU’s data from that project.
- Vikram Dendi, Senior Product Manager, Microsoft Translator