There are thousands of refugees in Pittsburgh, who had to leave their home country due to natural disasters, political and religious reasons, and so forth. These people are not usually prepared to live in a new place, where they typically meet cultural and communicational problems and their prior job knowledge and experience is sometimes no longer useful. These people must then learn new languages and be educated to have a job. There are many organizations to help these people settle in a new place and Jewish Family & Children’s Services (JFCS) is one of them. They pick up the Karen refugees at the airport when they arrive, help them find a place to live, arrange medical appointments and provide almost any kind of help they need. The refugees need help in many situations especially in the beginning of their settlement. But unfortunately, as mentioned before, few of them speak English and most of the helpers cannot speak the refugee’s languages. So they need an interpreter for communication not only between them but also between the refugees and helpers that cannot speak the language of the refugees.
For example, a refugee needs an interpreter when he/she meets a doctor to describe his/her symptoms and understand and answer the doctor’s questions. And an interpreter/translator is also needed in uni-directional message delivery because the message creator does not speak the language of the refugees. For example, when a helper wants to remind a refugee of a hospital appointment or notify a school-delay, he/she needs an interpreter to translate his/her words. JFCS is helping the Karen refugees, most of which speak Karen spoken in an area of Burma and there is only one interpreter. Since in most situations interpretation is needed, the interpreter is too busy to help everyone so they need other interpreters as soon as possible.
But because there are few English/Karen interpreters, it is not easy to hire more human interpreters. So they asked Language Technologies Institute(LTI) if we have helpful technologies in this case. In LTI, we have technologies for Natural Language Processing(NLP) and to get the best performance with minimal efforts in limited time, we decided to build a message distribution system consisting of an MT and a TTS system which distributes messages to the refugees on the phone. When an English speaking helper has a message to send to one or more Karen speaking refugees, he/she types in the message in English and the system translates it into Karen, the system synthesizes a Karen speech from the translation message and makes a phone call to the refugees to deliver the speech in Karen. Since calling/re-calling refugees takes a significant amount of the interpreter’s time, we estimate this simple but efficient system can help them substantially.
In this project, we achieved several goals. First, we helped the refugees. Since they were not prepared for the settlement and lack of interpreter resources, we built a system that helped them in the settlement. Second, we applied our language technologies to a real situation. We have learned a lot of language technologies but haven’t experienced much in the real world. By this project, we have been exposed to real world problems and tried to solve them using language technologies we have learned. Third, we built the system as general as possible so that we can apply this to other similar situations. For example, there are many cities where Karen people have settled. If they don’t have a good number of interpreters, our system can help them easily. This system can even help other language speaking refugees easily just by replacing the training data for MT and TTS, and accordingly the MT and TTS components in the system.
