Nectec helps overcome language barriers

Commercial ads 'Berlitz. Break the language barrier'

Commercial ads 'Berlitz. Break the language barrier'

Asian-language translations to follow Thai-English

Language barriers – particularly those between Thai and English speakers – may soon cease to be a major problem in Thailand, following the development of automatic speech translation technology by the National Electronics and Computer Technology Centre (Nectec).

The director of Nectec’s Human Language Technology Laboratory, Chai Wutiwiwatchai, said a Speech Translation Project at the centre had set out to solve the problem of poor English skills in the Thai workforce.

Services developed by the project will now allow speech-to-speech translation of Thai to English or English to Thai via the Internet.

“We started the project to support the tourism industry, since translation will help people to communicate better and create understanding between Thais and foreigners. The tourism industry has an important role because it can generate revenue from international tourism of around US$10 billion (Bt340.54 billion) a year,” Chai said.

Since it began in 2007, the project has successfully developed Thai-English translation for tourism operators. The technology can be either network-based, or a standalone application. At the same time, the project has developed technologies related to automatic speech recognition (ASR), statistical machine translation (SMT), text-to-speech synthesis (TTS) and speech-to-speech translation (S2S).

The Thai-English communication aids for the tourism industry have an accuracy rate of 40 per cent. They recognise more than 100,000 travel expressions and about 14,000 Thai and 21,000 English words.

The project plans to offer Thai-English speech-to-speech translation services for the public via the Internet next year.

Meanwhile, Nectec is also working with the Border Patrol Police Bureau to research and develop text-to-speech and speech-to-speech translation technologies in Thai and the Pattani Malay dialect. This is aimed at giving Thai police an ability to better understand and communicate with people living in the southernmost provinces. Similarly, current research and development aims to provide translation technologies between Thai and the Karen language, to help relations on Thailand’s northern borders.

Chai said the project’s next step would be developing technologies to support communication at international conferences and business negotiations.

“The system [now completed] is very useful for people to learn and understand Thai and English, so as to reduce communication barriers. It takes only five seconds per exchange to achieve the translation,” he said.

Nectec has also joined the Asian Speech Translation Advanced Research Consortium (A-Star), which has eight member countries using nine languages, and is collaborating in a network-based speech-translation communication initiative. The nine languages are Chinese, Japanese, Korean, Indonesian, Malay, Vietnamese, Hindi, English and Thai. The proposed speech-to-speech translation systems network will allow people in the Asian region to share knowledge and resources, including research exchanges, large multilingual travel-expression databases and system components. Moreover, the consortium plans to develop Speech Translation Marked-up Language (STML), with international Web-service standardisation.

The consortium is working collaboratively to gather collections of written or spoken Asian-language material in machine-readable form, create common speech-recognition and translation dictionaries, develop Web-service speech-translation modules for Asian languages and standardise interfaces and data formats for connecting speech translation modules internationally.

Last week, the consortium announced the prototype of the first Asian network-based speech-to-speech translation system. The system can provide real-time, location-free, multi-party communication between speakers of different Asian languages.

Chai said the centre expected that the system would provide language translation between up to four parties at the same time. There are nine possible languages, giving rise to a possible 246 different multiparty speech-to-speech language combinations. These will be developed and offered to the public over the next several years. The Nation