Published on Sep 6, 2004

The National Electronics and Computer Technology Centre (Nectec) is planning to further develop its existing translation software technology to reach a capability close to human translation in the next five years.
It also plans to make the technology capable of translating not only from text-to-text but also from speech-to-speech.
Translation software is regarded by MIT Technology Review as the top priority of 10 emerging technologies that will challenge the world this century. With this in mind, Nectec has put a focus on local translation software technology.
Nectec’s research team has already developed translation software called Parsit, to help people to translate text documents from English to Thai and Thai to English. As the software is just a prototype, it can only translate documents word by word.
Nectec’s director Thaweesak Koanantakool said the technology requires more development.
From the experience of developing Parsit, Thaweesak said the team could learn from errors in the existing version to continue developing the next version. Thaweesak hopes to improve the software by integrating artificial-intelligence technology to make the software learn from its mistakes and correct itself.
“This will improve the capability of the software, giving greater accuracy and allowing it to translate documents in more natural way,” he said.
Not only will the software translate content – mostly academic texts and news – the next version will also cover the translation of literature, feature stories, IT content, law and so on. “The software will become smarter, understanding how to translate each document to match the style of the story, so it can be applied to document translation,” he said. “We hope to build the software so it works very close to human translation. We want to make the software do complete translations and require less corrections from humans,” he said.
However, such a development is not easy and Thaweesak said at least five years is needed to develop the software to be smart enough. Thaweesak also hopes that not only will the software translate text to text, but also speech to speech. He said three key technologies are involved in this development. The first is speech recognition that will help convert voice into text that a computer could read. Then comes translation software, which will change the text from one language to the other. Finally, the third technology converts translated text back into speech, allowing users to hear the translation from the computer.
The three technologies have been developed in Nectec labs for various separate purposes and Thaweesak said if combined they could form a new type of translation software. “We already have speech-recognition software, software translation and text-to-speech, so it’s not impossible for us to ultimately make speech-to-speech translations,” he said.
Initially the translations will be done English-to-Thai as the software to recognise English language is more stable than Thai versions.
“We have to wait for improvements to existing translation software to make further developments,” he said.
With speech-to-speech translation, people’s lives will be easier and barriers to communications across languages can be eliminated. This technology, Thaweesak said, would make the use of IT a part of people’s lives.
pongpen@nationgroup.com
Pongpen Sutharoj
The Nation