June 09, 2009, 10:53 AM — Google has brought out a translator toolkit that combines the company's machine translation technology with online tools for manually editing content.
An earlier product, Google translate, allowed users to translate and read Web sites and plain text in a number of languages. The new translator toolkit targets users who want to edit and polish the translated text, and share or publish it, said Prasad Ram, center head and engineering director at Google India R&D, on Tuesday.
Google's statistical machine translation technology is trained using parallel streams of text in both the original language and the target language. Editing of machine translated text by users of the translator toolkit will help train the machine translation technology to a higher degree of accuracy, Ram said.
Google's Indian R&D lab developed the new product, which is the latest of a number of tools developed by Google in India to improve access to content in local languages, Ram said.
The products, like Google's transliteration application, are built in India, and are inspired by problems in India, but are designed to have global relevance, Ram added.
The translator toolkit supports 47 languages, including five Indian languages. "We have covered more than 98 percent of the world's population," said Michael Galvez, product manager at Google.
Instead of re-creating content in local languages, Google translate and the translator toolkit allow people to translate existing content, Ram said.
By an arrangement with Wikipedia, users can download a Wikipedia entry into the translator toolkit, translate it into a local language using machine translation, correct and polish the translation and post it to the online encyclopedia.
The editing can be done using a document layout that general users are familiar with, Ram said.
Professional translators can also upload personal files to the translator toolkit, and use the toolkit provided by Google, which includes a glossary, and dictionary, among other features, Galvez said. The local file and the translation can be kept private, he added.
A translation memory database in the toolkit provides lookups to previous translations of similar sentences and text, Ram said.
The toolkit is likely to be useful to publishers of news sites who now can target new audiences with local language editions at a lower cost, Ram said.
