The company has developed a trilingual translation system with voice recognition and synthesis that's compact and light enough to be installed in a cell phone. Unlike existing applications the software doesn't offload processing to a more powerful server on the network but performs the task inside the phone. That makes it quicker and avoids potentially costly data roaming charges when used overseas.

The software is a cut-down version of a PC application already sold by Toshiba in Japan and can translate freely between Japanese, English and Chinese.

In use the software first employs speech recognition to determine the language and what has been said. It then uses one of two translation methods -- statistical machine translation or rule-based machine translation -- to parse the sentence and break it down into its components. The translation is then performed and a speech synthesis engine outputs the result.

Statistical machine translation works on probability and word order to determine the construction of a sentence while the rule-based method utilizes a database of hundreds of sentences and phrases to extract the meaning.

The PC software uses both methods simultaneously but a cell phone lacks enough processing power to run both so statistical machine translation is used for English to Chinese and Chinese to English and Japanese while rule-based machine translation is used for the others.