New steps for global translation tools

If you've ever used the internet translation tool, you will probably realize how poor translation technology is compared to other types of computer programs. Punctuation errors, misplaced words and grammar errors can make sentences difficult to understand.

Translating language with words is even worse. No one can develop an automated translation system that can go anywhere like the ability of a translator or interpreter. Translation utilities can only translate certain predefined words and translation sentences just enough for a tourist to book a hotel but it is useless if you want to understand a newsletter.

Europeans want a common translation tool for everyone. With 23 official languages, European Institutes spend more than a billion euros a year translating documents and translating speeches. Business companies across Europe cost more than millions just to understand their business partners.

'Obviously there is a need for this type of technology in Europe and anywhere in the world . it saves time and costs for translation,' explains researcher Marcello Federico.

Federico leads the research team trying to achieve something even more ambitious. Working in the European-funded TC-STAR project, they are studying what is probably the biggest human language technology challenge ever: to receive words from one language and to say in another language .

Picture 1 of New steps for global translation tools

(Photo: www.indiamart.com )


Translate from speech to words

'For humans, translation is difficult. We have to understand both the original language and the target language, and machine translation is much more difficult, ' Federico notes. 'TC-STAR is the first project in the world to solve translation from words to unlimited words.'

For such a system that can translate from any word regardless of subject and context, there are three technologies used and these three are still incomplete. ARS automatic speech recognition technology is used to transcribe speech into text. Language translation technology says SLT translates the original language into the target language. Speech Text Technology TTS analyzes the results.

The TC-STAR project partners develop additional components to handle one of these tasks, creating the foundation for bringing the most advanced translation technology closer, similar to the translation by children. who performed.

One of their major innovations is to combine the results of several ASR and SLT systems together to make the translations and transcriptions significantly more accurate.

Based on the BLEU method, the method of comparing translation between humans and machines, the quality of translation has improved from 40 to 60% throughout the project while up to 70% of words are accurately translated, even if they are not put in the correct position in the sentence.

From words to Chinese newsletters

11 partners - pottery and telecommunications and entertainment companies like Nokia, Siemems, IBM and Sony - have worked with speech recordings from the European Parliament and translated them between English and Spanish. They also work with radio broadcasts and translate them from Chinese into English.

Although the system is still not the same as the translator's accuracy, Mr. Federico believes that, with further research, a speech translator into automated speech will be feasible within the next few years, at least for some simpler language pairs.

During this time, parts developed by TC-STAR project are ready with open source permissions. This project has at least launched a spin-off company and a follow-up initiative. (The spin-off company model is a company that separates part of its activities to establish a new independent company)

Called PerVoice, the company is providing remote automatic phonetic services to companies and government agencies.

'It saves them money and time for automatic transcription of city council meetings or meetings,' Federico notes.

The next project, JUMAS, focused on developing a similar phonetic system to record court cases in court.