Machine translation is on the cusp of delivering convenient yet reasonably accurate tools to decipher foreign languages, writes John Cradden
IT MIGHT seem like advances in language translation technology will soon cancel out the need for kids to learn foreign languages, but there are no signs that human translators will be out of a job just yet.
The field of machine translation (MT) has been around for years but it has improved significantly over the past decade. Now it is being merged with other technologies to give us the promise of highly convenient yet reasonably accurate tools that will allow us to decipher a range of foreign languages.
At the recent Mobile World Congress in Barcelona, Google chief executive Eric Schmidt demonstrated a new prototype of his firm’s visual search application, Google Goggles.
It works with the company’s MT technology, Google Translate, to make a smartphone application that can read a foreign language text taken by a camera photo, such as a menu or street sign, and get it translated instantly.
Google has also confirmed that it is working on a mobile speech-to-speech translation application that it expects to become available within a couple of years.
Using existing technologies in voice recognition combined with Google Translate, the firm aims to have a system capable of understanding a caller’s voice and translating it into something close to the equivalent in a foreign language. But it won’t be anywhere near perfect. It is widely acknowledged that, despite recent advances, automated MT is still crude compared to human translations.
Until recently, most MT technology worked on a rules-based approach to programming, teaching the computer the linguistic rules of two languages and inputting the necessary dictionaries.
This is being overtaken by a statistical approach. This is more like educated guesswork, aided by feeding in source-language data along with their human-generated translations in the target language. As well as huge amounts of data, it also demands lots of computing power, so it is easy to see where Google has spotted an opportunity to muscle into areas previously led by Microsoft, IBM and Babelfish.
But while the search engine giant has been busy capturing many of the headlines in the area of MT, research teams involved in the Centre for Next Generation Localisation (CNGL) have also been working on speech-to-speech translation technology and other MT-related areas.
The CNGL, which is funded by the Science Foundation Ireland, includes academics and researchers from Dublin City University, Trinity College Dublin, UCD and the University of Limerick.
“It’s doable already, especially in limited domains,” says Prof Andy Way, who leads the MT research group at DCU’s school of computing. “Surprisingly, given that speech contains many errors, false starts, hesitations and so on, machine translation quality doesn’t degrade much when confronted with speech input.”
The one variable that is proving a challenge, he adds, is the massive range of speakers’ voices and accents.
Way’s team is also working on smarter ways to translate using the data-driven approach, and merging MT with the translation memory software that many translators now use to improve translation quality and output. “In sum, there’s no reason to be fearful of Google,” says Way.
English-Irish is among more than 50 language pairs that Google Translate can work with, but one Wicklow-based firm has managed to get a headstart by creating its own statistical MT technology for this language pair.
“Machine translation tools are not widely available for this pairing, and there were even fewer when we started out a few years ago,” says English-Irish translation agency Traslán chief executive Donncha Ó Cróinín.
“So we developed our own, based on our own skills and experience in providing language tools and resources for Irish.”
However, advances in machine translation along with greater availability are causing mixed feelings among professional translators.
“On the one hand, translators continue to be sceptical about the quality of machine translation, but on the other hand they are fearful that they will lose their jobs as machine translation quality improves and becomes more pervasive,” says Way.
But he adds that translators will “never be out of a job” because MT quality is not good enough to replace humans. Nonetheless, it can be a useful tool when doing human or computer-assisted translations, as well as for getting the “gist” of something, he says.
“Human translators will soon have to use MT, as there’s a huge bottleneck in the amount of translation that needs doing, but we don’t have enough human translators to satisfy this demand,” says Way. “The economic situation that we all find ourselves in is also putting on huge pressure to cut translation costs.”
In what is clearly a sign of the times, DCU’s School of Applied Language and Intercultural Studies is launching a postgraduate MSc in translation technology. But senior lecturer Dr Dorothy Kenny worries that attention given to developments in MT may put off many language students wishing to become translators.
“If there is a perception out there that machines can translate adequately, the danger is that people will think there is no point in humans learning how to translate,” she says. “This would be very serious for the industry given that there is already a shortage of well-qualified translators in certain markets, despite the fact that pay and conditions are good.”
