In Hitchhiker's Guide to the Galaxy, a trilogy by Douglas Adams, different species communicate using the Babel fish, a leech-like creature that facilitates instantaneous universal translation. You would think that universal translation would lead to a world of understanding. Unfortunately, by removing barriers to communication, the Babel fish causes conflicts and wars. However, scientists are not letting that cautionary tale stop them from using computer technology to facilitate the translation of text and speech.
While the ultimate goal is to develop a real-time universal translator that can convert speech from one language into any other selected language, United Nations translators don't yet have to worry about their job security, according to Dr. Graeme Hirst, a professor in the University of Toronto department of computer science.
Great advances have been made in speech-to-speech translation over the last decade, and even greater advances have been made in text-to-text translation, he says. But the systems are far from perfect.
In Douglas Adams' The Hitchhiker's Guide To The Galaxy, the Babel fish allows numerous aliens species to communicate, but ironically creates strife and war
When it comes to speech-to-speech translation, several factors come into play. The first is that the device being spoken into must correctly identify the words being spoken. For instance, if the computer interpreted “how are you” as “cow are blue” the translation would be garbled. However, if the computer interpreted your words correctly, it still has to translate them into the selected language.
At IBM's Watson Research Center in Yorktown Heights, New York, scientists have developed hand-held devices that can translate in close to real time. For instance, a scientist can pose a question in Mandarin Chinese and a few seconds later the PDA can ask it in English. Translation devices that have been trained to recognize particular voices and focus on translating vocabularies of certain sectors can be quite effective and are being used in medical, military, tourism and other sectors. However, these devices work best when people speak clearly and slowly and ask simple questions or make simple statements. In other words, we are a long way from the universal small talk translator.
“Translation is a human linguistic and cognitive task that involves understanding ideas and aligning sentences in context,” says Dr. Hirst. Sometimes the beginning of a sentence cannot be translated until the end is known and sometimes a full sentence cannot be translated until the next sentence is spoken. In addition, live translators, let alone machines, can be driven crazy by people who do not always use proper grammar and who often use sentence fragments or change thoughts mid-sentence.
With that in mind, it is no surprise that “translating written work is a lot more accurate,” says Dr. Hirst.
Many text-to-text systems rely on word-to-word translations; however, more accurate systems contain large volumes of text that has been properly translated from the input language to the output language. Hansard, which contains printed transcripts of Canada's parliamentary debates in French and English, is often used in French-English databases. These systems compare words, phrases, and sentences of the text they have to translate to existing accurately translated text and produce a decent translation.
Dr. Hirst had cause to use Google Translate, an online translation application, to translate a Czech Republic high school course outline five years ago. “The translation system fell flat on its face in places,” he said. However, he did the same translation a year ago and the translation was “noticeably better through the refinement of the translation system.”
With that in mind, I used Google Translate to translate a simple passage: “My name is Paul.” This is the translation I received: “Mon nom est Paul.”
“The translation is accurate in terms of reflecting the meaning of the original text,” says Dr. Gaelle Chevalier, owner of SciDocs.com, an English/French translation service based in Toronto. However, she adds that the translation is inaccurate if you consider how a francophone would have written the corresponding sentence. “I would have written, ‘Je m'appelle Paul' instead,” said Dr. Chevalier.
In other words, a French speaker would have understood the translation, but would not have written or spoken the sentence as it was translated.
However, five years from now, text-to-text translation will be better than it is now, just as it is now better than it was five years ago, says David Nahamoo, a research scientist in IBM's Yorktown Height research laboratory.
He doesn't think we will see a real-time universal translator like the Babel fish any time soon, but he can envisage people who speak different languages conducting business over a video conference call using a combination of speech-to-text and text-to-speech technology. There will be slight delays between speaking and translation, but people will be able to discuss issues and come to agreements.
If an agreement is complex, live translators might be brought in to create a final draft of a document before it is signed, but the basis of the agreement will be formulated with the assistance of computer-based translation technology.
