Science-fiction almost unanimously takes one vision of our future universe for granted: that all peoples shall be able to communicate with one another.
The technology enabling this is generally known as the “Universal Translator”: enabling interaction, profitable trade agreements, peaceful resolution of conflicts and cross-cultural understanding.
At present, translation and interpreting remain time-consuming tasks, requiring the dedicated attention of highly-trained individuals, who each can only specialise in a limited combination of languages, or even fields. But even as a translator myself, if my job were to disappear overnight, I would indubitably rejoice at the utopian prospects the “Universal Translator” may realise: universal access to information, a level playing field for all! Égalité! But how close are we to achieving this wondrous goal, if it is even possible?
Machine translation has progressed a long way from simple dictionary-based methods pioneered in the 1950s. These have been replaced by context-based methods and advanced statistical tools, such as those used by Google Translate. These technologies promise to continually improve through increased integration of crowdsourcing principles, both imperceptibly in Google Translate, or obviously through Facebook or specialised platforms such as Transifex, which lists technology leaders such as Intel, Nokia, Firefox and redhat among their clients. However, the success of crowdsourced translation shows us this: translation technology only succeeds when it succeeds in making the process more human. Similarly, attempts by major translation companies such as Lionbridge or SDL to employ linguists as “post-editors” of texts lovingly-crafted by machines have been met with great scepticism, and the common remark that the humans must pick up the slack where machines have failed.
Problems will always arise when profiteers seek to separate translation from its intrinsic human element. Indeed, machine translation will never succeed as long as there is no natural language understanding, that is to say, as long as the machine does not understand the intricacies of meaning, grammar, dialect, emphasis, errors and cultural references in a text, it will remain unable to produce reliable translations.
I do not rule out that Universal Translator may one day come into being, but maintain that this development hinges on the simulation of all it means to be human: this machine must be able to interact with our society, understand jokes, think not just objectively but subjectively, creatively, and even sensitively to the character of its audience. It must think like a human.
I question that if we were to reach such an evolved level of human-machine interaction, whether inter-human interaction would remain our top priority? By comparison, research into communication with our fellow primates and earthlings has taken a long-standing back-burner following more exciting developments in space exploration, sending messages into outer space, and finally, artificial intelligence. When we finally develop a new “toy” capable of perfect machine translation of natural language, I doubt our desire for seamless cross-cultural interaction will remain as strong given the toy’s greater potential.
This article was originally written for the International Association of Profressional Translators and Interpreters and was published under the following title: Building the Universal Translator: a challenge for machine translation, human-machine interaction and human nature