Working with Machines for Document Translation_Shanghai Translation Company
E-ging Solutions is a world-leading Shanghai translation company with specialties in Document translation.
Decades ago, the prevailing opinion was that machines could master human languages if they could just memorize enough rules.
Beginning with that idea, programmers tried to cram artificial translation brains full of vocabulary, grammar, and syntax so that a source document could be rendered in a target language as a matter of simple substitution. The idea that languages can be substituted, one for another, on a more or less word-for-word basis is completely intuitive — and completely wrong. A substitution machine can do just fine with “This is my cat” but it would have trouble with Sinatra’s “That Sammy’s a real cool cat.”
Humans don’t master a language by first learning all the rules. We learn by inculcating some very general rules and then refining them as we go along. So too do today’s intelligent machines. In 1961, computer scientists tried to teach their mechanical pupils by feeding them enormous inputs. Today we know that machines, like humans, learn best through millions of small inputs.
If you’re looking to train a baseline MT engine for document translation, more is better. The machine can learn best if it has massive amounts of data to begin with. However, while you want to give the computer all the material you can, you should also keep your topic as narrow as possible. How you write about an industrial process is different from how you would write about a symphony. The machine will do a better job getting your language right if it can work within the same subject domain over and over, building upon what it takes from a massive source of basic information. Below are some additional tips to get you started in preparing a document for machine translation.
Word Choice:
While a rich vocabulary brimming with synonyms and metaphors makes good prose, it can make for horrible translation. Terms from your industry should be used consistently, even repetitively, to give the machine a firm grasp of your subject and its vocabulary. It’s a good idea to create a glossary of terms that occur frequently in your industry and make sure those terms are used consistently in all instances. You might also want to maintain a style guide to ensure that all written material conforms to established conventions in English. The machine will be more successful translating material that is written in the same style and voice consistently than it will be trying to reason through the diverse voices of multiple writers.
Flow:
The very elements that help amplify text for human readers can make it gibberish to machines. The machine wants to think in subject-verb-object simplicity. It assumes that what it reads is arranged in linear, propositional order. Anything that differs from that assumption will scatter the computer brain and result in a nonsensical document translation.
The machine cannot take the place of expert human translation. Tweaks and adjustments and the addition of idiomatic expressions, style adjustments, subtle clarifications, and the like — these are the proper purview of people, not machines.
A Bright Future:
It’s said that the past is the best predictor of what’s to come, and on that basis alone, we can safely assume that machine translation has a bright future. The sheer energy devoted to the field guarantees that it will continue to improve, reaching new milestones and setting new standards for accuracy and complexity. Already, what the most basic free translation programs can do far surpasses what the colossal IBM machines could do 60 years ago. Today’s most advanced translation programs defy the limitations of just a few years ago, and they are approaching a level of precision that makes science fiction practically science fact.
Still, language is as much art as science. The “real” human language needed for document translation might always remain slightly outside the computer’s grasp. The tips above will help your company get the most out of machine translation. That alone will get you started, but the human touch will get you where you want to be: fluent and effective in a world rich with diversity and endowed with a language inventory more intricate and expressive than any machine intelligence could possibly describe, much less match.