Using machines to translate languages has been a work in progress since 1947 when Warren Weaver, an American mathematician and civil engineer, wrote the “Translation Memorandum”, which suggested a simple word-for-word approach to programming translations. If you’ve been using A.I. for your translations, you know first hand that the technology hasn’t evolved much beyond that approach.
Many companies, including Google and Microsoft, have invested millions of dollars into online translation. This is consumer-level translation, intended to simplify daily interactions, and it is adequate for limited communication such as asking for directions or ordering at a restaurant.
Researchers at UC San Francisco studied Google Translate specifically, translating medical advice from doctor to patient. The majority of inaccuracies in translations were due to grammatical errors or colloquial terms. For example, a mistranslation happened when the doctor asked the patient to “hold the kidney medicine”, signifying him to stop taking it. When the algorithm translated it into Spanish, the instructions were interpreted as “keep the kidney medication”, and when translated into Chinese, the instructions told the patient to “keep taking kidney medicine”; a mistake the research team considered too life-threatening.
Why then, after so much money and time invested, are A.I. translators still not up to par? The following are three of the most common reasons professional organizations are advised to continue using human translators.
Translations are based on statistics, not an emotional vocabulary
Google builds their translations on statistics; they consider the “best” word to be the most “probable” word. In other words, Google selects a certain word for the translation based on words and phrases that have been previously translated in other documents. It cannot decipher tone, emotion, or feelings, so those principles are not taken into consideration.
There is a continuous struggle with complex grammar
When a translation is based on statistics, it struggles with complex grammatical concepts. English in particular is difficult for Google because of the past and imperfect tenses that we use. The past tense is generally used when speaking about an action that already took place, while the imperfect is for actions not yet completed. This is one of the biggest causes of mistranslations from Google.
The A.I. cannot write for an audience
Human translators understand that your work needs to be tailored for the audience. This includes reproducing the feelings and creativity expressed in the original document. When A.I. chooses a word based on statistics and there isn’t a clear understanding of complex grammar, it often leads to a very robotic, lifeless translation.
Machine translation certainly has its perks, mainly being that it’s free and available in a pinch, which really does make it a good choice for lost tourists. As you can see, however, it isn’t enough when translating legal, medical, or business documents. A good rule of thumb: the more you need to achieve with your translation, the higher the need for professional translation.
More information about choosing the best translation provider can be found in a recent article Choosing The best Translation Provider, which covers full comprehension, speed of translation, and how to ensure that the translation company has a thorough revision, editing, and quality assurance process. Click here to learn more!