In 2013, Joaquin Phoenix starred in Her, a science fiction film about a man interacting with his new computer AI, the voice of which was played by Scarlett Johansson. Johansson’s character could speak like any other human; she was able to learn extremely fast, display emotion and express humour, and was unrecognisable as a machine to the extent that Phoenix’s character developed strong feelings for her as the film progressed. It was a looking-glass into a world where humans and robots were equals and partners in a world where nobody so much as batted an eye.
It’s now 2019, and we haven’t yet reached the scary levels of interaction with computers that we saw in Her. However, there is no denying now that the language learning capabilities of machines have advanced to a level that none of us could have ever imagined.
With the near universal daily use of online translation services, the most popular being Google Translate, any person with access to a keyboard and an internet connection can now translate into every language at the click of a button – including Klingon. But how good are machine translations, and do they compare to a trained human?
The first commercially used system of machine translation was to directly translate individual words in order to form a new sentence. This format, named the ‘statistical method’, makes sense in theory, however falls short as it ignores the fact that texts are extremely complex structures which rely on a whole host of different variables such as semantics (meaning), syntax (word order and structure) and context. Having seen our own fair share of #GoogleTranslateFails, we have all seen how poorly these translations can come out. When things start to get even more complicated, such as with idiomatic, metaphorical or hypothetical language, these systems begin to fail completely.
For example, ‘the apple doesn’t fall far from the tree’ (a metaphor denoting similarity between relatives), translates word-for-word into German as ‘der Apfel fällt nicht weit vom Baum’ when translated through Google. Although grammatically correct, Google has missed the context here. In German, it is only the trunk (Stamm) of the tree which is referred to, opposed to the entire tree in English. It is elements like this that a human translator has always been able to pick up on and that which has always given them the advantage in the field.
In recent years, these rudimentary forms of translation have begun to change, notably with the rise of Deep Neural Networks as the preferred method of machine translation. In addition to the computer analysing a text word by word, machines are now able to analyse every word in a sentence (alongside the preceding and proceeding sentences) collectively. Moreover, software inclusive of a Long Short-Term Memory (LSTM) is currently in development. LSTM essentially forces the software to remember a sentence, run equations simultaneously to understand it, and use the overlapping results to produce a more accurate translation, which has shown to reduce errors by up to 60%!
So where does this leave translation companies? Are translators a thing of the past?
We say: absolutely not. But just as machines are learning and adapting to their environments, so must we.
Lingua-World has recently developed a translation delivery service model which draws upon both human and machine translation, namely the partial use of the world-leading translation tool, DeepL, and the benefits to our customers by offering this service are extensive.
After being run through DeepL, this rudimentary translation, still littered with contextual, grammatical and continuity errors, is used as a base model for one of our expert translators to work with and deliver a second-to-none translation. The potential time savings amount to around 60%-70% faster than a typical human-only translation, and costs can be brought down significantly too, coming in at around 20%-30% lower than our current model allows. All of this is achieved without a drop in quality. Of course, our principles remain unchanged – check, check, and check again via our three-step translation process to ensure that the end-product is 100% right, 100% of the time.
We may be a while away from having a regular conversation with an AI like in Her, but in a world getting smaller, faster and more digital by the day, agencies like Lingua-World need to keep their eye on the ball and adapt to the ever changing environment in which we work. Instead of ignoring it, Lingua-World leans in and embraces technological change, seeing the benefits to us and our customers instead: a tool we can use to deliver higher quality translations, at a lower cost, in record time.
Contact Lingua-World now for a precise and customised quote on all translation work.
Contact Lingua-World now for a precise and customised quote on all interpretation work.