Letting a machine translate or not?


Machine Translation (abbreviated as MT) literally refers to a translation process done by a computer without human involvement. The history of MT can be traced back to the 1940s.  

Each year, the volume of multilingual content and communication increases dramatically. Demand for delivering information or communicating in multiple languages is soaring. According to an ITU (United Nations International Telecommunications Union) report, the latest statistics show that the number of internet users will reach 3 billion by the end of 2014, and two-thirds of them will be from developing countries. Other current studies show that people are more likely to engage with and buy products and services that are marketed to them in their own language.

MT has developed quickly in recent years, and more and more free tools have become a part of everyday life at work and at home. Popular services include Google Translate, Skype Translator and so on. These services help with basic communication, improve translation productivity, and most importantly reduce time and cost.

Does all of this have any impact on the translation industry? Does it make us think it is time to abandon human work? My answer would be yes and no.

Yes, it does impact our industry. More and more translation companies are developing and customizing their own MT platforms. Transverbum’s MT system combines with website translation. If an update is made on a customer's website, we can extract it automatically and have it translated on-the-fly through MT. Because we are 'training' our MT system with proper terminology and other language resources (select online dictionaries, client-specific translation databases etc.), the benefits can be substantial—accuracy with very high efficiency in terms of time and cost.

So why is my answer also 'no'? As the Chinese often say: 'wanbian buli qizong', or 'the methods used may vary, but the principle stays the same'. The basic method and process of translation will not change. Humans still need to be involved since machine translation can be only used for certain steps of the translation process. There will normally be quite a bit of preparatory work before the machine takes over (pre-editing), and a lot of cleaning up after (post-editing). A nuanced understanding of the source text, the weighing of word and grammar choices and the consideration of a thousand shades of meaning are necessary to produce a good translation (and good writing)—and all of these are things that only humans can reliably do. Humans are irreplaceable in this context. If you try to skip the human process and overuse MT, it will cause quality issues that stop you from effectively communicating, rendering the whole translation exercise pointless.

Now let us go back to the title of this blog: how and when to let machines translate. It obviously varies with project and purpose. When you’re chatting or reading news online, or you need to extract some basic information from a webpage, the accuracy of translation details may be of minor importance and an online free MT service will work very well. I have friend who—at the beginning of her relationship with her boyfriend—managed to communicate with him through MT. The results may not have been spectacular, but they were good enough for understanding each other. The translations were pretty funny at times, but the couple could figure out what was meant, so it only added to the charm of the conversation. I am not sure how well a lover’s tiff would have been translated though—it could have gone horribly wrong!

And that is the problem with relying on MT alone. When the stakes are high, one wrong word can cause an avalanche of problems. If the translation is used for an important purpose where accuracy is a priority—think legal documents, brand localization, user manuals—MT can be positively dangerous. Only a real person can judge the nuance and particular meaning of a word or sentence and render it accurately in another language. Bad translations can damage your brand or even hurt your target clients, making you at best a laughing stock and at worst the subject of a court case. For this reason, MT should not be involved in some projects, or the translation should be fully reviewed word-by-word. This may actually take as long as a doing the translation manually in the first place, depending on the text and the MT engine used.

Another often overlooked issue is security. It should be noted that 'free' MT is not without risks. You must assume that both the source and target texts will be stored in the MT database, which brings about a host of confidentiality, intellectual property, and information security issues.

Examples of bad machine translation abound on the internet. You will find the '35 Hilarious Chinese Translation Fails' article amusing, especially if you know some Chinese.

Have you seen any other examples of sloppy machine translation? Do you have plans to try machine translation for your project? You are more than welcome to share your thoughts with dan.li@nescit.com.

Blog Event Calendar

How can we help?

You are welcome to request a proposal, send general inquiries to info@nescit.com, or phone us at +46 76 512 53 02 (English and Chinese) or +46 763 05 22 14 (Swedish and English)!