The idea of a future where people can communicate effortlessly with each other in different languages, translated perfectly by machines, is a seductive one. Wouldn’t it be so much easier if we didn’t have to go to the trouble of learning other languages any more?

Although machine translation technology is evolving all the time, it’s clear that we haven’t quite achieved this utopian vision yet. But will it ever happen?

Machine translation simply means the use of software to translate text or speech into different languages. The idea has been around for a long time. Researchers began working on machine translation projects back in the 1950s. The first commercial system appeared in 1991, with the first web applications following along a few years later.

Today machine translation technology ranges from massively used, free online translation services such as Google Translate, to cheap, on-the-go mobile phone apps such as Apple’s iTranslate, to rather more expensive, customisable, professional software packages such as SDL Trados Studio.

The benefits of machine translation are obvious:

  • It’s cheap – machine translation is much cheaper than human translation, and can even be done for free, depending on which service you use.
  • It’s quick – for on-the-spot translation needs, or time-critical web content such as breaking news, it’s hard to beat machine translation.
  • It’s improving all the time – researchers are continually striving to make machine translation technology better.

But the drawbacks are equally clear:

  • Lack of localisation – machine translation is just straightforward translation. There is little or no localisation process to put the content into an appropriate cultural context.
  • Lack of nuance – machine translation struggles to deal with the more subtle aspects of language such as humour and metaphor.
  • Lack of flow – machine translation often reads very awkwardly, producing badly ordered sentences that are difficult to read.

None of this might matter very much if the quality of the content being translated is either not that high to begin with, or ephemeral, such as discussions on a customer help forum or instant messages.

But if you’ve spent a long time crafting high quality content – for example, a persuasive marketing video or detailed product instruction booklet – then you really need human creative translation to do it justice and ensure that your translated material meets all the objectives of your original material.

Probably the most famous – and amusing – machine translation failure story is about the Chinese café that decided to provide a sign in English as well as one in Chinese. The Chinese sign (in Chinese characters) said ‘Dining Hall’, while the shiny new English sign said ‘Translate Server Error.’









No doubt in this case the global publicity gained by the café did it no harm at all, but in the serious world of business machine translation failure can lead to some significant problems, such as damaged reputation and the loss of customers. In extreme cases it could even result in legal action. It’s easy to see how cutting the translation budget could end up being a very shortsighted decision indeed.

For all of these reasons, we say that human translation is here to stay. There is definitely a place for machine translation, especially in those situations where immediacy is more important than accuracy, but in a professional context nothing will ever beat the impact of creative, precise, localised, human translation.

Source: Matinee Multilingual


  1. This one really sums up why computers cannot beat humans when it comes to translation. I work in a translation agency and I have seen my fair share of translation mishaps from translation programs that are sent to me to be corrected.There are just too many variables to consider: language is always evolving, one word has a lot of different meanings, computers cannot detect the “tone” of a text and so much more. Machine translation is good for basic things like greetings and simple words, but when you deal with things that count, nothing beats a translator who has lived in the worlds he/she is translating to/from. Real human interaction and experience is one that a machine can never have.

  2. Will Machine Translation Ever Beat Human Translation?
    The answer to this question depends on which pair of languages are we talking about. If we are talking about (English-Arabic), my answer would be “No” and that’s based on my experience with the currently available translation programs.
    There are several reasons for that:
    1- Sentence structure of both English and Arabic are different. This creates wrong arrangements of sentence elements.
    2- Differences between the two languages in many grammatical aspects including tenses, adjectives, and subject-verb agreement within sentences.
    3- Difficulty to reflect the exact meaning of some expressions especially if those expressions are used to reflect some kind of culture-related concepts.
    5- The subtle meaning of some linguistic expressions.
    6- Machine translation cannot convey the meaning expressed by most idioms in both languages.
    7- Machine translation gives word to word translation, i.e. literal translation. Mostly, this kind of translation does not reflect the exact meaning.
    3- Translation machines lack many vocabulary words of both languages.

  3. Until the AI programmers make a machine to win a turin test then it is still going to be a work in progress with machine based translation.

    You can achive simple things with pre-simplified phases optimised to fit the grammer and vocablary that the programmer is likely to have included in the system. This allows a gist level that with a lot of graphical and locational context can get a message though fairly successfully but if the programmer wasn’t into using the reverse side much so didn’t work so hard at it you can easily get a false confidance take English to/from Lithuanian for example with some online systems.

    The more complex the text the more complex the mechanics of translation become. It will be possible for machines to translate more complex things if enough increasingly complex proccessing is devised and programmed. The issue will be which is more effort: to teach a human two languages and how to relate them, or to study all the necessary complex thought done by a human who you’ve already had fully taught (in all probable translation scenarios) and then spend time and effort re-arrangeing these thoughts into machine algorithiums to be programmed into a machine that will mimic the human in the subset of tasks it is taught to handle [and have to repeat the process transfer for any new method or language that would be devised or learnt by the doner humans studied].

    At some point the law of diminishing returns or sprilling costs kicks in for most machine-translation projects which means they often stop short, compared to a human translators.

    The latest AI veiw was that if they could get enough background knowledge about the world into the system then hopefully they could get cognitive support in the process (ie that the computer actuly tries to understand what the orignal message is actually trying to say and mean – like human translators often do especialy when unstanderd gramer idoms and slang appear in texts).
    A couple of projects are trying to do massive knowledge compling and outside labs the most well known project that is attempting data for machine cognition is Google – most of its projects feed into this greater aim {from digital maps, directories, machine encoded books new and achient as well as learning from seaches humans do with its search engine interfaces (this really helps it build strong associative links betweens words and phrases when done with millions and millions of searches)}.

    It will take time but the best way to learn to trust something thats translates is to learn both languages so you can judge the results which ultimetly defeats the need for a translator! which is why there isn’t a lot of incentive to write good machine translators for those who’ve personaly had to learn both languages to be able to write them.

    Time will tell if things from AI will achive good translations but the big break will come when they start to actully cognate and then free think in the middle (between the languages) as then they can attempt to compete by trying to process as a human might, even if a bit juvinile to begin with.