If you’ve never needed a translation before, or if it’s been a while, it’s normal to be overwhelmed by all your options when choosing a language services provider (LSP). Some may wonder why it’s necessary to pay for a translation at all these days, now that Google Translate and other platforms have started to incorporate AI into their programming for many major languages. Others might ask why the fees at some translation agencies are so much lower than at others, and whether the service called “post-edited machine translation” that is offered by some translation companies is a good option. In part one of this post, we’ll take a deep dive into the linguistic aspects of the world of translation to discover some answers to the question, “Why can’t I just use a machine translation? After all, it’s free!”

You Get What You Pay For

Don’t get us wrong: we are very impressed by what some companies have done with AI and machine translation. It’s a feat of software programming, linguistics, and crowdsourcing, and for some translations, some of these programs work rather well.

The key phrase here is, “For some translations.” The problem is that if you don’t have a good grasp of both languages you are using, you are unfortunately not going to know which translations are good and which ones contain errors.

Out of curiosity, and in order to keep up with trends in the field, our translators occasionally run a few sentences they have translated through a machine translation program to see how well the program does. In our experience, some sentences are spot on, and some are completely incorrect, often in the same paragraph.

Why is that the case? Here are some of the main issues:

  • Machine translation programs often do not understand unconventional word order. Many languages have a word order that is fairly strictly adhered to (for example, modern English usually uses subject-verb-object word order, e.g., “I like you.”). Others, however, have a word order that is used most of the time, but that can sometimes be inverted or changed by the speaker or author without affecting the meaning of the sentence. We have found that machine translation programs often misinterpret this non-dominant word order, sometimes assuming that there is a typo in the text and supplying words that were not present in the source text in order to force the sentence to make sense as the AI understands it. This can result in serious errors, up to and including a translated sentence meaning the opposite of what it meant in the original text.
  • Machine translation programs frequently run into trouble with acronyms. Most technical and medical texts contain industry-specific acronyms that machine translation programs usually have no idea how to deal with. They will often simply leave the acronym as is in the translated text, or supply another acronym from its memory that may or may not be the correct one. Human translators often spend a good deal of time researching the correct translation of acronyms for a given document translation, and an AI is not programmed to do that.
  • Machine translation programs encounter difficulties understanding conversational language. If you are translating a conversation between friends, survey responses, focus group transcripts, or even business emails, your text is highly likely to contain slang and highly context-driven usage. A simple sentence that any native speaker of a language would utter on a daily basis and that is easily translatable by a human can give a machine translation program a great deal of trouble. For example, a commonly used sentence in Russian (–Да нет, наверное, я ему говорю.) that a human would translate into English as, “‘Probably not,’ I tell him,” is construed by Google Translate as of this writing to mean, “No, I guess I’m telling him.”
  • Machine translation programs have a hard time using the correct register for the context. Unlike a human translator, a computer program (even one that uses AI) translates one sentence at a time, and therefore has a limited understanding of context.
  • Only a human translator can tell you if there is something wrong with the source text. In the translation business, we frequently come across projects with pages that are out of order or have chunks of text missing. Sometimes, particularly in editable documents where a block of text may have been inadvertently deleted by anyone handling the file, this is difficult for even experienced translators to catch.

In part two of this series, we’ll examine the technical aspects of machine vs. human translation.

Already convinced that a document translation service that uses real humans with years of education and professional experience is right for you? Click here and fill out Apex’s simple quote form, and we’ll get back to you with a free quote!