For many businesses looking to cut translation costs, machine translation can seem a tempting option. Human translators are financially expensive and can also cost time to recruit and manage. Machine translation software, in contrast, can translate almost any amount of text in no time at all, and all for little or no cost. Despite those advantages, machine translation comes with some clear drawbacks. It may initially be cheaper, but the issues machine translation causes are much more costly in the long run.
Perhaps the most obvious danger of machine translation is translation errors. While human translators may occasionally make mistakes, machine translation software lacks cultural knowledge, background technical knowledge, and contextual understanding. The lack of these fundamental requirements of good translation means that for machines, errors are a constant, rather than an occasional, phenomenon.
Here are some examples of machine translation errors:
- A Siemens washing machine’s English label stated: “Made in Turkey.” This seems like a simple sentence that most machine translation tools would get right. The Arabic on the same washing machine label, though, meant: “Made by a Turkey.”
- The Malaysian Ministry of Defense announced plans to take “drastic measures to increase the level of any national security threat.” After making this unusual promise to make the nation more dangerous, the Defense Minister admitted that Google Translate had been behind the translation.
- Facebook’s machine translation software somehow turned one man’s Arabic language post of “good morning” into “attack them” in Hebrew. The mistake caused an arrest before the police realized their error and released the man.
Lack of Feedback
Review and feedback is often a vital part of the translation process. Translation is not just a question of translating correctly, but also of ensuring that the final product in the target language matches the tone, style, and terminology required by the client.
Review and feedback can be divided into two stages. The first stage is completed before work is returned to the client. After an initial translation is completed, it must be reviewed, checked, and polished to create a professional final product.
The second stage is after a translation is handed to the client. Tone, style, and terminology requirements can be complex and often change over time. Feedback from the client helps ensure the final product is just as the client needs it.
Artificial intelligence and deep learning may be slowly improving machine translation. But even supposing that machine translations greatly improved in accuracy, they would still effectively be producing first drafts. As every translator knows, an initial draft is only the first stage of a polished final product.
Cultural and Linguistic Context
Part of the reason machine translation produces confusingly garbled results is it lacks linguistic and cultural knowledge.
Common machine mistranslations happen because translation software lacks linguistic context. “Turkey” in English may refer to a country or a bird. It is easy for a human to tell which is meant through context, but for a machine it is much more challenging.
Another kind of context is cultural. Each of the thousands of cultures on earth has different stories, customs, and taboos, and these are often embedded with the language. Translation software lacks cultural knowledge, and so may make translations which are taboo, culturally insensitive, or which simply fail to localize to the target audience in the way that a human translator would be able to.
Author: Joseph O’Neill, Global Communications, Shanghai
CSOFT International works with an extensive network of qualified human translators to provide a range of cost-effective translation solutions. Companies including P&G, Sony, and GlaxoSmithKline have trusted us with representing their brand across cultures and languages.