Just a few years ago, you could often tell when a text had been machine-translated by how hard readers were laughing. Sentences like “This device will not tread in rust before the end of five years” left little doubt as to whether they had been produced by a person or a computer.
Today, however, you won’t find many laughing about machine translation (MT). Indeed, the past several years have seen it evolve from a primitive solution for emergencies into an alternative that has to be taken seriously in the translation market. This hardly means that human translators should start looking for new jobs, though – or that you’ll be able to leave your own translations to Google or DeepL in the future.
In this article, you’ll learn about the kinds of texts that are suitable for machine translation and which should be entrusted to a living, breathing expert. To find out which method will best meet your needs, simply read on.
Thanks to huge strides in artificial intelligence and computational linguistics, the performance of machine (or automatic) translation solutions has advanced significantly in recent years. In early 2018, specialists from Microsoft Research caused something of a stir when they claimed to have achieved parity in Chinese-to-English machine translation – meaning that their results were indistinguishable from those produced by a human.
While the team’s bold statements should be taken with a grain of salt, there’s no question that the quality delivered by machines is fulfilling ever higher expectations. This is mainly due to the fact that modern translation programs are capable of processing tremendous amounts of data (i.e. text) in just seconds. As a result, their output now bears little resemblance to the meaningless hodge-podge of words typically churned out by translation software a mere decade ago. In some cases, they actually do come impressively close to the work of a human translator.
The quality of the text produced fundamentally depends on the type of machine translation applied. Here, we differentiate among three different approaches: rule-based, statistical, and neural MT.
This method analyzes the source text and restructures it in the target language based on linguistic and grammatical rules, as well as dictionary templates. While its results are consistent, they can range from useful to somewhat silly depending on the text and language pair at hand.
Statistical systems “learn” how to translate by analyzing massive sets of data. The translation itself is then performed based on bilingual text corpora that provide statistical probability models. The results have a certain flow, but are less logical.
Neural machine translation (NMT) relies on complex neural networks. In this approach, associations are produced for each individual word based on those adjacent to it. The more training material a system receives, the more accurate its translations become.
While NMT is still relatively new, it’s becoming increasingly popular. And it’s not just private users and frugal companies that are taking advantage: More and more professional translators are using the natural-sounding output NMT can quickly provide as a way to increase their own productivity.
Automatic translation can be worth a look in certain situations. If you want to translate texts for internal use, for example, and the information itself is more important than how well it’s presented, a machine translation tool might be just what you’re after.
Here are some possible candidates for MT:
E-mails with colleagues abroad
Foreign-language press articles that you only need to boil down to the essential information
User-generated content such as reviews or comments on blog posts (a bit of post-editing is advisable to ensure comprehensibility)
Product information that only provides basic data (here, full post-editing is a good idea to make sure that all the key content has been properly adapted to the target market in question). Not to be confused with product descriptions, which serve more of a marketing-oriented purpose!
These example applications all require prior deep-learning training of the MT solution used in order to produce accurate results in the context at hand.
Just because someone speaks a language doesn’t mean he or she is good at writing it. The same applies to computers. By the same token, not everyone has a knack for blogging, writing copy, or crafting bestsellers. This is why texts that need to meet a certain standard of style – to evoke emotion or win the reader over, for instance – aren’t a good fit for a machine. Technical content featuring complex sentence structures and specialist jargon also requires a human touch.
Take a phrase like “tighten our belts”: While AI is getting better and better at making sound assumptions based on context, a translation program won’t always know whether a company’s revenues are down or waistlines are actually involved. Problems can also arise when key terms need to be translated differently within the same document. The German word Anfrage, for example, can mean “request”, “inquiry”, or “query” in English. An MT solution without the capacity for memory might not be able to determine which translation is appropriate in each case.
Meanwhile, there’s something else machine translation lacks: cultural insight. Particularly in the field of marketing, it’s important to account for the local idiosyncrasies of your target market. Imagine a U.S. manufacturer of spider spray wanted to advertise its product’s effectiveness against black widows and brown recluses in Europe. It could save itself the trouble of translating this message into one language, at least: These species aren’t common in Germany! An experienced translator would bring this to the company’s attention and perhaps suggest some poisonous pests that are actually prevalent in the country.
So far, we may have given you the impression that machine translation has no business handling content meant for PR purposes, but that isn’t exactly true. With relatively little effort, the raw translations produced with such solutions can be refined into texts that hold up to public scrutiny. This type of post-editing is an interesting option for companies because it sometimes enables them to avoid considerable costs.
Human translation can get expensive, after all – especially when large amounts of text are involved. This is where post-editors come in: These qualified translation specialists have in-depth knowledge of a particular language pair, which enables them to polish machine-translated texts to a shine. At Lexsys, our post-editing experts will leverage their extensive experience in translation, proofreading, and revision to ensure that your machine-translated texts read like they were written by a professional native speaker.
Whether you’re already making active use of MT or are still weighing up the pros and cons, we can help you transform your texts into appealing translations without stretching your budget. If you’d like to learn more about how we can meet your needs, contact us today to set up an appointment.