This article was originally published in April 2018 and has been updated.
If you’re at all familiar with the language translation industry, you’ve heard debate about the pros and cons of machine translation (MT), a technology that uses computer software rather than a human to translate text.
Though the technology arrived in the mainstream just a decade ago, it has roots dating back to the 1930s, when a Soviet scientist unveiled a very archaic version of MT to virtually no fanfare. IBM made strides in the 1950s and again in the 1970s, but it wasn’t until the most current decade that MT began making rapid strides in ability and precision.
MT is rapidly changing and evolving, but as the software becomes more accurate and reliable, some indisputable benefits have emerged. While MT isn’t right for every translation scenario — and has not reached the point where it can completely replace human translation — it offers a meaningful opportunity to speed up translation and reduce translation costs.
Faster Translation With MT
MT can translate documents in a fraction of the time that a human can, but the quality can range from excellent to very poor within a single document without the ability to guarantee a level of quality output.
If you consider just MT alone – the output is appealing for projects that need a super quick turnaround where overall quality can afford some variability. This is often termed as “gisting.”
Machine translation is also ideal when you just need to scan a document to find certain passages, sections, or pages to translate. Having a person translate the entire document when you only need a few sections here and there is laborious and a waste of time and resources. MT can skip right to the parts you need.
That being said, you’re almost always sacrificing considerable quality for those faster speeds and turnaround times, with MT alone. Translations will be largely understandable, but they won’t be precise or professional. Without a native speaker spending more time on the output, most readers would find the translated text rough and, most likely, awkward or “computer"-sounding to read.
MT Cost Savings
Adding in a human review of some sort is referred to as “post editing”. Post editing takes the “raw” MT output and improves flow, voice, tone, and style. While post editing adds cost and time, it still can be much more cost effective than “full human” or “human only" translation (typically about 30% more cost and time effective).
Another good solution is training or “curating” the MT so that the AI knows and understands your output, improving its “raw” or baseline quality.
By pairing post editing and training of the MT, organizations can build effective overall language workflows that significantly improve turnaround time and quality when compared to “raw” MT output and at a significant reduction of the cost when compared to human translation.
As a final consideration, also have a view to the security of the content you are translating. Raw MT output from Google Translate or Microsoft Translator gives a very effective “gist” or understanding of the content – but it also means your content has gone into the public domain. Many companies have started to prohibit the use of these public engines for fear that internal confidential content may inadvertently make its way to the public web. If it does, any small saving in translation cost will pale against data breach litigation costs.
Making The Most of Machine Translation
The key to maximizing the benefit of MT is to know where it can be applied and to focus on the type of quality and security you really need when communicating to your intended audience.
At this point, most companies opt for a mix of secure trained MT, plus post editing as a balance for quality, speed, and cost. We have a helpful case study about what Enterprise-level MT actually looks like, and we hope it can help you take a step forward into the opportunity that MT presents.
If you have any questions about MT or about how it can be integrated into your workflows and content, then please contact us and we will loop in our data scientists and MT experts to help provide guidance.
