Although machine translation technology has been in use for over 50 years, it remains a relatively unknown field to many translators. What's more, the relationship between human and machine has never been an easy one. Each technological advancement or hype in MT prompts a wave of defeatism, with some declaring that human translation is dead.
Given the fact that human translation is still alive and well, it’s easy to understand a certain degree of skepticism among language professionals. In fact, as translators become more accustomed to the task of post-editing MT (MTPE), and machine-generated results improve, MT is showing itself to be a genuine time-saving tool for many translators and a cost-saving proposition for customers.
Let’s look at some common perceptions about MT. Are they myth or reality in the current state of affairs? And how is the changing MT landscape positioned to address those issues?
Current verdict: Reality.
If you’ve ever found yourself looking at a machine-translated string of words and thinking, “I would be better off starting from scratch,” then welcome to the club. Thankfully, the problem of low-quality MT results has a finite lifespan.
As machine translation engines progressively receive more input from post-editing loops and fully integrate customer terminology, translation quality increases rapidly.
This isn’t to say that the results will be perfect anytime soon, but they’ll certainly become a solid foundation upon which to base high-quality, human-edited final results. In the meantime, it’s always an option to simply drop an unusable or unhelpful MT sentence and start over in order to meet the customer's (or the translator's) expectations of quality.
In this area, we can consider MTPE as a long-term investment. It will help the MT engine learn how to translate more accurately and consistently in the future. This is already the case with well-maintained translation memories (TMs). Translators far and wide know how much time and effort even partial TM matches can save.
The added benefit of MT is that translation engines can eventually produce similar results with completely novel content – strings that aren’t recognized in the TM, but which are identified by the MT engine.
Current verdict: Reality.
The correct and consistent use of customer-specific terminology is a must in professional translation, regardless of whether that translation was originally created by a human or a machine translation engine. But unless the MT system is trained on customer content, specific terminology is usually not applied. This means that the human post-editor has to be on the lookout for those terms during the machine translation post-editing process, which can entail substantial effort.
Tuning MT engines to specific terminological contexts is already possible, but it requires clean training data. If done correctly, MT engine tuning greatly reduces the need for post-editors to be as vigilant.
As machine translation technology evolves, term base integration and other functionality like fragment assembly will eventually become commonplace. In fact, these advances can accelerate the speed and accuracy of machine-generated texts to the point that post-editing will become a preferred baseline activity for translators. This will then allow them to focus their energies on transcreation and similar high-value, human-centric tasks.
Current verdict: Myth.
When one looks at previous industrial and technological revolutions, the source of this concern becomes clear. AI and automation are already replacing human labor in a number of areas, including customer service (chatbots), transportation (autonomous driving), medicine (automated diagnoses and surgical planning), law (AI-directed legal research), and more. This change in the nature of human labor has led to the fear that automation and digitization (machine translation, in our case) will produce a paradigm shift in the way we work.
While companies around the world are turning to machine translation (MT) to keep up with constantly rising global content volumes, the translation industry cannot depend solely on machines to accommodate this surging demand.
There will indeed be a change in the kinds of work that humans do, including in the field of translation. If you’ve ever worked with machine-translated texts, you’ll likely recognize that post-editing is a qualitatively different task than translating from scratch. It even differs from editing human-generated text: Machines tend to make different, sometimes bewildering mistakes when compared to humans.
It’s also clear that global audiences are growing accustomed to imperfect translation results in many areas. This in turn prompts customers of translation services to adopt a “good enough is good enough” policy when it comes to quality. Taken together, these developments make it is easy to understand why translators might be uneasy. That said, there is reason to be optimistic in this regard, which brings us to our next point.
Current verdict: Reality.
By taking a step back, it's easy to see that one of the other effects of MT is that it democratizes translation services in terms of potential customers.
Due to the continuing effects of globalization, small businesses and those in emerging markets are increasingly becoming customers for MT services on account of streamlined processes and costs. This means that while unattended MT may be taking a portion of the pie that once belonged solely to human translators, and post-edited MT may take another, the overall size of that pie is likely to increase drastically.
There will be an abundance of work for those who want to do post-editing, as well as significant demand for human writing and translation. This will be especially true in the ascendant areas of transcreation, MT-resistant technical translation, and creative copywriting.
Although machines are already positioned to assume much of the repetitive, formulaic translation work requested by customers, MT has its limits. Failing to post-edit can result in texts that are completely incorrect, even though they may look or sound fine at first glance. For instance, MT engines can't easily deal with character limits, handle embedded software code reliably, or adhere to a customer's complicated external instructions. Furthermore, the nuance involved in evoking a particular emotion, appealing to a specific target demographic, or accurately interpreting a joke or idiomatic phrase is still very much the domain of highly-skilled human translators – and promises to remain so for the foreseeable future.
When it comes to machine translation, we'd like to sign off with two closing thoughts:
1) The future is bright.
2) Resistance is futile.
The second point is a joke, of course. But like all jokes, this one has a kernel of truth: Machine translation will continue to grow as part of the language technology landscape. Ignoring its potential only puts yourself at a disadvantage. For translators, getting on the right side of the equation can make a career in translation and content services significantly more enjoyable and rewarding.
Here's some food for thought: Post-editing involves many of the skills that translators already use on a daily basis. With today's rapid rate of change in terms of technology and customer expectations, static skill sets are – for better or worse – a thing of the past. Translators who want to stay competitive in the age of machine translation would be best served by sharpening their skills in the ascendant areas mentioned above and adopting a flexible attitude in general when it comes to new, assistive technologies. In short, those who are able to adapt to the new reality in the world of translation will be the ones who thrive.