Versacom presents a series of articles on issues that matter to you!
For decades now, the impact of new technologies has been a central concern in the business world. Recently, that impact has taken on an entirely new meaning with the spectacular leaps made in AI and the resulting potential for automation in virtually every field, including translation.
What does all this mean for the language services sector? Well, it essentially implies a shift (still hotly debated) from computer‑assisted translation to machine translation. But what is the difference between the two?
Computer-assisted translation is when language professionals use tools, such as translation memories, to recoup previously translated content—which they can typically reuse as is or with minimal changes—rather than translate every line from scratch. In other words, it is the process of looking up previously translated material that is automated, not the translation work itself. The translation process is quicker and the result more consistent, but the job is still left to the language experts.
Machine translation is done entirely by a computer program, Google Translate being unquestionably the most widely known and used example. Language professionals sometimes review machine translations in a process called post-editing (a cursory review to improve the final product).
Google Translate can still be good for a laugh (or a cry, for those who’ve been burned!) with its unpredictable and sometimes ridiculous output, but recent AI advancements are really changing the game, and machine translation is evolving at lightning speed. New and established players are releasing applications powered by increasingly sophisticated technology at a quicker and quicker pace. Nevertheless, machine translation is still quite fallible and requires entirely legitimate precautions, but partially automated translation is now a fact of life.
Let us be clear: machine translation cannot compete with professional translation. That is chiefly because, unlike a translator, a computer program doesn’t understand what it is translating. That is a crucial distinction to keep in mind. Machine translation tools do appear to do an adequate job of certain types of texts, though. Now, figuring out how to gauge the quality of those texts and in which situations they are appropriate (or at least somewhat useful) is another matter altogether.
It’s hard to make an informed decision with such an overwhelming body of literature on AI and machine translation, most of it quite technical.
Any organization that is considering working with a language services provider would be wise to learn a bit about the main issues with machine translation. To get you started, Versacom will present a series of articles looking at the subject from five different angles which may be of particular importance for you:
- The current state of affairs and major issues (the topic of this first piece)
- Machine translation tools and information security
- The most popular and promising technologies
- Output quality in machine translation
- Safe, worthwhile uses of machine translation for businesses
As a language solutions and technology leader for close to 25 years now, Versacom is actively involved in all of the developments and issues that are defining and transforming the industry. Our articles are meant to make it easier for you to draw your own conclusions and come up with clear strategies suited to your organization’s needs. They’ll also get you thinking about the complexities of this changing environment.
The current state of affairs at a glance
Why is machine translation so popular?
It is (almost) instantaneous
We are in the midst of a virtual explosion of platforms and content. Communication needs are far more diverse, with content having to be interactive and published in multiple languages, formats and media, quickly and frequently.
It is lightning fast, so it is no wonder machine translation seems so valuable.
It costs (almost) nothing
The globalization of the marketplace is driving global communications, as is the fact that companies with vastly different cost structures from one country to the next are now on the same playing field. Even though the cost and effort are growing, this competition is driving prices to new lows.
It is dirt cheap, so it is no wonder machine translation seems so valuable.
It offers (almost) unlimited capacity
Not only are qualified translators inevitably slower and more expensive than computer programs (although their work is always far superior), but there are also too few of them to handle the massive volume of material requiring translation. There are hordes of amateurs out there who are selling their translation services but who just cannot uphold your quality standards.
With its huge capacity, it is no wonder machine translation seems so valuable.
Where does the debate stand?
Optimists say…
“Neural machine translation marks a new age in automatic machine translation. Unlike technologies developed over the past 60 years, the well-trained and tested NMT systems that are available today, have the potential to replace human translators.”
(Slator, April 2018)
–
“Machines were never so smart, but now they are made so smart that they can actually think for themselves.”
(TechGenYZ, July 2018)
–
“Within our lifetime I’m fairly sure that we’ll reach — if we haven’t already done so — human-level performance, and/or exceeding [sic] it.”
(WBUR, July 2018)
–
“Roughly, if you consider what machine translation is used for, and will be for the foreseeable future, accurately communicating meaning is probably good enough.”
(TRIBLive, May 2018)
–
“Microsoft announced a new way for users to customize neural machine translation systems […]. This enables additional context to generic translation models so that translations can reflect a company’s industry, tone and unique terminology.”
(MSPoweruser, May 2018)
–
“While many continue to moan about the quality of machine translation tools, we have already reached a point in human history where the substantial bulk of language translation is being done by computers.”
(CMS Wire, June 2018)
Skeptics say…
“Has AI surpassed humans at translation? Not even close! Neural network translation systems still have many significant issues which make them far from superior to human translators.”
(Skynet Today, July 2018)
–
“Machine learning has improved significantly in pattern recognition and prediction. Nonetheless, for it to rival the capabilities of a human brain, two things are needed: a full understanding of the human brain, and the computing power to replicate it.”
(Slator, February 2018)
–
“Various reports indicate that advanced machine learning systems will leave translators out of work in the near future. However, while there has been a significant improvement in neural machine translation, the technology still has a long way to go before it can match human translations.”
(B2C, July 2018)
–
“Recently, artificial intelligence and machine learning have made considerable progress with machine translation, which is very fast and economical to produce. However, in most cases, machine translation still isn’t good enough to be used as is for human audiences.”
(Markets Insider, November 2017)
–
“The progress we’ve made in machine translation is exciting. But, it’s not that exciting.”
(OBSERVER, February 2018)
–
“Robots Fail to Win Shoppers’ Hearts: How Man Beats Machine When Translating Retail Content”
(Retail Tech News, November 2017)
Versacom says…
Machine translation cannot replace language professionals, but it does help a reader get the gist of what a text says.
–
Artificial intelligence is literally artificial: it only simulates intelligence. Despite appearances, computer programs still do not possess the capacity for understanding.
–
Machine translation may or may not one day be an acceptable substitute for professional translation, in certain situations. In the meantime, it is one of the many tools leveraged by professionals and their clients to save time and money.
–
Machine translation is still much too risky for extensive use in organizational communications, but it can help you decide whether a text needs to be professionally translated and thus avoid the effort and expense of unnecessary translations.
–
When processing high volumes of previously translated content, new neural-based technologies sometimes fare worse than their statistics-based predecessors.
–
Machine translation is not an all‑or‑nothing debate. At this point it’s a matter of choosing the right tools and strategies for every type of communication to maximize quality and mitigate risk for your organization.
What do I really need to be concerned about?
Security
- Does the information entered in machine translators such as Google Translate remain confidential?
- How do these tools maintain confidentiality?
- How do you find out if a machine translation tool is capable of keeping information confidential?
- Are there any machine translation solutions that guarantee total confidentiality?
Food for thought
“On Sept. 3, the Norwegian news agency NRK reported that sensitive Statoil information—contracts, workforce reduction plans, dismissal letters, and more—were available online because employees had used the free translation service Translate.com, which stored the data in the cloud. […] Translation industry news site Slator then investigated, conducting its own searches of Translate.com documents. It reported ‘an astonishing variety of sensitive information that is freely accessible, ranging from a physician’s email exchange with a global pharmaceutical company on tax matters, late payment notices, a staff performance report of a global investment bank, and termination letters.’ Full names, emails, phone numbers, and other private data were publicly visible. Slator called it ‘a massive privacy breach.’”
Technology
- How does machine translation work?
- Do all of the tools work the same way?
- Are there tools that are more secure or effective than others?
- What are the technology’s main strengths and weaknesses?
Food for thought
“Originally centred around statistical machine translation, Google Translate worked by translating the required text first into English as an intermediary step language and then into the target language, cross-referencing the phrase in question with millions of documents taken from official United Nations and European Parliament transcripts. […] The original translations were inevitably imperfect, but served to reveal the broad intent of the original passage, even if Google Translate could not deliver the faultless, fluid translation expected from a human expert. But in November 2016, Google announced the transition to a neural machine translation premise – a ‘deep learning’ practice that saw the service comparing whole sentences at a time from a broader range of linguistic sources. This ensured greater accuracy by giving the full context rather than just sentence clauses in isolation. […] Processing these calculations over and over again allows Google to spot recurring patterns between words in different languages, meaning its chance of achieving accuracy is constantly improving.”
Quality
- Are the translations generated by these tools good enough for a business’s professional communications?
- What is post-editing, and how does it improve the quality of a machine translation?
Food for thought
NMT systems translate one sentence at a time. This means that when we ask an NMT system to translate a document, it’s really translating each of the sentences in isolation, then sticking the translated sentences together again. If you’re thinking that this sounds like a bad way to translate text, then you’re right – it is! Imagine attempting to translate – or to simply understand – a sentence without any context. […] So why do we train NMT systems one sentence at a time, rather than a whole document at once? The reasons are technical. Firstly, it’s difficult for a neural system to read a long document, store all that information compactly, and recall it effectively. Secondly, these systems take longer to run when the length of the input is long. Thus, we use single sentences for efficiency. Overall, the inability to incorporate wider context is a primary barrier to NMT’s success.
Best practices
- Under what circumstances would machine translation actually be helpful?
- What are the best defences against the risks of machine translation?
- Are there specific practices that have proven effective?
Food for thought
“Despite great strides in technology and addition of dozens of new language pairs, these free services are usable for “gist” or casual translation, but usually not for commercial purposes. On the other hand, commercial providers of MT technology have worked on improving their paid offerings and with customization such Machine Translation engines are finding commercial use in limited areas. […] The main use cases for machine translation are applications that require real-time or near real-time interaction, for assimilating texts and “chat”, and as a productivity tool supporting human translators.”
Keep an eye out for our next article
Our next article on machine translation will look at information security with machine translation tools. You can count on Versacom to help you understand the risks and the safeguards against them.
It is a informative post for me. Thank you so much.
This blog article shows the best balanced view on a machine translation argument that I’ve ever read.
Tech advance is unavoidable and beneficial for communication as well, for example, as in telephone history, but human engagement is also unavoidable, not only because the “deep learning” should still- and far more-be learning, but because human touch can deliver such delicacy that mom’s food is always unsurpassed, seasoned with warmth and love.