MT or poor human translation. Which works better?


Roman Mironov

This is a guest post by Roman Mironov of Velior, a single-language vendor located in Russia. Velior specializes in Russian translations and has been in the field since 2005. You can contact Roman through his blog at Views expressed in this post are strictly those of the author and do not reflect an endorsement by GTS or any of its employees.

Are you considering using MT to save on translation costs, but not sure whether it can fit any of your translation needs?

Aside from all controversy around MT, it has one clear advantage. The level of quality it delivers is the same or even higher than that produced by unprofessional translators who compete on price rather than quality.


Whenever I see a subpar translation, I feel very sorry for the client who—usually unknowingly—paid a lot of money for something so  worthless. Just a few months ago a translation agency asked us to translate an update for a number of learning programs geared towards managers. Since this was an update, we were supposed to re-use the translations made by the previous vendor. And those translations turned out to be among the worst that I have ever seen. I felt sorry for the client and the agency, which both got ripped off by the previous vendor, and I couldn’t help wondering whether using MT would have made more sense for both of them. The client would have paid much less, while the agency’s margin would have likely remained the same.

My conclusion

As a result of seeing such examples and the quality of the translations that we receive from some freelance candidates, I now believe that translators who offer rock-bottom prices tend to produce quality that is as low as raw MT output. And why would someone want to pay for a human translation, when a machine translation is so much easier and cheaper to get? With the progress of machine translation, I don’t think buying low-quality human translations makes sense anymore.

What I suggest is this: Sometimes, a client has lower quality expectations due to a limited budget. For instance, a company is submitting documents to the authorities for legal reasons, but no one is really going to read them. The company realizes this and is okay with the low quality of translation. In this scenario, going with the machine translation might be a better choice than spending time looking for the cheapest translator and then paying this translator.

Both have their disadvantages

Both machine translator and low-quality human translator have the same drawback: They make errors:

  1. Both translate literally, producing translations that are difficult to comprehend.
  2. A machine doesn’t have a clue about what it’s translating. The same is often true for a human translator who doesn’t make an effort to understand the meaning properly.

So, if both make errors, neither is better from the quality standpoint.

MT has clear advantages

While the quality is probably comparable, MT also has its benefits:

  1. MT can follow a list of a client’s terminology better (as long as you use MT software that supports uploading terminology lists).
  2. MT doesn’t make the types of errors that an automated QA is designed to detect, such as misspelling or wrong numbers.
  3. MT is often much faster.
  4. MT is often much cheaper.
  5. Buying MT reduces a client’s expectations. When the client engages a human translator, even at a bargain-basement price, it’s natural for the client to expect something worthwhile in return. But when he gets poor quality, the client tends to get angry anyway, because frustration about being ripped off is only human. But with MT, the client’s expectations are usually so low that it doesn’t matter.


When price is the most important consideration and you’re fine with subpar quality, using MT may be a viable alternative to a cheap MT a cheap and low-quality human translator.

So, are you using MT to cut translation costs?

Enhanced by Zemanta

Translation and Single Digital Market in the EU

European Union

European Union

Last month, I attended a LT-Innovate workshop in Brussels. In his presentation, Mr. Ruben Riestra gave an enlightening example of the status of the single digital market within European Union. In this article, I’d like to share this inspiring example with you.

Via the Internet, I can access digital content (e.g. web pages) in Spain in a fraction of a second. In contrast to digital products, the transportation of any physical products, like oranges, between any two EU countries will usually take several days. So, it is easier to sell digital content between the two countries than it is to sell an orange. Or is it?

Actually no, it is not. If I want to buy oranges from Spain, it is relatively easy to do so. An orange is an orange both in Finland and in Spain. There are plenty of sellers in Spain and it is easy to arrange the transportation too. Unfortunately, the same does not apply to digital content.

Between the most European countries there is a language barrier that prevents effective communication. In practice it means that I can’t use digital content in Spanish because I don’t speak that language.

For an effective single digital market, the following are required: common legislation, common currency and common language. We currently have the first two ones in EU. But we don’t and will not have a shared language within Europe. The only way to have a single digital market is to overcome the language barrier with translation, either by human translators or by machine translation. This makes language and translation technology crucial for Europe. Together they form the missing piece which will enable the European single digital market.

This is a guest blog by Niko Papula, Managing Director of Multilizer. Multilizer develops translation and localization technologies. Its latest innovation is award-winning machine translation quality estimation technology called MT-Qualifier.


Enhanced by Zemanta

Showcasing MT Technology: Day two of AMTA2012

Me and Don DePalma at AMTA2012

Day 2 of AMTA 2012 started with an excellent keynote address by Don DePalma, who is arguably the translation industry’s top analyst and overall expert. Don provided a lot of empirical data from Common Sense Advisory’s research which reinforces what most of us already know: that most of the content in the world is not being translated; that high cost prevents much of that material from being translated; that MT is applied for scenarios which require less stringent quality requirements; that more-and-more LSPs are adopting MT because their customers demand it; that freelancers hate MT.

There was one point that Don made in his address which I would like to highlight: Don said that at AMTA 2012 we are “preaching to the choir,” that we came to AMTA 2012 to talk shop with our “homeys.” Don’s recommendation: that MT vendors and LSP’s who sell MT services should go to other industry conferences and try and market these services to new clients and industry segments.  To expand the reach of MT in order to grow the industry segment. I think that is a good point. At conferences like AMTA 2012, it is easy to get mired in a quagmire of technical details like how to best train MT engines, what format of corpus should be standard, how to post-edit text, using controlled text, sentiment analysis and a bunch of other stuff that puts most people to sleep. The bottom line is that MT can be a cheap way to globalize products, to get new customers in new lands, to boldly go where no man has gone before. Let’s take it to the street. I thought is was an interesting point.

After Don’s keynote, there was an interesting panel on MT with Olga Beregovaya from WeLocalize, Chris Wendt from Microsoft and Wayne Bourland from Dell. Wayne made some interesting points about how Dell is using post-edited MT to localize their website into over 20 languages. One interesting point that Wayne made: It does not make any difference what linguists think about MT quality as long as the customer is happy with the result. And that as a buyer, Wayne does not care if his translation vendor uses MT, just as long as they deliver the translations at the required quality level.

There were a number of interesting talks that I heard, but I will not elaborate. The conference proceedings can be found online and anyone who is interested can probably get the slides/papers from the talks. I will mention that Sharon O’Brien from the School of Applied Language and Intercultural Studies at Dublin City University presented some interesting research on the usability of raw (unedited) MT. Without going into too much detail, it turns out that user instructions which were translated with Google Translate was of sufficient quality to be used by readers to perform step-by-step instructions. Having said that, I wouldn’t want to use raw MT to translate a manual for a medical device, where switching the wrong button can result in death. But for software instructions, raw MT may be a good option if nothing else is available.

There was also an MT technology showcase with some very interesting exhibits. In my opinion, Microsoft stole the show with their new Microsoft Translation Hub. I will dedicate a separate blog post to this product, which may set the standard for customized MT engines.

Busted Keynotes: the First Day of AMTA 2012

The first day of the AMTA 2012 conference went quite well, despite a few problems. With Hurricane Sandy wreaking havoc on the east coast, the weather in San Diego was just beautiful: clear skies and 75 degree temperature. And every room in the conference area of the Catamaran Resort Hotel has a beach view.

The first keynote speech was delivered by Caitilin Walsh, the President-elect of the American Translator Association. I was quite disappointed with this keynote. It was essentially a rerun of the speech delivered by her predecessor, Dr. Nicholas Hartmann, at AMTA 2010 in Denver. About how Machine Translation (MT) stinks and that no translator worth her/his salt would dare use it. The ATA as an organization has not changed its position in two years and still views MT as any enemy of the professional translator. What is even worse in my opinion, Ms. Walsh and the organization she represents seem to have no agenda at all concerning MT. At least she did not mention any such plans in her keynote address. Which means that the next two years will be pretty much the same. One question that comes to my mind is: if the ATA consistently refuses to acknowledge MT as a viable tool for translators; if the ATA has no agenda on MT; why invite the ATA President to deliver the opening keynote address at such an important MT conference?

The second keynote address was supposed to be delivered in person by Luis von Ahn of Duolingo. But Dr. von Ahn got stuck on the East Coast because of the storm and had to deliver his keynote via video conference. Due to technical problems, the video hookup did not work well so they just played a recording of a speech that he made last year at a TED conference (see it below).


Following the video, Dr. von Ahn took some questions from the floor and provided some more specific details about the Duolingo project. According to von Ahn, Duolingo is a poor man’s language learning tool. Why should people pay Rosetta Stone $500 to learn a language, when they can do it for free using Duolingo? Currently, the online learning tool only supports the FIGS languages. About 30% of the users are in the USA. And Duolingo plans to make money by selling crowdsourced translations to whoever wants to by them. All in all I would have to say that this keynote was disappointing by the very fact that it was a non-event. Every person in the translation industry has heard of Duolingo and I was eager to hear more about it from its founder. It would have been much better if Dr. von Ahn had been there.

The conference then split into three tracks: MT research, commercial MT applications and government applications. I mostly attended the commercial track and heard some very good presentations. Quite a few of them were case studies on MT post-editing projects. Many companies and LSPs are now using this workflow and its use seems to be increasing. LSPs are offering 20-40% price discounts on post-edited MT (compared with human translation), so the financial motivation for using MT is very clear. Another issue that came up time and again is the resistance of professional translators to participate in post-editing MT projects. Clearly, collaboration between translators and LSPs is essential to getting MT post-editing into the mainstream.

Another observation that I made is how many companies are implementing MT on their own, cutting LSPs out of the loop and working directly with MT vendors or even building their own systems. There were presentations by Adobe, Computer Associates and Nikon about MT systems that they had built and are maintaining. I also met representatives from Intel that has an in-house team that develops MT systems to publish their knowledge base in multiple languages.

With two more days in the conference, stay tuned for more of my blog posts. You can also follow the conference on Twitter using the #AMTA2012 hashmark.

Raising the bar: a translator’s perspective on quality and pricing in the translation industry

This is a guest post by Shai Nave, a English-Hebrew translator who resides in Israel. Shai specializes in Biotechnology translation and has been in the field since 2003. You can email Shai at

Disclaimer: Views expressed in this post are strictly those of the author and do not reflect an endorsement by GTS or any of its employees.

One of the biggest problems in the translation business is a lack of professionalism. This plagues both translators and LSPs. Some translators do not know how to conduct themselves in a proper business manner; indeed many of them do not even think of themselves as business owners. This plays into the hands of unscrupulous LSPs who take advantage of translators.

A lot of LSPs are now using Machine Translation (MT) as a direct method to push translation rates even lower. Even though MT post editing is more like re-translation and in many cases is not a much lighter task, LSPs ask translators to perform translation services for ridiculously low rates. As a result, some exploited translators are themselves leaning more-and-more to MT to accelerate their work while providing lower quality.

This is an issue of balance. I know of LSPs that constantly try to drive translation rates down for their vendors while not charging their clients less (or even charging them more). It is a real shame that translators and LSPs, who on the surface seem to share the same professional goals, have conflicting interests which mainly revolve around rates. The result of that conflict is affecting the entire translation industry. The clients find themselves paying the same but getting lower value. The standards of quality are getting lower by the day. Translators and even LSPs find themselves in financial troubles due to unhealthy profit margins and questionable payment morale, all because they ignore the business aspects of “doing business”.

Technology is great, but it seems as though the new advances in technology are just being used as another way to drive down the rates, with little to no regard to maintaining quality, business ethics and increasing efficiency. Technology is a tool. Professionals use tools to increase productivity because the tool helps them concentrate on the core work that they are doing and less on the process of doing it. Technology is no substitute for skills, professionalism and ethical conduct.

Both translators and LSPs should take a good look in the mirror and examine how they are contributing to this reality. LSPs and translators need to join forces to improve the profile of our profession while truly working together to provide efficient and high quality service that pays accordingly. At the same time, they should condemn unprofessional conduct at all levels. If not, the amateurs will only continue to grow in number while lowering the standards across the board, commoditizing the very essence of our profession. Translators need to stand their ground and insist on fair rate levels which will allow them to provide quality service.

Enhanced by Zemanta