Pages

Friday, June 20, 2014

The Expanding Translation Market Driven by Expert Based MT

There has been much talk amongst some translators about how MT is a technology that will take away work and ultimately replace them, and thus some translators dig in their heels and resist MT at every step. The antagonistic view is based on a zero-sum game assumption that if a computer can perform a translation that they used to do, it inevitably means less work for them in future. In some cases this may be true, however this presumption is worth a closer look.

While stories of MT mishaps and mistranslations abound, (we all know how easy it is to make MT look bad), it is becoming increasingly apparent to many in the professional translation business, that it is important to learn how to use and extend the capabilities of this technology successfully, as the technology also enables new kinds of translation and linguistic engineering projects that would simply be impossible without viable and effective implementations of expert MT technology. Generally, MT is not a wholesale replacement for humans and in my opinion never will be. When properly implemented, it is a productivity enhancer and a way to expand the scope of multilingual information access for global populations that can benefit from this access. 

MT is in fact as much or more a tool/technology to create new kinds of translation work, as it is a tool to get traditional translation work done faster and more cost effectively. While MT is unlikely to replace human beings in any application where translation quality and semantic finesse is really important, there are a growing number of cases that show that MT is suitable for enabling many new kinds of business information translation initiatives that may in fact generate whole new kinds of translation related work for some if not all translators. MT is already creating new kinds of translation work opportunities in all the following scenarios:

  • With high volume content that would just not get translated via traditional human translation modes for economic and timeliness reasons, and thus the use case scenario is either use MT or do nothing. MT is used to lower total costs that make content viable to translate without which it would have never been translated. This in turn has created new work for human translation professionals in editing the most critical content and helping to raise the average quality of expert MT output.
  • With content that cannot afford human translation because the value of the information is clearly not worth the typical human translation cost scenario.
  • High value content in social networks that is changing every hour and every day and has great value for a brief moment, but has limited value a few weeks after the fact.
  • Knowledge content that facilitates and enhances the global spread of critical knowledge.
  • Content that is created to enhance and accelerate information access to global customers, who prefer a self-service model as in technical support knowledge base databases which have new content streaming in on a daily basis.
  • Content that does not need to be perfect but just approximately understandable for exploratory or gist purposes.
One point worth clarifying upfront is that much of the interest in MT by global enterprises is driven by their need to face the barrage of product/service related comments, discussions and opinions that flow in social media and influence how customers view their products. This social media banter is very influential in driving purchase decisions, often much more than corporate marketing communications which are seen as self-serving and self-promoting. Also, as products grow in complexity it becomes important to share more information about power features and extended capabilities. The issue of growth in the sheer volume of information is increasingly clear to most but there are actually translators out there who think the content tsunami is a myth. EMC and IDC have well documented studies that show the continuing content explosion. 

Global enterprises who wish to engage in commerce with global populations have discovered that the control of marketing has shifted away from corporate marketing departments to consumers who share intimate details or real customer experiences. User generated content (UGC) such as product experience related comments in social media e.g. blogs; Facebook, YouTube, Twitter and community forums have become much more important to final business outcomes. This UGC content is now influencing customer behavior all over the world and is often referred to as word-of-mouth-marketing (WOMM). Consumer reviews are often more trusted than corporate marketing-speak and even “expert” reviews. We all have experienced Amazon, travel sites, C-Net and other user rating sites which document actual consumer experiences. This is also happening at B2B levels. It is useful to both global consumers and global enterprises to make this content multilingual. Given the speed at which this information is produced, MT has to be part of the translation solution to digesting this information, and conversion to multilingual modes, to influence and assist global customers in a time frame where it is useful. For those of us who understand the translation challenges of this material, it is clear that involving humans in the expert MT development process providing linguistic and translation guidance in this process, will produce better MT output quality. The business value is significant so I expect that linguists who add value to this conversion process will be valued and sought after.

While some translators see MT as a big bad wolf that looms menacingly around, they fail to see that the world has changed for everybody, especially corporate marketers, PR professionals, and any enterprise sales function facing customers who share information freely with details of personal customer experiences. An individual blogger brought Dell to its knees with a blog post titled Dell Hell. Some say it triggered a huge stock price drop. A viral video about careless baggage handling of musical instruments resulted in a PR nightmare for United Airlines and perhaps even a negative impact on their stock price. This user experience content really matters to a global enterprise and they need strategies to deal with this as it spreads across the globe and influences purchase behavior. As the infographic below (bigger version available by clicking on this link) shows, every time a consumer posts an experience on the web it is seen by 150 people, which means small improvements in brand advocacy result in huge revenue increases, and 74% of consumers now rely on social networks to guide their purchasing decisions. This means that non-corporate content becomes much more important to understand and translate since these experiences are being shared in multiple languages.

This graph details how negative experiences multiply in negative impact, as consumers tend to be much more invested in sharing bad experiences than they are about sharing positive experiences. Thus it is very important that global enterprises monitor social media carefully. This is yet another example of what content really matters and how social media drives purchasing behavior. 

So if all this is going on, it also means that what used to be the primary focus for the professional translation industry, needs to change from the static content of yesteryear to the more dynamic and much higher volume user generated content of today. The discussions in social media are often where product opinions, brand credibility and product reputations are formed and this is also where customer loyalty or disloyalty can form as the customer support experience shows. This is what we call high value content. MT is a critical technology that is necessary as a foundational element for the professional translation world to play a useful role in solving these new translation challenges. However, it is important to also understand that this challenge cannot be solved by any old variant of MT, especially the upload and pray approaches of most DIY (Do It Yourself) MT. This is challenging even for experts and failure is par for the course..


Where MT creates new translation work opportunities

image

Some specific examples of the expanding translation pie that MT enables and drives:

The knowledge base use-case scenario has been well established as something that improves customer satisfaction and empowerment for many global enterprises with high demand technical support information. To develop and improve the quality of the MT translations in knowledge bases, very special linguistic work and translations need to be done. And while we see many examples of translators commenting on the poor quality of the translations we also see that millions of real customers provide feedback to the global enterprise suggesting that they find these “really bad” translations quite useful for their purposes, and prefer that to trying to read a tech note in a language that is not as familiar. Thus, while MT is imperfect we have evidence that many (millions) find it useful. Generic users on the internet are information consumers who have to deal with a language barrier. They are often the customers that global enterprises wish to communicate with. Their growing acceptance of MT suggests that MT has utility in general as a way to communicate with global customers, even though it is clear that a machine’s attempt at translation is rarely if ever as good as a human translation.

We are now also seeing that social media content based sentiment analysis is increasingly being considered as a high value exercise by marketing groups in understanding global markets. To translate international social media content it is useful to understand core terminology and get critical language translations in place and steer expert MT. This is new kinds of linguistic and translation related work which involves understanding the behavior of language in specific domains and discussion forums and then building predictive translation models for them. This new linguistic engineering work is an opportunity for progressive translators. New skills are needed here, an understanding of corpus at a linguistic profile level, the ability to identify MT error patterns and develop corrective strategies by working together with experts. The objective here is to understand the customer voice by language and develop appropriate marketing response strategies.

We also see the growth of sharing internal product development information across language within large global enterprises. Rather than use a public MT engine that can compromise and expose secret product plans it has become important to develop internal corporate engines that help employees to share documents and presentations in a secure environment and at least get a high quality gist. This effort too benefits from skilled linguistic engineering work, corpus analysis, terminology development and strategic glossary and TM data manufacturing. 

Every large translation project that is ONLY done because the cost/time characteristics that expert managed MT lends to it will generate two kinds of translation opportunities that would not exist were it not for the basic fact that MT made this content viable and visible in a multilingual context:

  1. Post-editing of the highest value material in a multimillion word corpus
  2. Translation of content that simply would NOT have been considered for translation had MT not made it economically viable and feasible.

So the next time you hear somebody bashing on “MT” ask yourself a few questions:

  1. What kind of MT variant are they talking about as there are many shades of grey? Amateur DIY experiences producing shoddy MT output abound, and translators should learn to identify these quickly and avoid them. Dealing with experts provides a very different experience and allows for ongoing feedback and improvement. MT is a tool that is only as good as the skill and competence of the users and is not suitable for many kinds of high value translation work.
  2. Are you dealing with a client/customer who has a larger vision for expanding the scope of translation? There is likely a bright future with anybody who has a focus on these new massive data volume social media projects.
  3. Are you playing a role in getting information that really matters to customers and marketers translated? While user documentation is still important, it is clear the relative value of this kind of content continues to fall as an element of building great customer experiences. The higher the value of the information you translate to your customer, the higher your value to the client.
But I expect that there will still be many translators who see no scenario in which they interact with MT in any way, expert-based or not, and that is OK, as it is a very different work experience that may not suit everybody. The very best translators can still put machines to shame with their speed and accuracy. But I hope that we will see more MT naysayers base their opinions about MT on professionally focused expert MT initiatives, rather than the well-publicized generic MT and lazy DIY MT initiatives that are much easier to find.

"You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete." - Buckminister Fuller

43 comments:

  1. Kirti,

    I'm also upset about translators having an opinion about MT while many of them only know about Google Translate for instance. Nevertheless, I can also understand the negative feeling some of them have about MT: they are usually not the ones pulling in MT when they need it, but they get pre-translated jobs pushed to them where the low-fuzzies are replaced by MT of an unknown quality. Many times the price offered for checking and fixing the MT output is too low.

    Your blog puts things in the right perspective from a business view. And it is important this is being said.

    I'm convinced that in some years from now, when MT is even better than it is today, some translators will still not agree to your view. That would be a pity, because they will be blind for new opportunities. But that is their responsibility.

    Ours is to improve the MT engines and how these engines are being used in production.

    ReplyDelete
  2. Very interesting post, Kirti. In my opinion, the first point is what kind of company needs a MT service. And once it is established, what level of post edition has to be applied according to price, goals of the companies, etc.

    Celia Soria

    ReplyDelete
  3. Thank you for a very good explanation, Kirti. In March of this year I addressed the need for training translators in the concepts of MT and then on PEMT in a blog in TAUS, "Will you help me flip it." As you mention, there is a huge misunderstanding on the part of many translators about what MT really is all about. But translators do not have the resources to gain the understanding needed. Will you help me flip it?

    ReplyDelete
    Replies
    1. Claudia, thank you for your comment. This blog is an attempt to "flip it" by hopefully sharing information that creates more informed perspectives on what the technology is and what is possible.

      A key point to understand is that "MT" is not a monolithic single technology, there are many variants and unfortunately the ones that are most easily accessible (Generic and crappy Moses systems) tend to be the experience that is most common. (See my previous post on Shades of Grey).

      I think the disdain for "MT" given the frequent experience that translators have of correcting and post-editing generic MT and low quality Moses systems is quite understandable. Many "MT" proponents including TAUS, encourage the proliferation of low quality "MT" systems usually built by people who have little or no understanding of what they are doing, and thus triggering translator ire. While Moses is great for experiments to establish feasibility. I don't think it is useful for production use in professional business settings unless the users have real expertise and have many additional tools to make the process useful in professional settings. IMO It is important to understand that MT is complex enough that it requires real expertise and controls, especially if you expect translators to work with it. Translators actually understand quite well that they are being asked to fix crappy output for very low rates which is a no-win situation for them. I think if "MT" practitioners are more professional, and provide good system feedback mechanisms, and pay post-editors fairly, the "misunderstandings" would end.

      I think eventually many more higher quality expert systems will come into production and translators will learn to understand and discriminate between low quality DIY MT and more professional expert based MT systems.

      Delete


  4. Is "bullshit" allowed in here? Typical marketing mumbo jumbo for reducing costs and make people feel stupid.

    Ricardo Horta

    Tradutor (Translator) PT (EN/ES/DE)

    ReplyDelete


  5. @Ricardo, don't be too harsh here. The "MT generates new kinds of translation work opportunities" theme is not new. Now, however, there are two differences from when I first heard the song from AO in 2007. First, back then the song was only speculation. Now, there are actual case studies behind the "scenarios" from AO and others. Secondly, the tone has shifted from soliciting LSPs and buyers to soliciting translators.

    I actually like the term "lazy DIY MT initiatives." It parallels the imaging world. Have you ever noticed how some websites have "lazy digital images initiatives" with computer-generated images spliced & diced by DotNET/Java programmers, while others have images crafted by skilled professional artists? You know, something like the differences between http://www.asiaonline.net/ and http://www.uelsmann.net/

    My point is, translations are like the various types of graphic arts. There are use cases when computer-generated translations (MT) are appropriate. There are other use cases when only translations crafted by professionals will do. Unfortunately, most of the market attention in the last 8 years has focused on the MT tools for the first use case category.

    Why do I reference the digial imaging metaphore? Because digital imaging is an integral part of L10N. Every day, LSP's choose whether they insource or outsource digital imaging requirements. If it weren't for Adobe's wild-n-crazy venture into Photoshop in 1990, all LSP's today might have to outsource graphics requirements to digital imaging experts at graphics houses with $150,000 computers and software (Industrial Light & Magic?). Today, Photoshop is still a complex tool for professionals, but it requires a level of expertise that's achievable without a software engineering PhD for designing snaking algorithms.

    Likewise, SMT has come a long way since it's commercial debut 8 years ago. SMT tools (not the outsourcing portals) are still for professionals, like Photoshop is. However, with proper training and practice, technically-minded translators and localization engineers can and do learn how to achieve incredible results in about the time and cost it takes a professional graphic artist to learn Photoshop, i.e. no PhD required. This is what I call "MT on professionally focused expert MT initiatives."

    Tom Hoar
    Managing Director at Precision Translation Tools Co., Ltd.

    ReplyDelete
  6. Interesting perspectives in the debate. I would summarise it by saying that the MT industry as a whole is shooting itself in the foot (by living up to its own miserly reputation), and that you seem to expect translators to bandage the wounded limb (when you say "translators will learn to understand and discriminate between low quality DIY MT and more professional expert based MT systems"). Why should translators go to this trouble if the majority of the MT crowd continue to play the rip-off game? (i.e. offering low pay for horrible work).

    And I note that the word "translators" is still being misused by the MT fans. A person doing PEMT is NOT A TRANSLATOR (at least not for the duration of the PEMT job). This person is an editor, proofreader or copyeditor. By all means add the adjective "bilingual" to describe this role, but please stop using the word "translator" for this non-translation activity. Kirty, I note that you fall into this trap less often after my recent rant, but the temptation is still there. And the comments by Gert, Claudia and Tom make liberal use of the word "translators" when they are really talking about EDITORS. This misuse of a professional title is merely a symptom of a general disrespect for the REAL work of translators.

    And Tom, although you ask Ricardo to be less harsh, you do not address his actual complaint (that the potential earnings of translators are being squeezed and translators are disrespected, especially those who speak out against the MT/PEMT bandwagon).

    ReplyDelete
    Replies
    1. Victor, the response to "offers of low pay for horrible work" is very simple. REFUSE to do it.

      Also, as more translators understand how the average MT quality matters, and is (or should be) closely related to the pay rate, they will be less likely to accept crappy PEMT offers. Being able to understand this in any potential work situation is a useful skill IMO. So you know when to say NO even if you really need the work.

      It is in the interest or translator/editors to understand how to determine SPECIFIC JOB-RELATED MT quality to avoid getting involved with bad offers where they work for peanuts. With good MT output, translator/editors can actually make more money than doing regular translation work, in spite of the word rate being much lower. This too can happen. Do you know how to tell when this is a possibility? So why not learn how to tell, when this is a possibility. And I understand that for some translators, PEMT is an absolute "no-way Jose", and that too is fine.

      In some cases it is necessary to use translators, rather than pure editors, because the bad MT segments need to be translated from scratch like a very low fuzzy match. So this distinction that you want us all to live by does not always make sense practically. The fact is that PEMT can involve both total re-translation and also minor-editing in addition to approval of good segments, so PEMT is BOTH translation and editing. When the MT output is very good in mature domain focused MT systems developed by experts, it is even possible to use people who are monolingual (in target language) but they are also usually subject matter experts in the domain in addition to being competent in the target language.

      MT will tend to do best with highly repetitive kinds of content (like large manuals, BTW also where TM is most valuable) and it will reduce prices with this kind of material. MT does not tend to do well with completely new material that is not similar to what it has learned (trained on). Much of the content that businesses translate does not have a very long shelf life (i.e. nobody cares how well it is translated 1 months after the product is gone) so businesses who want to stay profitable will continue to have reason to find ways to do it efficiently. This is more related to business survival realities than it is to any general disrespect for translators. Business survival realities will demand that global enterprises continue to seek translation efficiencies because information about products and services has a half life that is always getting shorter, MT is a means to an end to get basic info out there in timely and cost efficient manner and not any kind of final translation glory or punishment for translators. But MT is not suitable for many kinds of projects and content, this will always be true, and there are many translators who carve out very comfortable niches with this kind of non-MT friendly material.

      Perhaps we all need to move beyond rants and personal bias, and figure out what matters from a professional standpoint, and learn how to discuss the use of these tools in appropriate ways so that there is more informed use and more win-win scenarios.

      Delete
    2. Language is powerful. You belittle my distinction between translators and editors as "this distinction you want us all to live by". In other words, you still regard PEMT as a sub-category of translation and are not able or willing to make a clear distinction between the two.
      If you were careful in your use of language, you would state that some jobs (apparently the worst jobs) require editors WHO ALSO have high translating skills or WHO ARE ALSO translators.
      Expert translators are sensitive to the use of proper terminology, and this lack of a meaningful distinction between the terms TRANSLATION and EDITING sends the message that language and terminology do not really matter (i.e. "Forget about the words, just get the job done").
      The terminological confusion is also a rhetorical device - it implies that PEMT is a sub-category of translation, and that translators should stop making such a fuss about proper terminology, the lack of job satisfaction and the abuses which are so often committed in connection with PEMT.
      As a contrast, it is interesting to note that in other areas you want TRANSLATORS (or do you mean PEMT editors?) to be the ones who make meaningful distinctions: by distinguishing between "good" and "bad" PEMT jobs. In other words, you want them to clean up the MT mess by spotting the bad jobs in advance, refusing to accept them and thus forcing the MT abusers to come back to the straight and narrow.
      In this context, I find it difficult to see any basis for your proposition of a "win-win scenario".

      Delete
    3. Victor

      The only PEMT jobs where there is no translation involved are for very mature MT systems where almost all the work consists of editing only because the output is of very high quality. ( A very small number of these systems exist.)

      Most PEMT jobs will require both kinds of tasks: editing and translating . It is possible to have different people do these talks but for the most part the the same person doing both. So my point is that is difficult to maintain this difference when in practice it is most common for the same person to do both. There is no disrespect to translators intended by this, and indeed there are translators who choose not to get involved with these types of projects because they have strong preferences for one kind of work or the other. The other comments in this thread seem to disregard the distinction you make (not because they do not understand the difference) but because they see that in practice PEMT usually involves both kinds of tasks. MT is not a replacement for translators.

      Delete
    4. On the use and meaning of the terms "translation", "PEMT", "editing", "translators" and "editors", we are essentially arguing about semantics. My position is that translation is the process from start to finish (i.e. the sum of the steps and skills involved in transposing the source text into the target language text), and a translator is a person who performs this process (or, according to an on-line Oxford defintion, a computer program which performs this task).
      This process, of course, involves a number of different steps. For the human translator these steps will usually consist of information gathering from various resources, e.g. from dictionaries and other literature, databases of past work, colleague opinions, MT engines etc. These steps and resources also include proofreading and editing. These individual steps and resources are not the translation process itself, they are parts of the process, cogs in the wheel. We should not create confusion by speaking about the parts as if they were the whole.
      A similar principle applies where MT is used as the central element in the process. The translation process is the whole shebang. This process has a number of constituent parts, one of which is likely to be the "PEMT" stage. The PEMT part of the process sometimes requires translating skills, but using the word "translator" as the standard term for people who carry out PEMT tends to promote confusion between the parts and the whole. At best, this is wooly thinking and careless use of terminology. At worst, it is a rhetorical form of NEWSPEAK which attempts to redefine the word "translator", suggesting that the "normal" translator is someone who carries out PEMT to fix broken MT manuscripts. At the same time, you admit that there are other types of translators who do not fit into this mould, although you still seem to regard this as the exception rather than the norm.
      This is the sort of socially and economically inspired semantic shift that I have written about in a couple of blog articles:
      http://language-mystery.blogspot.de/2010/11/still-building-babel.html (about the social and economic mechanisms of language change, by reference to the biblical account of the Tower of Babel).
      http://language-mystery.blogspot.de/2013/05/humpty-dumpty-and-taus-quality-concept.html (about the deliberate manipulation of language in the context of the MT and translation debate).
      Some people may be tempted to see my objections as "mere semantics", although that in itself would be an ironic footnote to the whole discussion, as if "semantics" were unimportant in a discussion between professional linguists.

      Delete
    5. Victor, my view (which may not be shared by other MT advocates) is that MT is a data transformation and linguistic engineering process whose intent is to approximate "translation" as you define it. It will by definition always fall short, and thus MT + Post-editing is an attempt to bring it closer to "translation". When there is real collaboration between the developers of the data transformation (MT) engines and the editor/translators who perform the post-editing services it can produce positive outcomes not only for the buyer of the translation services but also for the editors and translators involved. Much of the distress in the use of MT comes from incompetence in the development process and from agencies who try and abuse editors ("offers of low pay for horrible work" ) who do not understand the process and the technology. As more people begin to understand the technology, hopefully this will happen less.

      I agree that it would be useful to have a much clearer definition of the work so that everybody comes into this more informed.

      Perhaps this could be a future post that we (you and I ) could co-write or present in some kind of point-counterpoint format.

      Kirti

      Delete
  7. Excellent article. Thanks.


    Gloria K.

    ReplyDelete
  8. BrauerTraining .comJune 26, 2014 at 10:08 AM

    Kirti, very good article. Thank you! I would recommend translators to take the time to seriously read the whole article and think about what is being said.
    "...What used to be the primary focus for the professional translation industry needs to change from the static content of yesteryear to the more dynamic and much higher volume user generated content of today. The discussions in social media are often where product opinions, brand credibility and product reputations are formed and this is also where customer loyalty or disloyalty can form as the customer support experience shows. This is what we call high value content. MT is a critical technology that is necessary as a foundational element for the professional translation world to play a useful role in solving these new translation challenges."

    By BrauerTraining .com

    ReplyDelete
  9. I agree with Ricardo. Maybe the total market is growing, but the share of work done by human translators is decreasing at an amazing speed. The truth for us who are working in the translation field is quite obvious - human translators will soon be history (outside fields like poetry and fiction). I would honestly never recommend to a young person to get a degree in translation, it would be like training to be a typist 20 years ago...

    Still, if you know one of the smaller languages well, there is still work. It will take some time before results can be achieved with MT between two smaller languages that are as good as when you do MT between English and another big language (e.g. Spanish).

    By Peter Berntsen

    ReplyDelete
    Replies
    1. Peter, there are several active translator-bloggers who claim that they are enjoying ever growing workloads today, and see no threat from MT. I think it does depend on the languages in question and other skills and knowledge that translators have.

      Even for languages where even generic MT is very good (like Spanish) I do not see MT replacing humans completely. But I think there are some strategies for translators to build better future prospects e.g.

      1) Have clear subject domain specialization in addition to the basic linguistic knowledge. I think there will always be work for SME (subject matter experts) specialists even as random and generic translation work becomes less common.

      2) Get involved with high value content (from the Enterprise viewpoint) that Claudia refers to above

      3) Learn about the value added role that translators can have when they understand more about MT. The best MT systems will ALWAYS have high translator involvement.

      While typists did lose their jobs because of word processing technology, the new technology created better jobs (new eco-systems) for the most skilled typists who became office managers /administrators, marketing coordinators, executive administrators and even DTP experts. They often became power Office users who started using Powerpoint and Word. I think MT and collaboration technology are similar drivers of structural change in professional translation today where translators will help solve language related information problems like the social media examples i pointed out.

      Change is difficult as it does challenge and undermine old norms but it can also be a time to re-adjust and re-focus and build new skills that have greater promise.

      I feel that one of the most scarce resources in future will be "good" translators and I see that there will always be a role with economic benefit for these good translators.

      Delete
  10. This is a very thoughtful post. Although I learned my translation skills on a battered old typewriter (which still sits in a corner of my office) and enjoy an amble through the forest with my slow-moving bassetts, I have long realised that the speed of information transfer essentially changes the way people conduct their lives. This upping of the tempo of life is reflected in the social media, and who hasn't posted a photo on Facebook and received a dozen likes from distant relatives in other parts of the globe within seconds? The days of waiting 24 hours for the translation of a page of text are long gone. If a Chinese customer writes a comment on my hotel on my Facebook page I need to know if it's favourable or hostile NOW. MT will give me an instant idea of what my foreign customer thinks of my product or service. A Chinese travel company could be thinking about my establishment this very minute...

    By Terence Lewis

    ReplyDelete
  11. tool and how it is used, and I will tell you for whom it was developed. Generally speaking, Translation Supporting Technology is not developed to serve the interests of the translators, not with their input. The technology targets a different audience(s) and with their needs in mind. Can you imagine a medical device or new drug being developed without medical input? Can you imagine physicians using them if they are just being used to increase the profit margins of the insurance companies? Can you imagine a Law firm management software finding any customers if designed without their input and against their interests? Many time the Translation Supporting Technology it is developed by people with little to no understating of (or respect to) the profession, with no knowledge about a truly efficient translation problem, and without understanding what tools translators really need. This is an artificial solution to a semi-artifical problem that was devised to create a "need". If "productivity" was a true concern, there are many aspects, areas, and market inefficiencies that need attention.
    - Furthermore, many articles about MT are written for the attention of the potential clients of MT, trying to convince them. A general tactic to support that strategy is to paint the translators as outdated, misinformed, uneducated, and whiny that fear and fight for their relevancy against the superior technology, and therefore the clients and "smart" translators need to be saved by the technology developers and its proponents in the marketplace.
    - If you want more translators to consider the technology, demonstrate its merit and show them how it make *their* lives constantly easier, if you can that is.
    - Translators - like any other professionals - would be the first to adopt a tool that makes their lives easier, and just as easily reject inefficient technologies. Accept that the rejection of MT by many translators is not because they fear it, misinformed, or uneducated about it, on the contrary actually, they are rejecting it because they are educated and know the ins-and-outs of their work, informed about their clients' needs (which admittedly may not be the same clients that seek the MT refuge for cutting their costs), and just don't see how MT makes their lives any better.

    ReplyDelete
    Replies
    1. I agree MT is not suitable for all kinds of translation tasks - it makes most sense for repetitive and high volume tasks. And I do not assume that translators who do not wish to work with MT are misinformed or uneducated. PEMT is different from traditional translation work and many translators may wish to avoid it since they work in areas where MT is not a good fit or they simply do not wish to do any kind of post-editing work for other reasons.

      Delete
    2. By the way. I want to clarify that by using the pronoun "you" I don't refer to anyone specific (certainly not to you, Kirty), just addressing the reader this way in general as a way of expression.

      Delete
  12. If the technology developers and their associates want to engage in any kind of constructive dialog: stop insulting the the translation profession and the professional translators (as opposed to the para-professionals working in the bulk market who are the usual easy prey for many of the brainwashing that is going on); stop speaking as if the technology developers and agencies are the translators' patrons (so condescending and vile); stop speaking as if you (technology developers and agencies) represent the translation profession or had been authorized in any way, shape, or form to speak on behalf of it; and stop the insulting attempts to paint the rejection of MT by so many well-educated, well-informed, and experienced (yes! also experienced with technology. Some translators even come from the technology field and can identify technobabble from a mile away) professional translators after its current implementations have failed to impress them as a self-serving actions out of fear for relevancy. The pot calling the kettle black, because you are not working out of altruistic motives and are positioned to greatly benefit form selling MT-based solutions.

    ReplyDelete
    Replies
    1. There are situations where expert MT does provide value and benefit to some translators and hopefully we will see more of these situations in future. But the nature of the MT development process requires scale and data volume that require it be done at an organizational level rather than at an individual translator level. It is not done this way to exclude translators -- there are in fact many who do try with Moses but that is a difficult option that requires ongoing investment of time and money. It is possible in future that technology will reach a level that makes it possible for more translators to engage at an individual level but we all understand that it will need to offer more for this to happen.

      Delete
    2. I never say never, and some might find MT useful in a translator-centric workflow. But PEMT is not a translator-centric workflow, and it will never be one.

      Using the cost and complexity as an excuse to exclude translators is not a valid argument in my opinion. If you want to develop a professional tool, you need professional input and feedback, not least to understand how to design the tool to fit the professionals' workflow. Because translators are for all intents and purposes excluded from that process (while other stakeholders are not), and coupled with more than a single isolated statement about the superiority of technology over the translators, how they fear it, and the very public claims that it is intended to reduce costs by technology proponents and some who resell translation services, one can only draw the conclusion that none of this is developed as a professional tool for translators.

      The MT technology proponents are playing a double game: when speaking with translators they speak about MT as a productivity tool; when speaking with potential customers their main selling point is the claim and promise of reduced costs. They speak about the superiority of the technology, while at the same time work very hard to try affect and alter the perception of translation quality in alignment to what their "tools and workflows" can produce. They now started to claim that MT is just a tool, yet exclude the professional and show contempt to them and the profession almost on every step of the way.

      Maybe someday a trained MT engine will be used in a translator-centric workflow, and maybe some will find it useful. I never say never. But the whole tone of the conversation need to be changed. Technology developers and their associates need to understand their place in the market (they are not patrons of anyone, nor authorized to speak on behalf of the profession), and accept that many professional translators are rejecting MT on a well-informed ground. If they want to change that, show the professionals who do the actual work how this technology can be incorporated into their workflow, and how it can make their professional lives better.

      Delete
    3. While it is often true that translators are not part of the process, in best practice situations like the one used by ALT and described here
      http://kv-emptypages.blogspot.com/2013/04/pemt-case-study-advanced-language.html translators are very much part of the process of deciding when it is ready to pass on to larger groups of translators, and in determining and setting fair rates for the work. These translators also organize the feedback to adjust the MT so that it produces better quality and does not require painfully repetitive corrective work from the editors.

      Delete
  13. Hi Kirty, this is the full content of my first comment above that got cut-off:

    A couple of comments if I may, Kirty:

    - You wrote
    >There has been much talk amongst some translators about how MT is a technology that will take away work and ultimately replace them, and thus some translators dig in their heels and resist MT at every step.

    - Well, this is exactly the message that some LTA (Language Transformation Algorithm) proponents have very aggressively promoted (and some still do; some examples in the comments above), and the approach that many have (and still do) encountered "in the wild" as part of their daily encounters with some unscrupulous agencies in certain market segments. For every action, there is an equal and opposite reaction.

    - About the use of LTA on knowledge content: It is enough to read the original content to understand how low priority it is for this company. The original content is usually written very poorly - sometimes even on purpose to due to the use of the so-called 'controlled language' to make it MT Ready. It is probably just there for for the purpose of legally covering the company's behind. Most corporations have turned into crowd support in online fora of various kinds, in which more knowledgeable users help (usually in English) other users to solve issues, many times issues that are much more complex and deep than what the standard documentation has ever covered.

    - MT is just a tool as you say, but good and reliable tools are supposed to be developed for the professionals that will be using them, and with their input. Show me the tool and how it is used, and I will tell you for whom it was developed. Generally speaking, Translation Supporting Technology is not developed to serve the interests of the translators, nor with their input. The technology targets a different audience(s) and developed with their needs in mind. Can you imagine a medical device or new drug being developed without medical input? Can you imagine healthcare professionals adopting them if they are being used as devices to artificially increase the profit margins of the insurance companies on the expense of the healthcare professionals? Can you imagine a Law firm management software finding any customers if developed without their input, without understanding what they need, and for being used to ridicule them? Many times the Translation Supporting Technology is developed by people with little to no understating of (or respect to) the profession, with no knowledge about a truly efficient translation workflow, and without understanding what tools translators really need. This is an artificial solution to a semi-artifical problem. If "productivity" was a true concern, there are other aspects, areas, and market inefficiencies that need attention.
    - Furthermore, most articles about MT are written for the attention of the potential clients of MT. A general tactic in these posts is to paint the translators as outdated, misinformed, uneducated, and whiny bunch that fear and fight for their relevancy against the superior technology, and therefore the clients and "smart" translators need to be saved by the technology developers and its proponents in the marketplace.
    - If you want more translators to consider the technology, demonstrate its merit and show them how it make *their* lives constantly easier and better. How it improves their ROI, if you can that is.
    - Translators - like any other professionals - would be the first to adopt a tool that makes their lives easier, and just as easily reject inefficient technologies. Accept that the rejection of MT by many translators is not because they fear it, misinformed, or uneducated about it, on the contrary actually, they are rejecting it because they are educated and know the ins-and-outs of their work, informed about their clients' needs (which admittedly may not be the same clients that seek the MT refuge for cutting their costs), and just don't see how MT makes their lives any better.

    ReplyDelete
    Replies
    1. I agree MT is not suitable for all kinds of translation tasks - it makes most sense for repetitive and high volume tasks. And I do not assume that translators who do not wish to work with MT are misinformed or uneducated. PEMT is different from traditional translation work and many translators may wish to avoid it since they work in areas where MT is not a good fit or they simply do not wish to do any kind of post-editing work for other reasons.

      Delete
  14. I'll tell you why the majority of translators oppose MT and PeMT... they are just seen as a tool to reduce translators' rates and increase LPSs' profits, just like CAT tools. Just give me a decent output and a fair rate and I won't have any complaints... you also have to remember that you are asking professionals to help develop a system that will replace them to a certain extent. As I said, if we are paid fairly, that wouldn't be a problem, because there wouldn't be a drop in income. Some colleagues might refuse altogether because - let's face it - it's an inferior and demeaning task in comparison to translation... if all the "players" behaved ethically, it would be here to stay. But the big corporations are only interested in making money. Do you really think they care if blog or social sites' contents are translated or not? They sell lies to their clients, promising inflated and unrealistic gains in productivity, with all-encompassing systems that don't exist. Until we get to see a good, viable system, we will oppose it. I haven't seen one yet... I wonder why.

    ReplyDelete
    Replies
    1. I don't know if the majority opposes... I'm reading the numbers from the production system I built: it registers in detail what translators are doing, and all I can say: many translators definitely boost their production speed when they pull in MT. 30% speed gain is what almost everybody gets (compared to high fuzzies). And depending on how you measure, 300% is a real boost (that is compared to translating from scratch).

      4 remarks however:
      - I allow translators to pull MT. They decide and it does not influence the price of the job.
      - there is a huge difference between output of MT systems, and of course, not all source documents are fit to be translated with an MT system.
      - Speed gain does not say it all... Quality can be affected as well, but then again, if you use a defect centric quality process, MT is even helpful.
      - Some translators are not fit for post-editing.

      Last but not least: I don't believe MT will ever replace the human translator. MT is just a useful instrument in the translators'toolbox. And even when the MT output is perfect, you will always need human translators to validate this.

      Delete
  15. @Gert... the majority are opposed because they see it as a tool to replace them... this is obviously ignorance, but it's the general perception (same as CAT tools)... but there also translators that oppose it because they are not paid a fair rate... and this is the trend in the industry. I'm not against it, personally.

    You say you allow it and it doesn't affect the price of the job. Well, great! MT should be a tool to increase productivity without reducing the rates. Unfortunately, these days MT is "sold" as a tool to reduce rates and save money... to the clients...

    MT should be another tool - like CATs - which improve translators' output and allows them to do a better job... when you start reducing rates on the basis of it, quality is affected. We all know that. An unhappy translator works badly...

    ReplyDelete
  16. well written, too long, worth translating?

    Stylish and well argued, and paradoxically apt to challenge its own point.
    MT assisted translation of texts such as this might be considered feasible, these days. Unabridged.
    The result might be "the gist" but wordy, without the style.
    A poor result. But seemingly affordable, at MT depressed rates.
    In the past, it would "pay" to cut down the source word count, by getting the author involved before translating.
    Translating made sense with no MT, at a higher rate per word, if each word carried more weight.
    Crafsmanship all round...
    Less was more.

    ReplyDelete
  17. Fascinating. My take is that there are lessons to be absorbed and learned by today's translators if we want to grow with the tomorrow's market.

    There are opportunities to be had but there will also be casualties among those who ignore what's happening.

    Thanks for the insights.

    ReplyDelete
  18. I enjoyed reading your position paper. However, it fails to analyze what exactly MT (Machine Translation) is. MT in terms any 10 year old would understand is nothing more than an electronic dictionary. It should be marketed as such and not used to scam companies and translators alike.

    By using the term Machine Translation it leads corporate buyers to assume that they can bring down the cost of translation. MT is nothing more than an electronic dictionary that can be customized to include the client's terminology. Before it was given the improper name of Machine Translation it was called a "glossary". MT is nothing more than an automated search and replace feature. The human translator still has to unscramble the results and adapt it according to context.

    ReplyDelete
    Replies
    1. Michelle, you are quite mistaken about MT as your definition provides an additional type at a more basic level than what I have explained in this video which I created last week:

      The different types of Computer Generated Translation
      https://www.youtube.com/watch?v=Po0mEXwl9Dk

      Bilingual electronic dictionaries are another type of tool and every MT company I have worked for and know of has fought to make the separation distinct, including a very hot round table debate at AMTA1998 which led to the creation of the 16 versions of the Compendium of Translation Software.

      And there is still another type of MT (Knowledge-based semantics) which is a bit more complicated.

      Hopefully this video will demystify the fuzzy definitions of MT which keep floating around in the professional translator community, and among other industries as well.

      Jeff

      Delete
  19. I agree with everyone else that has replied regarding your content-to-volume ratio.

    And I understand that it would be great if MT could be applied to social media, user reviews, etc. But: OMG MT LOL RU4 real? With the possible exception of classical poetry, I can't imagine anything less suited as material for MT than social media and user reviews: rambling texts filled with slang, allusions, omissions, linguistic mistakes, typos, misused terminology, etc.

    I am not fundamentally opposed to MT and think that there are a lot of things that it can do better (also in terms of quality) than humans. I also agree that MT is capable of producing new markets in addition to taking over existing markets, and I think you've made an important point there.

    Personally, though, I have to question the expertise of any self-proclaimed MT expert who sees the key in post-editing instead of pre-editing/controlled language and - just because he has recognized the existence of a genuinely enormous market - starts rambling on about applying MT in a market that it has nothing to offer

    ReplyDelete
  20. Philippe EtienneJuly 8, 2014 at 9:15 AM

    I read in a magazine that machines are now as good as humans to recognize faces in a crowd, whatever the angle and facial expression. But the process to achieve this between a human brain and silicon is radically different.

    "Those algorithms work statistically on an exhaustive representation of information, without any actual reasoning behind it." (Science et Vie, free translation)

    Big data being the current big thing and all, I thought it suited MT perfectly: texts that don't need reasoning can be processed with MT with good results, just like face recognition.
    But put a hand on a cheek or mouth and the computer fails. Humans do succeed.

    To keep the parallel, translators will keep the publishing areas where texts are not simply strings of letters and require reasoning to be translated.

    Václav Pinkava wrote:
    ...In the past, it would "pay" to cut down the source word count, by getting the author involved before translating.
    Translating made sense with no MT, at a higher rate per word, if each word carried more weight.
    Crafsmanship all round...
    Less was more.

    Why be concise and quickly to the point if we can have it all machine-translated? A scary thought.

    Philippe

    ReplyDelete
  21. Giovanni GuarnieriJuly 8, 2014 at 9:27 AM

    I'll tell you why the majority of translators oppose MT and PeMT... they are just seen as a tool to reduce translators' rates and increase LPSs' profits, just like CAT tools. Just give me a decent output and a fair rate and I won't have any complaints...

    You also have to remember that you are asking professionals to help develop a system that will replace them to a certain extent. As I said, if we are paid fairly, that wouldn't be a problem, because there wouldn't be a drop in income. Some colleagues might refuse altogether because - let's face it - it's an inferior and demeaning task in comparison to translation... if all the "players" behaved ethically, it would be here to stay. But the big corporations are only interested in making money. Do you really think they care if blog or social sites' contents are translated or not? They sell lies to their clients, promising inflated and unrealistic gains in productivity, with all-encompassing systems that don't exist.

    Until we get to see a good, viable system, we will oppose it. I haven't seen one yet... I wonder why.

    Giovanni (Biscuit on your blog..above)

    ReplyDelete
  22. [quote]Michelle Kusuda wrote:
    ...Before it was given the improper name of Machine Translation.... [/quote]

    First let me say that I only skimmed the original post, because life is indeed too short, but I think Michelle is making an important point here. The whole taxonomy surrounding the issue is inaccurate. "Machine Translation"
    is not really translation at all. The most common form, statistical MT, simply involves calculation of word/phrase alignment frequency in a bilingual corpus. So in some respects it is like an electronic dictionary, in that it's just a tool that can aid a human translator. Although I would argue that stat MT is more flexible, powerful, and useful. It's somewhat like an automatic, recursive concordance search.

    "Post-editing" also seems like an inaccurate term. A human is still needed to read the source text and make a judgment about what the best translation is. That function comes rather close to the definition of "translation". Perhaps you could argue that it's a question of degrees and that if the MT output only requires light editing, or the client only cares enough to pay for light editing, then the "post-editing" stage is less like translation and more like review or editing. The problem is that it still requires the translator to read the source text (keeping in mind that it was not actually translated by a human) and make a judgment based on translation expertise, which is the real value that the translator provides.

    I am definitely not anti-MT, but I think this unfortunate taxonomy (which we are probably stuck with) will make it even more difficult for some clients to understand our value and the nature of what we do as translators. I use MT regularly in some of my work, not any of the free public services, but engines that I've trained myself using my own TMs and other previous translations. In some cases it slightly increases my productivity, and I pass along much of this efficiency as cost savings to clients. Hopefully there will be enough good clients out there who understand that if MT is really going to increase productivity, they will automatically see the benefit in lower costs, because after all, freelance translators have to compete with one another.
    There is no need to try and sell the so-called "post-editing" process as something new and different from translation.

    ReplyDelete
  23. @ Michelle Kusuda: My view (which may not be shared by other MT advocates) is that MT is a data transformation and linguistic data engineering process whose intent is to approximate "translation" as most translators would define it. It will by definition always fall short, and thus MT + Post-editing is an attempt to bring it closer to real "translation". It (PEMT) is ONLY useful if it helps work get done faster and is ACTUALLY useful to getting a translation project done more efficiently. MT is however, more than a dictionary. When there is real collaboration between the developers of the data transformation (MT) engines and the editor/translators who perform the post-editing services it can produce positive outcomes not only for the buyer of the translation services but also for the editors and translators involved. Much of the distress in the use of MT comes from incompetence in the development process, especially with incompetent do-it-yourself Moses efforts, and from agencies who try and abuse editors/translators ("with offers of low pay for horrible work" ) who do not understand the process and the technology. As more people begin to understand the technology, hopefully this will happen less and MT will take a place on a translator workbench as simply another tool that can sometimes be useful. We are still in the early days of professional use of MT as we are only now seeing it move from the role of free online utility for random internet users, to a professional translation production tool for high volume repetitive content. Most of the do it yourself development is not especially competent so it can be very painful for the editors and translators involved.

    @ Michael Wetzel User generated content is exactly the kind of content that MT can do reasonably well, as spelling errors, abbreviations and text acronyms like (LOL) can all be “learned” or listed in MT “dictionaries”. Remember the goal here is not a perfect human quality translation. MT will enable some level of “translation” or gist to be available for material that simply could not be translated any other way – because of sheer volume and timeliness reasons. Many hotel and product user experiences which heavily influence new customers behavior are translated by sites that encourage or provide platforms for these kinds of purchases e.g. TripAdvisor, Holiday Check etc.. Hundreds of reviews are going up every minute. There is very definite evidence that shows that these “translations” are good enough to influence what hotel reservations are made, and in the case of use case scenarios like technical knowledge base data (Microsoft) solves technical problems that would otherwise go to a tech support engineer (who is a much more expensive and slower problem resolution alternative). Thus as terrible as these “translations” may look to a professional translator, they actually work for many hundreds of thousands of consumers, and do in fact enhance international market customer experiences.

    @ Giovanni Guarnieri : Unfortunately, we are in a phase of evolution of this technology, where many agencies do use MT to simply push the rates down with little or no utility for the editors who have to clean up incompetent MT engine development. If translators learn to tell a bad MT engine (from the kind of output it produces) from a good engine they can then choose to work only on projects that offer 1) fair and reasonable compensation and 2) work that is not mind-numbing and tedious. More translators have to refuse bad PEMT jobs based on a real understanding of the specifics. But for now much of what you say is true so translator beware. Also many of the best MT systems are private and do not publish samples, and vendors are not allowed to share these systems which may ONLY work for a very specific domain and target style. The best ones can be so good that after some post-editing the final output is no different from what a standard TEP process would produce. I provide much more detail on these issues on this blog

    ReplyDelete
  24. Many other factors stimulate translation demand, but yeah, this is big too, imo.

    If people who demand translation services are more sensitive to better price-quality combinations available on the market as a result of MT aides, relative to how sensitive is the larger total capacity of the aggregate translation industry as a result of MT, then it is entirely possible that there will be a higher total volume of workable hours with wage increases across the entire wage distribution (although this could hide some specific winners and losers).

    Stated more easily, if the demand response to better price-quality combinations afforded thanks to MT is greater than the supply response of translators (who can then translate more words per hour at any given quality level with these technologies), then most translators stand to gain from technology, so long as they can maintain their relative position within the industry (but even if not, they could still gain).

    This is before considering that old players probably keep their clients as long as they deliver.

    So probably there aren't many losers at all. And there are likely to be many winners.

    Perhaps one day we will all wonder how it was that free and nearly perfect computer translation was just not the natural way.

    But in the decades or millennia between now and then, this industry will grow for a very long time.

    I think a primary advantage of MT that I hadn't thought of is if an organization regularly produces updated statements/releases, etc. in many languages, that MT would help to make turnaround faster in translation, and thus improve marketing/communications coordination and turnaround.

    ReplyDelete
    Replies
    1. Giovanni GuarnieriJuly 9, 2014 at 9:31 AM


      The problem is that this is not "translation"... it's "post-editing". So, translators will have to adapt to do a job which isn't their main job - the job they have studied for - to survive or earn more money (if that will be the case - which I doubt)...

      There might be more demand, but for a different type of job...

      Also, you are missing the point a bit... I agree that PeMT in some cases will increase translators' productivity, but that should be our choice - to use it or not to use it... when unscrupulous LSPs "force" bad MT on us, reducing the rates to increase their profits, we are supposed to take it and be happy? Not in a million years...

      At the end of the day, MT and PeMT are promoted by LSPs as a tool to increase productivity and save money... they will increase their revenues by passing some of the discounts onto their clients by simply paying the post-editors less...

      Unfortunately, after the CAT tools fiasco, translators are more and more wary of tools that should help do our job better, with better quality, but they are just used by LSPs to force the rates down and increase their margins... nothing wrong with the technology per se (and I include MT in his too), but not when it's used to make fat wallets even fatter...

      Delete
  25. Giovanni GuarnieriJuly 9, 2014 at 9:33 AM

    Kirti Vashee wrote:

    Also many of the best MT systems are private and do not publish samples, and vendors are not allowed to share these systems.



    This is part of the problem... Without proof, nobody is going to believe you... are you surprised that most translators are against it? The proof of the pudding is in the eating...

    ReplyDelete
  26. I had actually mentioned the German version of Microsoft help as a more or less positive example for good (= effective and useful) machine translation . However, Phil Hand replied that the Chinese version is painfully horrible, so who knows ...

    However, Microsoft Help is in many ways the polar opposite (relatively expert authors, relatively formal style, huge bodies of highly relevant professionally written parallel texts and professionally translated texts, subject-specific, purely utilitarian function, comprehensive professional glossaries) of an offering of something like "post-edited MT for user reviews" - even if a lot of the Microsoft content is user generated.

    However, if you have solid empirical market research in your hands that post-edited MT translations at TripAdvisor & Co. produce sales then there's not much sense in my speculating about whether or not consumers' behavior corresponds to what I consider logical.

    ReplyDelete
  27. If MT is so useful for productivity, can't the translator purchase an MT tool and use it to reap the benefits of productivity at his/her normal rates?

    Why should an agency force a translator to post-edit its MT output, that too at much lower rates?

    Why should the translator pass any potential MT cost savings to the agency?

    What value does the agency that sends out text for PEMT editing add, besides running the source text through an MT Tool and passing it to the translator?

    ReplyDelete