Aboriginal hand stencils in Red Hands Cave, Australia |
I take too many things for granted, things that seem small, not so important, until something goes wrong with them. I never gave my left thumb much thought until I cut it off one day when a donkey nudged my shoulder to remind me that she wanted some of the zucchini I was free-hand slicing for my sheep. After it was sewn back on I remember needing assistance buttoning my shirt and performing other minor tasks, but as the nerves slowly regrew in the years after and the numb sensation subsided into something approaching normality I forgot the experience.
My health classes in school covered a lot of subjects, among them the need for exercise, good nutrition (the definition of which keeps changing, but an experiment with rats in the sixth grade showed me clearly that man should not live by cornflakes alone), how metabolism works and how not to get syphilis, but I don't remember any discussion of occupational health and the need to protect our bodies from the consequences of work. Later safety instruction in the laboratories where I studied and worked emphasized short-term protection against accidents, such as chemical spills, sprays and gas releases, but ergonomics were never really a topic. I hope that has changed now.
As readers of this blog know, I have many disagreements with the promoters of machine pseudo-translation and especially with those pushing HAMPsTr processes. Aside from the fact that I cannot abide profiteering liars who overpromise and underdeliver with a technology that will never live up to its sales pitch except on the broken backs of the disposable human labor assigned to cover up the inadequacy of six- or seven-figure engineering investments, I am mostly concerned with the physical and psychological consequences of disastrous work ergonomics in the post-editing of machine pseudo-translation. Time and again I have presented data to show how dictation processes routinely far exceed the best productivity currently claimed for trained post-editors on optimized machine pseudo-translation systems. Some MpT researchers like John Moran are aware of these productivity issues and helpfully point out that there are ways in which these different technologies can be usefully brought together (by the use of speech recognition to reduce strain injuries for keyboard-using post-editors, for example), but so far I see little willingness on the part of capital investors to invest in human health protection as long as custom and the law allow cheap laboring humans to be tossed like coal and wood scrap into the furnaces that fire their profit engines.
I feel a little sick sometimes when I read discussions of speech recognition for translation which emphasize possibly greater speed of production. I've started a few of these, and my little video demos on YouTube which show some rather impressive translating throughput in three-minute tests certainly promote that topic of discussion. But the best value in this alternative or complementary technology to the keyboard is the potential for improved digital health... the "digital" in this case referring to the digits on one's hands.
When that digital health is degraded to the point that holding a small dictaphone up to catch some important words from a conference speaker becomes unbearably painful after a few minutes, it is easy to focus on the disability and the disadvantages of it. These can be fixed or not, by therapy, surgery or perhaps even time. But what I cannot forget is how lucky I am to have these numbed hands, and while they do not always do my bidding, they do enough. I have met people who have lost their hands or parts of them in accidents or have been born without them, and while I would of course adjust to the loss of mine, I am grateful that this is not my present need.
When I was 15 I had an experience that I remember often when facing some incapacity. I lost my sight. For a week I was blinded after an accident with a hot nitrating mixture in the lab ate my face and sealed my eyes in the swelling of burns. I accepted quickly that my sight was lost, there was nothing to be done about it after the 3:1 concentrated sulfuric and nitric acid mix did its work. I sat in bed for most of the Easter holiday, listening to the television and wondering how I would go about learning to read Braille. Somewhat uncharacteristically, I wasn't thinking all that much about what had been lost, but when, after a week, the seal on one eye dissolved, fluid ran down my face and I saw light - just a little and nothing in focus, I felt a gratitude that I will never forget for that gift of limited sight.
And so, whether they serve me well or with difficulty, I am grateful for my hands, which serve me well enough and which I hope will always have enough strength and dexterity to serve others in some useful way.
Kevin - Thank you for the shout out. It helps to have hard data to justify funding, especially on the engineering side. So I would ask any translator who dictates to take a few minutes to answer this questionnaire by a colleague in Leeds.
ReplyDeletehttps://docs.google.com/forms/d/1_WWoIoQxt8smlyclpnhmdzpkO1BTiF3n77MXrffv4YA/viewform
(By the way, any translator who has tried Dragon for more than a few hours and stopped working with it please contact me at transpiral 'at' yahoo.com. This is the most interesting demographic for me as a researcher.)
Dragos was kind enough to show me the intermediate results of his survey and they confirm my own findings from many unstructured conversations with translators over the past decade. ASR helps some translators achieve greater productivity, but it does not help everyone and it is not equally suitable for all material. Moreover, many translators are not sure of what that gain is. However, when it is suitable, with practice, productivity gains from ASR can be dramatic. But that is beside the point. The benefit is indeed digital health (great phrase by the way!).
The problem is - as you have pointed out - you don't know what you've got 'til it's gone. As the only non-medic from a family of six (incl. three physiotherapists) I have always had the benefits of preemptive action drummed into me. Nonetheless, despite being a keyboard warrior (variously a translator and programmer) I have been lucky never to develop RSI. It is hard for me to imagine sore hands. My motivation for learning to dictate my translations would be based on productivity. I see no harm in using this as a driving force.
However, the productivity argument is a means to an end, not least because economics may balance out the gains in earnings over the (very) long term. Reduced keyboard use is a worthwhile goal in itself. It simply MUST reduce the risk of certain forms of RSI like carpal tunnel. I know too many translators (and a few pianists though my wife) who suffer from this syndrome to doubt that it is a very serious occupational hazard. Why else would translators swap tips on creams and pills to ease finger pain on Facebook after long keyboard sessions?
However, it is going to be a long road. Here’s why.
Even in the languages it supports well, Dragon is not for everyone but every translator who works in a quiet environment should try it out. That is easy to write - but here is the thing that frustrates me. I speak to the four or five freelancers I work with via Transpiral regularly on Skype. They know I am working on a Ph.D. focused on translator productivity. I tell them truthfully that their colleagues are working with Dragon, quoting very high hourly throughput and suggest they should try it out. I even offer to pay for the software under the understanding that they repay me from the productivity gain or return the license so we can try it out with someone new (a trick I learned to convince good translators to work with Trados many years ago). I even shouted at one (politely using CAPS LOCK on Skype). It doesn't work. Old habits die hard. They are too busy just now.
So here is what I propose - I already know that the speed reports in iOmegaT are useful to measure and improve MT on large documentation projects. I have yet to meet a translator who is not curious as to what he earns per hour and per client and I don’t think translators will mind sharing raw speed data (no linguistic info, no client info, no price info) with entities that are in a position to improve that speed (e.g. MT companies or end clients).
ReplyDeleteIn fact there is a precedent for this in the latest version of MemoQ.
If the other CAT tool companies can be persuaded to follow Kilgray's lead and implement speed reports that remain private to translators but sharable at their discretion (e.g. with a company that provides consulting on how to set up and use Dragon for maximum accuracy) we might see greater take up. These user activity data logs can be much more sophisticated than the current Kilgray report and they can even be used for other kinds of training and optimization of CAT tool settings (e.g. the optimal length in characters for autocomplete).
If we can measure the impact of training in dictation for translators that are a few years into their career and allow productivity gain data to percolate through the community this can only help the cause. Think of time series graphs that look like share price reports published by translators who are learning to work with Dragon on their blogs and you will get the picture.
There is a lot of negativity around MT, not least because the messages around it are coming from corners of the industry that don’t pay well. I have the utmost respect for the skeptics. I was one. Now I am excited about new technology.
Allow me to riff - I picture my now four-year old son at my age spending the morning beside a pool with VR glasses dictating a medical text on a new transplant technique from Polish to English with the support of MT. From afar he looks like a kitten playing with string as his hands move in the air to separate out and recombine complex clauses in long sentences (this is based on your comment about using your hands while you dictate to demarcate phrases).
I imagine him going for a walk along a country lane after lunch interpreting text to speech on the fly for an easy text (no MT here) and sitting down later to fix it up (or paying a junior colleague to do that). Most of all, though he won’t be aware of it as it is facilitated by the RSI he DOESN’T suffer from, as he leaves his poolside office at four to practice guitar with his band, I imagine him thinking he is vaguely glad his dad studied computational linguistics. ‘Cos the MT was really useful that morning and traffic at rush hour is a bitch.