I couldn’t have been more pleased with the reaction to my last post a few weeks ago. Lots of people have told me, here on my blog and on social media, that they were delighted someone had started talking about the translation process. Some were even kind enough to share their processes with me and other readers. There were slight differences of course, but it seems to me that the most significant divide was between those who, like me, make a very rough first draft and then polish it in successive rounds of revision, and those who put much more into their first draft to make it as near as possible to the finished article.
As I said in that post, I had started to wonder whether it was really sensible to put so much time into a first draft which did little more than extract the meaning of the original and still needed a great deal of work to turn it into polished English. Was there a way of getting to that stage a little more quickly so I could invest more time and effort into what I could see were more important translation stages?
A possible answer came to me when I was working on a revision job where the English was poor and the original was full of convoluted sentences. Just to be sure I hadn’t missed something, I ran some paragraphs of the original through a commercially available machine translation service to how the result matched up to the English I actually had. I wanted to make sure no meaning had been lost, and this I did, but I was amazed at how similar the results were to what my own first translation draft of that piece of text would have been. What if, I wondered, I used machine translation to provide my first draft? Wouldn’t that be the solution I was looking for? And that is what I have been experimenting with for the last few weeks.
It was a big step for me as up to now I have always vociferously opposed translators using machine translation. At the ITI Conference a couple of years ago, I absolutely agreed with Chris Durban when she urged us to “Eschew MT” and not only because I love the word “eschew”. So I’m well aware that what I’m talking about here is going to be controversial and I realise that I even risk losing a few friendships over it. With that in mind, before going through the results I’ve achieved and the lessons I’ve learned from my experiment, I want to make a few things absolutely clear.
- I’ve been working with a paid-for machine translation service. That means there are no confidentiality issues and the service doesn’t get my translations. In any case, I haven’t experimented with any confidential texts.
- I do not believe machine translation is going to provide an adequate translation of anything for me. What I have been using it for is to obtain a very rough first draft, of a similar quality to my own rough drafts which were so sketchy that I’d never have shown one to anybody.
- I have not been doing post-editing on my machine translation, I have been using it as a basis for a human translation. If MT has improved since I last looked at it – and it has – that improvement means that it is now not always quicker to start from scratch when working with it. But often it needs thorougly rewriting and reworking. It also needs checking with the original source. So I see my new second stage (the first one being the MT) as a translation stage, not a revision stage. I’m just using the MT as a guide to the meaning. But how that meaning is eventually expressed is up to me, not the machine.
- I’ve saved very little time overall by using MT, I’ve just changed the way I use the time I have available. Instead of spending the bulk of it producing a very rough draft that still needs lots of work, in my experiment I’ve found that not only do I have more time to spend on the second and third stages of my process, I’ve even been able to introduce an extra stage. So, after the first draft, now done by machine, I do my second translation stage, working in MemoQ with the MT to try to produce something as near perfect as possible, and then my third stage, which is a revision of my translation also carried out in MemoQ. I then spellcheck, quality check and export the document and then carry out a final revision stage in the original format before spellchecking again and then running PerfectIt.
So, to sum up, am I saying that MT produces good, usable translations? No, but it does produce raw first-draft material similar to what I was working with before. Nor am I saying that its translations can be made usable with a little bit of post-editing. When I am translating while looking at the machine’s output, I have one eye on the original all the time and I’m looking critically at what the computer has produced. Is anything wrong? Has anything been missed out? Could I express that better or this another way? So it’s nowhere near the finished article, it’s just a first step in the translation process.
So what are the advantages?
- Time. Using MT I get my first draft in a few minutes rather than taking several hours over it. That leaves time for a much more thorough second draft, a third to polish the English even more, and even a fourth in the final format. I’m using that speed to improve the end product, not to produce poor quality work faster and cheaper.
- Inspiration. We’ve all sat and agonised over finding the right word without being able to bring it to mind. Quite often the MT will throw up precisely that word. And if it doesn’t, I’ve got more time to look for it.
- Perspective. As I’ve mentioned before, I find it difficult to simultaneously extract the meaning from a sentence in one of my source languages and find the best way to express that thought in English. But when I’ve got the meaning set out in front of me, my mind is freed up to play around with the alternatives in English. Do I need a different word here? Should I turn that sentence round? What WOULD we really say?
What about the disadvantages?
- Mistakes. You do have to check the MT carefully against the source because there are mistakes. It’s not to be trusted absolutely.
- Inconsistency. Contrary to what you might think, computers do not provide consistent translations. Quite the reverse. Looking blindly at one segment at a time, they’re quite capable of translating the same word three or four different ways in three or four paragraphs. In my view, a little flexibility is a good thing, but mostly clients are going to want consistent translations, which means I have to work hard to iron out the MT’s little foibles.
- Clunkiness. Just like my own first drafts, MT can sound awkward and clunky. In short, formulaic sentences where there aren’t many options, that maybe doesn’t matter, but for anything longer I have to make sure none of this comes through in my final version, which means treating the machine’s version as a starting point and not an almost-finished product.
My overall verdict on my experiment is positive. I’ve felt much better and less frustrated about the work I’m producing. And because the machine translation, unlike my own first draft, is not my work, I find it much easier to be ultra-critical and turn it upside down or throw it out entirely if necessary. My clients seem happy with the translations they’re getting and I’ve had positive reactions from revisers, none of whom actually knew I’d changed my process. One even complemented me on a word choice which had actually come from the MT.
I realise, of course, that I’m not the first translator to do this, and that lots of people have been using MT in this or similar ways for some time. I’m writing about it for two reasons: firstly honesty – after what I’ve said in the past, I don’t think it’s right that people should go on thinking I’m against machine translation if I’m actually using it. And secondly, I want it to be talked about. Because of the climate of controversy existing around MT, I think people feel they either have to be wildly enthusiastic about it or stubbornly resistant and there’s no room for a proper discussion of when and how it should and shouldn’t be used. I’m not sure that it would benefit everyone in the same way as it does me, for example, depending on the method they use. So I hope this post is a little step forward in breaking the silence.