A Pragmatic Eye

Threlford Lecture 17 September 2016

I live in the northern-most reaches of London so I figure I really ought to make the effort when translation-related events are being held somewhere I can reach with my Oyster card. Even so, I nearly didn’t get to this lecture on account of being slow to register and having to join a waiting list, so my thanks to those who were decent enough to make their tickets available rather than simply not show up.

After a wander around the back streets around Ludgate Hill (deliberately, I mean, having a gawp at relics from the Olden Days; I wasn’t lost or anything), I ambled down the steps into the Stationers’ Hall, which is a splendid building, all dark wood panels and leaded windows and heraldry and lists of people who have held various obscurely-named positions dating back to the 16th century and discreet, sprucely turned-out staff. Upon arrival, a table with several neat rows of wine glasses raised your correspondent’s hopes, the sun being well over the yard arm by this time, but investigation revealed the contents to be water. I knocked a couple back anyway and had a chat about the new incarnation of Chartered Linguist status.

We all drifted into the main hall for the afternoon’s action. After a review of the Institute’s year, it was time for the headline act, Professor Dorothy Kenny from Dublin for a “critique of contemporary translation technology”. I can but recommend you go and listen to her talk if you get the opportunity.

While CATs and speech recognition got a mention, the talk was mainly about MT and its consequences, including the need for post editing.

It made a pleasant change to a) hear someone talking about MT and PEMT, as against reading about it or watching a video, and b) hear about MT and PEMT from someone who appears relatively disinterested, being neither trying to sell us something or railing against its evils. Yet it was not a solely academic approach taking MT in pure isolation; for instance, the issue of confidentiality was raised given the cloud-based nature of most systems, as was crowdsourcing to fix MT output.

I’ve no intention of summarising the whole thing, but quite early on she drew an interesting parallel between the ideas of reusing versus recycling (e.g. paper), and the ideas behind CAT/TM systems and SMT (i.e. statistical MT). If you reuse a bit of paper, it stays a bit of paper, identifiably so, as you do so. Whereas if you recycle paper, it will go through a stage where it is pulp before becoming paper again. CAT tools are thus like reuse – the inputs and outputs are what they always were. Whereas SMT is more like recycling, and the reason the output looks the way it does is because of the “pulping” stage the texts go through. This idea was new to me, and does seem to explain why MT output can be unexpectedly odd.

Prof. Kenny also raised a couple of interesting questions.

Firstly, does the acceptance of (or attempts to get translators to accept) post-editing of statistical MT output take resources away from more productive or useful areas that are excluded as a result of an assumption that PEMT is the way forward? One obvious area would be to improve MT output in the first place (apparently neural networks are on the horizon as far this goes). Another area would be improved handling of terminology. A third might be speech recognition to replace any need for MT at all.

Secondly, and again relating to PEMT, is there a psychological effect from consistently working to a lower quality standard, given the current market that uses PEMT appears universally satisfied with “just good enough”? Assuming that for ethical and job-satisfaction reasons, we strive for excellence at all times, can this be reconciled? To paraphrase the good professor who posited the idea without descending to the vernacular, is it possible to be “excellent at being just good enough” without your brain exploding?

I’ve thought about this last question a good deal since. Translators are a funny bunch. On the one hand, they swear blind they could never, ever possibly countenance even considering the idea that they could “compromise on quality” and that every piece of work they ever deliver is beyond reproach, with due account taken of the maxim “errare humanum est” and the fact spell-checkers don’t catch everything. On the other hand, they also love that quality-time-price triangle and the famous “pick any 2 from 3” tagline, which implies heavily that quality will suffer under the wrong time/price combo. Given we’ve surely all done the occasional job that was either for less money or delivered sooner than we would have liked, or even both (whether out of desperation or expediency), then was the quality of that work as good as it would have been if you had chosen the rate and deadline yourself?

I’m going to assume not. But was the quality good enough for the client’s purposes…? To go back to the original question, then:
a) in theory, surely, it must be possible to be excellent at anything, if you define the parameters;
b) two weeks later, my brain continues to wrestle with the abstract idea of being excellent at “just good enough” which can obviously apply to many fields – football, golf, driving a car, maths homework, cooking, and so on. In each case, it seems there is an unsaid aspect, namely “good enough” for what?
c) a quality of “just good enough” is therefore rarely a standalone requirement, and for translation it usually comes with constraints on price and time. If you include those aspects with the need of “just good enough”, then I would think it is possible to become excellent at meeting the needs of such clients. I’d see a potential parallel with the likes of Lidl or EasyJet, who are excellent at providing a fairly basic level of quality;
d) in terms of PEMT, which is where this question started, then in future, it seems likely that MT will be available on standalone desktop computers, without being mainly controlled by cloud-based LSP giants or wannabes, with the concomitant pressure on time and cost. Under those circumstances, there will be no need for PEMT to frequently imply “just good enough” quality; MT could be used by a freelancer as and when he or she sees fit (and indeed with no need for clients to know MT was used to produce the first draft) and thoroughly post-edited at the translator’s leisure (if you see what I mean). Question dodged rather than answered, perhaps.

Anyway, back to the event. There wasn’t much time for Q&A after the talk, and we quickly moved onwards and upwards into yet another panelled room, with a chandelier and everything. These old-time stationers certainly seemed to like having plenty of space and didn’t skimp on fixtures and fittings. ‘Course, maybe Rymans’ head office is just like this even now. This time the table with rows of wine glasses was not an anticlimax, both red and white being on offer (I’ve mostly been a rosé man since the mid-90s but I was prepared not to quibble on this occasion). The mingling and networking session seemed to be a success (my experience was entirely agreeable, at least) and ran over the allotted time, until we eventually dribbed and drabbed out into the tea-time tourists taking in St Pauls.

All in all, highly recommended. Do go next time.