Facebook Twitter LinkedIn E-mail

Machine Translation Gold Rush

The dozen or so companies in the translation industry that are investing heavily in machine translation tools have a major opportunity for about 1-2 years to make a lot of money out of post-edited machine translation (PEMT), and then the margins will shrink.

Let me explain my thinking.

The Chinese Anomaly

About 10 years ago, typical pricing in the translation industry for Western European companies looked a bit like this (using very approximate rounded numbers to illustrate the point):

En > FIGS, Portuguese etc.: £100 per k [fair enough: not necessarily the cheapest combinations - although some of them are fairly cheap - but lots of supply and demand]

En > Eastern European: £120 per k [fair enough: cheaper combinations to source, but perhaps lower demand, so at least some explanation or higher prices]

En > Scandis + Asian including Chinese: £140 per k [mostly fine, but what is Chinese doing there? Lots of demand, lots of supply, and very cheap supply at that]

So there was an anomaly in pricing for Chinese. Margins were massive: £20 per k for a (good, professional) translator, maybe £5 per k for proofreading, but buyers willing to pay £140+. How did that work (and indeed does still work for some companies and some buyers)?

Well Chinese and all Asian languages were seen as being scary and distant and exotic. The internet was young, and the world of commerce hadn’t yet shrunk in everybody’s minds. But then gradually some LSPs started pricing Chinese (and other languages) based on cost of procurement, rather than on cost that they could get away with, and then much of the industry quickly followed suit, and this goldrush on Chinese margins was over.

Machine Translation Gold Rush

So back to machine translation. Companies (and freelance translators) that are using PEMT very effectively and ahead of the industry curve right now are producing high quality (often even better than pure human quality) output with PEMT workflows, but charging rates for full human translation. Buyers (either end-buyers or LSPs buying from savvy freelancers) are happy and none the wiser because they are receiving the quality that they require, and the PEMT users are keeping quiet about their techniques.

What’s the difference in margin between delivering high quality human output and high quality PEMT? Let’s forget about the costs of the MT tools (which obviously you can’t, but they are an overhead that can be spread over several years and clients potentially), and look at a corporate buyer of translations buying from a PEMT enabled LSP using PEMT without telling the client.

Human translation: Let’s say the corporate buyer is paying £160 per k for T+P. The LSP is paying a translator £60 per k, and an editor £20 per k. That’s a margin of 50% – and a profit of £80 per k for the LSP. Good, but not atypical by any means.

PEMT: The corporate buyer is still paying £160 per k. The LSP puts the content through their MT tool. We’re ignoring the overhead, but let’s assign a cost of £10 per k. Then we send it to a PEMT-er, who is paid £30 per k. We could even give it to a second editor to be on the safe side, who we might pay less as they are doing more of a review task, so let’s pay that second person £10 per k for a quick check. Our profit is now £110 per k for the LSP, and this is for a pretty solid workflow still involving two humans. And the output of this, if done properly, will still be very high quality.

Before too long, that £30 per k margin gap will shrink and disappear as PEMT workflows become more normal, and pricing adjusts to reflect this norm.

So, anyone using PEMT now and not telling their clients: make the most of it, or be the ones to tell you clients loud and clear and force the industry to follow you!