This is a post I have wanted to write for a long time – that time has come! From the moment we made teh decision to move away from our core CMS technology and morph into a translation company we have been driven by a belief that focusing on improving translator efficiency was a far more logical way of improving the cost of translation. As we watched different crowdsouricing platforms emerge (and get funded with not insignificant amounts of money), and existing translation agencies try and tweak their existing business models we kept our head down and focused in on solving a number of really difficult problems centered on the belief that improving translator productivity was the right approach.
The reason we haven’t pushed into the public domain is simply because from our view we didnt have enough data to show, without doubt, that its possible to improve the words/hr a translator can achieve aand that subsequently you can have a far greater impact on improving the cost dynamics of delivering a human translation than by other methods (crowdsourcing, workflow optimisation etc). The good news is that we now do have that data – and a lot of it (north of 10M words translated/58 different languages/almost every content area imaginable). If your interested in findingout what we have worked through – read on:
Changing The Economics of WHat Determines the Cost of Translation
We don’t (and have never) believed crowdsourcing is the right model to pursue if your main aim is improving the cost of translations. Break out the translation vakue chain and no matter which way you cut it the “Translator” segment makes up the bulk of the cost. Yes you can optimise content workflows (using API’s, upload tools etc), drive down translator rates by aggregating the jobs betwen lost of translators (and assuming teh more desperate will offer teh lower rate).
So first up a few statments of fact:
1. There is no correlation between translator efficiency (words per hour translated) and the quality of the translation. Nada, none, zero. Its a complete misnomer – overlaying 1000 datapoints across speed and quality (measured by spot checks, client feedback, rework%) shows there is zero correlation.
2. Languages do not matter. The consistently fastest languages are Arabic, Spanish, Chinese and French. Slowest Italian, German and Korean.
3. Post Edit is vital. Machine pre-translating the content improves translation %’s by at least 40% for most languages.
4. Workbench enviornments count. We make changes it has an effect. There are usability factors that improve translator efficiency inside the workbench.
5. Content Familiarity is Golddust. Nothing (and I mean nothing) imapcts translator performance more than being able to work on the same clients content. The more familair they get the faster they go