So you'd like to save 20%-50% on your translations, take out the middleman and work with freelance translators directly.
Not only will you save money, you can also speed up delivery by ensuring direct and smooth communication. Last but not least, you get full control over your project.
Here I will tell you how you can build your own localization department.Setting up an internal localization department
The biggest problem with translation is that everybody can call themselves a translator. There are many bad translators, and there is no clear and safe way to separate the good from the bad. There are some indicators which you can use though. Note: this text is long enough as it is, and even then I had no choice but to base the below on a few generalizations - of course there are always exceptions to the rule.The basic criteria
Anyone offering a rate less than 0.07 euro per word for western languages should at least be approached with caution. Unless you live in India, this is not a rate you can make a living with, indicating that the translator is desperate (no client wants him) or merely works for fun (no professional experience). On the other side, anyone offering a rate higher than 0.14 euro might be out of your league - these people mostly work in other fields with much higher rates (think patent translations and the like). Normally a game translator should charge 0.08-0.13 euro per word. Once more: this is for western languages only. Languages like Japanese and Chinese are much more expensive.
I have been translating full-time for 27 years and am still learning. When I look at translations from 3 years ago, I still see room for improvement. I'm quite sure this trend will continue the rest of my life.
As a rule of thumb, a translator should have at least 5 years' experience before starting to work for direct clients. Translators with less experience are better off working for a translation agency, which may have procedures to proofread their translations and give feedback.
A professional translator should have several references available - clients who have outsourced relatively large-scale projects with satisfactory results. Check these references. Many translators put Microsoft on their resume as soon as they receive a Microsoft manual translation from a local translation agency, but this does not always mean they have directly worked with Microsoft before. Ask for both direct and indirect references.
Note that in the world of game localization, some agencies forbid their translators to mention any games they have localized for them in their resumes. Sometimes this is because of non-disclosure agreements, but often it's merely a way to safeguard their position as a middleman.
An average translator translates 2000 words per day (western languages) in total. Good (as in fast) translators can translate 4000-6000 words per day. Higher quantities are highly unlikely or cannot be maintained for longer periods. If a translator translates less than 2000 words per day, they are either very slow, only works part-time (no professional experience) or simply has never worked as a translator before. In that case, they are merely making a wild guess as to how many words he/she can translate. If a translator translates more than 6000 words per day (in total), he/she will definitely not be able to maintain this speed for longer periods at a stretch, delivers translations of very dubious quality or is once more making a wild guess.
A good translator either has a strong technical background (for example: a game development curriculum) combined with linguistic experience (for example: writing software manuals) or is someone with a strong linguistic background (for example: an MA in translation) combined with technical experience (for example: playing games).
CAT (Computer Assisted Translation) tools are special tools developed for translators, to ensure that they use consistent terminology and style (these are not Machine Translation tools like Google Translate, which give horrible results). There are many tools out there, and most of them are compatible. If your game contains many menu options and references to these in the manual, a CAT tool will come in handy to ensure the translator always uses the same translation for a certain button. If you expect many future repetitions (that is: you expect future translations will often contain sentences that have been translated before - so-called "matches": think manual updates, version updates, UI updates, et cetera), CAT tools are a must, as it's customary to offer a discount on repetitions. Good translators own at least one CAT tool. The costs vary from a few hundred to a few thousand euro, so CAT tools also indicate whether the translator has made any serious investments in their career (read: takes themselves seriously).
CAT tools are a difficult subject and especially in the beginning, you had better leave the inner workings and analyses to the translator. But they should at least own one and use it. Save future price negotiations about repetitions for later and first obtain some insight in how these things work. This and this are the CAT tools I use, by the way.Gathering quotes
So, now you know the basic criteria. The following procedures are a bit time-consuming, but will only need to be carried out once in a lifetime. So don't cut corners and consider this as a long term investment. If you follow this procedure (which will take you about 1 or 2 hours per language), you will be able to save 20%-50% on translations for the rest of your life, by taking out the middlemen. A bit like Dell, really.
The next thing you do is gathering resumes and quotes, for example using LinkedIn.
Now offer a potential job. Specify your source/target languages and some basic criteria (experience with games, a sample text they need to translate) and wait.
After a few days, you will have received several job offers. Check whether all of these meet the above-mentioned criteria. If not, throw said resume away. The people who are left are the people you are going to test (not meeting the basic criteria means the translator is not the right candidate for the job - it does not mean that all remaining translators are good translators).
If you read the above well, you had all translators translate a short sample text (300 words should be enough to give you a good impression). Make sure you provide some context so that the translators know what the test is about and don't choose a passage that is too easy. For example, if your game uses much dialogue, you can pick a passage with word plays (not always translatable, but see how creative your translators are) or lots of obscure slang.Judging the sample translations
When an agency in New Zealand outsources a translation from English to Danish, you can bet they have no in-house people to check whether the result is actually satisfactory. In those cases, the agency has the translation checked by a second external translator, and sometimes even a third, to make sure the second translator does not grab the opportunity to run the translation into the ground merely to obtain a new client (this unfortunately happens very often).
Language is not like mathematics. Some translators have a great writing style; others not. The problem is that even if a translation has correct spelling and grammar, this does not mean the translation reads like an original. Unfortunately, you can't win discussions about style, as you simply can't come up with any rational arguments. You either like a translation or you don't. This is why during the first phase, you should concentrate on grammar and spelling only.
Have the translations checked by other translators who responded to your quote (this will cost you a bit of money - per language the costs should be about 30-40 euro for one review). If you don't have any budget for this, skip this step and go to "The final test", but realize the outcome of your evaluation will be less reliable. Otherwise, continue with this step.
Make very clear that the reviewers should focus on spelling and grammar only to prevent yes/no discussions, and let them come up with arguments for every mistake they find. An argument is not: "this is wrong" or "I don't like it". An argument is: "this and this dictionary states this is wrong" or "according to Dutch grammar, this verb should end with a t". Instruct your proofreaders and give examples of what kind of arguments you would like them to use (feel free to use the examples above).
Now, hand out the same translations to a second reviewer and repeat the procedure. You should now have two proofread translations of each sample translation. Let's take one somewhat simplified example:
Dutch Translator 1 - Sample translation
Dutch Translator 2 - Evaluation. Translator 2 liked translator 1's translation and found no mistakes.
Dutch Translator 3 - Evaluation. Translator 3 did not like translator 1's translation and claims it is infested with errors. No arguments are given.
Dutch Translator 2 - Sample translation
Dutch Translator 1 - Evaluation. Translator 1 did not like the style of translator 2, but realizes he has no arguments. He found 1 spelling mistake, nicely pointed out with a dictionary reference.
Dutch Translator 3 - Evaluation. Translator 3 did not like translator 2's translation and claims it is infested with errors. No arguments are given.
Dutch Translator 3 - Sample translation
Dutch Translator 2 - Evaluation. Translator 2 found 12 spelling and 2 grammar mistakes, all pointed out with a dictionary reference. She strongly advises you not to use translator 3.
Dutch Translator 1 - Evaluation. Translator 1 found 14 spelling and 3 grammar mistakes, all pointed out with a dictionary reference. He strongly advises you not to use translator 3.
Who is your ideal candidate? It's translator 1. Translator 3's translation is infested with mistakes (according to both translator 1 and 2) and has shown very amateur behaviour. He claims both the translations of translator 1 and 2 are terrible, but doesn't come up with any arguments, despite your instructions.
Translator 2 comes second. Translator 1 found 1 mistake, explained why this was a mistake, and says he doesn't like the style of translator 2 very much (always take remarks about style with a little grain of salt, as this is very personal and no argument - treat it as extra info). Translator 3's judgment can't be taken seriously (see above).
Translator 1 comes first. Translator 2 found no mistakes and has no negative remarks about the style. Translator 3's judgment can't be taken seriously (see above). Also note that translator 1 was able to find more mistakes in translator 3's translation than translator 2, indicating he works more accurately.
So that's how it works. You now have a pool of 2 translators for Dutch. You try to use translator 1 as often as possible, and use translator 2 as back-up. The more translators you test and the more often you have the same translation proofread by different translators, the more reliable your evaluation will be. In this example I've tried to keep things simple though.The final test
Now it's time to do the final test. Show the sample translation of translator 1, 2 and 3 to your user base or agents and ask whether they like it. Since your users/agents are no professional translators, you cannot expect them to make actual corrections and come up with arguments, but at least you should be able to get a general idea about the quality.
Remember that hardcore gamers will often claim they don't need a translation or that the translation is terrible just because "it doesn't sound right" or "because they are not used to it". For more information about this particular phenomenon, see the FAQ (General concerns - Dutch is uncool) on this site. Tell them you are not interested in their opinions about translation itself - you merely want them to judge the translation as it is.
Normally, the result of this evaluation will be the same. Your users will state that the translation of translator 3 is a disgrace, the translation of translator 2 is so-so and the translation of translator 1 is perfect. If they don't like any of the translations, you have a problem. In that case you will need to test more translators. Once more make sure they give their opinion about the translation as it is though, and not about the phenomenon of translation itself. Also make sure your users don't run a translation into the ground because it contains one minor error. These are the same type of users that run your game into the ground because it contains one minor bug.
And that's it. You now have your very own translation agency/localization department and can start using the services of your freelancers to your heart's content!Regular check-ups
It's always a good idea to check whether your translators are maintaining high quality standards. Some of them spend hours on a sample translation, only to rush through things once the real work is coming. Therefore, especially in the beginning, you should pick 300 word samples from the translations they delivered and have them checked regularly (first weekly, then monthly, then yearly) by the translator who is second on your list or your user base.
Never attack a translator if other translators or users claim the translation is bad. Always give the translator an opportunity to defend themselves. Be diplomatic and state that so-and-so says this-and-this, that you don't immediately believe them, but that you're interested in hearing the translator's response. Make sure the translator comes up with valid arguments though. And once more: don't forget there's no accounting for tastes (style). Focus on grammar and spelling only. Of course you can't use a translator if your user base doesn't like their style, but that doesn't mean you have the right to attack them or claim they are a bad translator. If the style does not fit your audience, "things simply didn't work out".
Copying and pasting translations
Never, ever cut corners by copying and pasting previous translations from your translator. This can have absolutely disastrous results. I once did a translation about a space ship simulator - in this context space is translated as ruimte (space as in universe). The client copied and pasted the word ruimte to a message popping up every 3 seconds, instructing the user to press SPACE (as in space bar) to jump, without consulting me beforehand. The result was that the user was instructed to press the universe every so many seconds. The end result, reminiscent of Zero Wing (All Your Base Are Belong To Us), was run into the ground by all review sites. Instruct your programmers to never, ever touch translations if they don't speak the target language and always consult your translator.
Using CAT tools
Ask your translators to send you regular updates of the translation memories of their CAT tool (preferably in the tool's native format), containing all translations they have done for you before (it is not reasonable to ask for translation memories with translations done for other clients). If you don't know what you can do with these memories, read up about CAT tools on the internet first (try Googling a bit). Even if you won't use CAT tools yourself in the beginning, they might prove very useful in the future. It's not a good idea to actually start using translation memories yourself until you have a very good understanding about the pitfalls and possibilities. Until then, just make sure you have a copy of them for future use.
What is a slip of the tongue, and what is actual malpractice?
Translators are humans and humans make mistakes. There's no reason to fire your translator because of one or two mistakes. As a general rule of thumb, the Dutch Association of Translation Agencies says that more than one error per 500 words can be considered as substandard. Personally I believe that a translation should definitely not contain more than one error per 1000 words. I strive for 1 error per 5000 thousand words.
If your translator delivers a translation with more errors, they should at least apologize, or even better: offer a discount. Based on your previous experiences, you can then decide whether you accept this discount or not. If the translation is clearly substandard (say 1 error every 100 words), you can tear the invoice apart. That's not substandard, that's malpractice.
Respect your translators and treat them as you would like to be treated yourself. You get your wages every month - therefore make sure the translator gets paid within 30 days. Translators have a family to take care for too. Note that translators exchange blacklists with defaulters using all kinds of venues. If you think it's perfectly acceptable to ignore three reminders in a row and pay after 3 months or so, you will end up there, which will make it virtually impossible to continue your business. Nobody wants that.
Always proofread and playtest
A short addition translated by a programmer instead of a translator, a mistake made while copying and pasting a text into the game... you do not want to know how many perfect translations are completely messed up by the client themselves. In fact, this happens in the vast majority of the cases. Always have the end result proofread and playtested by your translator, and don't let things go wrong in the final stage of the localization. Remember: a poor localization is worse than no localization at all.
Loek van Kooten
Your English/Japanese-Dutch game translator