History

The translation industry has certainly progressed during my time in it. When I first started, I remember one older gentleman who would bring me a translation scribbled onto loose pieces of paper. I had to hire a typist, who would fix some of his blatant mistakes, and then a proofreader to polish up the language.

For a long time I would have to take a 45 minute bus to the city center to deliver a nicely printed translation in a plastic sleeve with a floppy disk in it. This was not fun when the job was only three pages, but at least I had a fax, saving me from having to pick it up as well. Then the need to also have a printer made entry into the industry a little more expensive.

Eventually the big floppies were replaced by an amazing technology, a hard, smaller version. Things were progressing at lightning speed.

At that time the rage was a dos-based software called T602, a local production. It even had the amazing capability of exporting the text based file into Rich Text Format (RTF). There was no mouse and you had to press TAB a lot of times when trying to format a document.

Then along came the amazing world of Windows 3.0. Later on Windows 95. I had a computer friend back in Vancouver, Canada, and boasted over the phone how I had a whopping 256MB of memory (no, not ram). He sent me by post some programs on two boxes of disks, one of which compressed everything, even my Windows operating system, such that I could now have more than 500MB of data on my disk! I was amazed, but my friend laughed that I will use that up soon enough.

Then there came modems. I was excited at the opportunity to send files to my customers over the phone instead of taking the long bus. I taught all my customers how to use it and set it up on their computers. One of them commented I was like a missionary.

Eventually along came the internet. At this point I was still working next to a massive bookshelf of dictionaries, within arms reach as I spent much of my day fumbling through pages of different books, looking for a particular term.

Of course eventually came digital dictionaries, and I presume you can imagine the rest.

Translation memory software

Lets say you work for a real estate agency and translate a lot of rental and other contract. Such as:

Bob hereby agrees to furnish Sue with the property located at 115 Deepdene Drive, Prague, Czech Republic.

And later you get another translation job:

Tom hereby agrees to furnish Sam with the property located at 417 Sussex Ave, Prague, Czech Republic

A good translation memory (TM) software will take your translation of the plain, non-bold text and replace the bold text with the new bold text. The software does not know that they are names that do not need to be translated. And even if they were not names and did need to be translated, you would only need to translate the bold text.

Another big advantage is that the software usually has an internal dictionary which you can build up yourself, so every time a word in the dictionary appears in the source or target file, it will be marked in some way and the translation of that term shown in a little window. This saves having to go to some other software, type in the word and so on.

There are more advantages but these two are the main ones.

Different software

Over the years I have tried all sorts of software (because different agencies use different software and you are forced to install/purchase if you want work from them).

The most popular seems Trados. Easy to set up a project and quickly get statistics on a file, such as the total number of words and the fuzzy matches.

Fuzzy matches are the percentage of words or sentences in the document that duplicate, or are similar or the same to your previous TM. In our rental contract example above, that looks like it would have a fuzzy match of roughly 80% (for that particular sentence). Meaning that roughly 80% of the sentence has already been translated and you only have to worry about the remaining 20%. In our example, you would be lucky because you could just quickly look at it and realize it does not need to be translated at all, because it concerns names.

Here is a table showing the standard in the industry concerning payment of fuzzy matches:

<

Obviously the higher the percentage fuzzies, the less you would get paid from the base price.

A lot of agencies demand their translator to have Trados, and because of its relative ease of use, those are probably the primary reasons why it is the most popular. I tried it a few times but personally found it rather slow, comparatively to my preferred software.

When I was in the market of buying such a software, I had landed the greatest mega project < of my career. 3.5 million words in eleven languages, distributed to just over one hundred translators. I did some quick research and consultation with an expert and eventually decided on the below. I’ll go over some of its advantages and how I use it. Not as a sales pitch for the company but just to show some of the features that such software should have. It was expensive, but I was not looking for software for myself as a translator, but as a project manager, which is quite different.

Termstar Star Transit <

The main reasons I chose this software over Trados were the following:

• much more powerful from a project manager’s point of view
• there was a free version for the translator, whereas the translator version of Trados was still fairly expensive. I did not want to limit my pool of translators to only those who had this software

Later on when I reverted back to me personally translating, and forced to use Trados because of some customer, I found Transit overall much faster to work with, and with more tools.

Below I will list some of its qualities. Could easily be overkill considering your actual needs as a translator, but at least you can see what such a software is capable of, for your own personal consideration. On a side note, there are numerous software which work to some degree with Microsoft Word, meaning that the agency can prepare files in Word so that the translator is not forced to buy the same software.

Continuation from Part 1 of How to Manage Large Translation Projects

Start the project

Now that you have put together your glossary and translated your repetitions, you can begin the main part of the translation. Hopefully you will have accomplished all this during the first month or set up period, while you continue to receive and assess applications, and develop a clearer picture of the quality levels of your chosen lead translators and proofreaders. Them having already spent the month assessing samples, working on and discussing the glossary, and translating the repetitions, they have also become seasoned and more ready for main project launch. During this initial period you may also discuss extensively with the customer about terminology and other matters. This will greatly increase their confidence on your handle of things, and better prepare you for the main launch. You can also work on the webpages to explain all particulars to the remaining translators once you start to use them.

For a project of this size and calibre, I like to parcel out files with size of about 2,500 words. This should be accomplishable within about one day by the average translator. You can assign them several files to keep them busy. For such projects I set up a dedicated server with its own IP address. Project managers can log in and work with the TM software to create these smaller files. They save these files on the server, from where the translators can download them. As translators are assigned new files, their assignment can be shown on the server (visible on the internet because of your dedicated IP address, password protected if you so choose) so they can check they are downloading the right files, or use the system (explained later) so they can only download the files they are assigned.

A simple snapshot of the accounting in Excel, showing which files were assigned to whom, with the total word count for each file. But this can be easily copied into a webpage, or better yet, to use the system so translators can only download files they are assigned, to prevent any mistakes.

For example, the customer may send you files averaging about 50k words in size. You break those up into consumable 2.5k sizes (that is, 2.5k of NEW words, not including the repeating text which have already been translated during the initial phase). If the file is called Technical_details.doc, you might create files called Technical_details_01.zip, Technical_details_02.zip and so forth. This makes it easier to understand and manage, for everyone.

When parceling out files like this, you will also want to put some thought into it. Some sections of a file may be more unique than others. If you spend some time studying the material, you might find that certain parts are more appropriate for certain translators. Perhaps those translators are more technically adept. Even though they might charge more per word, you might decide it is better to invest more to have it done properly, and save the easier parts for the less expensive translators. Parcel things out strategically to produce a top notch translation.

Test the translators as you proceed

As I mentioned before, it is not uncommon that a translator produces a sample not reflecting their regular work. Make sure to warn them that you expect the same quality as their sample, otherwise penalties will result. When they send back their first translation batch, have it quickly assessed. If it does not reflect their sample quality, you can ask them to stop, have it proofread, and show them the changes made. If the changes are extensive, you can deduct those additional costs from their agreed rate. If they are not happy with that, you can pay them in full but tell them they are no longer part of the project. This can be a tricky issue. Some translators can be so brazen as to resort to google translation for a quick buck.

Once you are confident in their consistent quality, you can continue to assign them more files.

Proofread the files and import them into the TM software

As your first files are translated and start rolling in, have them proofread and finalised by the vettors as soon as possible. You can then import them into the TM software, adding to the memory. This will result in additional savings that were not found in the initial analysis of repeating text, increasing your profits. Use your filters to prevent the proofreaders from proofing text that is repeating and already proofed and finalised previously – additional savings and higher profits. This is another reason why it is good to farm out the work in smaller chunks. Not only can it be imported back into the software to reduce costs by finding more repeating text/fuzzy matches (which will only materialise when you produce new files AFTER you have imported more files), but it will keep you in contact with the translators on a daily basis and give you a clearer picture of progress and potential issues.

Communication System

One of the important aspects of a successful project is the ability for translators to communicate with the lead translators, or proofreaders, or project managers. For this I have developed a system where I can assign translators to a particular project, they are only able to download files they are assigned, and can communicate with other members of the team, according to my choosing. For example, I might make it possible for the individual translators to ask questions to the lead translator, but not be able to communicate with one other. Divide and conquer, as they say. When they communicate with one another, everything is anonymous and they do not know the email address of the person they are communicating with. I can assign my own usernames, such as Dave_LT, where LT would stand for Lead Translator, or Sam_P as a proofreader or John_PM as a project manager – to give everyone a clearer picture who they are communicating with.

All correspondence can be easily downloaded and searched for messages where translators attempt to establish direct communication with one another, if this is important to you.

Accounting System

The same system described above also has an accounting portion. When translators are assigned files, each assignment record also show the number of words, with fuzzy matches. Different fuzzy matches are usually paid different rates. I personally use the following standard.

For those who do not know what fuzzy matches are, it might be where 85% of a sentence or segment is similar to a previous translation. When the translator gets to that segment, a window appears showing how it has been translated previously, with the new words in a different colour. The new words might be a different name or numbers. Such as:

Bob agrees to these conditions for a payment of $500.
George agrees to these conditions for a payment of $600.

The software automatically keeps the translation of “agrees to these conditions for a payment of” and inserts the name George and the new number $600 into the proposed translation. In this case neither would need to be changed. But if George was actually a word which needs to be translated, the translator would only have to translate that one word. 85% of the sentence has already been translated for them, for which reason they should be justifiably paid less. But because they only have to translate one word, they get paid more for their time, while you save from having the rest of the sentence translated again.

This is the advantage of translation memory software, and for such large projects, as you continue to import the translated files back into it, you will find that the fuzzy matches will continue to increase, translators will be earning more for their time spent, and you will be saving more (increased profits). So everyone benefits (except of course the translator in the sense that they will have less work, but this is simply the nature of this dynamic industry).

Again a snapshot of the Excel accounting file showing who which files were assigned to, the fuzzy numbers, and the total word calculation in bold. A part of the calculation formula is at the top. The translation project management system will show each translator only those figures associated with files assigned to them.

So as the project progresses, each translator, when they log into their account, will see a summation of their accounting. They will see which files they have been assigned, when they delivered it, how many words and total payment (broken down into fuzzy matches etc), when they have been paid for which files, and how much they are still due. This greatly increases the confidence of each translator. If they were to have to manage all this themselves, not only would it cost them precious time they should be allocating for the actual translation, but they might feel somewhat in the dark and unsure. They can always confirm the figures themselves, but having it all displayed like this will give them the confidence to continue without hesitation. Trust me, it is not easy finding good translators, and you want the best ones to work full throttle.

Payment

This in itself can be a time-consuming migraine. One easy solution is to request that all translators accept payment by paypal, or some other service such as moneybookers/skrill. With either it is easy to create a simple text file using the accounting system above to make a mass payment to everyone at once. Some banks also offer the option of making mass payments. Using some text file that can be created from the online accounting system. Most translators these days have a paypal account and using such a system makes management a lot easier. The accounting shown to the administrator of the project would be obviously much different and more thorough than that shown to the individual translators.

Vetting and Project Management

Once the files are imported into the TM software and exported back into the format provided by the customer, they need to be vetted, to make sure that nothing is missing and that everything is according to the customer’s instructions. This is not a difficult process and I always managed to find people on the net willing to do this for around $3 an hour (I can help you with that as well). They might be located in Nepal, as some were in my past projects, where such a wage is good for them. These vettors can be assigned communication rights with the translators, or only with the lead project manager (PM). Whose job would then be to forward any messages to respective translators to fix any issues. Or possibly only to reliable translators assigned the task of finalising the files. If there are too many mistakes in a particular file, the original translator could be penalised and a file sent to them showing the changes required for finalisation. This process will keep the translators on their toes and ensure higher quality.

An example of a vetting job, comparing the final Word translation against the PDF source file. The vettor does not necessarily need to know both languages but count lines, points, paragraphs and compare numbers and formatting to make sure nothing is missing and that the formatting is correct. If something seems odd language-wise, they can tag it and send it to a trusted translator to look at.

Once the different files have been finalised, they can be pieced back together into the large file chunks originally provided by the customer. By then hopefully it is absolutely perfect. By parceling up the files into smaller chunks, you can create a rolling production where it does not take too long before you start delivering the files back to the customer. They too will want assurance that everything is top notch and on schedule. You can also show them a portion of the accounting displaying the progress, how many files are currently farmed out, how many words total are left, and how many of the remaining words are repetitions that do not need to be translated, etc. This will greatly increase their confidence, and also give them an opportunity to give some feedback to your translation early on in the project. It is perfectly normal that the customer will hire their own translators to look over your work. These translators might make some requested changes. Get them involved right from the beginning with the glossary development, and keep the customer constantly in the loop (as you inevitably issue it more invoices). It will understandably only want to pay you for delivered and finalised files, which can take some time considering all the processes mentioned above. While your translators will not want to wait long for payment of their files. It is a tricky juggling act. You will definitely want a reasonable advance payment to get the ball rolling, considering your time spent putting together the project and the glossary and TM building well before you deliver your first file. Feel free to explain the process to the customer so they understand the initial delay.

Synopsis

As you can see, it is not an easy matter to organise such large projects. In my early years of managing large translation projects, before I developed all the online software and got a handle on the processes and the TM software, I spent so much time emailing directly with the translators, proofreaders and vettors, paying people, figuring out accounting, that I barely got any sleep. I developed my system over many years and continue to do so. I now prefer to work on this development, as I consider it creative and challenging, while leaving the project management and other tasks to others. If you’ve taken on a project larger than you can chew, I would be happy to help you with it.

The industry is constantly evolving and long gone are the days when I first started, when I had to resort to translators who would leaf through their paper dictionaries, were proficient in their field but would produce work on a typewriter or hand written, well before the days of internet. Bringing their work on floppy disks – does anyone even know what that is anymore (the original floppy disks actually were floppy and flimsy)? It’s been fascinating watching the industry develop, but that is the nature of the free market. If you want to stay in the game, especially for such large projects, you need to utilise the latest technology and know how.

The use of translation memory software, digital dictionaries, project management software and even machine translations have made the industry leaner and meaner, but it has also provided the tools to increase quality and consistency, especially for such large projects.

To manage small ones requiring only one translator is fairly easy: assuming you have a translator you trust, you send them the file, they send the translated file back, and if you want to make sure it is perfect, you forward it to a proofreader, and perhaps afterwards to a vettor (someone who checks that nothing is missing and that the formatting correctly reflects the original).

But with large translation projects requiring several translators, all this emailing back and forth can get quite complicated, files may need to get broken up into smaller, more consumable chunks, glossaries may need to get developed to help the translators unify terminology, while the proofreader will need to unify the styles. Especially large projects will require more than one proofreader, so a head translator/proofreader will need to be assigned to try and unify everything in terms of style and terminology.

I have had the good fortune of having to manage several large projects, my largest about 5 million words in 11 language combinations and more than 100 translators which needed to be completed within a single month! And I had to accomplish this with no experience with most of the translators! I’ll explain below exactly how I went about it.

Find your translators

When you receive a large translation project, chances are your existing pool of tested and trusted translators will not be enough to handle it. In my cases I had to delve into language combinations I had absolutely no experience with. But I’m good at logistics and I always find a way.

If you do not already have enough tested and trusted translators to accomplish the job, you should give yourself roughly a month to test new ones, choose a head translator, develop a solid glossary, perhaps even build up core translation memory, before you start farming out the bulk of the files to the remaining translators.

For this I posted the project on the two major translation portals:

• proz
• translatorscafe

There could be other ones and it doesn’t hurt to try them. You can create different email addresses for each account to give you an idea of how effective each one is.

Prior to this though I would put together a webpage(s) explaining about the project in detail, if I felt the need, and which would be used as a base of operations, so to speak, once the project began. For example, glossaries could be downloaded from there, instructions updated and so forth. Basically a one-stop shop where the translators can upload or download files and get the latest info on the project. New translators (I found I had to keep posting for more translators as the project progressed) could study those pages to get in the loop. Or you could have a special section only to introduce the project to new translators, where translators who are in and have passed the tests might require a special login to access the project pages. What is important though, before launching into such large projects, you really need to sit down and brainstorm for a few hours or even a day, to plan things out well.

For each of the portals you post your project on, you can use a different email address or intro webpage to keep track of which translators responded to which portal in order to help you decide which ones are the most effective. Note that I would be happy to help you with all the technical particulars. Just email me so we can discuss.

If using the email address approach, I would create an autoresponse explaining the basics of the project, perhaps with a link to a webpage, and ask the translators to apply to my agency through the automated application form, and submit a translation sample. I can also let you use my application form and the various webpages I have developed over the years to automate as many processes as possible.

I found that a 100 word sample was a fair request to ask for a free translation sample. You do not want too large of a translation sample, because this could discourage a lot of translators from applying (especially the better quality ones who tend to have enough work), and it will cost you that much more to have assessed. And besides, this is only an indicative test and assessment. It is possible that the translators will try harder or consult with others for the test, so the real testing will begin once they start translating. Pick a 100 word sample which accurately reflects the project, but something more difficult to truly test their skills.

Applicants are able to submit a translation sample to any number of projects.

Now that you have a steady stream of applicants, you will want to go through the applications to pick out the cream of the crop. Study their information, perhaps their rating on the portals by their customers, and get an idea who you think the best ones are (assuming you do not already have tried and tested translators that you trust in those language combinations). The system I have set up allows to download all this information, which I then paste into an Excel file, which I can then easily add comments to offline, colouring the individual cells to visually help me get an idea of the quality of each translator.

Numerical ratings are on the right with the name of the assessor in the column heading. To the right of each rating is a comment, if the assessor left any. Purple is where the assessor assessed their own sample (unknowingly).

Once I have chosen someone who I feel is qualified to assess the quality of the translation samples, I ask them to log into the system to assess them. Once they are done I download their ratings and paste it into the Excel file I have created. I like to use a 20 point grading system, with 1 being the lowest quality and 10 the highest, but half points possible, such as 9.5.

Now that the first assessor’s ratings have been pasted into a column showing their numerical assessment of each translator (a second column would be reserved for their comments, if they felt a need to mention something), I have my first glance at a potential overall assessment. In Excel it is easy to sort the records/rows according to individual columns, such as the numerical rating. The header of each assessment column would show the user number of the assessor. You would be surprised how they often assess themselves, not knowing it was their own translation. Easy to forget when you assess dozens of samples.

After this first assessment I would have a clearer picture of who the better translators are. In conjunction with my study of their profiles, I would start to pick out other translators for a second and further assessments. It should only take a few hours to assess a lot of samples, so this process can be quick.

To help choose quality assessors, you can refer to the assessment of their submitted translation samples, the data they submitted when applying to your agency, and their online presence. Such as my proz profile above, where you can see ratings by previous customers, or contributions to the Kudoz glossary terms. 

In the end I would usually pick at least three assessors, often up to six, or whatever is needed for me to feel confident about the quality of the samples. Lets say you end up with about six assessments for each translator. When you compare them next to one another, you might find that three of the translators’ samples have been highly graded and that each of them have assessed the translators with a consistent grading, while some of the assessors’ ratings seem sporadic and inconsistent. At this point I might decide that some assessors are not doing a good job and stop asking them to assess more samples, as they continue to flow in. By now you will start to get an idea who the good assessors and translators are, and begin to form a team of core quality translators, and eventually the lead translator.

Start building the glossary

Once you feel confident that you have found a quality translator/assessor, you can begin working on the glossary and translation memory.

When I received my first large translation project to manage, I spent some time trying to decide which translation memory tool to use. After some research and consultation, I ended up deciding on Star Transit. It seemed the most powerful from a translation project management perspective, while the individual translators were able to use the free version. Here are some of the reasons why I chose this particular software:

• the translators could use the free version, easily downloadable
• it uses RegEx (Regular Expressions) to protect certain segments from being tampered with by the translators. For example, one large project I worked on was the localisation of a game. It was created in some software and then exported into a text file to be translated, but it was littered with special characters, such as <+3> that could not be changed, otherwise the software would not run properly. By being able to tag such character strings to protect them, they could not be changed by the translators and was understandably a strict condition by the software developers in order for me to take on the project. Otherwise, for projects in the hundreds of thousands of words, you can imagine that cleaning up any mistakes would add significant time before the localised versions could be rolled out. Be warned though that writing the correct RegEx formula can be quite a challenge, and I can help you with that as well.
• during the proofreading phase, as the project advanced forward, an increasing amount of the translated text was already translated and proofread earlier, as per the translation memory. It made no sense for the proofreaders to proofread it again, but the software allows that to be filtered out, so they end up proofreading only newly translated segments.
• I find the software quick and easy to work with as a translator. I can scroll down through both source and target windows at the same time with the down arrow cursor, I can make global Search and Replace changes much like in Windows, and there is absolutely no lag or sluggishness to the software, like I found in other TM software I tried. More details about the free translator’s version here (these same advantages apply to the paid/full project version I use while I translate).

What the working environment looks like, with glossary terms marked red and a fuzzy window bottom right. There are several views you can choose from, or custom make your own.

You can use software to find out the most common terms and sort them in that order. You should also ask the customer if they can create a list of terms which they feel are important to always translate the same. In this way you can start to compile a glossary, but thankfully by now you should have a lead translator you trust to start translating that.

After the glossary is translated, you definitely want to ask at least one other quality translator (and also the customer) to have a look at it. This is the very core of your project and you want to make sure you have this right before proceeding. In the fantasy game software we localised, there were many terms, such as sword, special skills and so forth which would have to be translated exactly the same as they are used throughout the game. For this I found google docs rather useful. You can set up a spreadsheet with the source term in one column, proposed translation in other columns, a comment column, and assign viewing and editing rights to different people, so they can discuss amongst each other the individual terms through the popout comments feature and chat windows.

Online google spreadsheet, where you can set different access rights for different users (View Only, Editor, Administrator). Editors can chat with one another through a little chat window, and leave popout comments. This can be prepared in Excel offline and then uploaded.

Once the glossary has been translated (note that it is a living being and evolves as the project proceeds), import it into your TM (translation memory) software. In Transit, all glossary terms are marked as red in the individual project files. Before starting, the translators have the option to “insert all glossary terms” into their assigned file, or you can do this yourself as the project manager. This effectively automatically translates all those words in the target text, so that the translator does not have to type them out (less mistakes that way, and makes it possible to arrange for a price discount from the translators). Those terms still remain marked red in the source text window (usually the top half of the computer monitor – example two pictures above).

In Transit the lead translator can make changes to the glossary, or open up a particular term, like above, to add special comments.

When translating, each translator has the option to suggest additions or amendments to the glossary terms. These amendments and suggestions are incorporated in their exported file, which they send back to you, and can be studied by the lead translator(s), who then decides if they should be incorporated into the glossary.

Start to build your translation memory base

Once you have begun to build your glossary, use the TM software to create a file of repetitive text. In our games example, one repeating segment might be “Hey, I shot you, why are you not dead?” You can get your best quality translators to start working on this. Obviously you will want this thoroughly proofread as well, because it will become your second layer core. For a project of five million words, the repetitive text could easily end up being in the hundreds of thousands of words, which will be a project in itself. Make sure it is as perfect as possible, because the other translators will see it injected within their project files, so it will serve as examples for them (not to mention that it should be perfect anyway).

In the project settings, you can choose to Create Translation Extract > Only with frequently recurring segments. You select this and create the extract file when you first import the text into Transit. This becomes your repeating text. You translate that first, proofread to perfect, which then gets automatically translated moving forward, wherever it repeats in the remainder of the text.

As you continue to test translators and assessors, you may bring more on board to help with the repetitive text. Use your lead translators and proofreaders to make it as perfect as possible. In your initial analysis (when you were giving your initial price quote to the customer) you should have determined how extensive this repetitive text is (I can help you with this). This is where you can make most of your profits. Not only do you not need to pay translators to translate those phrases over and over again, but you are also ensuring consistency and higher quality. On the other hand, TM software is becoming so pervasive that you probably had to include a certain percentage of these savings in the original bid price, because the other agencies bidding on the same mega project are using the same approach. Nevertheless, if you use the TM effectively, it should result in higher profits for you.

When you first import a file into Transit, you can save a report showing the fuzzies and what has already been translated (Pretranslated). This can be due to previous translations, or after translating and importing the repeating text extract. Once you know this information you can make your competitive price quote. The report can also be copied into Excel to calculate the total payment to the translators based on the fuzzies and pretranslated text.

Continue to
Part 2 of How to Manage a Large Translation Project