Letting the CAT out of the bag

 

Is translation memory more trouble than it's worth, or a tool no translator can afford to be without? Michael Benis tries to answer some of your questions.

 

Setting the stage

 

The most common question now asked about translation memory (or Computer Aided Translation) is not whether it works, but whether it's worth the investment, especially in view of the differential rates many translation companies now impose. The question generally goes "I'm thinking of buying a TM package to get more work, but will it ever pay off?"

 

The problem with this question is that it's based on a series of almost entirely erroneous assumptions. So, having set the stage - or table - let's take things one step at a time and make a meal of it.

 

Cutting through the jargon

 

For all its much-touted sophistication, translation memory (TM) is in fact an almost absurdly simple idea. It consists in nothing more complicated than filing away every single sentence you've ever translated, with the source and target texts aligned in a database so that you can retrieve the latter should you ever encounter the former again, or anything remotely like it, during your translation travels of the future. If you find a sentence in your translation memory that's exactly the same, it's called a 100% match, whereas sentences that are very similar but have one or more words that are different are called fuzzy matches and their "fuzziness" is indicated as a percentage of less than 100.

 

TRADOS got the ball rolling with their "Analyse" feature which gives a breakdown of the various matches found for any given translation in your translation memory and can sometimes be a useful indicator of the time you are likely to spend on a project. A number of translation companies use Analyse or similar features in other programs to implement a differential rates policy whereby the translator only receives their full rate for matches of, let's say, less than 50%, whereas 100% matches are only paid at 50% of the translator's full rate and sometimes less.

 

Leaving aside the history, merits (or otherwise) and consequences to the profession of this practice, many translators have quite naturally questioned whether the productivity gains of translation memory and the additional work it might enable them to secure could ever outweigh the cost of buying the program/s and the time and money spent on learning how to use them, especially if they find themselves effectively having to reduce their rates.

 

Forking out for fuzzy thinking

 

As I stated from the outset, however, this whole scenario is based on a number of false assumptions, foremost among them the idea that the productivity gains offered by translation memory are chiefly a result of the 100% or high fuzzy matches generated. This may well be true when the text you're working on is highly repetitious or largely a revision, but in most cases the real gains are to be had from not needing to use other tools (such as dictionaries, glossaries and the Internet) to research or verify terminology while you otherwise work pretty much as you would have done before the advent of our brave new world. Depending on the system you use, significant gains can also be achieved during the quality control stage, with Déjà Vu and Transit offering particular strengths in this area. So, there are advantages to translation memory irrespective of the whole business of matches. The only exception would be a highly-experienced specialist who no longer requires any external assistance where terminology is concerned and never makes the occasional mistake.

 

On the other hand, practice clearly indicates that differences in fuzzy match percentages have practically no impact on the time taken to edit or rewrite what the system suggests in order to achieve a satisfactory solution. Indeed, increasing numbers of TM-experienced freelance translators are now refusing to offer discounts for anything other than 100% matches, arguing that in all other cases it takes no more if not less time to simply type or dictate a new translation. The same translators likewise contend that this can also apply to 100% matches, particularly if they occur in isolation and, therefore, in a completely different context, meaning that they will need to be changed - and this, of course, isn't just an argument about money, but about quality, too. It's also an argument whose merits have been recognised by the TM companies themselves, such that the latest versions of TRADOS, for example, now include a feature known as XTranslate, which compares new translations with their previous versions rather than the translation memory database and then only takes into consideration those 100% matches that occur in strings of several at a time. Atril's Déjà Vu X follows a similar approach with its Guaranteed Match function.

 

Underlying the whole debate is a further assumption: that the productivity gains offered by translation memory can be quantified reliably from project to project or reflected in reduced rates. Strong marketing inspired this pipedream, but real-world usage over more than a decade has shown that it is quite simply untrue apart from in certain special cases, such as software localisation (where tight time-to-market pressures demand that the often highly-repetitive interface, online help and documentation are written and translated in tandem with the development of the product itself, proceeding through a whole series of different versions) and industries with modular product ranges or that share common components (such as the automotive industry, where a number of different models can have the same engines, for example, with just minor differences).

 

The above doesn't mean that translation memory doesn't have its uses. Indeed it is almost essential to ensure terminological consistency on large projects to tight deadlines that require a whole team of translators to work simultaneously for the job to get done on time. But it does explain why a large number of the translation companies that believed they could afford to fund their growth by price cutting on the back of assumed TM efficiencies have found themselves with much tighter margins than anticipated. This isn't simply because the productivity gains aren't as high as expected, but because ongoing investments are required in product and database maintenance (never mind staff training), not least of all to ensure quality doesn't suffer from inappropriate translations being recycled endlessly through the system.

 

The cherry on the cake

 

All this is a long way off from the heady days not so many years ago when numerous companies were convinced that day-to-day productivity gains of fifty and more percent were within their reach. At the moment, the problem is that just as translation companies are outgrowing this view, more and more of their customers are asking for discounts on 100% and high fuzzy matches. At the same time, others are becoming disaffected with the quality offered by the translation industry, partly because of problems with "recycled TM garbage" that are not solely due to poor translators (in every sense of the word) and poor quality control, but also the fact that the systems and how they are used make it more difficult for translators to depart from source language syntax, sentence and paragraph structure to arrive at translations offering a high standard of clarity and impact. As a result, many are demanding that translation companies deliver both higher quality and lower prices at a time when translators are becoming increasingly resistant to the differential rate trend, in many cases voting with their feet.

 

Clearly, a major rethink is on the cards. What's more, this isn't just my opinion. I heard the same views confirmed by a director of one of the largest translation companies in the world, amongst others, only days before I wrote this article.

 

Where does all this leave the freelance?

 

As you can see, none of this offers any easy answers to a freelance translator asking themselves whether it makes financial sense to invest in translation memory, except that having neatly thrown out a number of unfounded paradigms with the bathwater we can recognise our baby again. To do so, we need to shed one final assumption: that any tool can be a magic pill with the power to transform our careers.

 

Translation memory is merely one of a number of aids that can help the translator. It is by no means essential and certainly no substitute for rigorous research and a scrupulous insistence on "writing" one's translations to the highest standards of the target language - both qualities that are at a premium because they still tend to be sadly underrepresented in the profession. That said, research takes up a lot of our time and translation memory is the most effective way of ensuring that once we have hunted down a particular term we will never have to do so again.

 

Making your own choice

 

Investing in TM in the hope it will bring you extra business is a risky step that could nevertheless pay off. I generally advise people against it unless they have plenty of money to spare, not least of all because most people I know of who have done this didn't gain any new customers as a direct result. Likewise, I would be cautious of buying a particular TM package for a given job on the promise that plenty of other work will follow, although the chances of success here are possibly higher. The situation is obviously different if a translation company with which you have been working successfully for many years is about to switch over to using a TM system and invites you to do the same. But even then, there's no need to feel forced into actually shelling out any money of your own, especially now that many of the major products have free versions which allow translators to work on projects created by others. These provide the basic interface (and no more) that you need to get the job done. However, if you can't resist the risk of buying TM to make yourself more attractive to translation companies, at least make sure you buy one of the more popular systems. TRADOS is the most widely used, followed by Déjà Vu and Transit.

 

It naturally helps if you have some basic familiarity with how TM works, making it easier for you to learn any given "new system". Some of the "satellite" free versions actually come with sample project files that will allow you to learn the ropes in preparation for future opportunities, but you will gain more from working with a "full" version that allows you to create and experiment with your own translation memories and terminology databases. The cheapest way of doing this is to buy Wordfast, which although not the most elegant of solutions is nevertheless admirably complete and significantly less expensive than its direct competitors. Metatexis costs even less, but handles fewer formats.

 

This is, of course, assuming that you're only interested in using translation memory to make yourself more marketable, although doing so to increase the efficiency of your own translation processes actually makes much more sense, if only because you can at least be sure you'll achieve your goal. Using translation memory for all your work is a relatively pain-free way of building up an enormous (and ultimately enormously helpful) glossary and concordance over the years, while in many cases making it faster for you to check your own translations and ensure the correct terminology is used consistently throughout. What's more, it's much easier to reconcile your own needs with those of your clients now that an increasing number of systems are able to use more than one database at a time.

 

You'll need to bear a number of factors in mind to identify the system that best suits your needs, including which languages you translate from and into, the speed with which your specialist terminology changes and your need to track these changes, which types of files you work with most frequently, whether or not you use speech recognition, which system (if any) your customers and close colleagues use, as well as, finally, your own personal preference regarding the interfaces on offer.

 

Where the latter is concerned, the field is divided between TRADOS on the one hand and almost everyone else on the other. The TRADOS philosophy attempts to always show the text you are working on as it will finally be seen by the end user, so that you'll generally be working in Word, PowerPoint or Tag Editor (which allows you to view the HTML files you are working on in Microsoft Internet Explorer). A similar approach is followed by Metatexis and Wordfast, while everyone else provides a custom interface that at first sight looks less user-friendly because it provides less contextual information, although you can of course still obtain this from a printout or by displaying your source document on a second screen. But these systems offer benefits of their own, namely the simplicity of working in a single interface for all the work you do and that certain functions - such as terminology entry - are faster to use and therefore more likely to be exploited, while advanced quality control features can also be provided. Many also allow you to work on and compare all the files in a project at the same time, even if they're in different formats.

 

Studying the product specifications on the manufacturers' web sites with these points in mind and asking questions on the related Internet user groups will help you eventually make the right choice, assuming that you're not so computer shy that you'll never actually take the plunge and use the product you have bought. This has happened to more than one well-intentioned translator, so don't dismiss the suggestion out of hand. Take a long hard look at whether you currently use your computer and its related software to the full. If you've never discovered how to optimise Windows and don't, for example, know how to defragment your hard disk, have never worked out how to alter paragraph settings or tabulation in Word and are plagued by disappearing toolbars, then the likelihood is that you'll never get on with translation memory. The best policy in these cases is that if you're doubtful you can cope, sign up for a course where the software is supplied and try it for a day before parting with more money. The course will definitely cost less than the software if you decide not to go ahead and give you the best of all possible starts if you do.

 

A discount roadmap

 

As for the whole business of discounts, you can always just say no or negotiate them as appropriate, bearing in mind above all that whether or not a match is going to save you any time depends on the quality of the database your client has supplied. What's more, if it's your own database and the fruit of years of research and expertise, you may well decide that in most cases you'd rather not offer any discount at all. Remember that the demand for translators who can deliver quality work to deadline still outstrips supply, so make sure you don't put all your eggs in one basket, market yourself intensively to create a broad customer base and then narrow it down to those who are willing and able to reward you best for a job well done, whichever systems you use.

 

Good luck, and let me know how you get on!

 

Michael is always happy to receive feedback on his articles. You can e-mail him at: michael@michaelbenis.com

 

[Panel]

 

Which one's right for you?

 

The first step in choosing an appropriate TM product is to browse the manufacturers' web sites. Having narrowed your choice down, follow the online user groups. Most participants will be happy to answer questions providing their own insights to help you decide.

 

www.atril.com

The home of Déjà Vu, perhaps the most powerful program on the market with an enthusiastic following. A new version, DVX, has just been released.

User group: http://groups.yahoo.com/group/Déjà Vu-l/

 

www.cypresoft.com

This is where you'll find TransSuite 2000, an unpretentious but reliable tool that is very easy to use. A good choice for the technophobe.

User group: http://groups.yahoo.com/group/ts2000_users/

 

www.metatexis.com

A user-friendly system that integrates with Word. Very reasonably priced.

User group: http://groups.yahoo.com/group/MetaTexis/

 

www.star-transit.com

Transit is a strong contender that is widely used in the automotive and localisation industries. Excellent terminology management.

User group: http://groups.yahoo.com/group/transit_termstar/

 

http://www.sdlintl.com/sdlx

This is the site for SDLX, the translation memory suite developed by SDL, a leading international translation company based in the UK.

User group: http://groups.yahoo.com/group/sdlx/

 

www.translationzone.com - www.TRADOS.com

The industry leader with a distinctive but fragmented approach that offers a complete solution to almost all needs. Version 6 of the product has just been released. The former site is especially for freelance translators.

User group: http://groups.yahoo.com/group/TW_users

 

www.wordfast.com

A relatively new arrival that is admirably complete and excellent value for money.

User group: http://groups.yahoo.com/group/wordfast/

 

[End of panel]

First published in The Linguist, 2003.