|
How the memory measured up
Extended comparative review of Translation Memory products part 2
[panel]
Now is the time that many of us finally invest in ourselves, caught up in the general fever of Christmas spending. Here's the information you'll need if this means you finally succumb to the temptation of Translation Memory software.
Horses for courses
Thankfully part one of the Translation Memory review went down very well with ITI members, non-members and even the manufacturers themselves. That notwithstanding, many of you have taken the time to ask me which I think is the "best" product, or (not generally known for my reticence) accuse me of sitting on the fence.
I didn't come out with the ITI Bulletin equivalent of a "Which? best buy" for a number of reasons. Firstly, because I wanted to gather feedback from readers to provide added depth and breadth to any recommendations I might make. Secondly, because first impressions can be misleading, and finally, because it’s important to understand that as much as I have tried to be as objective is possible, in the end we're talking about one translator's personal preference. The detail of both parts of this review means that anything I say now in summing up will be within a context that allows you to evaluate whether or not you're likely to agree. Much of this detail is in part one, however, and it may help to re-read that before proceeding, since I didn’t want to waste space or your time by repeating everything that’s already been explained there.
When it comes down to your final choice, there is no single best of the bunch, but it is likely that one product will suit your needs more closely than any of the others. That may not necessarily be the one I prefer. After all, the only way you'll feel the same about any of these products as I do is to become me, and that's not frankly something I'd recommend. Nevertheless, the conclusions to this review will be brutally frank because that's the only way they can be really useful. So, strap yourself in tight and enjoy the ride.
Who bought a lemon?!
One of the most surprising facts to emerge from member feedback was that several of you have bought Translation Memory products and never used them. Frightened into feeling you would be relegated to translation prehistory, regarded with contempt by your much more technically-sophisticated colleagues and treated with disdain by potential work providers, you rushed out and spent over £2,000 on a box, dongle and floppy disks that some of you have never even got round to installing. You never needed Translation Memory because you were quite busy enough as it was thank you very much, because your work arrives by post or fax and because none of it's particularly repetitive, if at all. It doesn't matter how good a tool is, it's wasted money if you don't need it.
Well, I don't know if it's much consolation, but at least you now know you're not alone. Others in this group include those who do fairly repetitive work but for one reason or another believed it would be worthwhile scanning printed/faxed texts to convert them to files. They without exception found they spent more time correcting the output from their Optical Character Recognition programs than they saved with Translation Memory. As I reported in part one, it just doesn't work that way. Suffice it to say that the only real use of OCR to translators is in scanning source texts to obtain word counts. They’re generally great at recognising a word is a word, but start straining as soon as soon as you ask them to tell you what that word actually is.
If you can identify yourself with either of these categories, you might as well stop reading this now. Go away and empty your piggy bank on speech recognition, mince pies or champagne instead. Any of these three are more likely to increase your productivity.
A third group is formed by those who do use Translation Memory but have been disappointed by the productivity increases achieved. In many cases this is simply because, like the former, they do not receive enough work in electronic format for Translation Memory to you make a significant impact on their earnings. In certain cases, however, it is also because they have been misinformed about the scale of productivity increases that are possible.
One of the problems is Translation Memory companies (no names!) who behave more as if they were used car salesmen in the time-honoured tradition, quoting one-off instances of productivity increases (in some cases over 60%) as if they were the rule rather than the exception. Don't bank on getting more than a 30% increase overall - and that's not on your total income but just for those texts that are suitable for TM.
So how do you avoid buying a lemon? Fortunately, I believe there is an acid - not to say citrus - test. If you already suspect Translation Memory could help increase your productivity it probably will. If, on the other hand, you read through part one of this review without some sort of a penny dropping that it might have anything to offer you... then keep a tight hold on that penny. There are other ways of increasing your productivity, foremost among them dictation and speech recognition - also considered in this issue. Indeed, it’s my opinion that most ITI members will gain more from speech recognition than from Translation Memory, which is why I shall also be highlighting how well the various different Translation Memory packages work with the best speech recognition solutions on the market. When they work well together the productivity increases can be very good indeed.
Who pays the Piper?
Both parts of this review refer to the fact that certain clients and translation companies in particular may require you to use the same translation memory system they do (or a system that is compatible with it) to work with them. A number of ITI members consequently considered the purchase of a translation memory system to be a useful marketing tool for attracting new clients. Indeed, the marketing tactics of certain TM companies encourages this.
User feedback indicates that this is not, however, the case. If a company promising regular work insists you use a particular package for a specific long job, maybe even encouraging you with a discounted purchase price, then you can weigh up the pros and cons, and make a choice that is based on known elements. Very few translators who followed this route ended up being disappointed. Conversely, most of those who speculated and invested in a package believing it would help attract new business have not achieved a return on their investment, including those who did succeed in attracting new clients requiring the product concerned.
When a 100% match is not a 100% match
One thing that has to be borne in mind is that many clients apply a differential pricing structure based on the percentage fuzziness of the matches they obtain when pretranslating/analysing a job. Some, for instance, will be reluctant to pay more than 25% of your per-word rate for a 100% match. Such an arrangement might mean that you actually suffer a fall in income working on certain projects, since it is wise to check your 100% matches at least once to make sure there is no database corruption, that there are no errors or poor translations in the database and, finally, that the 100% match is actually appropriate. "How can it not be?" you ask. I'm sure, however, that it won't take much head scratching for you to imagine potential situations for your language and subject combinations in which a given sentence may need to be translated in different ways to suit a different context.
Likewise, some clients only want to pay 50% of your normal rate for a 50% fuzzy match. Don't accept it. The time you spend editing a 50% fuzzy match is unlikely to be less than that required to translate the same segment anew. If you're dictating text and have a well-fed terminology database (for those odd moments when your memory needs quick refreshment to remember some exquisitely rare term), you even may find there's very little difference in your productivity for 0% matches and 75% matches or higher.
There are two lessons to be learned: be careful about how you negotiate any such deals and don't count your chickens until they're hatched when it comes to the return on investment you can expect from increased productivity gains. There's no doubt these products can increase your income, but also remember that if there isn't going to be an immediate gain based on your current situation, buying TM may have an effect on your credit card statement alone.
The low-down
The main purpose of splitting this extended review was to receive feedback from users. This feedback consisted in over 30 e-mails, a number of phone calls, one letter (from a resolute typewriting veteran who managed to encounter every known weakness in Windows 98 in less than 98 hours) and many informal conversations at ITI events such as the Conference, Seminar on translating for the European institutions, Open Forum and Weekend Workshop.
The nature of this feedback was very varied. Most users seem to acquire the minimum working knowledge they can "get away with" to process texts using Translation Memory. Only a few dig below the surface to maximise the productivity gains achieved. As may be expected, these tend either to be translators with a very high throughput of repetitive technical texts or translation company managers and their staff. I'll be reporting on their and my own findings system by system below.
Atril Déjà Vu
A virtual community
Déjà Vu received by far the most enthusiastic support of all the various different Translation Memory programs. There were a number of reasons for this. In a market dominated, at least until very recently, by packages costing around £2,000 without including alignment software or filters for the big DTP programs (allowing you, for instance, to work on Interleaf or FrameMaker files), Déjà Vu distinguished itself with a philosophy that combined a "complete product" approach with what was generally perceived as a much "fairer" price.
The single most vocal complaint against Translation Memory companies in general is that little user support is provided and that it is generally considered to be inadequate and overpriced. Atril scores heavily here, providing very prompt e-mail support and often working on a user's files to resolve a particularly thorny problem - sometimes quite obscenely late at night and over the weekend. One user in particular, now in-house and "forced" to use a competitor's system, contrasted his very positive “Déjà Vu experience” with the inferior performance and backup provided by one of the program’s much bigger commercial rivals who actually charged when he reported a bug to them. All Atril's user support, on the other hand, is provided free of charge.
There are two main Internet user groups for Déjà Vu, both administered by Atril, on which it maintains an active presence, one for general users and one for beta testers. Both of these are far more active than their equivalents for other products mainly because they don't just discuss how to use Déjà Vu, but also make suggestions on how it can be enhanced and expanded. This becomes a dialogue for the simple reason that practically every suggestion for which there is some sort of consensus gets incorporated in a new release within a matter of days (when it can be downloaded free from Atril's Website), and its implementation then discussed and possibly improved. Many user's loyalty to Déjà Vu (and this includes translation companies) comes precisely from the fact that it has been customised to their needs.
This sometimes has its drawbacks, particularly for those using the beta versions, who may find that some cherished feature, which worked perfectly well in the past, suddenly stops working until the problem is sorted out. But there is a simple solution to this - only experiment with a new release when you don't have an urgent project on the boil (unless you also want to experiment with your ability to cope with rapid changes in blood pressure). One way or another, this constant feedback and group problem solving means Déjà Vu users form a virtual community that is without parallel for any other translation memory product.
An effect of this is that Déjà Vu has developed much more rapidly than any of the other systems in this review - and that's notwithstanding the fact that Atril are currently working on a completely new version of the product due for release late this year or in early 2000. The latest version (2.3.86 as I write) now offers filters for PowerPoint, runs much faster, features a number of automated troubleshooting routines, a repetition analysis function, TMX filter, a very handy filter option that works like "Fold" in transit (allowing you to revise or analyse all the occurrences of a particular "problem" en bloc), powerful SQL features for database management and the first implementation of Shallow Translation Memory.
How shallow can you get?
Let's just consider that last one for a moment. What exactly is Shallow Translation Memory? Translation Memory systems break your source and target texts up into corresponding segments, such that when they meet the same or a similar (fuzzy match) source segment in a new text they will match it with the old target segment. What happens, however, if the new segment is exactly the same as part of an old segment? Depending on your fuzzy matching settings it will be retrieved for you to edit. But if the new segment contains two sub-segments, both of which already exist as separate sub-segments of two different segments in your database, even the best fuzzy matching will retrieve either one or the other and you will have to edit the result to bring them together. That takes time. Not much, but it mounts up in the course of a job. Shallow Translation Memory technology aims to bring these two sub-segments together to form a new segment without your having to do any work.
Although all the main Translation Memory companies are working on this technology, Déjà Vu is currently the only product to offer it, albeit in simple form, on the market. First steps in this direction can also be seen in the way Transit is able to automatically integrate Termstar entries in fuzzy matches. Likewise, Trados has a function called “Substitution localisation” that automatically replaces dates, times, measurements and user-defined variables such as product names. Both these functions can in some cases transform fuzzy matches into 100% matches.
From the user's point of view, however, gaining the full advantage of Déjà Vu’s shallow translation memory technology requires a little patience. That's because the current versions require you to manipulate the three different types of databases it uses in a project (project lexicon, terminology database and memory database) to achieve this result. At the moment, Déjà Vu basically implements shallow Translation Memory technology by assembling the translation units out of sub-segments you yourself create using a number of semi-automated routines.
You are best off experimenting with this only after you have gained experience in using the less sophisticated features of the product. The old adage of learning to walk before you run was never truer than in Translation Memory and very definitely applies to Déjà Vu. Take things methodically one step at a time, making full use of the excellent advice on offer in the Internet user groups and you'll make steady progress while maintaining a constant smile. Ignore this advice at your peril! I have received many late-night phone calls from hysterical users who tried to do too much too soon, were rudely surprised by the results of their experiments and found themselves in a real panic about how to meet deadlines. The biggest surprise for me has always been meeting them subsequently, their sanity restored and - miraculously - their hair still showing youthful vigour and colour. Nonetheless, I'm convinced the risk of sudden hair loss is real. Avoid it if you can.
Talking of sudden shocks, Déjà Vu is one of the most stable systems, along with Transit. Unlike the latter and indeed the rest of the pack, however, it saves your work as you move from cell to cell, so that should you experience a crash you will at the very worst only lose the last segment you were working on.
Coming of age
Those users who have mastered this three-way database assembling technology have reported that it makes a significant difference to their productivity gains and I can confirm this. No other system on the market currently allows you to get so much out of your databases. The new SQL features allow you to leverage this further and are proving particularly useful for project managers.
As if that weren't enough, the forthcoming Version 3 will also include filters for Corel Catalyst and Excel (which will probably also be available for version 2 before the end of November), a new more flexible user interface with re-sizeable cells that can be edited in situ, user-configurable menus, tool bars and keystrokes, together with Wizard-driven assistance for beginners and other enhancements to improve user-friendliness. There will be a new editor for Windows dialogue boxes that automatically re-sizes all objects in the dialogue box to accommodate translated text lengths. The program will also support all Windows languages, including right-to-left and Asian languages as both source and target. It will be possible to use multiple, hierarchical databases for each project and import different file formats in the same project. Format support filters will be plug-ins so that third-party companies can develop their own. Batch alignment of existing translations will be possible, semi-automatic glossary building will be provided and the system will include a back-end for Access databases and, optionally, for SQL Server and Oracle databases. True shallow translation memory technology will integrate example-based machine translation principles to yield much more accurate results when assembling translations, meaning that Version 3 of Déjà Vu will mine your past translations much more deeply. Atril states that productivity gains are expected to reach up to 5 times the current version's figures, which are in themselves amongst the highest for any current translation memory software. The final bit of good news is that anyone who has bought Déjà Vu after October 1st 1999 will be able to upgrade to version 3 free of charge.
Last but not least, Déjà Vu is also one of the very best Translation Memory programs when it comes to integration with speech recognition technology. Dictation inside it is no problem and nearly all the standard editing commands work faultlessly. It is also relatively easy to create macros that activate its very many keystroke shortcuts for increasing productivity. Our Bulletin Editor, Antonio Aparicio, was reporting productivity in the region of 1,000 words per hour on a technical manual revision within a matter of days of integrating the two.
In short, whereas Déjà Vu used to be a product that held its own with the best while offering the advantages of an attractive price and "complete solution" philosophy, it has now very much come of age. While continuing to be a complete solution - even allowing you to work on projects for translation companies that use competitor products - it now offers an array of features, filters and productivity tools that none of its competitors can match. Its product support and user satisfaction also seem to be much better than its competitors. What's more, if the forthcoming Version 3.00 is half as good as promised it will zap everyone else straight between the eyes. Definitely one to watch and probably the one to buy, but read on before you make up your mind.
|
