A simple idea.... The idea behind translation memory is fairly simple: to speed up repetitive translation work by hoarding all your previously translated source and target language sentence pairs in a massive database. This database can be built in one of two ways: “aligning” the sentences in existing source and target documents or splitting a source document into its component sentences, translating these sentences one by one and then feeding the sentence pairs into the database. Once a database has been formed, you can search for matches with the database sentences to assemble the translation of a similar new text. This is clearly ideal for manuals containing simple instructions or updates of such manuals, and it should come as no surprise to learn that the software for performing these tasks was initially developed within the software industry itself. The aim was twofold, to save time and money by avoiding duplicated translation and to increase consistency by feeding the translator approved terminology.

... made simple Simple as the idea may have been, the first programs weren’t particularly user-friendly and alignment was often a nightmare. Déjà Vu arrived on the scene in 1993, created with the aim of making alignment easier than previous packages and facilitating the translation process itself, since however much a database contains, the translator still has to check and edit the output as well as translate everything that’s completely new. Déjà Vu made the translation side much easier by, amongst other things, offering the option of integrating its TM functions in an interface with the market’s leading word processor, Microsoft’s Word 6. It soon won a strong following among freelance translators and agencies alike thanks to its combination of ease of use and outstanding flexibility, delivered in a complete package at a very competitive price. (Then, as now, most translation memory packages only offered filters and alignment programs as costly ad-ons). Déjà Vu also had its problems, however, mainly related to slow search speeds once the databases had grown sufficiently large to be really effective. 16 bit coding and slow 486 processors (if you were lucky) simply took ages to wade through the vast amounts of data involved. This was naturally a failing common to all TM programs, however, and is indeed still encountered on many today.

The latest version These problems were addressed in Déjà Vu 2, released in 1997. As far as I’m aware this is still the only true 32 bit translation memory program currently available on the market. This, together with Pentiums, has gone a long way to solving the problem of speed. Déjà Vu now runs very quickly and reliably (the database system used is based on Microsoft’s Access, thereby also offering the advantage of greater potential compatibility than the proprietary systems used by other TM systems) and doesn’t slow down even for projects containing many, many thousands of words. You’ll find very few translators with any experience of Déjà Vu’s competitors who are prepared to say the same thing. One of Déjà Vu’s strong points is that it’s continually developing in partnership with its users, with free upgrades available for downloading from Atril’s Website. The features provided are constantly being improved and expanded in response to feedback from freelance translators, translation companies and other users. As a result, Déjà Vu incorporates a number of functions not offered on any of the other leading packages. Perhaps the simplest yet most important of these is the option of actually choosing how the sentences are split up after they’ve been processed by the program. A sentence is a sentence you may say. It starts with a capital letter and ends with a full-stop. Well that’s what most translation memory packages think, anyway. But ask an experienced translator and they’ll wax lyrical about the myriad changes required by syntax and sentence order to obtain a clear “naturally” flowing translation. One way or another, you need to chop source sentences up and join them together in a way that’s determined by context and individual judgement. That not only applies to the first version of a manual, but also to the second and the third. Manuals you want TM to eat its way through without a second thought. Especially if that second thought is yours, when you’re forced to go editing the output afterwards. And that’s only half the story, because cunning choices about how long sentences in particular are broken up into units can ensure you have a far more useful database for solving similar problems in the future. Then there’s “fuzzy” matching. That’s what happens when the system finds a sentence in its memory that’s similar to one you want to translate but not identical. All the current programs offer a facility of this kind, incorporating routines that enable the system to cope with simple differences intelligently, but not all implement it as effectively. Such differences may involve simply substituting digits and perhaps converting the decimal marker from a comma to a point. Or they could require changing a word or two. In many cases Déjà Vu will do all this automatically. Translators who successfully use translation memory sometimes confess that their productivity increases are often due to the terminology management features offered on TM programs more than anything else. These functions allow you to search for particular terms or phrases and some, including Déjà Vu, will allow you to use your databases in a stand-alone terminology management package that can be run alongside your standard word processor. Déjà Vu also offers enhancements of these basic functions that make a big difference, more about which in the paragraph after next. At the same time, however, Déjà Vu stands this database-in-a-word processor thinking on its head, incorporating advanced word processor productivity enhancements within its operation, including cut and paste, drag and drop editing, search and replace, a customisable spell check, capitals toggle and AutoText. These are joined by a function called Propagate, which allows you to repeat a chosen translation and permutations of it throughout a project. You can even choose to have Déjà Vu do this for you automatically as you work. A similar but more powerful function is known as Autoprocessing. This comes into play when no fuzzy matches for a complete sentence can be found and basically builds a translation for you from sentence elements in the memory database (updated with information from sentence pairs previously translated within the same project), together with the words in your terminology database, gradually narrowing its focus down to the level of individual words. When set up properly, Autoprocessing can produce usable suggestions that require surprisingly little editing. But even when it doesn’t, it nevertheless very usefully jogs your memory regarding the terms and/or phrases required. All these features could not be offered from an interface within Word and this aspect of the program was dropped. So, gain some, lose some, as they say. The filters work extremely well, however, even with complex formatting and I have only ever had a significant problem once, with a file that indeed had problems of its own. What’s more, within days of having informed Atril of the problem they had incorporated a solution in the latest upgrade on their Website. An added benefit of the change is that you don’t have Word’s memory-hogging overhead to slow down your system. To be honest, however, I never really missed Word much, since so many word processing functions have now been built into Déjà Vu and the new functions work so well together, speeding things up considerably.

So what does it look like? When you’re working in Déjà Vu Interactive (the main component in the system) the screen is divided into four areas: a pull-down menu and button bar strip across the top and a vertically split screen below, divided horizontally into two areas. The first of these is divided into sentence cells showing the first few words of each sentence—one of which is highlighted—while the bottom part of the screen contains the full sentence selected. You simply select your source file and let Déjà Vu feed it into a project, splitting it up following the rules you yourself have selected. (Don’t worry you can leave the settings alone until you have a clearer idea of what you want; the defaults work perfectly well). The next step is a “pretranslation,” which is when Déjà Vu rushes its way through the whole lot, digging all the answers out of your chosen database. You then hop down through the text cell by cell, merging or splitting these sentences further as required. They will either be 100% matches, partial matches or no-match sentences, colour coded so you can recognise them at a glance (and yes, you can customise these colours to suit your taste). In theory you can ignore the 100% match sentences, though I confess to nearly always checking them, the victim of a suspicious and neurotically perfectionist disposition. Partial matches may only need checking if the fuzzy matching magic has been working well, but otherwise they’ll need editing. As for the no-match sentences, you can “scan” the database to check for individual terms or phrases or let Autoprocessing have bash at it for you first. That could once again leave you just editing the result. If not—gulp!—you’ll simply have to translate it yourself. The whole process is fast and intuitive and the cells even scroll automatically for you when you reach the last sentence displayed.

What will it do for you? So, those are the features. The question is, do you need them? Well the answer in some ways has very little to do with Déjà Vu. Because the real question is... what kind of work you do? First of all, you’ll need to receive the work as a file. Forget the idea of scanning texts in, since you’ll more than likely lose more time correcting OCR than you gain with Déjà Vu later. The next consideration is the subject matter itself. Instructions—ranging from lengthy assembly and disassembly operations to surgical procedures, as well as repetitive product descriptions—are ideal fodder for translation memory. While parts lists and the like are a dream come true. But move off into anything more discursive and you’ll find the time spent preparing databases and/or working on a sentence-cell-by-sentence-cell basis will never be repaid. The borderline for me was a very long project on bird watching. Déjà Vu helped me remember the different names of all the species and their habitats, while also coping well with repetitive information like clutch size, plumage and wing span. But, on balance, the real savings were on the terminology management side. The sentences were too varied for Déjà Vu to offer many suggestions that it wasn’t quicker to simply rewrite rather than edit. That said, these time savings where real and increased productivity in what proved to be a challenging project.

Facts and figures In my experience, Déjà Vu handles electronic device catalogues and manuals, technical automotive texts, surgical devices and all repetitive technical descriptions and instructions very well. Productivity increases aren’t consistent because they can be massive for lists, very repetitive procedures and updates but considerably less in other cases, ranging from productivity losses when starting a database to gains of anything between 10 and 90%. I’m a little sceptical about giving percentages because, frankly, I’ve never translated anything both with and without Déjà Vu to make a real comparison. That said, I’ve achieved a record of over 20 thousand words done and dusted in a day using the system. And when I say a day, I mean eight hours not a 24 hour translation marathon.

The agony and the ecstasy A word of caution however: this sort of output is not, unfortunately, an everyday occurrence and when it does happen make sure you’re prepared. These levels are only achieved when you have a massive database that’s able to cope with pretty much everything a new translation throws at it. And that can mean frighteningly intensive mouse work. Ensure you have adequate wrist support and/or a good trackball. A cavalier approach could mean massive productivity gains followed by repetitive strain injury and several weeks off work. It’s less a matter of swings and roundabouts than diving spectacularly into an empty pool. It has in fact been calculated that the majority of RSI cases amongst computer operators are caused by mouse work rather than bashing away at a keyboard. That’s a figure I don’t find hard to believe because I was once languoring sadly in their ranks. After precisely two days of the 20,000 words per day I mentioned. The latest version of Déjà Vu cuts the amount of mouse work significantly, however, thanks to features like Autoprocessing, Autopropagate, automatic digit copying and decimal conversion etc. So, as far as TM packages go, Déjà Vu 2 is good news all round. It’s significantly cheaper than its competitors, faster, more stable and more flexible. It offers productivity enhancements that the others don’t and its alignment program is not only easy to use but comes bundled with the package. Go for one of the other famous names and you’ll find their alignment add-ons are inferior and in some cases cost more than twice as much as the whole Déjà Vu package. What’s more, it’s very easy to use Déjà Vu with one of the various speech recognition systems from Dragon, IBM and Kurzweil, which will increase your productivity gains still further and leave your hands feeling much happier at the end of the day.

To buy or not to buy So, should you buy it? If most of your translation texts are supplied on paper or bear little resemblance to one another where sentence structure is concerned, forget it. But if you translate the right kind of stuff and want to experiment with translation memory, you won’t regret giving Déjà Vu a spin. It will set you back less than any of the other packages and most likely work better. All it takes is two or three clients that are suitable for TM and you’ll quickly earn on your investment. It’s also easy to learn and allows you to discover its more powerful features step by step as you become acquainted with it. If you’re not convinced, download and try the fully functioning demo (www.atril.com), which will give you a month to make up your mind and won’t cost a bean. If you’re a translation company, once again yes (unless you have a major client demanding compatibility with one of DV’s competitors). Déjà Vu offers as many if not more project management features than any of the others. What’s more, forthcoming versions will enhance this with the possibility of integrating local and remote databases using the Internet. And because it costs so much less than its competitors while providing superior performance, it will also be much easier to persuade your translators to use it.

Stop the Presses! The following item was received after the Translation Journal had been put on the Web: The latest version of Déjà Vu can now import, work on, and export Translation Manager and Translator’s Workbench files. The forthcoming version will also be able to build databases from files produced using these packages.

See also the review of four other translation memory programs in this issue of the Translation Journal.

Michael's website:www.michaelbenis.com

© Copyright 1997 Translation Journal and the Author
URL: http://accurapid.com/journal/03TM1.htm