|
Much more than memories
Michael Benis reviews Atril's new Déjà Vu X
Once more into the TM fray...
During the recent ITI Conference at Warwick University I was keen to dispel a widespread misconception by pointing out that modern translation memory products offer very much more to help a translator in their work than just translation memory. Perhaps the most common misconception about translation memory products is that they are only of use to people who work on repetitive projects. In actual fact, nothing could be further from the truth. Certainly, translation memory was originally designed for just such a scenario, where it is unquestionably of great help in achieving time and cost savings while simultaneously increasing consistency. But as the years passed, both freelance translators and translation companies gradually realised there were other benefits, too. Whenever you use translation memory you are feeding a system that provides you with support in every area from terminology to problem constructions, automatically showing you all the various solutions adopted in the past. What's more, almost all of this happens without you having to do anything special in the way you work, unlike glossaries which take time to create, maintain and actually use. So you get more support than from a glossary and for less work, too. But there's more to the picture, since the products have gradually accumulated a whole series of features and functions designed to help increase productivity and quality, ranging from autotext to automatic functions for checking formatting tags, numbers and terminology.
A new Déjà Vu
Déjà Vu 3 was consistently in the forefront of these developments, above all thanks to Atril's close relationship with its large and enthusiastic user base. For many years, any useful features suggested would often be implemented within days. As a result, Déjà Vu became one of the most feature-rich translation memory programs on the market and though its users were aware that the long-awaited latest version would include new technology for getting more out of their databases (see example-based machine translation below), everyone was a little mystified about what else could be added to the package to make it work better. Indeed, many were rather fearful that any changes to a product that had grown and matured so well would not be improvements at all.
When the new version, baptised Déjà Vu X, was released with a completely different interface and a little too early in less than stable form, there were gasps of disappointment all round. It wasn't just fear of the new. Déjà Vu 3 was (and is indeed still is) one of the most stable translation memory products on the market. Everyone suddenly found themselves with an unfamiliar interface and a program that was much less stable than they were used to. Part of the panic was generated by the fact that more than one competitor product has spent years never mind months sorting out all the initial bugs that can occur with an entirely new design. There was considerable fear that the situation with Déjà Vu would be the same.
Well, you can all relax now. Bulletin is pleased to report that the situation is firmly under control and in all apart from a few aspects Déjà Vu X is now a stable and reliable product (and believe me, I did the most horrendous things to try and trip it up). In fact, the development team's round-the-clock dedication to ironing out the initial wrinkles in Déjà Vu X means they're almost entirely out of fire-fighting mode and that we'll soon be seeing a steady stream of enhancements, some of which will be announced in the course of this review. So, now that you know Atril's new baby provides a wealth of features and is stable to boot, let's move on to take a look at the details.
Mining your memories more effectively
The big leak during the years when Déjà Vu X (DVX) was under development concerned the new morphological engine used for its shallow translation memory technology, which Atril now refers to as "example-based machine translation". Don't despair, this is about as technical as things are going to get in this article. So, just to take a deep breath and swallow the medicine.
Put simply, this means that the new Déjà Vu goes even further than Atril's previous "assemble" technology at combining terms in your terminology database with sentences and even small portions of sentences in your memory database to make your fuzzy matches less fuzzy or indeed even "repair" them to form a 100% match. In short, the system uses rules a little bit like machine translation to form solutions that are however based on your own or your colleagues work, stored in your databases. Hence it is "example-based". This makes the core translation memory technology more effective at recycling old work to increase your productivity. That said, the results in practice are somewhat less than earth shattering compared to mature versions of Déjà Vu 3 (DV3), though nevertheless significantly superior to Déjà Vu's competitors. That includes STAR's Transit, which is the only other system on the market to my knowledge that is able to use its terminology databases to perform the relevant substitution for word differences. Had that been the only improvement in DVX, there would still be a story, but as things stand that's less than the half of it.
An improved interface, whichever way you look at it
As I mentioned at the beginning of this article, Déjà Vu has accumulated a wealth of features over the years. Consequently, the interface in DV3 was the result of a gradual process in which a variety of separate windows showed different information, and different interfaces were used to maintain one's terminology databases, maintain one's translation memory databases and customise the user-configurable SGML filter respectively.
The new Déjà Vu interface has been completely redesigned from the bottom up, and although it will cause existing users a little culture shock when they make the transition, I honestly can't imagine very many not agreeing that it is an improvement. Firstly, it is cleaner and less cluttered than its predecessor. Secondly, you can set it up so that the sentence you're working on is always displayed in the middle of the screen, giving you preceding and subsequent context without any need to constantly scroll to adjust your position. But best of all, it's highly customisable, allowing you to choose precisely how much of the massive array of information that DVX can provide is actually displayed on screen, as well is how it is displayed.
Easier on your eyesight
Even the colours used to denote the different types of matches found or translation problems encountered can all be personalised by the user. In addition, these colours are now displayed down the centre of the main window and not over the text itself, which will be welcome news for those who found it difficult to read over a coloured background in DV3.
Another very nice touch allows you to choose a bigger font size for the text you're actually translating and a smaller font size for your reference material - the database terms, sentences and portions displayed - enabling a lot of information to be provided while simultaneously keeping your eyes more rested because you spend most of your time working with bigger characters.
Easier on your fingers
Déjà Vu X inherits all the time-saving keystroke shortcuts of its predecessor, Although some of the actual keystroke combinations have had to be changed you can easily change them to anything you like. Salient features include the ability to split or join sentences easily to achieve clearer or more effective translations, as well as to feed the terminology database on the fly by just selecting rather than typing in the terms as you go. Database search results from both translation memory and terminology databases can also be incorporated in your translations with a short keystroke combination.
While I'm on the matter of saving time typing, DVX is also perfectly usable with speech recognition, although it would benefit from a little work to allow users to simply say the names of the menus in order to drive them, rather than to speak the keystroke shortcuts as is now required.
Meanwhile, benefiting the typists among us, Déjà Vu's existing Autotext, which automatically puts in the full words or phrases for user-defined abbreviations you enter to save typing time, has now been joined by an AutoCorrect function similar to Microsoft Word's that corrects any misspellings as your fingers trip their way over the keyboard.
Enhanced quality control
Following this logic, the spellchecker can now be configured to run automatically as you move from one sentence or segment to the next, as can the quality control checks on formatting codes (tags), numbers and terminology. These checks can, of course, still be run in batch mode when your work has been completed.
Quality control also benefits from a new "Comments" feature that allows the translator to note any doubts our findings relating to their translation of a given segment, either to facilitate subsequent research, database and project management or flag points requiring team decisions.
Other new interface features include a more comprehensive Analysis function along the lines to be found in TRADOS and a new type of match, known as a Guaranteed Match, used to designate 100% matches that occur one after another in any user-definable number and are therefore likely to be more appropriate than isolated 100% matches. These guaranteed matches can be locked to prevent them from being edited, very much in the same way as TRADOS' new XTranslate matches. And before anyone's hackles rise in anger at the thought that this is simply a pretext for ensuring translators can't charge for 100% matches, don't forget that it can in certain circumstances be very useful for the end client or translation company if it is only their reviser who checks and edits such matches, although the translator ought and should be paid to at least read them. One way or another, though, these matches must be checked. Just one example of why might be when the source language does not require a sentence to have a subject, but one or more of the target languages in the project do. If your string of Guaranteed Matches is not long and lucky enough to register that they apply to a product with a different name (because the name isn't present in at least one of the source segments), leaving them unchecked would leave all the occurrences of the old product name in the translations (where they are the subject one or more sentences) unchanged. We are, however, starting to move away from the differences you see on screen to aspects that make it very clear the new Déjà Vu offers very much more for project managers, translation companies and corporate translation departments than any of its predecessors. Because the differences in DVX also go way beyond the interface, as you're about to discover.
Differences that are more than skin deep
The crucial differences between DVX and DV3 extend beyond its new morphological matching engine to the fact that it is now able to carry out batch processing (including pre-translation) on files of any type within the same project, increasing productivity and making for more effective quality control using Déjà Vu's "Filter" function which allows you to select and analyse, check and revise all the segments in a project (irrespective of whether they are in the user interface, help files or manual, for example) that contain a certain expression or term in either the source or target languages. This is an enormously powerful tool for ensuring quality and consistency that is only available in Déjà Vu or STAR's Transit, where it is however limited (as it used to be in Déjà Vu 3) by the fact that projects can only contain files that were imported from a single given format. What's more, the new DVX interface also allows you to use this function on translation memories.
Multiple everything
In a way, the core theme for the new developments in DVX is that you get more of everything and multiple everything. In addition to being able to mix different file types in a single project, you can also use a number of different databases in your own order of preference - in Déjà Vu X workgroup - even assign different read/write attributes, allowing you to fine-tune the system to deliver the greatest possible support for any given project.
Continuing the multiple theme, Déjà Vu projects can now contain an unlimited number of target languages as, indeed, can the memory and terminology databases, significantly facilitating database management and project quality control, especially for large corporate users. This multiple language capability is enhanced by full Unicode support.
Last but not least, Déjà Vu's terminology management system is now much more sophisticated than before and allows you to add as many term attributes (including synonyms and antonyms) as you require.
Easier project/file management
Getting back to file management, it is now easier than ever to add, export or remove files from a project as it proceeds and your projects can contain files from a large number of different directories. What's more, it is also easier for project managers to send out a given selection of files in individual translation projects accompanied by the relevant memories, creating what DVX calls "Pack and Go" packages, which bring everything an individual translator requires together into a single, highly-compressed file to dispatch over the Internet. Several of the program's filters for different file formats have also been improved, including the PowerPoint filter that now imports PowerPoint notes as well as slide content. Unfortunately the Word filter still doesn't import the text in Microsoft Draw text boxes, a benefit that is still the prerogative of Transit. Also, while filters are provided allowing you to work on TRADOS and IBM Translation Memory projects in Déjà Vu X, they are not available for SDLX or Transit projects.
The option of temporarily exporting source and target text for review in a two-column Word file continues almost unchanged, with the exception that you can now also export projects for "External view" (as this feature is now called) in the form of an HTML file as well.
Corporate support
Completing the new features of interest to translation companies and large corporate departments in particular, DVX includes a number of security functions and levels that allow you to track and control who can make given changes to projects and databases. Last but not least, the Workgroup version comes with a COM+ server capability that significantly increases database access speeds on networks (I can't tell you by exactly how much, but it was certainly noticeable). Indeed, for those worried about database access speeds, particularly considering the fact that DVX's new Unicode databases are approximately twice as large as its predecessor's, we're happy to report that DVX is consistently faster, with the exception of situations in which you have decided to use many different databases at the same time. Even in these cases, network performance was almost invariably faster than DV3 with the new COM+ server components installed and enabled. The only thing that definitely takes longer is, strangely enough, loading the list of subject fields. On the subject of servers, it's worth noting that the corporate version of DVX (Workgroup) also features a programmable API and that a remote server solution is currently under development.
One thing you won't currently find in DVX, or more correctly accompanying it, is Atril's new TM Builder, which is however due for release in the not too distant future and will be supplied free of charge to all past and future purchasers of the DVX Workgroup version, while also being offered as a standalone option for all and sundry. The aim of the TM Builder is to incorporate similar batch processing capabilities to those now offered by the new interface into a powerful alignment tool that also uses the translation memory and terminology databases to improve the accuracy of its automatic alignment results. As things stand, the older Déjà Vu alignment system continues practically unchanged, although the interface is, as you can imagine from the foregoing, now clearer than before.
Quick tips
Ideally, your operating system will be Windows NT, 2000 or XP to get the best out of DVX. It's not that Déjà Vu X won't run under Windows 98 or ME, but the results are less predictable.
Irrespective of the operating system you choose, if your PC has a USB port, I'd recommend you go for the very dinky little USB dongles that Atril now offers to enable and protect Déjà Vu. They're not only more compact and more reliable, with a handy little green LED that indicates all is well, but they're also much easier to switch from your desktop to your laptop than the bulkier parallel Port dongles of the past.
Flies in the ointment
The weaknesses in DVX are more than anything the inevitable teething problems of a new product or minor shortcomings for which in almost all cases a remedy has either already been found or is under development. Examples of this include the less than complete implementation of Active Accessibility for smooth control of the program using speech recognition software, irritating minor inconsistencies with the naming of Pack and Go files, program crashes if you do a couple of spectacularly silly things when setting up projects or building the lexicon, and above all the fact that the memory and terminology databases can only show one target segment/term at a time. The latter is particularly irritating if you want to check the previous and subsequent segments around a particular problem term or construction, investigate previous translation choices for filtered segments in the translation memory database or when engaged in database maintenance. As stated, however, the developers are already working on overcoming this problem, which is less simple than one might imagine when you think that the new databases can, of course, comprise many different languages. In short, although there are still a few bugs left in DVX, the overall stability and smoothness of operation offered is already average for software in the sector. Hopefully it will soon be even better and on a par with its predecessor.
Competitive position
For a number of years now, Déjà Vu has consistently offered more features than the competition, particularly where support for the translator is concerned. Déjà Vu X takes things very much further, while adding a wealth of new functions that will make it significantly more attractive to corporate users. There is currently no translation memory system on the market that offers the same level of support in terms of increased productivity from database leverage and word processing features combined with reference information, editing and quality control tools. The only translation memory product that comes anywhere close is STAR's Transit, which has a highly versatile and practically incorruptible system of reference files instead of translation memory databases, although this can be a drawback if you need to maintain tight control over how these resources are deployed in a large organisation. Transit's interface also allows certain character formatting (bold, italics and underline) to be displayed rather than shown as codes, which many users may prefer.
Others may find the TRADOS approach preferable, especially if they do all their work in Word. ExtraTerm could further attract corporate users to TRADOS, offering more than Déjà Vu's terminology extraction tool, although TM Builder might yet prove an even bigger magnet. On the other hand, whereas TRADOS presents a variety of different interfaces, Déjà Vu has the advantage that you use just one and can batch process any number of files in different formats from within it. Although many lament the laboriousness of working in TRADOS' TagEditor, the preview function that allows HTML files to be seen as they would be displayed in a browser window is quite popular.
|
|
First published in ITI Bulletin, 2003. |
