This article was written by Michael Benis (www.michaelbenis.cwc.net) and first published in ITI Bulletin (www.ITI.org.uk).

There can be few readers of the ITI Bulletin whose curiosity hasn't been aroused by translation memory. Yet most translators still don't use it. The ITI Rates & Salaries Survey reported that only 15% of respondents used translation memory tools, although this rose to 40% and 50% for those earning £50k-£54,999 and £75k+ respectively. If you're not in those groups, read on to find out what you're missing. If you are, read on to find out whether you could be doing even better. Because we've now reviewed the five leading translation memory packages on the market to help you make 1999 your most productive year yet.

So what's all the fuss about? Why do you keep seeing adverts for translation memory products? Why are increasing numbers of agencies trying to get you to work with these products? What's more, why are increasing numbers of freelance translators using them without any prompting? We'll be answering all these questions and more in the world's first extended review of translation memory products.

The idea of translation memory is basically very simple. It's to stop you scratching your head and losing time while you do it. That's the reason for the enigmatic title of this review. No longer will you bring digits to cranium in the hope of stimulating your grey cells to provide an answer to the question: "Oh, how did I translate that last time round?" Instead, your translation memory program will instantly remind you, prompting the satisfied exclamation: "Ah, so that's how I put it last time round!"

That's basically the long and the short of it. From "Oh" to "Ah". Translation memory is nothing more than the computerised equivalent of two massive parallel filing cabinets containing every sentence you have ever translated, both in the original and in your translation.

This, as you can imagine, does not have any obvious appeal to someone translating poems, advertisements, newspaper articles and the like, but is obviously highly exciting for anyone involved in fields where there is a certain repetition in the procedures or practises described. Consider the operations involved in adjusting the valve clearances on a car engine, or installing a new PCI card in a personal computer, describing the procedures for selecting items in a pull-down menu, the symptoms of a given disease or social service case histories and you'll realise that there are in fact quite a number of fields where there is a high probability of sentences or phrases being repeated.

Then there are fields, like software, where time-to-market is such a crucial factor in the success of a product that it's often essential for work on the product documentation to commence while the product is still in the final stages of its development. This can involve a large number of very small changes to the product literature. If most of them can be performed automatically or semi-automatically, this reduces the time-to-market still further, on the one hand, while also increasing productivity.

Moreover, there are many fields plagued by frequent product updates that aim to ensure a competitive position is maintained. Here, again, the best translation memory software will be capable of immediately translating everything that is the same, allowing us to just focus on what is different. Some translation memory software can even make at least some of these changes automatically.

Completing the picture, more and more product ranges now feature shared components, modules or subassemblies, which means that a translation memory user who has completed a translation for one product may well find that subsequent documentation on a sister product contains substantial chunks of text that are exactly the same.

The software localisation industry cannily recognised all these potential benefits around ten years ago, mainly with the aim of making the localisation process as quick as possible. As the systems became increasingly user-friendly, technical translators and translation companies were quick to follow suit, realising that the productivity increases achieved when these systems are used properly translates into happier customers and more money in the bank.

Translation memory is like most things in business. If you want to make more money, you have to invest some first, and none of these systems really comes cheap. Indeed, any of them are likely to be the most expensive software you buy.

Before you start fingering your credit cards, you need to ask yourself a number of questions in terms of how fast a return you are likely to get on that investment.

Firstly, there's no point in considering any of them unless a sizeable proportion of your work arrives as files on disk or via the Internet. Although optical character recognition software has improved tremendously over the past few years, it still requires substantial correction, resulting in productivity losses that may very well not be outweighed by using translation memory.

Secondly, consider the type of work you do carefully. If you can identify areas in which it is repetitive (and you need to think of sentence construction rather than just terminology), then you should seriously consider translation memory. Take another look at the examples in the previous section. Remember that these do not have to be technical areas. Book translators engaged in projects with extensive specialist terminology, for example, will also benefit. One such project of my own was a CD on birdwatching in Italy. There was a vast number of "specialist" terms to describe the different body parts and feathers of a myriad of birds, their shape, colouring and so on. What's more, the phrases used in these descriptions all followed very similar patterns. Translation memory made that project much less laborious, more enjoyable and profitable.

If you have just one or two regular large customers that fit the bill, so to speak, you are more than likely to see a return on your investment within a year, and maybe as little as three months or even your first large repetitive project. Every productivity gain after that is a kerching! on your cash register.

All the different translation memory packages are in fact suites of software products that integrate to varying degrees. We tested the offerings of 5 different manufacturers: Atril (Deja Vu), IBM (Translation Manager), SDL (SDLX), Star (Transit) and Trados (Translator's Workbench).

All of these offer the same basic types of programs:

This allows you to create translation memory databases from existing translation files. If, for example, you have been translating motorbike manuals for five years and kept all your source and translation files, one of these programs will enable you to build up a massive database immediately, allowing you to exploit the full potential of your system from the word go. They work by creating your first database - the computing equivalent of a printed parallel text translation. On one side there's a source segment (which can be a clause, sentence or even paragraph), while on the other there's the target language equivalent. It's obviously important that all these segments are lined up properly and that's what the alignment program does.

There are, however, big differences in the alignment programs offered. Transit, for example, offers a highly automated program that will save you a massive amount of time. Trados offers WinAlign, a semiautomatic program with a graphical alignment function that allows you to spot and rectify any alignment problems very quickly and simply. Both these alignment tools will save you a lot of time (so you can make more money) but they are also costly extras. Deja Vu, SDLX and Translation Manager are all supplied with their own alignment programs at no extra cost. The most cost-effective choice will depend on the volume of files you have to align immediately and the likelihood of your having to align a large number of files in the future. Trados caters for clients who only have an immediate short-term requirement by also offering WinAlign on a short-term lease basis.

Life isn't predictable. Everyone makes mistakes. The most carefully researched terminological choices may need to be changed, sometimes for no other reason than that there has been a management change in the company you work for. Power cuts and equipment failures can cause computer crashes and corrupt your cherished databases (yet another reason for having an effective backup strategy!). What's more, your databases can become enormous over the years and need pruning. A database maintenance program enables you to cope with all these eventualities. Only SDLX doesn't come with a database maintenance program. Its databases are, however, in Access - meaning you can edit them if you have Microsoft Office Professional.

These are slightly more sophisticated versions of the electronic glossaries many of us have built up over the years. Moreover, they can import these glossaries and indeed other dictionary files. IBM Translation Manager already comes with terminology dictionaries, while rather more sophisticated dictionaries are available as extras for Transit and Translator's Workbench. Some of these terminology programs are very complex including fuzzy matching algorithms that enable them to search for all words with a given root, so that they could - for example - find "logic" when "logical" is the word you want but isn't in the dictionary. That said, unless you're a beginner working in-house with suitably verified glossaries, you stand to gain little from these programs, and whether one is superior or inferior to another should not influence your choice of package too much.

Building up and maintaining glossaries takes time - more time than they are ever likely to save you, particularly since all the better programs allow you to search for individual words in the translation memory, displaying them in context - which is much more useful. What's more, you don't have to go through any special operations to get the terms there. Two programs are slight exceptions to this rule: Deja Vu - which will "assemble" translations from the largest available units (but settle for individual words if it can't find anything else), and Transit - which is able to insert words from its glossaries in fuzzy matches (see below) when these words constitute the only differences between a segment in a translation and a segment in its memory.

This is the program in which you actually carry out your translation. It generally takes the form of two parallel windows, with the source text on one side and the target text on the other. In addition, there is often a further window showing the segment/s in the translation memory. The segments displayed will either be 100% matches, meaning the system has found exactly the same sentence in its database and you can simply import this directly into your translation (several of the programs can do this automatically) or "fuzzy" matches - which are segments with a number of small differences. Sometimes, of course - especially in the beginning - you won't get any matches at all.

All the programs allow you to set the degree of "fuzziness". Some people choose a very high setting, considering absolutely any suggestion helpful to jog their memories, while others find that anything requiring more than minimal editing is simply a distraction. Finally, most document editors also contain a window to the package's terminology program.

Translation memory features aside, the document editor is basically a word processor. Some of them are pretty primitive - offering little more than cut and paste, search and replace, while others show full formatting information or offer an extensive selection of AutoText features. The big exception here is Trados' Translator's Workbench, where you work directly in Microsoft Word alongside a second window with panels showing the translation memory matches and terminology matches.

These are used to convert files from one format to another, whether to import them into an alignment program or document editor, or export them from the document editor into their original format. The reason for this is that whatever format a file is supplied in, you will of course work in the document editor/translation environment of your translation memory product. This is in itself a productivity benefit in that you always use the same interface and don't have to cope with re-remembering different menu and button locations, and different keystroke short cuts.

One difference between the various different packages is that some come with a full selection of filters as standard (Deja Vu, SDLX and IBM Translation Manager), while others only offer them as costly extras (Transit and Translator's Workbench). Note, however, that all the packages come with a filter for Microsoft Word as standard.

Before buying a package, you need to carefully consider what you need from these four basic components of any system. The pros and cons of the different systems are considered in the individual reviews below.

Opinions are divided on the quality argument. There's no doubt that translation memory can help increase terminological accuracy and consistency. The more experienced and specialised the translator, the less of an advantage that will be. It's spectacularly helpful for junior in-house translators, however, who don't need to spend any time building up translation memories or glossaries, and immediately benefit from the many years experience embodied in them. Many comment that using a translation memory system is like always having a pet senior translator on hand to help them out.

When it comes to style, however, we find ourselves on the other side of the coin. Only three of the programs (Deja Vu, Transit and Translator's Workbench) allow you to split and join segments entirely at will. Not being able to do so can tie you too rigidly to the source text - more so than an experienced translator would allow themselves to be in a "free translation". It is also easy to be distracted from the "flow" of a translation when working on it in segments. It is of course true, however, that the influence on fluency of working in segments decreases as you become used to it. Nevertheless, many translation memory users find it useful to have a quick read-through upon completion, editing the final document to increase its fluency and style in general. This of course needs to be factored in when considering any potential productivity gains.

This one's the big daddy of them all, developed by IBM to satisfy its own localisation requirements. For a while, it ruled the market but gradually slipped from pole position due to a combination of slow development and poor marketing. Version 2.5 is a full 32-bit program and the only product in this review to be supplied complete with an alignment tool, terminology dictionaries and filters (for most popular word processor formats excepting Word 97 - only Word 95 - and downloadable free Microsoft Word plug-ins for PageMaker and QuarkXPress). A wide range of languages are supported in addition to the usual Latin languages. You can find the latest information on the Website detailed in the contacts panel.

That said, IBM Translation Manager is in many ways the simplest of the programs, with what barely appears to be any special user interface. This can be confusing at first, since you have separate windows for every function, which can be resized and positioned as you prefer. Selecting an imported document for translation or selecting the "Translation Environment" option in the "Window" pull-down menu does, however, cause all three relevant windows (Translation, Translation Memory and Dictionary) to be displayed simultaneously in more or less the same way one finds with Deja Vu, Transit and SDLX. The plus side to all this is precisely that you can customise things pretty much as you like and, above all, that this program is blisteringly fast to open and work in.

What you actually see when working is the source text itself, with bracketed codes indicating the formatting information of the original file, together with the text that you need to overwrite as you would in your normal word processor. The Translation Memory window shows any matches found, and the Dictionary window shows the relevant terminology available in the quite acceptable basic dictionaries that IBM supplies.

Searching and database management is simple, fast and logical. Moreover you have the option of searching the dictionary for words and the translation memory for sentences without leaving the Translation Environment. The one big disadvantage here is that you can't search for individual words in the translation memory, meaning you can't consult the possibly very many sentences containing the word that's causing problems without shutting down your project and opening the database separately. Although this doesn't take very long since the various items of Translation Manager all open and close so fast, it nevertheless means this useful information is several mouse clicks away. What's more, you can't survey it whilst considering the current sentence and are therefore more likely to forget the various different alternative solutions possible. You will also find it more difficult to envisage how they can actually be incorporated in the segment in question. Compare this with the Scan, Fold and Concordance functions offered in Deja Vu, Transit and Translator's Workbench respectively.

Like most translation memory programs, Translation Manager offers very little in the way of word processing features. You can select from three fonts in terms of how the text is displayed, plus a number of different colours to distinguish between translated text, untranslated text, the active segment in which you are working etc., but that's it. Forget WYSIWYG. All the formatting information in your files that would be used to show italics, bold, tables and the rest is displayed as codes between brackets. A number of options are available that allow you to protect this information (accidentally changing it would change the formatting of your translation) and display it in full or abbreviated form. This is the aspect most likely to cause confusion in new users. Most segments will at least be enclosed between these code tags, while you may find others around certain words (such as an underlined word, for example) or characters (like the superscripted 2 in pr²). You naturally need to pay attention to ensure that these codes are around the equivalent words in your translation. Translation Manager does nothing special to make this any easier than any other program, but doesn't present any obstacles either.

What many users will miss, however, is the possibility of using the AutoText features that most word processors now provide to help increase your productivity. All you get in Translation Manager are the basics of cut and paste.

On the other hand, a useful feature that could be refined further is the possibility of joining segments together. This can be helpful when translating into languages that tend to use longer sentences with a more complex structure than is usual in your source language, or in which information tends to be presented in a different order. A classic example of this is the way English instructions tend to follow the temporal order of the operations concerned, while French or Italian instructions first describe the principal operation and then the other operations leading up to it. Unfortunately, however, Translation Manager won't allow you to do this in reverse, since although segments can be split, you can't do this into units smaller than a sentence. It's worth noting, though, that not everyone finds such a function useful. Many would say that it's simply easier to enter a number of sentences in the target segment or allow one sentence to flow across several segments. Especially when the system (as is the case with Translation Manager) doesn't have any automated functions that allow these smaller/larger segments to be used automatically.

IBM's alignment tool - which goes by the perfectly descriptive but rather unglamorous appellation of Initial Translation Memory Tool, works fast enough but is not automated in any way. Nor does it allow you to customise the segmentation rules to suit the punctuation practice of different languages or clients. You can join segments, but once again, you can't split them into units smaller than a sentence.

So what's it like to work with? Well, it's undoubtedly the least user-friendly of the programs. If you've never used a translation memory package before it will take a couple of days before you're using this one confidently. This isn't helped by the documentation supplied (Online Help and a series of hypertext documents with very little in the way of illustrations), which is comprehensive but not particularly well organised and doesn't hold your hand through the initial operations needed to create a database and complete your first translation. Otherwise it's fast, relatively uncomplicated and bug-free, though without the bells and whistles you'll find on many of the other programs. If you want to try it, you can download a fully functioning demonstration version from the IBM TM Website for evaluation.

The price is about average at £1495 for the full version, which makes the whole package quite competitive if you consider that a wide range of filters and dictionaries are included. Moreover, if you don't need the non-Latin character sets (e.g. Cyrillic, Arabic etc.) or Initial Translation Memory alignment facility, you can get the Personal version for just £525. So, if you have a big customer that particularly wants you to use IBM Translation Manager you won't have too much to regret. It will keep your customer happy and improve your productivity. But if you're making a choice to satisfy your own needs only, Deja Vu, Transit or Translator's Workbench are all easier to use. What's more they also offer additional productivity-enhancing features that will give you a bigger bang for your buck, as they say over the pond in the land of the Big Blue.

SDLX is a newcomer on the scene, developed by SDL, one of the biggest localisation companies in the United Kingdom. Having looked at the various translation memory packages on the market they decided they'd rather develop one themselves, tailored to their own requirements. Indeed, it's not the first time they've done this. Many of you will remember Amptran, their first venture into the field, reviewed in the ITI Bulletin a couple of years ago.

SDLX is a development of Amptran, but also a far more sophisticated product with a number of unique features of its own. Unlike Deja Vu and Transit, which integrate practically everything in a single interface, the SDLX suite is installed as a series of discrete programs. SDL Edit is the one in which you do your actual translation. This features the usual customisable three window layout showing the original text, translation and translation memory. As soon as you load and work on a translation, however, you'll find there are two important differences. Firstly, SDLX offers WYSIWYG features, with font formatting actually being implemented rather than displayed. As a result, things look more like they do in a normal word processor. Bold isn't shown as a word with codes around it but as a word in bold, and so on. This makes the whole environment look considerably more familiar than almost all the other packages. Having translated the segment, you simply select the word or words in bold (italics, underline etc.), click on a paint icon and select the corresponding target text to copy the format. Formatting buttons or keystroke shortcuts would often be more convenient, however.

The other thing that makes life look more normal is the absence of codes in brackets. All you see is the text itself and an array of bright colours. These colours take the place of the codes and can be painted across in the same way as the formatting information. This doesn't actually save you any work, but it certainly makes life more colourful and uncluttered. Many will undoubtedly find this less intimidating during their initial approach to translation memory. It's an attractive solution that means you can concentrate on the translation before going back to check on the codes later. When you do, you can see if anything is missing at a glance because you just have to check the colours rather than the weird and sometimes lengthy contents of the codes. SDLX also incorporates a check to make sure you can't export a document if formatting information in the original is not also present in your translation.

These very nice features apart, SDLX is a disappointingly spartan program offering no more in the way of AutoText or other productivity functions than IBM Translation Manager. All you get is cut and paste and an integrated glossary program. You can't search for words and phrases in the translation memory and indeed you can only apply it to perform a pre-translation when you import your documents. What's more it won't subsequently process 100% matches automatically. You can't join or split segments either. What you can do, though, is customise the segmentation rules. The only drawback here is that you have to write a file of your own, which many users may find a little daunting.

Making life more complicated, you can't import documents directly into SDLX Edit. You have to convert them first using SDL Convert. This also has the drawback of only being able to convert from a limited range of formats (RTF, HTML, MIF and TXT). The same separate conversion process is required when you want to align files. The alignment process is carried out in a program called - yes, you've guessed it - SDL Align, which is visually more attractive than the IBM offering but not a great deal more sophisticated, though it does have an automated feature for checking existing alignments. All in all in all it's a competent tool, but nothing special. You can join segments in it and are supposed to be able to split them but I could never get this to work properly. Given the speed with which SDL generally works, however, I'd expect that little glitch to be fixed up pretty soon.

Completing the line-up comes SDL TMX, which allows you to convert the Microsoft Access databases that SDLX uses to the Translation Memory eXchange (TMX) format developed by LISA (Localisation Industry Standards Association) to allow translation memories from different applications to be exchanged. You can also use SDL TMX to convert databases from and into Trados Translator's Workbench 2.00 format and delimited format files.

Once you've worked out how all the various different bits slot together, SDLX is a swift, predictable and satisfactory productivity tool. Surprisingly, however, for a package developed by a translation company, it offers very little in the way of project management tools, making it unsuitable for use by other translation companies. Combine that with a number of little bugs that indicate we're only just into the alpha stage of the software, plus a general lack of more sophisticated productivity functions, and you can see why I get the feeling we're looking at a promising product that nevertheless has a good deal of its development still to come. Considering the standard version sells for $1,899 (+ VAT), I'd say that unless you work regularly with SDL (in which case you also benefit from a discount) you'd be better off buying one of the more sophisticated established products on the market.

Deja Vu is a longstanding favourite with freelance translators, not least of all because it has always been extremely competitively priced in a market where the competition have generally been perceived as grossly overpriced. That's not the only reason for its success, however. It's always been an extremely versatile product, offering a large number of features. What's more, it's a product that's been developed in conjunction with its customers. Emilio Benito, the brains behind this memory, is not only extremely approachable in a generally faceless market, but also quite indefatigable in his determination to improve Deja Vu. Many is the customer who has approached him with a suggestion only to find it has been incorporated in a new version of the product a few days later. Which is why Deja Vu is more feature-rich than practically any other translation memory system on the market today. It's also fast and reliable with it.

Like the previous two packages, Deja Vu comes complete with its own alignment program, terminology program, translation environment (Deja Vu Interactive - DVI), spelling dictionaries and filters. The latter are, however, an extraordinarily comprehensive set that will allow a freelance translator to handle practically anything thrown at them, from Interleaf to Translator's Workbench files. Alignment is still largely a manual process (though a number of developments are in the pipeline), but the segmentation rules can all be customised and the segments can be joined or split as desired. Alignment is clear, smooth and unproblematic. But it's when you get into DVI itself that you realise what all the years of work with many different freelance translators have produced.

In its simplest form DVI doesn't look very different from SDLX Edit, with the centre of the screen split into two vertical windows containing the source text on the left and target text on the right. This section of the screen is divided into lines or cells, each of which contains a segment or the first line of longer segments. Moreover, the height of the DVI cells can be adjusted, enabling you to select how much context you see. The full segment is shown in a further window below, which is where you actually enter the text.

When you work in Deja Vu, you start by following a similar procedure to IBM Translation Manager and SDLX, importing your text into the translation environment and carrying out a "pretranslate", which brings in all your full matches (green) from the translation memory, any fuzzy matches (magenta), any assembled matches (blue) and any multiple exact matches (dark green) Unlike the former two programs, Deja Vu then offers you a number of options. You can leave all the segments for which no translation is available empty and simply type in your own translations, pulling across any codes (displayed as curly brackets around numbers). You do this by either dragging them across using your mouse, or placing the cursor where required and clicking the right mouse button or CTRL+D to bring across the next code in line. Likewise, you can just press that keystroke shortcut for each code as you type in your text. If you prefer, you can also populate all the empty cells, copying the source text into them either cell by cell or for the whole document. You can also do this for the whole document and then overwrite it - either manually or using the pretranslate function. This can be useful if there are a large number of lines that remain unchanged, such as code strings or menu figures. These are just some of the many options that can be set in Tools/Options/General.

As you translate, you can also choose to send every translated segment to the translation memory and then at a given point carry out a pretranslation again, to recycle any translated segments that are repeated later in the text. A feature called Autopropagate will do the same thing for you automatically without sending anything to the translation memory if you prefer. You can also propagate manually for a single segment. Not only can your pretranslation insert fuzzy matches with a percentage that you select, but the Propagate function can also make changes for any minor differences it encounters. What's more, the terminology searches can themselves also be fuzzy searches if desired. Last but not least, the segments themselves can be split and joined as required, while DVI also makes it easier to customise your choice of segmentation rules than the other programs.

The list doesn't end there. A further feature called Assemble will attempt to put a translation together for you using the translation memory and terminology database when the system can't find a 100% or fuzzy match. This can be totally useless, a handy memory aid or produce something that only requires minimal editing to be used. Deja Vu can also perform this automatically every time you move to a new cell.

Yet another function, called Learn, can be used to find the translation of an individual term in the database. It does this by comparing all the instances in which the source term occurs and filtering out the other terms in the various different segments. A derivation of this function (Build/Resolve Project Lexicon) makes it much easier and faster to create your own terminology databases with Deja Vu than any other program.

Finally, a function called Scan searches the database for occurrences of a particular term that you highlighted in the source text, displaying it as it occurs in context in one half of the window with the translated segments in the other. This is an extremely useful translation aid that boosts productivity considerably by showing a large number of usage examples at a glance, reminding you - for example - not just of the correct noun but also of the correct verb to use with it etc. You can then click one of a number of buttons to paste the segment from the memory either alongside the source text/fuzzy match that's already in your translation window, or directly over it. Making this feature even more useful, you can choose to let Deja Vu continually display the scan results from both the translation memory and terminology database in two separate windows, providing you with this reference information all the time without your needing to do anything about it. Helping you keep tabs on your translation memory, you can also delete segments that you're no longer happy with from the Autoscan window, although you can't currently edit them as you can in Transit and Translator's Workbench.

All these productivity features can be pretty overwhelming at first. This is occasionally not helped by the fact that any documentation supplied is often one or two steps behind the development of the program for the simple reason that it keeps developing so quickly. What used to make life even more difficult is that there was no comprehensive manual supplied with Deja Vu. That's now a thing of the past, and you get a comprehensive manual that's very easy to follow and could well serve as a model for many other companies in the field. Accompanying it you get a short tutorial on floppy disk with a separate tutorial manual to guide you through your first tentative steps in the use of this powerful software. The manual and tutorial can also be downloaded from the Atril Website.

Just in case you're still worried, Deja Vu also allows you to disable many of its more complex features until you have become sufficiently confident in using its basic functions (Tools/Options/General). Many of these are also productivity aids, though of the sort with which you are likely to be familiar already, such as an AutoText feature similar to that provided with Microsoft Word.

Deja Vu is a stable and reliable product that automates a large number of processes (if desired) so that the translator can concentrate entirely on the task in hand. It even saves your segment translations as you move from one cell to the next, so that should the worst come to the worst and the program crash, you won't lose any work. Not that I want to create the impression that it isn't a stable program. Indeed, it's better than most. Working in the program is a pretty straightforward, step-by-step process. While the window layout showing the full text of the active segment plus the first lines of each segment (and a significant amount of context if the cells are resized) helps you keep track of the flow of your translation.

DVI also offers an array of quality control features when you come to the end of your projects, enabling you not only to check your spelling and the codes, but also to check your terminology against the terminology database. The only other program offering this feature is Transit.

Although Deja Vu originally started out as a tool for freelance translators, it is now used successfully by an increasing number of translation companies and their input has been incorporated in a number of new features that give the product significantly enhanced project management features.

All in all, Deja Vu can hold its own with the very best translation memory programs on the market. It currently loses out to Transit in only being able to use one database at a time (though multiple databases are planned), and has a less automated/more laborious alignment tool than either Translator's Workbench or Transit. Considering, however, that it comes with everything including an alignment tool and comprehensive filters for the extremely attractive price of euro 990 (including all taxes, on which ITI members get a 20% discount), there's really nothing to beat it for the translator who is interested in using translation memory to increase their own productivity rather than comply with the requests of a particular client. What's more, now that you can import Translator's Workbench files, you can keep your customers and your wallet happy at the same time. If you're interested in evaluating Deja Vu either to consider using translation memory in general or to port across from an older system, you can download a fully functioning version on 30-day trial from Atril's Website.

Atril's customer service is generally excellent, also providing very prompt troubleshooting. I know of at least one case where they helped a translator get back on track with a project on a Sunday morning. They also offer a series of Internet mailing lists for translators to exchange information and advice on TM in general as well as Deja Vu in particular. In addition, they have recently added a knowledge management database ("Support Center") to their Website where common questions and answers are discussed.

Now that you've got this far in the review, you're probably getting an idea of how most of these programs work. You have a vision of a massive central database, like an index card system of segments, through which the systems scroll looking for matches. That's perfectly accurate, in essence, where almost all the systems are concerned - except Transit. Transit is the dark horse of the translation memory world. It does things differently. Not everyone is quite clear how. But line the systems up side by side (as we have done), and you'll suddenly find Transit streaking down the home straight just when you thought the race was over.

Transit has a number of surprises up its sleeve - all of them winners when it comes to squeezing every ounce out of your previous translations and increasing your productivity.

The first surprise is undoubtedly the one that will surprise you most - Transit doesn't have a translation memory. There is no massive database. Hold your breath and I'll explain.

Transit takes every source file and every target file you've worked on, strips out all the extraneous formatting and font information, and just keeps the text. It stores all these files with the same filenames but different extensions. The extensions indicate the language concerned. You can store these files in any particular directory structure that appeals. You could, for example, have a root directory (folder) for information technology, a subdirectory for PC systems, a further subdirectory for software and hardware, and additional subdirectories for, say, scanners, printers and modems. These could in turn have different folders for different manufacturers.

Let us suppose that you now want to use the information in these directories to translate a manual for a scanner made by WonderWoof. If you've already translated a manual for a WonderWoof scanner - or indeed several for several models - then the most useful files would be contained in your WonderWoof directory. Nevertheless, you could find useful information in the directories for other scanner manufacturers, such as their competitor CatNip, or any of the other directories. You may not want to use some of these because, for instance, one manufacture uses some rather peculiar terminology of their own.

Okay, here comes the exciting part. With Transit you can select exactly which directories are used to form your translation memory and even which files. Transit then builds what it calls an Associative Network of references (the files extracted from your old projects). If it can (and this is entirely dependent on how much RAM you have installed) it will pull all this information into shape in your computer's system memory to ensure maximum speed (otherwise it's forced to use a swapfile that has to be accessed on your hard disk). Let me emphasise that this massive physical database is never actually created. What happens is that Transit simply establishes the connections between and priority of the various different directories and files specified. It's as if Transit was installing a data link to them rather than actually bringing them "in-house".

Having been informed that one way or another you end up with a large amount of segments available to the translation memory program, I can imagine you asking: What on earth is the point of that exercise? Read on.

Firstly, you never get a database that's any larger than you need it to be. That has two advantages: searches take less time and are less likely to churn out useless information.

Secondly, searches are even quicker because the Associative Network is resident in RAM (if you've got enough of it).

Thirdly, you can always tailor your database precisely to your needs. Even if you're half way through a project before realising that you keep wasting your time with fuzzy matches that come from your CatNip Scanners subdirectory, it's still not too late to simply remove that subdirectory from the Associative Network. This helps ensure that the suggestions you get from your translation memory are always of practical use for your work on that particular WonderWoof scanner. The alternative is rather like a charming but senile member of the family with the most extraordinary memory who keeps volunteering fascinating but totally useless pieces of information.

Fourthly - and this is the most important bit - it makes your system far more reliable. I'm told that all sorts of things happen in an ideal world. I'm not in a position to know. I've never been there. What I do know, however, is that if things can go wrong, they eventually will - generally at midnight when I'm slaving away to a tight deadline. That is not a nice time to find out that your translation memory database has become corrupted half way through a project. Quite what it means will depend on the circumstances, ranging from weird glitches to an interesting selection of utterly useless suggestions and even a database that quite simply refuses to open. None of these problems are insoluble, but they're never pleasant and can well cause you to miss a deadline, never mind a heartbeat. The good news with Transit is that these problems can never arise because there is no database with all its various sector, client, language etc. tags to become corrupted. Think about it. The more you do, the more you'll like this system.

Fifth - and this is of particular interest to translation companies -it means you can store a large number of segments in much less space than conventional translation memory systems.

Transit's terminology program, called Termstar, offers a similar sort of versatility in that you can load multiple dictionaries in it to tailor the terminology provided exactly to your needs. Termstar also differs from the other terminology programs on the market in offering an extremely user-friendly interface that presents itself in the graphical format of a hardback dictionary. It also offers a level of integration with the translation memory that is only rivalled by Deja Vu.

Whereas you can carry out the usual operations to replace terms in your translation with terms in the terminology database and send pairs of terms there from the translation environment windows, Transit can at the same time incorporate words from the terminology database into its fuzzy matches to provide you with a 100% match. This is less all-embracing than Deja Vu's Assemble feature, but often more reliable since it is more modest. This isn't a criticism of Deja Vu, since most translators work in it with the target cells empty until they translate them, meaning that if Assemble comes up with gibberish you can simply highlight the whole thing and overwrite it, during which time it may have usefully jogged your memory. In Transit's case, the whole process goes on in the background, so you're not aware of it. My first encounter with the program for a real job was rather amusing as a result.

I had no problem getting used to the interface, which operates on very similar lines to that of Translation Manager, SDLX and Deja Vu. Basically, it's split into four parts with a layout that you can customise (with the codes shown as tags). In addition to this, you actually change the contents of the windows depending on what you need to do while you're working. But when you start, you generally have the source text in one window, your translation in another, dictionary information in a third smaller window and finally - in the smallest window of all - what Transit calls a Scratch Pad. This is a very useful on-screen equivalent to the sheet of paper I normally make notes of queries and uncertainties on. The advantage of the Scratch Pad is that the notes will stay there on screen, in context, reminding you of the problem until you have solved it and deleted them. Whatever your own personal favourite, including an array of little yellow Post-it notes all-round the screen, nothing is quite as reliable as this Scratch Pad for the simple reason that it can't get blown away by the wind, eaten by the cat or lost behind the back of your desk.

One word of caution: Transit features an AutoSave function like Word, but it doesn't automatically save your translated segments as you move from one segment to the next. As a result, performing operations that involve switching from one window to another can cause you to lose the segment translations completed since the last time you saved. As a precaution, always save before switching windows to carry out Fold or search operations etc.

But back to my strange confession. I battled my way through the manual (very comprehensive but not terribly clear, lacking a "tutorial" section that takes you through things step by step). Then, having converted an old database into two Transit reference files, I set up an Associative Network for the first time and imported my file. The database I had used to prepare the Associative Network was very large and the file I received didn't seem particularly unusual, but I hadn't got any exact matches. Not expecting this (it's been a while since I've never got any 100% matches for this client) I assumed I hadn't installed the Associative Network properly. So I did it again. And got the same result. Fiddling around with the program a little more, I found I was however getting fuzzy matches further down in the document, meaning that I had actually succeeded in loading the Associative Network properly the first time round. So I set to work, cheering myself up by writing a series of macros for Dragon NaturallySpeaking Professional to see whether I could control everything by voice. That all went very nicely, so I leant back in my chair, stuck my feet up on the desk and carried on rattling away. Just as I'd started taking everything for granted, the program suddenly went beep and jumped to the end. Infuriated, I sat upright again to try and find out why it had suddenly done this strange thing. The reason was simple. It had automatically recycled material that I'd translated in the meantime and completed the rest of the translation for me. Notwithstanding the prosaic (but clear) interface, I suddenly found myself rather liking this program.

The romance had actually started earlier. One of the most boring things you can do when using translation memory tools is to align old files. It's a process that can seem as if it's taking forever and if you have a busy schedule and are always in demand the only thing it can be compared to is standing under a cold shower tearing up five pound notes. Transit automates the alignment process by comparing a number of attributes of the source and target segments, including their length, formatting information, use of symbols, bookmarks and so on to do a lot of the work for you, stopping when it reaches a threshold of certainty that you yourself set. The results are spectacular, enabling it to finish alignments in well under half if not a quarter of the time of most other programs. Unfortunately a lot of the preparatory work hasn't been automated, making the whole thing a rather more fiddly process than it need be.

This criticism also applies to a number of the other features offered by the program. Fold, for example, is Transit's equivalent to Scan in Deja Vu and Concordance in Translator's Workbench. Basically it's like folding a piece of paper to see only the parts that interest you. Its strength is that, unlike any of the other programs, you can use it in absolutely any of the document windows to compare segments sharing a common attribute in your source text, target text and any or all of the files in your Associative Network. There's no doubt in my mind that it can be a really useful feature when you have a difficult problem to solve - especially one that isn't just a matter of terminology. What's more, you can use it to revise all the segments containing a particular term or phrase one after another and even to perform global replace operations in all the reference files opened. That's very handy when you find you have to change the verbs you've decided to use with a given noun.

The irritating thing is: here is Star with a brilliant and versatile feature that you can't use by simply selecting a series of words with the left mouse button and activating the function with the right button. Instead, you have to either cut and paste the phrase using your mouse or enter it by hand in a dialogue box. There is in fact a good reason for this. Namely that the dialogue box offers you a number of choices that make this function even more powerful. Still, most people are likely to use the same settings again and again, and it would be quicker to do the whole thing with just a mouse click when that's the case. Similarly, if the Associative Network window isn't displayed, you have to click on a menu to bring it up and obtain the equivalent of Scan or Concordance.

Gripes apart, though, Fold offers yet another big advantage - if you find something in the Associative Network that's incorrect, you can change it immediately without having to go into database management. The only other package to offer this is Translator's Workbench. Having to put off tasks like these means they're much less likely to get done.

Moving on to more prosaic functions, Transit also offers an interface that can be customised to operate like a number of different popular word processors, although it rather shows it's age here: you have a choice of Word for Windows, WordPerfect and WordStar. The Word for Windows interface - which is the one I used - offers the full array of AutoText and cut and paste functions, and enabled me to proceed at a very decent rate. Combined with this is a very nifty little macro editor that you can use to record a series of keyboard (but not mouse) operations and then give them a single keystroke short cut.

You always work in overwrite mode, meaning that you only need to rearrange codes but not to copy them, and help files containing lots of screen emulation characters in Courier are automatically copied across for you. The same goes for code strings and the like. The documents aren't displayed as cells containing segments (like SDLX and Deja Vu), but as a continuous series of segments that you can choose to have each begin on a new line and are highlighted as you move down them. This means that a fairly large amount of text can be displayed at once, helping you keep your flow.

Although the tags for codes are normally write-protected, you can choose to remove this and then delete them to join segments as you prefer, or insert separator tags to split them. Only Translator's Workbench and Deja Vu offer you the same freedom of choice (although it's easier in those programs, since you only have to press enter for the former or press a toolbar button or keystroke combination for the latter).

I expect you're starting to get the picture by now. Transit is quite simply the fastest and most powerful Translation Memory program on the market, but it's not the most user-friendly. It has a logic of its own and you'll take a little time to get used to it, but as far as productivity gains go only the best programs come anything like a close second. Helping to make your first steps as easy and productive as possible, Star UK will install your system for you and provide you with a free initial training session.

A new version is expected to be out later this year and its performance should be quite terrifying, even if for no other reason than that it will finally mark Transit's move to 32-bit code (at last allowing you to use long filenames). Considering the powerful core features of this program it is very much to be hoped that the new version also tidies these things up around the edges. The hot news is that this new version is now going into beta.

The evaluation software that Transit provided came with a full set of filters, almost all of which work faultlessly without your needing to do any special preparatory work on the files (this is not always the case with Deja Vu and Translator's Workbench - particularly where Interleaf is concerned - which require files to be exported from their native program in a format that clients don't always understand).

Although I'm not really in a position to comment personally (my languages are English, French, and Italian), Transit has a formidable reputation for being able to handle non-Latin character sets better than most. I would, however, recommend anyone with any such special requirements to post queries on Lantra and FLEFO, to hear what their colleagues worldwide have to say. Also make sure you get a proper demonstration from the companies or evaluation software to try. I've heard too many stories of companies claiming their software will work perfectly well with certain languages only for the translator to find the program delivers much less than promised.

Transit, like Deja Vu, offers a three-way final quality control procedure for checking spelling, codes and terminology.

An additional "filter" is a little DOS-based program that goes by the enigmatic name of Trados.exe. You can use this to convert Translator's Workbench databases, either to port the existing databases of new clients, or to work on Word files using databases that have been sent to you by clients who use Trados. You would then send a Word file back for them to do their own alignment and import the results into their database. This naturally relies on the goodwill of your clients in the latter case, particularly if they are translation companies. I know of several translators who nevertheless use both programs to provide a complete service. Trados.exe means that you can do this but work in Transit all the time if you want to gain speed and don't require the user-friendly features of the Translator's Workbench. At least in theory, however, you can keep clients that use both programs happy while only having to buy one of them. I say in theory because Trados.exe will only work with old *.tmw files and not the latest ones.

So what's it going to cost you? Transit comes complete with Termstar and some perfectly usable demonstration dictionaries that you can use as the basis for your own, as well as a comprehensive collection of filters for many popular file formats (FrameMaker, QuarkXPress, Ventura etc.). The PageMaker filter differs from the others in that it requires you to have the program on your system. That basic package will set you back about the same amount as the other major packages (with the significant exception of Deja Vu and now Translator's Workbench), namely euro 2,173 (in the region of £1,489 + VAT). The alignment package (which integrates completely in Transit, a little like the alignment package in Deja Vu) costs almost as much again: euro 2,020 (around £1,384 + VAT). The filters for PageMaker and Interleaf are supplied as extras for euro 486 and 997 respectively. Considering the sharp fall in demand for Interleaf, I wouldn't advise anyone to shell out the money for that filter unless they already have a customer that regularly provides them with work in this format. That said, Transit's Interleaf filter is the best in the business where speed and simplicity are concerned.

The alignment package would certainly be worth buying if you have a large number of files to align for the simple reason that it can save you so much time. It's also a helpful tool to ensure you start out on the best possible terms with new clients who are able to supply you with old files. That said, it's not essential. You can happily align files with a text editor, either manually or with the help of a couple of macros. It's not just freelance translators who do this either. I've heard of more than one translation company that has created its databases manually rather than buy the rather expensive Star (or Trados) alignment tools. In an effort to make life easier on the wallet, however, Star does offers an alignment service at an hourly rate.

One final consideration regarding money matters is that although Star don't offer an ITI discount, you can get Transit at a lower price through a number of the translation companies, such as LMI, that use it and encourage their freelance suppliers to do so too. Last but not least, don't forget that if software holds no fears for you, this is the product that will have the biggest effect on your productivity.

Now that you've got your head round how Transit differs from most TM programs, I'm afraid you're going to have to go through a similar process for Translator's Workbench, which is equally unique. Its distinctive feature is that the translator can work in their usual word processor (either Microsoft Word or WordPerfect) and doesn't have to worry about getting used to some unfamiliar translation environment. What's more, Translator's Workbench gives you full WYSIWYG functionality, so that instead of being confronted by all sorts of exciting codes in brackets, you see the text displayed in different font styles and characters, in bold, underline, superscript and all the rest, just as you would when working on any Word document.

Because you have so little to learn, you can take full advantage of the power of Translator's Workbench immediately. That means you'll be making productivity gains while people using other systems are still losing on the swings what they gain on the roundabouts, taking time to familiarise themselves with a new interface.

It was this user-friendly advantage that quickly took Translator's Workbench to become the market leader, which in itself now constitutes a further benefit of the program. A large number of translation companies in particular use Translator's Workbench to increase their throughput and help ensure consistency, and will often specifically state that they prefer working with freelances that use the Trados product. A number of agencies make similar requests regarding Transit and - to a lesser extent - IBM Translation Manager and Deja Vu, but Trados is still in the lead. If you want confirmation, simply note how many of its competitors offer the possibility of handling Trados files or databases. One way or another, if you have Trados you place yourself in a strong position with a large number of leading translation companies. And of course there's no point in having a productivity tool if you aren't going to get a lot of work.

Translator's Workbench is undoubtedly the most professionally-presented product, coming in a solid branded box that contains three A4 ring binders, each a manual for the three main components of the package: Translator's Workbench itself (the translation memory program), MultiTerm (the terminology management program) and WinAlign (Trados' visual alignment tool). The manuals are comprehensive and easy to understand but not - ironically enough - always perfectly translated (a problem shared by Transit). Personally, I'd prefer a smaller format like Transit's A5 ring binder. Minor gripes apart, though, the Trados manuals set the standard to which others should aspire. Otherwise, you get a few floppy disks to install the programs and a dongle - a small component that plugs into the back of your computer and authorises operation of the software. Without the dongle the software only works in demo mode, a "feature" shared by most of the other products in this review, designed to prevent unauthorised (i.e. unpaid) use of the software. This can be a bit of a nuisance if you have other programs that also use dongles (such as older versions of the CD Encyclopaedia Britannica). I had a line of 6 plus the plug to my printer poking out of the back of my computer during this test and practically had to build an extension to my office to fit the whole lot in. Well, not quite, but you get the idea. In fact each one only occupies about an inch and shouldn't cause you terrible problems. Even with six in a row, I didn't have any problems with them operating correctly during the course of this review.

When using Translator's Workbench, you have to start its individual components separately. Before starting work you would therefore launch MultiTerm, Translator's Workbench and Word (or WordPerfect). MultiTerm sits in the background, with the Translator's Workbench and Word windows on top. The Translator's Workbench window occupies the top third of the screen and is itself split into three areas: a panel on the right showing any terminological matches found in MultiTerm and two long panels - one above the other - on the left. The panel on the top is your source window and shows your active sentence (the one you're actually working on), while the window below it shows any exact or fuzzy matches found. Alongside this window is a button that allows you to select the degree of fuzziness without having to pull down any menus etc. A nice touch. As are the cute little clickable language flags.

When Translator's Workbench installs, it copies a series of macros and templates to Word, leaving you with a Trados pull-down menu and a row of new buttons on the toolbar for opening the database, moving on to a new segment, importing a fuzzy match, copying across the source text, accepting the translation of a segment that you have completed and so on. The document itself looks the same in Word as it would do normally, with the exception that as you move on to each segment it will split into two differently-coloured paragraph panels, the first of which continues to contain the source text while the second is empty for you to enter your translation. If you get an exact match, simply clicking the appropriate button will accept the correct translation, close that segment pair so that the English translation remains above, and then open the next segment in two coloured panels as before. Alternatively, if the translation memory offers a useful fuzzy match, you can accept it (once again by clicking on a button), edit it, send it to the translation memory and move on to the next segment. The last two operations can be done simultaneously with a single button click.

One way or another, you end up clicking away through the text, increasing your translation memory as you proceed and recycling any segments that are subsequently repeated. One of the great things about Translator's Workbench is an Analyse function that gives you precise information on what this involves in advance. As soon as you receive your file you can find out how many exact matches there will be in the database, how many fuzzy matches and what degree of repetition, enabling you to calculate precisely how much work will be required and what sort of delivery/price you can offer. Although almost all the other programs offer some form of statistics (with Deja Vu coming a very close second for completeness), they can't take repetition into account until the translation has actually been completed. Only Translator's Workbench does this.

Similar buttons are provided for importing terms from MultiTerm and incorporating what Trados calls Placeables. These are non-translatable elements in the source segment such as tags, graphics, fields, numbers and dates etc. These normally stay the same in the target segment, although their position may change, making it necessary for you to decide where to place them. Positioning the cursor as required and flicking between the various buttons allows you to transfer these Placeables as required. The process is the same as the codes you have to juggle in the other programs, only they look less intimidating - partly because there are less of them with the formatting codes out of the way. What's more, Translator's Workbench can automatically convert some of them as it works, namely date formats, times, numbers (decimal places from commas to points and thousands from dots to commas), measurements from metric to imperial and even acronyms etc.

A further button starts a function called Translate to Fuzzy, which proceeds through a series of 100% Match translations until reaching the next fuzzy match. This is generally the fastest way of proceeding through translations where there are lots of exact matches.

It must be emphasised that buttons aren't the only way of working. The same options can be selected in a pull down menu or the functions can be activated using keystroke short cuts. Keyboard users will prefer the latter, but if you're dictating and using macros to control the program, set them up to activate the buttons or menu options.

The last remaining button is for the Translator's Workbench Concordance function. Select anything from a word to a phrase in a segment, click on the Concordance button and you'll be presented with user-set number of segments in which the word or phrase occurs. I've already discussed the usefulness of this for a number of the other systems reviewed, so I won't go on about it again. Suffice it to say the function is quick and easy to use, and the information is displayed very clearly. Moreover, as mentioned elsewhere, you can edit the translation memory from the Concordance window. This can also be done for individual segments in the Workbench window displayed as you work, and from the dedicated translation memory maintenance window. The latter is clear and simple to operate but suffers from not being resizable (you can't maximise it to occupy the whole screen), which is a disadvantage if some of your segments are very long (like this sentence). The only desirable feature missing from Concordance is the option that Transit (and only Transit) provides of being able to do a global replace without having to go into Database Management.

The business of how much text you can see at a time is a further area in which Translator's Workbench is at a slight disadvantage. When the program is up and running, you actually see each segment 5 times: twice in Word (source and target) and three times in the Translator's Workbench window (once in the source window and twice in the translation memory window, which shows the source and target segments one above the other). This doesn't leave much space in Word's window to display the segments that come before and after the active segment, making it more difficult for you to keep track of how the text flows. If you don't use MultiTerm, Translator's Workbench won't open the third terminology window, giving you a longer Source and Translation Memory window, which means you can resize the Translator's Workbench window to make it thinner and minimise the space it robs from Word, thereby alleviating the problem.

The advantage to all this is that someone starting out in Trados can quite happily rely on its default settings, and proceed through their translation one step at a time by clicking on the appropriate buttons, all of which bring up Word's little yellow ScreenTip labels to remind you of their functions.

This is one of the biggest single strengths of the Trados package. There are no codes or tags to confuse the translator. You work in your normal word processor. You use the same spell checker, AutoText and other productivity functions that you already know and love. In short, you really don't have to learn anything special (apart from the basic principles of translation memory). This makes Translator's Workbench the ideal solution for the computer-shy translator.

Until, that is, something goes wrong. This has nothing to do with how stable or reliable Translator's Workbench is. I, like many translators, have heard the odd rash statement that circulates on the Internet, but can happily reassure you that the latest version is perfectly stable. I threw the most excruciating combination of problems at it and it never missed a beat. Users who do have problems should consider upgrading to the latest version (the files for 2.2 can be downloaded as a single executable over the Internet), installing Service Release 1 and 2 for Microsoft Word (which can also be downloaded over the Internet) to improve memory usage, or checking their Windows installation in general.

No, the things I'm talking about go wrong for other reasons. Translator's Workbench makes extensive use of tags and hidden text to integrate in your chosen word processor and they can sometimes get damaged, although TWB takes a number of steps to protect them (one of which has the irritating side-effect of meaning you can't use backspace to delete characters in the Spell Checker). Even using speech recognition program macros to execute the keystroke shortcuts will damage the Trados tags. The package includes a number of automatic features to fix this up, but sometimes they simply can't solve the problem. If you've been working properly and saving each segment to the database (it happens automatically if your press the right button), this shouldn't be a disaster even if you don't have the technical knowledge to go into the file and fix things up yourself. Assuming you've saved a copy of your original file precisely so you can cope with such an eventuality, you should be able to re-translate everything you had already completed using the Translate to Fuzzy function. Nevertheless, if things do go seriously wrong, the fact that you've not climbed any sort of learning curve to work with or recognise the tags used, means you'll be much less likely to have the knowledge or confidence to fix them up on your own. Unfortunately, there are swings and roundabouts to everything.

Trados uses databases that make intensive use of attributes and compression. This has a number of advantages. You only activate the parts you actually need in order to keep searches as fast as possible and ensure the results are relevant for the clients and subject area concerned. You can even keep track of which translator has inserted any given segment. This is only one of a massive number of very useful project management tools this program offers its corporate users. The downside is an increased likelihood of data corruption. Now don't get me wrong, I'm not saying this will never happen with any of the other programs (with the exception of Transit). What I am definitely saying, however, is that keeping regular backups is definitely a good idea! Don't forget your databases will quite literally become one of your most valuable assets.

If you're a computer-shy translator who's careful to follow these precautions, however, you'll love Trados. You can, for instance, split your segments by simply editing them in word. One of the very few disadvantages to TWB is that because it has to cope with running three programs at the same time - and Microsoft Word's memory overload in particular, combined with the macros for revealing and hiding hidden text as it proceeds, Translator's Workbench is the slowest of all the translation memory suites tested. That doesn't necessarily mean it's the least productive, but that it takes longer to trawl through its memory database (particularly noticeable in long files and with very large databases), and longer to process its way through large series of 100% matches.

One of the past gripes about Translator's Workbench concerned Trados' horrendously laborious and overpriced DOS-based alignment program. If that's the last thing you heard about alignment tools from Trados, prick up your ears and get ready to start smiling. WinAlign, which has recently been further improved (now in version 1.03, build 48, which can be downloaded on the Internet), is a highly effective and extremely user-friendly alignment tool that's light-years away from their previous offering.

For a start, it uses a series of parameters such as the presence of tags, numbers, formatting and document structure etc. to automate the alignment process. Unlike Transit, it doesn't simply whizz through the files, stopping at points where the program considers the probability of a correct alignment to be below a certain user-set percentage. Instead, it proceeds through the files, indicating how segments have been aligned using a graphical system of links between individual segment icons (which is why Trados call it a visual alignment tool). You are then required to scroll through the document checking how the segments have been aligned. The program is generally so reliable that you just need to look for zigzag links where a segment on one side has been aligned with two or more on the other. Most of the time the program gets things right, but sometimes you have to fix the links up. After a certain point, you can get WinAlign to process the files again, incorporating the information available from the way you aligned the segments up to the point you stopped. It's an effective process that makes alignment much faster and much less of a strain. There's no fiddling about to set the process up and no special operations that need to be performed when you've finished. In addition, there are a whole series of little detail touches that make everything as simple as possible, including an Outline Area that shows all the various different files you want to align, and a display that synchronises the aligned segments as you scroll down so that they are always graphically aligned horizontally across the screen.

WinAlign does have its limitations, however. You can only align Word and Help files in RTF format or hypertext files with the extension htm or html. Other formats can be handled using the ITP filter pack and S-Tagger (see below). More seriously, if you have to align very large files, you will lose any editing (but not your alignment decisions) if you do not complete the alignment and export the database in one session. Earlier versions didn't allow you to customise the segmentation rules, but the latest WinAlign now provides the same options as Translator's Workbench. Finally, segments can be joined, but they can't be split.

All in all, WinAlign is by far the most user-friendly alignment tool of the bunch, though it lags behind Transit in sheer speed. That said, it doesn't come cheap. Trados do offer a special starter kit that bundles it with Translator's Workbench, and ITI members benefit from a special discount. You can also hire WinAlign to perform your initial alignments if you prefer. That said, the same price considerations and workarounds mentioned regarding Transit apply in this case too.

Price always used to be an issue that worked to Trados' disadvantage. That, too, has now changed where freelance translators are concerned. WinAlign and S-Tagger (which creates tagged files for Translator's Workbench from FrameMaker MIF files and ASCII files exported from Interleaf) continue to be relatively expensive extras, but - and this is the BIG news - Translator's Workbench is now available in a "Lite" Freelance version that costs significantly less than the Full version. What's the difference?

Wait for it...

You can't use it on a network.

That's right: that's all. Recognising that a freelance has exactly the same needs as a Translation company but will be working on their own, Trados offers us full functionality and simply restricts it to use on a single computer. What's more, you can now work on a large variety of different file formats without having to pay a penny for the filters. You can also download the ITP filter pack for html, Ventura, PageMaker and QuarkXPress files free of charge from the Trados Website. And as if that weren't enough, the latest release of Translator's Workbench (2.2) comes complete with Trados' new Tag Editor for tagged files and Workbench RTF files generated by S-Tagger and the ITP filter pack.

Tag Editor is a handy little tool for localising html files in particular that is still not entirely bug-free, being at the "release candidate" stage. Two problems that I noticed are that JavaScript scrollit instructions are misrecognised as scrolling messages, and that the handy Microsoft Internet Explorer button that allows you to preview html files causes Tag Editor to crash. Otherwise it's a nice little program that should bring relief to those starting out in localisation who have little knowledge of hypertext markup language.

What this all adds up to is that assuming you don't want to buy the alignment tool, you can get the full functionality of Translator's Workbench - including filters to work with most files - for £960 (+ VAT).

Our original idea was to provide a report on the productivity achieved by all the different translation memory systems on the market when used by the same experienced user for the same clients. Despite following a rigorous method of allocating work of similar length on similar subjects for similar clients to each of the different systems, this turned out to be unrealistic for a number of reasons.

Firstly, although I went to tragic lengths to import data between the databases to ensure their contents were always the same, it proved impossible to take "luck" into account in terms of how many matches a text would turn up in a database or whether it had a large amount of internal repetition.

Secondly, the speed differences between the various systems were much more marked than I had anticipated, making it quite clear which products would offer higher productivity assuming all the systems were equally easy to use.

Finally, the fact that there is such a big difference between computer-confident and computer-shy translators, paralleled by a similarly large difference in the level of user-friendliness of the products, it became clear that not every user would be able to obtain the maximum potential of every system.

As a result, we've described all the systems in sufficient detail for you to make your own informed choice, while nevertheless providing some summarising guidelines below.

I'm afraid there is no answer to the question of which single product you should buy, since it depends on your individual circumstances. If you're a software localiser and do 300,000 words a year for one of your clients and they suddenly turn round and say they want you to use one of these products and that one only, you'd be ill-advised to tell them to take a running jump. The long and the short of it is that once you've got used to any of these products they'll increase your productivity and consistency while keeping that big customer happy. And that's what big customers like. You did it their way. And more, much more than that, you probably made yourself more indispensable into the bargain. (Apologies to the late Frank Sinatra).

Let's suppose you have a free hand, however. You can take the lead and offer them the benefits, while also reaping the rewards yourself. Which package should you go for?

I'm sorry to be so awkward, but again it depends. Did it take you years to master your word processor? Do you have terrible trouble finding files on your computer? Do you have a crisis every time you have to install a program?

If you answered yes to all these questions, cross IBM Translation Manager and possibly Transit off your list. If you never work for SDL, cross SDLX off too. Save your money and relax with Trados. Or if you want a full range of filters, an alignment tool and more speed at the expense of some user friendliness for around the same price, download the Deja Vu demo and see how you get on with it.

On the other hand, if you're a software localiser who's used a number of these tools for several years and are perfectly at home messing around with tags in all their glorious forms, you should seriously consider Transit. There's nothing to touch it for sheer power, speed and guaranteed data integrity. Alternatively, if you like to see a little more fat on your wallet, go for Deja Vu. It comes a very close second.

So what do you need in the way of hardware if you're getting involved in Translation memory tools? The answer is nothing particularly glamorous in this day and age. The computer I used to test all five systems (installed at the same time, each with their own databases) featured an AMD K6 233 processor, 128 MB RAM and a relatively small 1.2 MB hard disk. What's more, I also had Dragon NaturallySpeaking Professional installed and could happily run it to enter text and control all the various programs with no speed penalty. The only exception was Word (even with its various service releases) with Translator's Workbench. The results using NaturalWord were too slow to be acceptable during translations of long files, although the fault here probably lies more with Word than with Translator's Workbench. The solution was quite simply to use NaturallySpeaking's NaturalText instead, which is more than sufficient to dictate a segment of compound words at a time.

The one big recommendation (which really applies whether you use Translation memory software or not) is to go for the biggest screen you can afford. Don't forget that you're going to have several windows all displayed at the same time. A 17-inch monitor is really the minimum and a 19-inch model would be even better, especially considering the shrinking price differential. But if you really want to treat your eyes to something approaching a decent life, splash out on one of the new LCD screens. The dent that makes in your credit card will undoubtedly prove a wonderful incentive to increase your productivity in leaps and bounds. And after all, isn't that what a flexible friend's for?

One thing's certain in the world of productivity tools - if you get all your work on paper, translation memory won't make a world of difference. Once you start using translation memory tools - particularly if you translate a lot of technical or other procedures that tend to have a repetitive aspect to them - you'll start looking for more customers that provide you with work on file. Some translators even offer a discount on suitable files or apply a surcharge to documents sent by post or fax. The client benefits from a lower price, faster turnaround/time to market and the likelihood of improved terminological consistency (although there is the possibility of the reverse - indeed one of the manuals of a translation memory company that shall remain nameless includes an incorrect translation that occurs with impressive consistency throughout).

At the same time, there is an increasing practice - increasingly frowned upon by translators - for translation companies to pay translation memory users a varying percentage of their per-word/line rate depending on the percentage match of each segment. A variation on this theme is to send out a file of disconnected segments from which all the 100% matches have been removed. Increasing numbers of translators can't be bothered with the latter, just as many are prepared to give a discount for 100% matches but not on anything below that, claiming that the editing required can actually lead to a fall in one's productivity compared to a "straight translation". The consensus of opinion is that these programs require a financial investment to buy them and follow the associated learning curve, and that while they also represent productivity aids, the real payback for the customer is in an additional quality guarantee and faster service.

A further area of contention is in the matter of copyright when a translation company builds up translation memories from the work of its freelance translators. A number of translators claim to receive increasing numbers of translations in which the 100% matches have been weeded out or, alternatively, a steady fall in demand from companies using translation memory products. The finger of blame is often pointed at the use such companies make of the translation memories they have built up with their freelance translators' help.

But we mustn't forget that even if we assume translation memory products are nothing more than productivity tools, they are precisely that - tools. They aren't the only tools in a translator's tool chest and aren't always the most appropriate, but they shouldn't be undervalued because of that (in the same way you wouldn't look down on a screwdriver because own a claw hammer). How they are used is a matter of who uses them and doesn't reflect any inherent quality of the tool itself. The legal implications concerning copyright are, however, far from clear - though opinions abound. This side of the question will be considered in detail during the ITI 1999 Conference.

The actual productivity increases offered by these tools is another area where people are quite happy to venture ridiculous comments. The answer really depends on far too many variables, including subject area, the use of the documentation, the writing style of the technical or other author concerned and so on. Productivity levels approaching the miraculous are possible, and I have on several sadly infrequent occasions translated thousands of words in a matter of minutes. My own average figures, however, which are confirmed by a number of translation memory companies and the ITI Rates & Salaries Survey, are around the 30% mark.

Apply that to the annual volume of work you receive on file that's not topical or event-related, and you'll get an idea of the change TM can make. At which point you can decide to spend less time in front of a screen and more with your family watching TV, or boost your income to whisk the lot of you out of the house and off to a country where it isn't raining. Bon voyage!

Most of the better speech recognition packages available on the market today can be integrated with translation memory tools to maximise your productivity gains while reducing work stress to a minimum. What's more, if your translation memory package has macro functionality (see ITI Bulletin comparative review of speech recognition products), you can control practically everything by voice - meaning you can sit back while you work, forget the keyboard or fumbling through dictionaries and simply get on with the job in hand. Not only will you produce more work during the day, but you'll also feel better at the end of it. I was able to control all five programs tested programs using Dragon NaturallySpeaking Professional. NaturallySpeaking has no problems dictating directly into any of the programs using Natural Text. ViaVoice, on the other hand, generally works better if you transfer the text from its own dedicated speech-to-text word processor. The macros I created in ViaVoice also worked faultlessly.

One word of warning, though, for Trados users: set up your speech recognition program to control Translator's Workbench using the toolbar buttons or menu options and not the keystroke short cuts, since this can mess up your tags and even Fix Document won't be able to help you out.

Apparatus layouts

Assembly procedures

Balance sheets

Birdwatching

Building specifications

Case histories

Central heating systems

Concert programmes

Conference programmes

Contracts

Courses of study

Disassembly procedures

Sectional furniture systems

Hotel descriptions

Installation procedures

Interrelated patents

Interview procedures

Medical procedures

Modular systems/products

Operator manuals

Parts lists

Product/service updates and revisions

Recipes

Technical product descriptions

Tenders

Travel directions

Trademark descriptions

and more.....

Get the message? We probably do a lot more repetitive work than we realise.

Our aim has been to make this review as comprehensive as possible without going into so much detail that the ITI Bulletin turns into a paperback book. To help bring you all the information you need, we're extending this comparative review to build in your feedback on the Translation memory tools you use.

Provide it! Please....

If nothing else, it will give you an opportunity to get your own back on your company of choice - or say thank you. You can either e-mail me directly (michaelb@pavilion.co.uk), including the words "translation memory" in the subject line so that I can filter all messages to a dedicated folder, or come along to the ITI 1999 Conference and join the seminar and discussion we'll be having on the subject.

I'll be taking up all reports with the companies concerned and attempting to replicate any bugs on the systems they've provided for this comparative review. I'll also continue to monitor how they perform over the long term. Then I'll report back to you with a closing article in the Bulletin.

http://www.atril.com

http://www.qsoft.de/ibmtrans.htm

http://www.star-ag.ch

http://www.trados.com

http://www.sdlintl.com