MT promises

 

Review of Systran Version 2.0 Professional

 

Let's get paranoid

 

After all that's what normally happens when translators discuss machine translation, doesn't it? We shuffle around from one foot to the other smiling knowingly. Thoughts of how misunderstood we are and future unemployment run through our heads. We regale each other with stories - some real and some apocryphal - of the many translation howlers unleashed upon the world by machine translation systems of one kind or another. They'll never replace us, we smile gleefully before changing the subject. And of course, we're right. The only thing is we have, somehow, missed the point.

 

In fact, we're so expert in feeling threatened by new technology that missing the point can become a habit. Translation Memory is another example. As I point out in the second part of our extended review, also published in this issue, Computer Aided Translation tools are more than a means to simply increasing our productivity and maybe consistency on certain types of projects. They can also make it possible for individual freelances to acquire large direct clients whom we would otherwise have been unable to serve.

 

Let us, for example, take the case of a specialist software house that only requires translations into English. They can write their manuals while developing their products and we can translate them at the same time, working on the help files and interface in the same way. We can be faster at this because these manuals are generally sufficiently repetitive for us to gain in productivity using translation memory. More important, every time they revise an aspect of the product we can very quickly process the new revised files. We are thus able to help them meet the fast time to market requirements of the highly-competitive software industry. Once the mature product is on the market and undergoing continual development, we can, once again, help them ensure the localisation process does not delay each new release. If we did not have translation memory we would probably not be able to service their needs on our own and the job would go to a translation company.

 

 

Developing your service

 

So, it's not just a matter of the tools you use but how you use them and market them. We see a need, we find a way to service it and we let people know. That's not just marketing, that's developing our service so that we actually have something to market. All very well, you say, but where does machine translation fit in with all this?

 

Well, what other needs do to our clients have? Think hard. Think four o'clock on a Friday afternoon. Yes! They want to keep up-to-date in their sector, they want the information fast, generally at the last minute and they may never even find it relevant never mind do anything with it, so the translation doesn't need - at least initially - to be of the highest quality. What's important is that the terms used are correct and that the meaning is clear enough for them to get to the drift ("gisting" as its known in geekspeak).

 

That is precisely where we come in. Anyone can buy a machine translation program, but it takes a translator to make one work. That is what we can sell to our clients: quick for-information-only translations generated and checked by a professional as part of a complete service. We charge less for them because we spend less time doing them and can always offer a small discount on a "proper" translation of the same text should our client subsequently require that. We thus not only offer a complete service but also add an extra pinch of "feel-good factor" into the bargain. So what do you actually have to do to provide this? Read on and find out.

 

 

What it takes

 

The easiest way to understand how to get the most out of a machine translation program is not to see it as a single translator but as a group of them, like a mini translation agency. A job comes in and you think: who am I going to give it to? You need to think not only about the subject matter but the type of text. Maybe John and George both have an intimate understanding of particle physics, but this is a press release and whereas John makes everything read like a shopping list, George could take a shopping list and make it sound like an aphrodisiac recipe. We obviously want George on this job!

 

Likewise, if you're after a machine translation package, you want one that allows it to specify a large number of different subject matters and types of texts. In Systran Professional you select the former by choosing what's referred to as a Topical Glossary. What's more, because a very wide (specialist) range is available, you can actually choose up to four different glossaries and give them a hierarchy. This simply means putting them in order to let Systran know that if you have, for instance, selected both "Automotive" and "Mechanical Engineering" for an automotive translation, it had better give preference to those terms in the former rather than the latter.

 

Similarly, the Options Menu also offers you a choice of document types ranging from "Abstract" through to "Colloquial", "Parts Lists" and "Patents", for example. But that's not all. Let’s think back for a moment to John and George in our little translation agency. It will occur to us that both these fine upstanding young members of the translation community would benefit from specific client glossaries to help them, since even the noblest customers with the most regular payment records can have their little eccentricities. Systran has this one covered too, with what it refers to as CSDs or - wait for it - Customer Specific Dictionaries. These differ a little from what we'd actually be giving good old John and George (not to mention Paul and Ringo) because they're not entirely customer-specific. The whole idea of machine translation is obviously to automate things as much as possible, and yet, particularly in the beginning, Systran will come across words that aren't in its glossaries. You can obviously enter these manually every time you have to do any work for that customer, but it's naturally more effective to automate the process. Systran allows you to do this by generating what is ingeniously referred to as a "Not Found Words list", which you of course turn into a found words list, alias CSD.

 

So, the first time you use Systran you select one or more topical glossaries and a text type. You won't be able to select a CSD because you first have to create one, which you do from your Not Found Words list. Once these CSDs have been created you can edit them and create new ones based on them using a nifty little utility known, surprisingly enough, as WinCSD (which can also be accessed from the main Systran window). This makes it easy to cater for new clients operating in subject areas for which you already have an existing CSD. Another way in which CSDs can be used is to enter compound nouns or phrases, referred to as "expressions", for which you can specify the principal word in the source and target languages, as well as part of speech and gender. As you would expect, you can also specify words that are not to be translated, such as product names.

 

You then feed your text through a dual window arrangement with your source text in one window and the new target text in the other. A wide variety of file formats can be handled, including IBM Translation Manager and Trados Translator's Workbench files. Déjà Vu's new SQL capabilities mean that it can now also be used with Systran, although the process of exporting and reimporting these no or low-match segments is a little more complicated. Conversely, the Trados Analyse facility makes it particularly easy to use with machine translation. Systran can be used from within Word and integrates with it very nicely, but also offers a number of text editing features from within its own dedicated environment for those who don't have or prefer not to use Word.

 

 

What it gives you

 

Well, there's no doubt that it does occasionally give you a laugh. But that should generally be the beginning of things, not the end. As the smile fades, you ought to set to tweaking your CSD to try and prevent this from happening again in the future. I found the best policy in the long run was to work iteratively on my file/s, working on all the "not found words", mistranslated words and word combinations that could valuably be entered in the CSD as expressions. This not only gave me good results for my current project but also meant I could rely on achieving a similar level next time round. In most cases doing this didn’t seem to take a great deal longer than it would have done using search and replace. I would then, perhaps, do some minimal editing to rearrange a few sentences, generally dictating them in to save time. And that was that, the next step simply being to e-mail the files back to my clients.

 

So, in conclusion, Systran Professional is a complete and stable package offering good compatibility. After a short "learning curve" period for each customer, it allows you to generate understandable files for "gisting" very quickly, thereby adding an additional customer service to your arsenal. A machine translation product without all the features listed above would not be able to achieve the same quality standard.

 

I can report that most of my customers have been pleased with it, partly because they always have the option of contacting me if a particular section is confusing for one reason or another. They have recently also come to use it more frequently. There's certainly no way I can claim that it has become a major source of income and it would take just over a year to amortise the investment (around £500 with one language pair) if demand continues at its current rate - except of course that it may continue to increase.

 

Although Systran has enabled me to satisfy the occasional last-minute demands of existing clients, this hasn't yet translated into new clients contacting me specifically because they've heard I can offer this service. But then this could quite simply be another one of those cases where you never know how long it takes for word to get round. One of my biggest advertising clients, for example, came from some marketing work I had done three years before they first contacted me.

 

Another area that I have to explore more fully concerns the impact that integrating Translation Memory and Systran could have on my productivity using the former. The speed of developments in shallow translation memory technology makes this particularly uncertain.

 

Naturally, I'll keep you posted on all three counts, but there's no need for you to wait, you can easily give the system a try yourselves. Contact Omega First, who very kindly supplied ITI Bulletin with its evaluation copy (and are also an excellent UK source of foreign-language Microsoft operating systems, Office components and proofing tools). They can supply you with a free evaluation CD that will give you 15 runs in each of the language pair combinations offered (including Chinese, English, French, German, Italian, Japanese, Portuguese, Russian and Spanish). It's not really enough to go for some full trial runs with your clients, but it will probably give you a clear enough idea of the usability of the system. The rest really depends on your client base and business plans. Systran on its own is unlikely to make a radical difference to your business as a whole, but it could prove very useful in tipping the balance of things in your favour.

 

[Panel]

 

For further information contact:

 

Omega First Limited

Kingston House

Portsmouth Road

Thames Ditton

Surrey KT6 5QG

 

Tel: 0181 410 5050

Fax: 0181 410 5215

 

www.omeganet.co.uk

First published in ITI Bulletin, 1999.