Mendez Rolls its Own

NLP software has the reputation of being tough to market. But why are a surprising number of companies developing their own linguistic tools, particularly in the translation world? Is the demand not satisfied with the supply? Herewith the profile of one do-it-yourself toolbuilder.

Plenty of translation companies are beginning to exploit translation technology but few take the adventurous step of developing their own software. Mendez Translations is an exception. This Brussels-based company commissioned the development of its own terminology management and translation memory package to support its translation and localization activities. Called the Mendez Translation Memory (MZTM), this DOS-based system has been under phased development since the beginning of 1992 and is now routinely used within this high-tech translation company.

While the first version of MZTM required users to work in heavily tagged text files, version two of the system is configured to work directly with Microsoft Word and WordPerfect. Mendez works with both in-house and freelance translators; and Word and Word Perfect together are used by more than ninety percent of European translators. This was an obstacle to using IBM’s package TM/2; it runs under os/2, and the company cannot expect freelancers to have state of the art of equipment. “We’re happy if they have 386s,” says Joe Mendez, the company’s decidedly un-technophobic co-director, who is as easy to reach by Internet these days as he is by telephone.
    Similar to the IBM and Trados packages, MZTM creates a database of sentences against which it compares new texts. In contrast to the those packages, however, it does not attempt any kind of fuzzy matching. Either sentences match or they do not. For production services manager Rudy Tirry, Mendez’s MZTM specialist, fuzzy matching is not that important, partly because the kinds of materials that Mendez handles are highly repetitive. The current system makes substantial gains as is. He adds, however, that in a subsequent version, he would like to see what he calls “lazy matching.” This would treat numbers, proper names, and terminology as variables within sentences. Exact matching does have its advantages; Mendez also uses the system for quality control — to check consistency.
    MZTM boasts an elaborate scheme for preserving the formatting of word processor files. It attempts to detect which codes are in a fixed position (ie, indexing and TOC codes) and which codes can shift within a sentence (typographical codes). The latter are stored alongside the sentence. The system then converts formatting to a tagging scheme based on SGML before running the text through the matching process.
Inevitably, tags get modified or deleted accidently but MZTM has a utility for verifying tags.
    A unique feature of MZTM is its use of Word’s reference fields. These substitute repeated terms with a dynamic pointer to the first occurrence of the term. This is good psychology; translators tend to forget a few pages later how they translated a previously occurring term. This does the job for them. Thirry says he is looking for ways of doing the same in Framemaker, an increasingly popular package in the documentation world on account of its cross-platform file format compatibility (especially under Unix).
     MZTM was written in ADAPT, a data conversion and string management language developed by the Canadian-Belgian firm Destin (Destin also developed MZTM for Mendez). With ADAPT, a new file format filter can be written in two weeks, which Tirry finds most acceptable. This means that Mendez is by no means limited to the file formats MZTM currently supports, a not insignificant advantage. MZTM represents an investment of BFR five million (US$ 150,000) on the part of Mendez, the bulk of which went into developing the conversion routines.

HHaving evaluated the offerings available, Mendez chose to develop its own translation memory system because it could not find what it needed. However, the company is not excessively attached to MZTM. Rudy Tirry says that Eurolang sounds very promising, particularly the rumored integration with Word version 6.0; if it should prove to be more effective than MZTM, he says he would have no compunction about advising the company to step over to it.
    Using MZTM in day,to,day operation, the Mendez translation coordinators create a document profile for each new text. In addition to various pieces of house, keeping information, the profile also indicates which existing databases may be used. As Tirry explains, projects frequently have different terminological requirements and new texts are checked only against certain other projects. Moreover, terminology remains the property of the customer. And there is the matter of confidentiality as well. “If, for example, we are localizing Compaq DOS,” says Joe Mendez, “we can’t simply include our Microsoft databases. That’s not allowed. It’s up to the customer to make arrangements to share resources.” After a text has been run through the system, the parts that need to be translated are sent to the translator with previously translated segments. Translators do not come into contact with the system. After translation, the layout is restored.

FFor the kinds of materials Mendez handles — mostly but not exclusively computer,related — the repetition rate can be surprisingly high, not only within a given document but also between new and previous versions and across large projects. Thirry has measured repetition rates of more than twenty percent among sentences within documents, what he says is “except’ ionally high,” and he has even measured rates as high as a whopping seventy’percent between versions. In view of these figures, the company, not surprisingly, sees great potential in automatically aligning existing translations, and hopes to work together with both the Swiss research institute ISSCO and TRANSEARCH in Canada on alignment and other issues.
    In economic terms, who reaps the benefits of such a repetition processing system? The customer or Mendez? Joe Mendez says the customer does; but the customer is billed for the time Mendez personnel spend operating the system. Clearly, however, both parties stand to gain: Mendez can offer competitive prices and the customer reaps consistency — something that is always hard to put a price tag on.
     Mendez Translations is run by the brother and sister team of Florita and Joe Mendez, whose father founded the company in 1970. While its main operations are centered in Brussels, Mendez also has offices in Cologne, Paris, Madrid, Milan, and, more recently, in St. Petersburg. Together, the company employs a total of slightly under a hundred people. Traditionally, it has been difficult for companies to achieve economies of scale in the translation business. For Joe Mendez, the main added value his company can offer is project management. While translation is done locally at Mendez’s local offices, Brussels coordinates the projects, many of which involve more than one target language. For a substantial localization job, this can range from copy editing and layout to quality assur’ ance, testing software, and fixing bugs — things which Joe Mendez says companies are eager to outsource.

TThe translation industry is gradually maturing, partly because needs are changing. Localized products are competitive advantage, and when a company such as Microsoft releases a product like Windows NT, it wants it available around the world as soon as possible for obvious strategic reasons. “Our clients have big translation appetites these days,” says Mendez. “Four, thousand pages into four languages at the rate of four, hundred pages a week is not so uncommon. That means you have to have somebody just managing files full time.” For this kind of volume, software publishers need highly reliable partners; and this means that the rampant price wars of earlier years may be on the wane. Joe Mendez says his company has ten steady major customers for software localization. They are compatible too. That means, for example, “Microsoft bu t not Borland.” The translation industry might be stabilizing; Microsoft, for example, has settled on two strategic vendors for each major language. (Mendez is a Microsoft strategic vendor for French and German.) This relative stability means that a company like Mendez can make the medium,term investments in technology that it needs. An important part of the equation remains translators, both internal and extern, al. How does Joe Mendez keep his translators happy?” ”We pay them well and on time [a rarity in the translation world], and we give them interesting projects to work on, like Windows NT.” In St Petersburg, adds Joe Mendez, “we’ve hired ten “exceptionally well qualified people — unemployed professors and the like.” Is there much demand for Russian localization? “Everyone is talking about doing Russian,” comments Mendez, “but few actually do it.
Microsoft is an exception.” The ten are currently working on a project for the giant of Redmond.

Financially sound, technologically advanced — Mendez may be the most successful translation company in business today. Mike Anobile, director of the Localization Industry Standards Association, knows many of the players in the business, and he speaks highly of Mendez. “It’s a top,drawer company,” enthuses Anobile. “Mendez has some of the best customers in the IT industry — and it knows how to keep them happy.” He adds that Mendez is one of the very few translation companies he knows that steadily books a twelve to fifteen percent profit margin.

Mendez Translations, Avenue Franklin Roosevelt 8, B-1050 Brussels, Belgium; Tel: +32 2 647 2700, Fax: +32 2 647 5550