Translation Issues 4-99 (December 1999) by Ansgar Knauf

 

Introduction

Product life spans and documentation production times are becoming increasingly short and the expenditures for documentation are rising simultaneously with increasing product complexity. Hence, translation projects are becoming more costly as the parallel increasing documentation complexity.

The volume of translation in EU countries will more than quadruple from 1995 until the turn of the century from 500 million to 2 billion Euro. (Germany contributes to approximately one quarter of this sum). The requirements of harmonizing Europe-wide guidelines are only a single, though large factor that contributes to this growth. Increasing world-wide expansions and global presence, even in smaller organizations, not to mention the Internet, contribute as well.

But the requirements on the specialized, linguistic and technical data processing competence of translators become ever more demanding with the increasing technicality of documentation production. Capacity, availability, hardware and software, compatibility, organization, liability, risk of failure and the cost of human translators have led to the call for computer based-alternatives. When one considers these diverse variables, all of which have to be kept in mind when searching for competent translators, one can easily see why there were such high hopes and such incredibly high sums of money invested in the development of translation memory systems.

 

Development of EDP at the Translator Workstation

Since the first PC came on the market at the beginning of the 80s - and possibly earlier - text processing has belonged to the standard tool of the translator. The path from there to fully automatic machine translation was and still is a long way.

Though machine translation (MT) is as old as Electronic Data Processing, MT systems have been developed since the early 50s, but up to now they have not been really successful. Early projects, such as SUSY, a research project of the University of Kaiserslautern, or TITUS, an MT system of the German Text Association for literature analysis, were discontinued early, partially due to lack of funds and partially due to their inefficiency. But even in such systems as the now discontinued "Metal" (Siemens), whose successor is T1, or at Logos, a basic problem surfaces again and again. No matter what the language is, human expression is so complicated that no software has been developed that can sufficiently understand the human written word. That is, software that can adequately analyze, structurally and semantically, so that a grammatically, syntactically, semantically and stylistically equivalent expression can be formed in the target language.

Even in the department stores and in the Internet today programs are available that promise "fully automated immediate translation" at cut rate prices. Yet it is still a long way away until we will be able to really use them practically. No more than a "rough idea" of the approximate meaning of a text is possible by using them. Whoever would like to see for themselves can test the online translation of Babelfish at http://babelfish.altavista.digital.com or www.babelfish.com.

The first practical use for electronic DP at the translators' workstation, which was specially targeted at translators, was seen in computerized glossaries and terminology management. Since about 1985 corresponding databases in electronic form were common, yet were without a close connection to the text processing and so of only limited use. Only the use of full text databases (translation memories, also called TM systems), which could store entire sentences or text segments, brought a decisive breakthrough in this decade.

Therefore, the basic principle of TM systems is simple: the prerequisite is that the source text is not delivered on paper anymore, but as a file so that the translator can process it. This is unfortunately something that only became reality in the last few years, and even then is still not always the case. Correspondingly, TM compatible formats are also required.

Text information is completely separated from file formatting code by using a special filter, and is put back together after the translation is completed (this would otherwise also be necessary in a pure machine translation). Whoever has looked at the source code of an HTML file can imagine how difficult this undertaking is in considerably more complex data formats such as Word for Windows, Pagemaker or Framemaker. In extreme cases, the necessary post-formatting can be just as expensive as the actual translation.

The translator then produces a copy of the source text that has been exported from the original file and imports both files into a MT translation program so that the copy can be written to directly, while the write-protected source text can be viewed in a second display window. The software aligns the original and translation sentence by sentence and stores this as translation memory. As soon as an identical sentence appears in a text, the program suggests the translation found in the translation memory. The translator only has to confirm acceptance.

This system surely is not suitable for translating literature, because it is characterized by linguistic and terminological diversity and creativity. However, In technical and scientific translation possible ambiguity should be eliminated. The advantages of TM systems can be seen when in the same text or the same words, specialized terms or sentences appear in successive translations, and perhaps the only differences are numbers or measurements.

An example from the field: a machine manufacturer fits pumps for hydraulic machines in large quantities, but each is individually tailored to his clients in each of the EU countries. The use of these devices may be for the most part identical, but the measurements and unique details are different. During translation of the product documentation, a pre-translation can be produced from the translation memory, which can be provided to the human translator with the source text. Even if this pre-translation is produced to a large extent with the help of computers, this is not true machine translation. Moreover, one can speak of recycling the translation already performed by the human translator (transcycling). The translator makes the necessary changes and can complete the project in a much shorter amount of time.

But TM systems can do much more. In association with electronic terminology databases, they optimize user friendliness by immediately suggesting the defined terminology from the client glossary to the translator, and if necessary they offer alternatives or point out different meanings without the translator having to switch between programs.

Presently there are a number of practical solutions offered by different companies, such as "Translator's Workbench" by Trados and "Transit" by Star. These are the most widely used, together with the corresponding terminology management solutions such as "MultiTerm" (Trados) and "Termstar" (Star) and a series of filters and plug-ins, which make importing the most essential electronic formats possible. Other systems such as Déjá vu (Atril), TM2 (IBM) or XL8 (Globalware) are also common, but due to Microsoft's share in Trados, Trados products currently have a decisive competitive advantage with regard to marketing and development.

All of these solutions, which favor a combination of computerized and human translation, come quite close to true machine translation in their degree of automation, but they are not true machine translation and are, therefore, the best of what is possible without loss of quality. In any case, "as much EDP as possible, with the least amount of necessary human labor" is more cost efficient and linguistically practical than a careless, full automation.

 

Profitability of TM Systems

Aside from increased efficiency, the largest advantage of TM systems is their profitability. Since the translator can work faster, lower costs result. Many translation companies have partially or entirely passed these savings on to their clients. On average, these transcycled translations cost half the price of a new translation, depending on the layout program and the structure of the layout.

Despite all of the savings, the hidden costs of these translation systems must not be forgotten, namely higher costs for hardware, software, maintenance and, of course, staff training. There also are costs involved in building the glossaries. At the beginning of a project using a TM system, these costs can be considerably more than the savings, but after the consistent use of TM systems follow-up projects almost always provide an attractive return on investment.

 

Prerequisites for the use of TM Systems

The technology discussed here offers many opportunities, but also requires a great deal from everyone working on the project. The translator must be adequately trained, the client's original must be in a TM supported format, and the project manager of the translation company must be familiar with a multitude of areas in order to anticipate possible problems and to keep the team focused on the project goal. In this regard, the requirements put on translation companies in recent years has risen markedly.

Those translation companies, which have a great amount of competence and experience using TM systems, require the active support of the client to use this solution effectively. The "GIGO principle" (garbage in, garbage out) applies here as in all other areas of the industry. The savvy client integrates graphics and glossaries into his language management concept for the coming translations as early as during production of the source text.

A professional formatting software that is supported by the translation memory software is a part of a translation-friendly source text. In addition, the expansion factor of individual languages must also be considered. That is, there must be enough space in the formatting for longer texts in the target language (for example, a French text is on average 30-40% longer than its English source). Unfortunately many clients forget this detail, which later leads to increased costs for postformatting or printing.

Finally, great expectations are also put on the editor. The better the editing job, the greater the usability and savings in the TM-assisted translation. Editing for translation means succinct text. It contains short and simply structured sentences, many repetitions, and the use of standardized terminology, tables and graphics (with text that can be edited!), and a universal data compatibility between all of these elements. By the way, all of these measures also lead to improved source text quality at a lower cost for translation. Some translation companies now even offer training for technical editors in this area.

 

Conclusions

There are three distinct methods of translation: true machine translation, purely human translation, and TM-assisted translation. The first is, for the most part, impractical for the near future, since it is unreliable and due to costly re-editing remains expensive. In any case, automatic "human translation" with the help of translation memory systems and terminology databases are more cost effective than a purely human translation, provided the source texts and formats are conducive to translation. In the coming years, the market will continue to belong to translation tools rather than to pure machine translation, though a further integration of both is going to happen. The translation tools will contribute to faster, better and generally more rational translations, but will not replace the translator.

"transline Deutschland - Übersetzungsdienst für technische Übersetzung"