One Company's Efforts to Improve Translation and Localization.
Medtronic is the world's leading medical technology company specializing in implantable and interventional therapies. The company does business in more than 120 countries. Translation is a critical issue, particularly in Europe, where regulatory requirements and local laws mandate that product documentation be available in local languages. Currently, the company translates documentation into more than 10 languages:
* French, Italian, German, and Spanish (user interfaces and documentation)
* Swedish, Dutch, Danish, Greek, and Portuguese (documentation only)
* Japanese (handled out of Japan)
* Chinese (some products)
Most of the documentation is in hardcopy, although we supply Adobe Acrobat[R] PDF files on our intranet. We also do some online help, and we are preparing for electronic distribution. The paper manuals range from about 400 words to over 200,000 words. The help files include approximately 30,000 words. The number and complexity of products is growing, as is the number of languages needed. As new languages are required, legacy documents for more than 2,000 active products must also be translated.
A centralized technical communication group in the U.S. writes the product manuals in English; a Medtronic department based in the Netherlands coordinates translation for Europe and Latin America. These translation coordinators also do some translations themselves, especially for late-breaking changes. (Another Medtronic department in the U.S. coordinates translations of marketing, educational, and corporate material, but this article focuses mainly on the manuals.) Three translation agencies, one in the U.S. and two in Europe, do the majority of the translations. We are just beginning to handle Asian languages. In all cases, the translation process includes a review by sales and support staff in countries where the target languages are spoken.
We author our user interfaces and source documents in English. However, products are released outside the U.S. months or years before they are released in the U.S., so translations are needed at the same time as or before English. Furthermore, to minimize total cycle time, translation begins before the English source is stable, especially for complex products.
Over the past few years, we have done several things to improve the process and quality of translations.
1. LISA[TM] (Localization Industry Standards Association)
Realizing that we needed to improve our translation process and that we might benefit from benchmarking, we joined LISA in 1995 to learn what other companies and translation vendors were doing to improve quality, save time, and reduce costs. We send representatives to each of the U.S. and European meetings, which take place three or four times a year. Although LISA originated in the information technology industry, the conference sessions are applicable to other industries. The conferences provide contacts and opportunities for discussions with other companies. The frequency of meetings has ensured steady progress on discussions and topics, particularly in this tumultuous field.
The LISA meetings gave us:
* Confirmation that some of our problems are not unique
* Knowledge about tools such as translation memory
* Contacts with vendors and publishers
* Quality metrics
* Grounding in "best practices"
The conferences also have afforded managers from the technical writing and translation groups extended time together, with opportunities to explore ideas about improvements. We believe LISA membership to be very valuable. Organizations without the resources to join LISA might benefit from participation in STC's International Technical Communication Special Interest Group.
2. Common publishing platform
In 1994 Medtronic technical writers were using six different word processing programs and publishing tools. The translation group was using a completely different publishing tool. This lack of standardization meant completely redoing desktop publishing after translation, as well as problems with training and teamwork. Choosing one standard publishing platform was an obvious first step in improving the translation process. It eliminated a lot of rework, especially in desktop publishing.
We chose Adobe FrameMaker[R] because of its robustness, its strong long-document features, its strong ability to move files across different hardware platforms, and its conditional text feature. We knew that FrameMaker was available in European languages; we neglected to find out how widespread its use was in Europe. (Initially we had difficulty finding support for FrameMaker and staff with experience using it in Europe; this problem is gradually disappearing.)
A small team consisting of a writer and an electronic publishing professional designed the FrameMaker templates. One important goal was that they should facilitate the translation process. For example, we used FrameMaker's Variables feature for often-repeated text such as chapter, so that it would only have to be translated once. The designers were in the U.S.; their counterparts in the Netherlands reviewed the templates.
3. Page-true translations
Initially we designed templates to maintain "page-true" translation. In page-true translations, each translated page contains the same information as its English source. We designed different templates for the U.S. and for translation, but with the same paragraph tag names. The translators could apply the translation template relatively easily. To accommodate the fact that text in most European languages expands 25%-30% from English, we used slightly larger type size in English, along with extra line leading and white space.
Below are the advantages we expected from page-true translation:
* It would make telephone support easier if the support person could tell a caller to look at a certain page number in the manual, no matter what language each is viewing.
* It would facilitate communications about changes during the translation process.
* It would give translators the option of translating the table of contents, instead of regenerating it.
* It would give translators the option of translating and resorting the index, instead of translating all individual index markers and regenerating it.
We discovered that, despite our planning, we were not completely successful in maintaining page-true translations. The translators found it better practice to regenerate tables of contents and to translate all index markers. Otherwise, one risked having the table of contents and indexes not precisely match the body. Also, having to apply a translation template is an extra step. Therefore, we now are developing templates to be used for all languages.
The wisdom of page-true translations aside, standardizing the publishing platform has significantly improved the efficiency of the translation process.
4. Business process improvement team
In 1995 Medtronic's European management ranked the translation process as one of five processes most critical to business success and most in need of improvement. Therefore, they formed a multi-country, cross-department team to do "business process improvement," using the methodology from Quality Network, Inc. The team did a fishbone analysis to identify causes of problems and ranked the problems. Figure 1 shows the fishbone analysis. Figure 2 shows the team's ranking of the problems according to three criteria: quality, time, and cost. Note that one of the important dimensions of cost for proofreading is potential lost revenue.
The task force recommended the following actions:
* Generate awareness about the translation process and issues among the company's businesses and distribution organizations.
* Hire and train full-time in-country reviewers, at least one per language.
* Evaluate the existing processes for translation and reduce the number of processes.
* Develop a standard comprehensive glossary, standard process measurements, and standards for the creation of materials.
The recommendation about hiring in-country reviewers was not funded. Despite that disappointment, I believe the task force was useful in providing widespread understanding and consensus about the causes of problems, and in laying a foundation for continuing improvements.
5. Web-based glossary
One of Medtronic's biggest problems with translation was the inconsistent use of technical terminology. Dozens of people author source documents, including user interface developers, technical writers, and writers of marketing, educational, and promotional material, many of whom are outside contractors. We use different translation agencies for marketing material than we use for manuals and user interfaces. (We have found that agencies usually focus either on user interfaces and documentation or on promotional literature, but not both.) Therefore, more than one agency works on material for the same product. In turn, the agencies use many freelance translators. Finally, in each country many different people review translations.
The number of people involved in authoring and translation led to many inconsistencies in terminology. Also, we often discovered too late that terminology chosen by engineers was not suited to users. This terminology was included in the user interface early in the product design cycle and was codified in manuals before the marketing material was written. We often did not discover the problem until the marketing material was being reviewed. At that point it was often too expensive and time consuming to change the terminology in completed material. As a result we often had inconsistencies between the technical and marketing material. Furthermore, we realized that discussions about terminology were adding time to the critical path for product release. In at least one case, a terminology problem delayed release of a product in one market for three months.
To help solve this problem, we developed a comprehensive intranet-based, worldwide, Medtronic glossary, one of the first of its kind. We worked with Dr. Alan Melby of Brigham Young University to design the architecture, develop the initial data models, and prototype the database. We piloted the glossary in one product line, and are still expanding it to other product lines and businesses.
The master glossary is stored in a Multiterm database. To date, it contains 3,800 terms, including definitions and examples of usage in context. It is delivered to the intranet using MARTIF (Machine Readable Terminology Interchange Format), which is an ISO standard. The intranet version of the database can be searched in any of the languages covered in the database. It also has attributes for different product lines. It is designed to be used on the company intranet for internal users, including in-country reviewers (sales and support staff in sales offices around the world). Because of the definitions and contextual examples, the database is also useful for new employees and people new to a particular product line.
Unfortunately, for something as interactive as a glossary, the poor performance of our transatlantic intranet lines precludes much use of the Web version by translation agencies and in-country reviewers. Therefore, we output versions to different electronic formats or to hardcopy for local use. The fact that MARTIF is based on SGML (Standard Generalized Markup Language) makes it very portable and compatible with any number of different publishing and translation systems.
We hired a full-time terminologist to maintain the glossary and to help drive the process of defining terminology early in the design cycle. With guidance from the terminologist, technical writers identify terms in existing documents to build the initial glossary. Technical writers also write definitions and extract or provide examples of usage in context. On an ongoing basis, writers identify the need for new terms early in the product development process. Marketing and development staff review and approve the terms and definitions before they are translated. Then sales and support staff in the country of the target language review the translations of the terms. The goal is to establish the terminology early in the process so that once we are in the throes of translation, everything proceeds relatively smoothly.
Based on project post mortems that included surveys of the people involved in the translation process, we have good evidence that the use of the glossary has significantly improved consistency and turnaround time at the end of the projects. On one project that used a 1,000-word glossary, we calculated a 25% reduction in cost compared with a previous similar project. Although not the sole innovation that led to this cost savings, the glossary was widely agreed to be a major factor.
6. Web-based tutorial
Employees in our technical communication department have long realized the importance of carefully preparing the English source text to smooth the translation process. In the past, a translation manager taught two stand-up classes in writing for translation. However, new employees and contractors also need to learn about writing for translation and for non-native speakers of English. Furthermore, much of the marketing and advertising work is outsourced. The knowledge level of those writers varies widely, and they haven't the time or inclination to come to a class, even if the training resources were available as often as needed.
As part of an industry/academic affiliation program, technical communication students in the Rhetoric Department at the University of Minnesota agreed to do a class project to help us meet this ongoing need. They developed a Web-based tutorial for writers, user interface designers, and multi-media developers. The companion article in this issue by Patricia Flint, Melanie Lord Van Slyke, Doreen Starke-Meyerring, and Aimee Thompson provides details about this project. Although it is too early to assess the tutorial's success, we hope that it will help individual writers learn about writing for translation when they need to, in as much detail as they want.
7. Better relationships between people
Fons Trompenaars' (1993) research showed a cultural continuum of universalist and particularist cultures. The U.S. is one of the most strongly universalist cultures. Particularist cultures place more importance on relationships than do universalist cultures. Thus, for example, Americans often need to pay more attention to relationships when they are working with people from other cultures than they otherwise might.
The relationships between the people involved in the translation process can be strained by the situation. Often our tasks are on critical paths, as we try to incorporate last minute changes in time for product release in many languages, with daily or hourly communications by fax, phone, and e-mail. It may appear to translation coordinators that technical writers make unnecessary stylistic changes to text that should be stable, or to technical writers that translation coordinators do not understand the technical details or the pressures of product development, system testing, and scheduling. However, the delivery of information - in all languages - is one large process. I believe that it is imperative that writers and translators work together well.
Medtronic has done several things over the years to help build relationships.
We find it is much easier to work with someone long distance when we know who the other person is and have built trust. I believe that no other form of communication is more effective in building those foundations than face-to-face communication, especially if it includes informal social interactions. We have committed to having at least four to five people from each site visit every year.
All the team leaders have visited their counterparts at least once or attended a conference together. A few of the other writers and translation coordinators have also exchanged visits to work on strategies, projects, and issues. The visits also afford the opportunity to see the work space, the pressures, and the processes of the other group.
Regular telephone meetings
In addition to face-to-face time, the managers and team leaders have regular telephone conferences to compare notes, keep each other up-to-date, and discuss issues. The frequency of these meetings varies from weekly to monthly. In addition to giving us a chance to work out problems, these regular calls help cement relationships.
Initially we also tried videoconferences. Some people who had never met enjoyed seeing each other in addition to hearing their voices. However, in the long run we've found telephone conferences more convenient. At least with our equipment, some of us found the slight time lag between the audio and video disorienting. We have also found that operating the camera can be distracting, especially if more than three to four people are in the room.
In addition to the contact at the manager level, writers and translation coordinators are in daily contact about their projects, both through phone calls and through e-mail.
In the U.S., we have invested in formal training in intercultural relations. Several people attended a class on working in business with people from other cultures. Others have been to short sessions on other cultures. A training vendor customized a one-day course for us on working with people from the Netherlands because we work most closely with the Dutch. That class covered some basic concepts about three continuums along which cultures are measured. It also provided a very brief introduction to Dutch geography and history.
TABLE 1: TIME SAVINGS FROM USING TRANSLATION MEMORY ON ONE PROJECT In-country Translation review Total Initial 8 days 14 days 93 days Subsequent 5 days 0 days 42 days Time savings 3 days 14 days 51 days % time savings 38% 100% 55%
The course also led to the exchange of photographs between the U.S. and Dutch teams. We posted our pictures in a remote corner of our intranet. Visitors also have taken photographs of people and offices so that those who have not had an opportunity to travel can get a sense of the working spaces of their colleagues across the ocean.
Finally, we try to remember to ask what works and what does not. We hold post mortems after big projects, most recently using our intranet. However, more frequent and less formal feedback is also helpful. For example, during one videoconference we learned that for our Dutch counterparts, it is often better if we send e-mail than if we telephone. For most of them, who learned English primarily by classroom study, it is easier to read English than to hear it, especially considering individual accents, the tendency to use slang in speech, and speaking speed.
While the benefits of building relationships are inherently difficult to quantify, we believe it has meant quicker responses to urgent requests, better cooperation, more sharing of ideas, and more pleasure in working together.
8. Translation memory tool
About two years ago, we began investing in translation memory databases at our translation agencies. The basic idea behind these systems is to automate the reuse of previously approved translations, usually at the sentence or paragraph level. The system contains chunks of text, usually sentences or paragraphs, in a source language (American English in our case). In the database, those chunks of text are aligned with their approved translations. Later, when a new or revised document needs to be translated, it is compared with the database, which automatically replaces any 100% match. The systems also contain linguistic algorithms that allow them to detect "fuzzy" or close matches as well. The translator can then accept or modify the fuzzy match. Thus, the translator can concentrate only on those parts for which the database can find no close match. In theory, this technology should vastly improve consistency, quality, and speed of translations.
On a recent project we needed to produce manuals for several models. First we translated the most comprehensive manual completely and had it reviewed by the in-country reviewers. We put the approved manual into a translation memory database. We then used the database to derive the translations for the other models. Table 1 shows the time savings from using translation memory, including two of the most important tasks. In addition, we saved about 60% of what we would have paid for the project had it not been completed using translation memory.
On the other hand, we have some problems with the translation memory databases. Keeping the format of the source document intact is a problem, because of the lack of direct filters from FrameMaker to the translation memory system. In our experience, the tagging for tables, index, cross-references, and conditional text gets corrupted, causing problems either when we reach the publishing stage or when translations are put into translation memory.
Complete matches are not always identified reliably by the system because of format coding and punctuation changes. Also, if the translators work in RTF (Rich Text Format) rather than directly in a translation memory environment, they may not be able to see the color coding that indicates the type of match, and thus may rework material unnecessarily. Furthermore, inconsistencies in English sources may get into the database. We find it difficult to decide at what level of granularity to divide up the database, weighing on the one hand the potential for poor performance from too large a single database, and on the other hand the difficulty of managing too many small subsections.
Setting up a translation memory database is a big hurdle, and maintaining it is also a lot of work, especially given many late-breaking changes. Yet overall, because we have so much overlap in content from one model, generation, or product line to another, translation memory makes a great deal of sense.
9. In-country reviews
As they are in many companies, in-country reviews tend to be a critical bottleneck in the translation process for Medtronic. It is not uncommon for raw translations to take only a few days to complete, whereas the review may take weeks because the reviewer needs to fit the task into his or her schedule.
The purpose of the in-country review is to ensure, first, that the terminology is appropriate, and second, that the translators have not inadvertently introduced technical inaccuracies. The translators are linguistic experts, and they may have some basic training in medicine and medical devices. In-country reviewers are native speakers who are product experts, as well as being familiar with the customers.
The problem is that the best (indeed often the only) people qualified to do in-country reviews are company employees in the countries. Usually, their primary job is selling to or supporting customers. In fact, we found that they often review translations of manuals in the evenings, on planes, in whatever free time they can find between their primary tasks. Until a few years ago, in many cases their managers were not even aware they were doing these reviews. Therefore, a task that is a vital part of a critical business process - reviewing translations - was being done "in the dark."
An important recommendation from our Business Process Improvement Task Force was to hire dedicated in-country reviewers, at least one reviewer per language. The reviewers would be trained extensively on products. They would be located in the sales offices, where they would have some contact with customers. Furthermore, they would have access to sales and support personnel and could solicit their assistance during peak workloads. Unfortunately, the additional headcount necessary to implement this recommendation was not feasible. One country manager who had led the task force believed in this recommendation enough to implement it in his country. We are still collecting data about the success of the arrangement.
In another country, the sales and support personnel share the review work. Before a major release, they schedule one or two days in a hotel conference room, where they work together until the whole job is complete. In yet another country, a translator who previously worked on the materials has been hired as a full-time reviewer.
Finally, for at least one product line, the task of reviewing translations has been added to the reviewers' job descriptions, and their performance evaluations include that task as well as their other work. Anecdotal evidence indicates that this has improved the timeliness of reviews.
Many of the sales and support people who review translations believe it is useful because it gives them an opportunity to learn about new products. However, they are understandably annoyed if they have to review the same material repeatedly, as when different models or generations share content.
Currently our strategy is to continue to raise awareness of the importance of in-country reviews and to find ways to reduce the amount of work for the reviewers (such as through the use of translation memory) or to take it off the critical path (such as by refining the glossary early in the process).
Despite some progress, we continue to struggle with four issues related to the translation process.
1. Rogue processes
The company's organization is heavily matrixed rather than hierarchical. Many of the business units were acquired by the corporation, and in some cases, the acquired companies had existing translation processes. The various business units are not required to use the standard processes developed by the central groups, and the charge backs to cover our costs are a deterrent to some of them. Yet we believe there are hidden costs to the seemingly cheaper alternatives, including:
* Poor terminology
* Unexpected rework
* Higher-than-necessary production and printing costs
* Confusion for reviewers, who get materials from too many contacts in varying formats and packaging
2. In-country reviews
Despite our efforts, we need to further streamline in-country reviews, not only to shorten this critical path task, but also to make courteous and efficient use of reviewers' time, thus freeing them to concentrate on customer and support tasks.
3. Interface between publishing and translation memory systems
The vendors of translation memory systems have made the investment to provide an integrated interface to the most widely used word processing programs. This makes sense considering that most agencies use freelance translators who work out of their homes, and those translators generally use word processing programs rather than more expensive publishing programs.
However, there is no integrated interface for FrameMaker, the publishing package we currently use. The agencies use a third-party package called S-Tagger[TM] that strips out formatting coding prior to processing by translation memory and then reinserts the coding after translation. This process works, but having to use it adds labor-intensive steps to the process and opportunities to introduce errors. On the other hand, we need the sophisticated, robust long-document features offered by a publishing package as opposed to a word processing package. It appears that migrating to SGML would help solve this problem.
4. Redundant work
Despite advances in international template design and translation memory, we are still handling many pieces of text and graphics multiple times. In part this is because of the repetition of content between models, generations, and products, and between technical material and promotional material. This latter redundancy becomes particularly apparent to in-country reviewers.
Below are some of the things we plan to further improve our process for writing and localizing documentation.
* Develop common visions, goals, and strategies between the technical writing and translation departments, despite our geographic and organizational separation. Using part of Joann Hackos's model for strategic planning (1997), we brainstormed a list of strengths, weaknesses, threats, and opportunities as a basis for developing such common goals.
* Implement a source control system, to facilitate check-in and check-out. We believe this would also help with change management, which is currently manual and labor intensive.
* Implement a component and document management system integrated with translations, based on an SGML/XML database. We hired a consultant to assess our writing and translation process and tools. He confirmed our analysis and helped refine a plan to migrate from FrameMaker to an SGML/XML database. Ideally, we would align the components in the source language and target language, and use a single common structure across all languages. Then, we would send only new or revised components through the translation process. We are currently finishing the business plan to justify the significant investment required to implement this solution.
* Consider whether to bring control of translation memory in house rather than outsourcing that function to translation agencies. We talked to two companies who have brought translation memory in house, as opposed to having their vendors maintain the database. This would be another significant investment, not just in acquiring the technology, but also in ongoing staffing, and development and retention of that staff. To bring translation memory in house is against the trend toward outsourcing whatever functions are not part of the company's core competencies. Yet we believe that potential improvements in cycle time, costs, control, and flexibility make it worth considering.
* More closely integrate glossary and authoring systems. Currently the glossary is separate from the authoring systems. Making it easier for authors to consult would be one important step. Another step would be to code terms in English source text, to ensure that they are translated with the glossary. Finally, we would like to investigate an automated checker.
* Consider integrating a controlled or simplified English checker to further improve the translatability of English source text.
As an international company committed to diversity, Medtronic considers translation and localization to be of significant strategic importance. I believe that identifying and ranking problems help to lay a good foundation, even if that ranking is based on the subjective opinions of experts. I also believe that LISA membership and relationship-building have significantly helped to improve our outcomes. We have solid evidence that the glossary, common platforms, and templates helped save time and money. Although still problematic, translation memory has also reduced time and costs. Page-true translations were, for us, not worth the extra time and steps. It is too soon to assess the results of the Web-based tutorial on writing for translation, or to determine the best way to structure the in-country review process.
As a result of all our efforts, our average cost per new translated page has dropped steadily over the past few years, from nearly $90 per page in fiscal year 1990 to about $60 per page in fiscal year 1998. For all pages, including maintenance and revised pages, the average cost is down to about $30 per page. Although some of that decrease is attributable to increased volumes, a significant part is due to process improvements.
Axtell, Roger, Tami Briggs, Margaret Corcoran, and Mary Beth Lamb. 1997. Do's and taboos around the world for women in business. New York, NY: John Wiley & Sons.
Axtell, Roger. 1993. Do's and taboos around the world. New York, NY: John Wiley & Sons.
Business process improvement, improving process effectiveness and efficiency. 1994. Rochester, NY: Quality Network Inc.
Hackos, Joann T, 1997. "From theory to practice: Using the information process-maturity model as a tool for strategic planning," Technical communication 44:369-381.
LISA[TM] (Localisation Industry Standards Association) Web site. 1998, www.lisa.unige.ch.
Melby, Alan K. 1998. "CLS framework" section of Web site about Translation, theory & technology. www.ttt.org.
Trompenaars, Fons. 1993. Riding the waves of culture. London, UK: The Economist Book Ltd.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Medtronic translates documents into more than 10 languages|
|Date:||May 1, 1999|
|Previous Article:||GIFAS Rationalized French: A Controlled Language for Aerospace Documentation in French.|
|Next Article:||Going Online: Helping Technical Communicators Help Translators.|