After welcoming participants, Michael Anobile introduced the Workshop's organizer and presenter, Prof. Alan Melby of Brigham Young University. One of the pioneers of translation technology and terminology interchange, Alan had done a fine job with his team of organizing the Workshop and, in particular, producing complete documentation on the day.

Alan then opened the Workshop and introduced the keynote speaker, Dimitri Theologitis from the European Commission's Service de Traduction (SdT). Dimitri emphasized this very large body's positive experiences with introducing translation technology. He defined three main user needs: reference documents in electronic form, pertinent terminology, and access to previously translated texts to prevent duplicated effort.

Three major types of tools are in use: administrative systems for ordering and billing, document management tools containing past texts, and linguistic systems for terminology (Eurodicautom and Trados), MT (Systran) and TM (Euramis and Trados Translator's Workbench). Users access the systems via a variety of different interfaces, which are to be integrated over time. Another medium-term goal is to influence the authoring process, and hence improve translation input quality, by offering truly user-friendly systems.

The CEC's Euramis system, conceived in 1995 by Jean-Marie Leick as a backbone for most of the linguistic tools required, basically comprises a linguistic resources database (aligned texts and dictionaries) and an e-mail batch server. Users send texts to the system from the user-friendly interface and receive files containing the relevant terminology and bi-text memory entries/MT sentences back. Benefits of the system include terminological and phraseological consistency and productivity increases of c. 60% for 25% of texts. User satisfaction was high, despite teething troubles. Problems and issues still to be faced concerned data management, the gap between authoring and translation, Web interfaces, verification tools and search engine integration.

Summing up, Dimitri said that we are now entering the era of text crunching, whether we like it or not. Translation today can be done by a mixture of human and machine output - the art is to position each document correctly within this continuum.

Next on the agenda were presentations by the various tools vendors: AlpNet (TSS and Win-Joust), IBM (Translation Manager), Star (Transit), Trados (Multiterm, Translator's Workbench and S-Tagger), Carnegie Mellon (the Kant MT system), Systran, and Logos. Each developer had to answer the same three questions: "How do your tools fit into the document workflow, what are their major features, and how do your tools interoperate with others?". Following this, there was a panel discussion in which the different vendors answered questions from the floor. Key issues were terminology (e.g. support for phraseology and neologisms), scalability, DBCS and code page support, and the Web and security.

In the next session, users of the different tools described their workflows, environments and needs, and how the vendors' tools fitted in. Here, too, standard questions were used. Siu Ling Koo from localizers L&L, an AlpNet user, said tool selection criteria at L&L included speed, networkability, compulsory dictionaries, term extraction, repetition files, format preservation, Dutch support and cost effectiveness. Now they had AlpNet tools, they felt no reason to change - all the technology was very similar.

Susan Daly of financial software publishers JD Edwards, a TM2 user, explained that the company's publishing system was the back-bone for all marketing, training and documentation information, and was based on reusable Interleaf files. TM2 had had the best interface to Interleaf of the tools they had tested, and its network facilities also allowed translators to share information instantly, and so only do things once too. Tools would change dramatically over the next five years, and the company had to deal with the HR implications of this. Translators were seen as becoming blue-collar staff, and were in danger of losing motivation.

Mark Gravez from Bentley Systems Inc., a global engineering and drafting company producing the MicroStation design product, was next. Responsible for localizing documentation produced in Framemaker, Bentley works with BTS, the US Star Transit distributors and services provider, on localizing texts for Asia. Good project management is the most important factor for successful localization. The formula for producing gains from technology is tool quality plus the ability of the operator to work around the box, plus publisher support. It was very important for publishers to understand the concepts behind the technology in order to interpret vendors' statements.

Sharon O'Brien from IT localizers ITP spoke about the company's experience with Trados and Multiterm. ITP had in fact developed S-Tagger for Framemaker and Interleaf (it was not limited to use with the TWB). Factors influencing the need for translation technology were the increase in partially completed jobs and update requirements which ITP was being asked to handle, pressure of time, the re-use of translated material, improvements in technology and customer expectations. Trados was chosen because of its Windows/Word environment, network capabilities, standard hardware and software requirements, stability and features such as fuzzy matching and concordance. Investments in licenses were only a small proportion of the total: recruitment of skilled staff, tool development, training and technical support were also major issues. Unresolved technical issues were the lack of support for bi-directional languages, word counting discrepancies and unexplained alignment problems, while generic issues were translator payment, the quality of previous translations, customer expectations and the freelance-dominated industry.

Dr. Christine Kamprath of heavy machinery manufacturers Caterpillar then talked about Kant, a knowledge-based MT system developed by Carnegie Mellon University's Center for Machine Translation for multilingual document production in an industrial setting. Caterpillar had c. 15 document types and c. 350 current products, with liability being an important issue. The average product age was 17 years, and translations into many languages were required. A new authoring environment offered support for controlled English input and reusable "information elements", in order to increase the accuracy and consistency of human and machine translation. An average of 5 full-time equivalent employees per year had been involved in the Kant project on Caterpillar's side for the last five years. Benefits so far included increased consistency of English texts, rapidly increasing reusability, a productivity gain of at least 2:1 for French, and heightened awareness of language issues. Unresolved problems were evaluation metrics, the degree of stylistic leeway acceptable, translator reluctance, and the need for terminology work and good terminologists.

Ann Adams of Xerox Corporation said that the translation group in Rochester, NY (USA) has been using Systran for many years. Work includes service manuals, customer guides, multimedia training and sales materials, and software and hardware interface texts. The Writer's Corner Website offers the very large terminology database, plus guidelines for writing for international markets and controlled English checkers. Controlled English rules have been reduced from 28 to 4, as authors found them too complex. General problems include staff turnover and how to influence developers, plus the increased complexity of the business and number of products, a fact not appreciated by management. It was hard to quantify the benefits because Systran was an ingrained part of the process: it was difficult to imagine life without it.

Françoise Densmore of Canada-based CIT vendor Mitel Corp. then described how she uses Logos in a distributed environment for French translations. Initial cost for the investment was USD 95,000 for 800,000 words annually; current annual maintenance is USD 7,350. The time required to customize the Logos dictionaries had been 9 months for Alex, plus ongoing work on the Windows client. Benefits included improved QA, standardized glossaries and pattern matcher files, remote dictionary set-up and dictionary merge features, fast, toll-free submission and retrieval of texts to the server, and gains in productivity. Unresolved were the availability of Samantha for Windows, to improve semantic rules locally, and remote control of the Logos Scheduler. A facility to translate text contained in CorelDraw illustrations in a DRW-compatible format was also needed.

In the question and answer session that followed, discussion centered on the various pricing models for fuzzy matches (even 100% matches entail editing and proofing), the ability of tools to cut DTP costs, and the relationship between translation memory and machine translation. This was no longer an either/or choice. Which should be integrated with which depended on the environment and users' experience. While TM was generally more comfortable for translators, MT was identified as such and it was thus easier to find mistakes.

Following the user presentations, vendors had another chance to present the tools, after which there was a wrap-up session. Asked what was unique about their products, translation memory vendors answered as follows:

• Star: A single-source approach to working on files; no database means no redundancy.
• AlpNet: Flexibility: AlpNet doesn't sell its tool on the open market, but develops solutions to customers' workflow problems.
• Trados: The database approach means no administration is necessary for storing files, keeping directories, etc. Also, integration with the translators' standard Windows environment and with Multiterm.
• IBM: An industrial strength partner, and a proven system, plus more language pairs (including bi-directional ones).

Asked to comment on their installed base, vendors gave the following answers:

• Kant: One
• Systran: Several, including the US government and the European Commission.
• Logos: A large number of high-end industrial users. The precise number of users varies depending on how you count it: OSRAM had one installation but 500 users.
• IBM: Over 20 internal translation centers plus commercial users.
• Trados: Over 10,000 installations, not counting the CEC sale.
• AlpNet: 33 offices plus affiliates, plus those people who bought it ten years ago.
• Star: 300 internal users, plus c. 2,500 external installations (including the Light version).