ISSN 0965-5476………………………………………………………… Issue no.6, September 1993

John Hutchins, The Library, University of East Anglia, Norwich NR4 7TJ, United Kingdom. Fax: +44 (603) 259490; Email: L101@uea.ac.uk

AMTA: Joseph E.Pentheroudakis, Microsoft Corporation, Redmond, WA 98052, USA. Fax: +1 (206) 936-7329; Email: josephp@microsoft.com

EAMT: Tom C.Gerhardt, CRPCU/CRETA, 13 rue de Bragance, L-1255 Luxembourg. Fax: +352 (44) 73 52; Email: tom@crpculu.lu

AAMT: Professor Hirosato Nomura, Kyushu Institute of Technology, Iizuka, 820 Japan. Fax: +81 (948) 29-7601; Email: nomura@dumbo.ai.kyutech.ac.jp

Bill Fry, Association for Machine Translation in the Americas, 2101 Crystal Plaza Arcade, Suite 390, Arlington, VA 22202-4616, USA. Tel: +1 (703) 998-5708; Fax: +1 (703) 998-5709.

The Fourth Machine Translation Summit, held in Kobe, Japan, on 19-22 July 1993, lived up to its theme, "International Cooperation for Global Communication." The conference was attended by 220 participants from 15 countries. Both the keynote speech and a concluding panel addressed the theme specifically, while reports of several multinational projects highlighted the importance of international cooperation and attested to an increasing trend in this direction within the MT field. As Conference Chair Makoto Nagao pointed out in his opening speech, the International Association for Machine Translation has provided a framework within which such cooperation can be advanced and in fact accelerated. IAMT's impressive growth since its establishment two years ago at MT Summit III is clear evidence not only of interest in MT throughout the world but also of the strong desire of researchers, developers, and users to cooperate at both national and international levels.

MT Summit IV marked a new beginning in more ways than one: it was the first Summit to be organized within the framework of the International Association for Machine Translation, and it was also the first to be held again in Japan, bringing the tradition of biennial MT Summits full circle and starting a new rotation between the three regions of the world. In addition, IAMT gained a new president at the close of the conference when the Association's founder and outgoing president, Makoto Nagao, turned over the gavel to Margaret King, president-elect of IAMT and president of the European Association for Machine Translation (EAMT).

The Hotel Okura Kobe site, facing the magnificent sweep of Osaka Bay, was a fitting venue for this major event. Indeed, one of the special treats of the conference was the welcoming reception, which took the form of a twilight cruise around the harbor.

In his keynote address, Makoto Nagao stressed the importance of MT being utilized appropriately. Citing surveys of MT use by the Asia-Pacific Association for Machine Translation (AAMT) and the Japan Electronic Industry Development Association (JEIDA), he pointed out that a sizeable number of translation services count on MT to handle as much as 1,000 pages a month. The studies revealed that successful user sites tend to have the following characteristics: input is already in electronic form, the subject matter is highly focused, and thoroughly customized dictionaries have been built up over time. The break-even point for cost-effectiveness seems to come after about 10,000 pages a year. There continues to be a problem with the quality of both input and output, but the advantages of MT appear to outweigh these problems. Current emphasis is on greater utilization of networks, the development of filters between different word-processing and publishing systems, and the improvement of pre- and postediting facilities. He also noted the increasing use of MT on personal computers for information purposes only. By way of conclusion, Nagao outlined future tasks for all concerned with MT. Users should build dictionaries of specialized terminology, exchange experiences with others, collect typical corpora, cooperate in the development of guidelines for evaluation, and share all this information through user groups and clearinghouses. Developers, in turn, should clarify the information they give to the public, with specific statements about what their systems can and cannot do, and they should cooperate in the definition of common format codes for the exchange of texts and dictionaries as well as standardized evaluation procedures. And finally, researchers should explore new approaches such as statistical and example-based MT, cooperate with counterpart teams on a worldwide basis, and at the same time continue to press forward with discourse analysis, syntactic and semantic disambiguation, and the build-up of knowledge sources.

The conference featured three invited speeches. The first of these, by John Hutchins, covered the latest developments in MT technology. Since the end of the 1980s the most striking change has been the emergence of new approaches and methods in what is broadly called 'corpus-based' MT, particularly the increased use of statistical methods and explorations of example-based MT. However, the rule-based approaches characteristic of systems in the 1970s and 1980s have also witnessed considerable changes: the widespread adoption of constraint-based and unification formalisms, enabling the development of general-purpose shells for NLP and MT; the move towards 'lexicalist' approaches (illustrated at its most extreme by the 'shake-and-bake' approach), the increased attention to lexical acquisition, and the greater emphasis on the generation of idiomatic output. Other significant developments include multilingual generation from non-textual databases, research in systems for monolinguals not knowing target languages and numerous domain-specific systems developed by companies for specific purposes exploiting well-established methods and techniques of NLP and MT. He ended by speculating about the types of MT systems which may emerge from the various and sometimes divergent methodologies seen at present. He suggested that during the 1990s we may see the appearance of a "third generation" of systems, founded on a linguistic rule base (although less abstract than those typical of previous interlingua-transfer systems) incorporating a mixture of dictionary-derived lexical information, databases of domain-specific knowledge, aligned corpora of bilingual texts giving examples of translations, and the use of probabilistic approaches to lexical and structural transfer and selection.

In an invited speech on the current state of machine translation usage, Muriel Vasconcellos reported data from a worldwide study of MT users. The survey yielded 40 responses, 23 of which gave annual production figures. The data from this subset alone showed that users are enlisting MT to produce more than 170 million words (680,000 pages) a year. On the basis of these figures and the number of other known users (about 80, most of them with smaller installations), Vasconcellos proposed that a very conservative estimate of MT use in the world would come to a total of 300 million words (1.2 million pages). Of the volume reported in the survey, 60% involved the translation of technical manuals, software, and other materials related to localization. At least one of these users found MT "indispensable," and those that reported on productivity cited increases ranging from 30% to 50%, which for them meant both reduced costs and faster turnaround. They were happiest with the fact that MT keeps terminology uniform and saves the need to re-enter format codes in each translation. The most frequent complaints had to do with document preparation, the quality of input texts, and postediting. [A revised version of this paper is reproduced in this issue on pages ... ]

The third invited speaker was Toshio Yokoi of the Japan Electronic Dictionary Research Institute. His topic was the problem of constructing very large-scale knowledge bases, not just for MT but for a multitude of natural language processing purposes. He began by stressing the crucial role of natural language for the representation of knowledge, not just as the interface between knowledge databases and their users. Major requirements of large knowledge bases must include: a) the capacity to expand without difficulty, to enhance the quality during expansion, and to absorb information from a variety of different sources; b) support for interpretation by both humans and by computers, for integrating representations in many types of languages (natural and formal), images, diagrams, sounds, etc.; c) demands on the development of new system architectures, where natural language is the core medium of the computer system; d) the ultimate goal of full understanding of natural language in reliable processing. The EDR electronic dictionary project was given as an example of progress towards these ends. In one respect, the electronic dictionary was itself a large knowledge base, and here Yokoi stressed the relationship between the EDR concept dictionaries and the text base providing example definitions. More important perhaps was the experience gained during the project in lexical knowledge acquisition, providing techniques and methods for building knowledge bases in general. He concluded by describing plans for the "Knowledge Archives" project, the creation of a very large database of 'knowledge documents' - in free or controlled natural language, in formal knowledge representation languages, in pictures, images and sound.

The EDR project was the theme of other contributions - by Seibi Chiba and Yoichi Takebayashi - who together brought participants up to date with its current status. Hiroshi Yasuhara (another EDR member) concentrated on a description of the research on example-based MT using text data gathered for the dictionary. Research and development on text corpora was the central topic of Yorick Wilks' presentation. He gave a broad survey of developments in message understanding (SRI, BBN, Massachusetts), tagging of corpora (Lancaster, AT&T, Pennsylvania), alignment of bilingual texts (Brown et al., Church, Kay), stochastic grammars, machine-readable dictionaries (LODCE, CoBuild, etc.), connectionist models, large text corpora (Lexical Databank Consortium, the European Corpus Initiative, etc.), ending with speculations about the role of corpus-based approaches in MT, which was dealt at greater length in his paper in the proceedings which concentrated on an assessment of the achievement of the IBM Candide project.

As in previous years, the MT Summit provided a forum for the description of major MT and NLP projects. Susumu Funaki described the status of the CICC multilingual project involving teams from Japan, China, Thailand, Malaysia and Indonesia; and Meying Zhu (in a joint paper with Hiroshi Uchida) gave a detailed account of the structure of the interlingua being used. Ahmad Zaki Abu Bakar described the Malaysian contribution to this project and also outlined other MT activity on the translation of textbooks and of news bulletins for the Kuala Lumpur Stock Exchange.

The ATR project was described by Tsuyoshi Morimoto and Akira Kurematsu, with particular emphasis on its experimental speech translation system. In earlier versions more than 90% of utterances were recognized and translated accurately in about 25 seconds processing time. The latest version (ASURA) has been greatly extended in vocabulary and sentence types, but accuracy has dropped to about 60% and processing time has increased to 50 seconds. The speakers were confident, however, that hardware improvements will enable the achievement of near real-time processing. In brief, ATR is claimed to have demonstrated the technical feasibility of an "Interpreting Telephony System" in the near future.

The third major project is also focused on spoken language translation. This is the Verbmobil project described by Wolfgang Wahlster of the German Research Center for Artificial Intelligence. The aim of this long-term project is the development of a portable device for aiding the translation of spontaneous spoken language in face-to-face negotiation dialogues. The first version will aim to provide translations on demand for participants (German or Japanese) who both have passive knowledge of English but are not fluent speakers; i.e. where English is used as a common language in business or technical discussions. The project is planned for 8 to 10 years; in the first four years funding will amount to 60 million Deutschmarks. International collaboration is planned with ATR and with three US groups at Carnegie Mellon, Stanford (CSLI) and Berkeley (ICSI).

A more general view of European activity in MT and related spheres was provided by Loll Rolling (Commission of the European Communities). He described the various projects which have been supported by the Multilingual Action Plan, the Eurotra project, the Language Research and Engineering (LRE) project, the ESPRIT programme, EUREKA projects such as GENELEX and GRAAL, and the ECLAT programme in language technology.

A panel on the evaluation of MT gave rise to a lively discussion with considerable audience participation. Under the chairmanship of Margaret King, some of the key issues in this area were addressed by Jaime Carbonell, Hirosato Nomura, Loll Rolling, and Muriel Vasconcellos. The panelists agreed that there is no single "right" way to evaluate MT and that there should be a variety of methodologies, much like a cook's batterie de cuisine, from which several can be chosen in combination, depending on the circumstances. They went on to discuss the differences between progress, adequacy, and diagnostic evaluation, as well as the best approaches for each. A heated debate ensued over the relative merits of glass box vs. black box perspectives. The conclusion seemed to be that a view inside the glass box is essential for progress and diagnostic evaluations and that a "gray box" may be best for evaluations that focus on adequacy: even though users may be more concerned with black box considerations, it is always important to know why a system fails.

Future developments in MT technology were the subject of a panel chaired by Hiroshi Uchida. Six experts took on the challenge of this topic: Christian Boitet, Pierre Isabelle, Hwee Boon Low, Sergei Nirenburg, Christian Rohrer, and Junichi Tsujii. The general agreement was that future MT systems will be 'hybrids', combining the best features of rule-based approaches, whether linguistic rules or knowledge bases, with the more recent stochastic and example-based methods. There would be more attention to specific user needs in the design of actual systems (i.e. fewer general-purpose systems) - selecting the 'best' methods for the purpose. Isabelle argued that corpus-based approaches were more appropriate to machine-aided translation than rule-based methods; indeed, research should concentrate on tools for aiding translation rather than systems for producing translations. Rohrer saw a major role in the future for multilingual text generation from databases; and Boitet speculated about the future for speech translation, the place of MT in multimedia communication (pointing to experiments in TV subtitling as an example), and aids for simultaneous interpretation.

In keeping with the theme of the conference, the third panel brought the substantive program to a close with an overview of current international cooperation in MT and plans for the future. Y.T.Chen, director of the NSF information center, gave the US view, emphasizing the advances in many areas related to language technologies and the growth of the global economy, multinational companies, telecommunication networks, and the appearance of new information- and knowledge-based products. He then outlined the support of the US government to research in these areas, through ARPA and NSF, and the encouragement of international cooperation, particularly the sharing of data and tools, the establishment of standards and evaluation methods, through demonstrations of technology, workshops, joint sponsorships, and the establishment of the national information infrastructure. The Japanese view of was given by Makoto Nagao. He highlighted first the areas which are still problematic and where international cooperation is desirable, even essential: discourse, selection of texts suitable for translation, terminology problems, multilingual text corpus, learning mechanisms, parallel architectures for MT, methods of evaluation, networking MT systems, common text and dictionary formats for interchange. He ended by surveying current Japanese government involvement in the multilingual cooperative project (CICC), cooperative development of terminology dictionaries (EDR), and support for MT and NLP information exchange centres. A weakness in the Japan picture was the poor progress in building large text databases, with the main impediment being the problem of copyright. He thought also that far more discussion on standardisation was desirable in Asia; there was much activity here in Europe and the United States but Asian countries had lagged behind. Loll Rolling sketched previous CEC involvement in MT, and then went on to describe the present activities of EAGLES on cooperation with text corpora, lexical resources, evaluation and raising of awareness. He emphasized the need for projects and plans to be adequately reported and distributed, and believed that MTNI could and should play an important role in the dissemination of information about international activity. Yang Tianxing described the support for R&D in MT in the People's Republic of China, surveying briefly the current centres and their research activities, and describing China's involvement in the multilingual CICC project. Toshinori Saeki of MITI (Japan) stressed the large-scale support for CICC project by MITI, but he also pointed out the relatively low level of telecommunication networking in Japan and the need for its improvement if international cooperation is to be encouraged.

The final event was the Second General Assembly of the International Association for Machine Translation. The minutes of the Assembly are published on page -- of this issue.

[Copies of the proceedings are available from the MT Summit Secretariat c/o AAMT, Akasaka Chuo Mansion 305, 7-2-17 Akasaka, Minato-ku, Tokyo 107 Japan. Tel: +81-3-3479-4396; Fax: +81-3-3479-4895.]

For three days (14-16 July) in the week preceding the MT Summit, researchers gathered in Kyoto for the Fifth International Conference on Theoretical and Methodological Issues in Machine Translation (TMI '93). Its theme was announced as "MT in the Next Generation" and, being held just one year after the previous TMI conference in July 1992, it was perhaps inevitable that papers would take up the issues raised in Montreal, in particular the 'confrontation' between the well-established methods of rule-based approaches to MT system design and the newer corpus-based approaches (stochastic methods and example-based MT).

The first three sessions (on 14 July), were devoted to the newer methods. The opening session was started by Jing-Shin Chang and Keh-Yih Su with a description of the BehaviorTran (previously ArchTran) MT system at the National Tsing-Hua University, Taiwan ('A corpus-based statistics-oriented transfer and generation model for machine translation'). Its combination of rule-based and corpus-based methods illustrates an increasingly common option in MT system development. They were followed by Hiroshi Maruyama, Shiho Ogino and Masuru Hidano on 'The mega-word tagged-corpus project' for building a large, high-quality text corpus for Japanese. The main emphasis of the paper was a description of the Japanese morphological analyzer. Eiji Komatsu, Jin Cui and Hiroshi Yasuhara of the Electronic Dictionary Research Institute then reviewed the EDR project and its relevance to interlingua-based MT ('A mono-lingual corpus-based machine translation of the interlingua method').

Five papers on example-based MT (EBMT) followed. First, Sergei Nirenburg, Constantine Domashnev and Dean J.Grimes (Carnegie Mellon University) described 'Two approaches to matching in example-based machine translation' (comparing performance using different passage lengths); then, Satoshi Sato of the Japan Advanced Institute of Science and Technology spoke on 'Example-based translation of technical terms', describing the latest version (MBT3) of a series of systems exploring 'memory-based translation' using an external database of translation examples, a Japanese thesaurus and an English thesaurus. The last talk in this session was by Stephen Richardson (in a paper written jointly with Lucy Vanderwende and William Dolan) on 'Combining dictionary-based and example-based methods for natural language analysis' which described experiments at Microsoft towards 'hybrid' NLP using rule-based methods for structural information and the extraction of information from dictionary definitions to create a large database of examples illustrating semantic relations. The final two papers of the day were devoted to aspects of preference scoring for selection in EBMT. Eiichiro Sumita, Osamu Furuse and Hitoshi Iida (of ATR) reported on 'An example-based disambiguation of prepositional phrase attachment'; and Kiyoshi Yamabana, Shin'ichiro Kamei and Kazunori Muraki (of the NEC Corporation) also spoke about resolving ambiguity problems in their paper 'On representation of preference scores'.

On the following day (July 15), it was the turn of the 'rational approach' in the first session. Jörg Schütz and Bärbel Ripplinger began with research emanating from the Eurotra project in a paper on integrating computational terminology in a constraint-based NLP/MT environment ('Machine translation supported by terminological information'). They were followed by Nili Mandelblit from the University of California, who argued for the building of interlingua MT on a cognitive linguistics basis by incorporating a knowledge base of language usage ('Machine translation: a cognitive linguistics approach'). The proposal was illustrated by translations from English into Hebrew. The third paper by Margherita Antona and Jun-ichi Tsujii put forward a rule-based approach to tense and aspect, involving a bilingual translation knowledge base ('Treatment of tense and aspect in translation from Italian to Greek - an example of treatment of implicit information in knowledge-based transfer MT').

There followed a session for papers describing other new approaches in rule-based paradigms: Koichi Takeda (IBM Research, Tokyo) on 'An object-oriented implementation of machine translation systems' (essentially a knowledge-based rule-driven approach within the research on Shalt-2); Hagyu Lee and Yung Taek Kim (Seoul National University) on 'An idiom-based approach to machine translation' (Korean-English); and Tetsuya Nasukawa (IBM Research, Tokyo) on 'Discourse constraint in computer manuals' (an examination of the sublanguage in this area, and using no extra-linguistic information).

The third session of the second day began with a description of 'Recent advances in JANUS: a speech translation system' presented by J.Tsutsumi on behalf of M.Woszczyna and others from Carnegie Mellon University. It was this system which was involved in the demonstration of spoken MT by ATR in February [see elsewhere in this issue]. This was followed by Pierre Isabelle with a joint paper with others from CITI, Canada (previously CWARC) on 'Translation analysis and translation automation'. He argued that the analysis of previously translated texts can be analyzed as structured databases of great value as aids for human translators. He described TransSearch (a tool for bilingual concordances), the development of TransCheck - a system for detecting deceptive cognates ('false friends') - and research on TransTalk for facilitating speech-to-text transcription of dictated translations.

The final session of the day was devoted to a panel discussion on 'Translation equivalents' with contributions from Christian Rohrer, Jaime Carbonell, Yorick Wilks, Pierre Isabelle, and Makoto Nagao. The discussion was dominated to some extent by the theoretical question of whether 'equivalence' as such is possible, and reached the obvious conclusion that in MT a pragmatic approach must be adopted (thesaural, statistical, example-based, etc.) The discussion did raise some major issues: the inevitable finiteness of databases, the limits of compositionality, the abstract nature of dictionaries, the most appropriate unit of equivalence, the role of examples, and so forth.

The third and last day of the conference (16 July) began with three papers on source language analysis. Masaki Murata and Makoto Nagao (Kyoto University) discussed the 'Determination of referential property and number of nouns in Japanese sentences for machine translation into English'. Then, Satoshi Shirai, Satoru Ikehara and Tsukasa Kawaoka (NTT Laboratories) described experiments on a method for rewriting modal and tense expressions and expanding coordinate structures in Japanese input sentences, for implementation in the ALT/JE system ('Effects of automatic rewriting of source language within a Japanese to English MT system'). Satoshi Kinoshita, Miwako Shimazu and Hideki Hirakawa (Toshiba Corporation) reported on methods for deriving reliable semantic data from texts, which can be applied in the analysis of ambiguities in other texts ('Better translation with knowledge extracted from source text').

The second session of the last day was devoted primarily to questions of MT evaluation. Masuru Tomita and colleagues from Keio University spoke about attempts to evaluate MT performance on translating English TOEFL (Test of English as a Foreign Language) texts into Japanese ('Evaluation of MT systems by TOEFL'). Evaluation by student examinees was not in general reliable, in part because the translations themselves were poor, but some guidance for devising evaluation tests was derived. This was followed by Adriane Rinsche who reflected on her experience in evaluating a number of MT systems on behalf of the Commission of the European Communities ('Towards a MT evaluation methodology'). She concluded that few truly quantitative measures are practicable and that some subjectivity is inevitable in this area. The third paper in the session was by Kwangseob Shim and Yung Taek Kim (Seoul National University) who described a statistical approach to structural disambiguation of verb phrases ('Towards a machine translation system with self-critiquing capability').

The final session of the conference considered problems of extracting information from corpora - thus returning neatly to the topics of the first two sessions on the opening day. Hideo Watanabe (IBM Research, Tokyo) described 'A method for extracting translation patterns from translation examples'. Shinichi Doi and Kazunori Muraki reported on NEC's statistical method for selecting translation equivalents ('Evaluation of DMAX criteria for selecting equivalent translation based on dual corpora statistics'). Finally, Teruko Mitamura, Eric Nyberg and Jaime Carbonell described the approach at Carnegie Mellon University towards the building of large knowledge bases for interlingua systems such as KANT and its successors ('Automated corpus analysis and the acquisition of large, multi-lingual knowledge bases for MT').

For those participants spending time in Kyoto before going on to the MT Summit in the following week, there was the fascinating glimpse of traditional Japan provided by the spectacular Gion Matsuri procession on the next morning. It was a most fitting climax to what was undoubtedly one of the best conferences in the TMI series. The next TMI conference will take place in 1995 in Louvain, Belgium.

Copies of the proceedings of TMI '93 are available from the organizers: TMI '93 Secretariat, HS Building, 7-1-9 Nishi-gotanda Shinagawa-ku Tokyo 141 Japan; or Hozumi Tanaka (Dept. of Computer Science, Tokyo Institute of Technology, 2-12-1 Ookayama Megura Tokyo 152 Japan) and Yuji Matsumoto (Graduate School of Information Science, Advanced Institute of Science and Technology, Ikomashi, Nara 630-01 Japan).

Translation Fair was held from December 15 to 17, 1992 at Science Museum in Kitanomaru Park in Tokyo, under the theme "Future Electronic Communication." The Fair was supported by various ministries and agencies, including the Ministries of International Trade and Industry (MITI), Foreign Affairs, Labour, Education, and the Science and Technology Agency.

The purpose of the Fair was to explore the future state of communication through discussions on various techniques, to find a way of effectively using machine translation systems, which are tools for electronic translation, and through the presentation of up-to-date information on research and development.

The Fair was co-ordinated by the Japan Translation Association (JTA), with special cooperation by the Centre of the International Cooperation for Computerization. These organizations were involved so as to encourage the participation of a diversity of groups, ranging from researchers and manufacturers to end users of MT systems. The Japan Translation Association is a corporate body under the administration of the Ministry of Labour. The Association trains a large number of translators and gives translator certification tests as well.

The Center of the International Cooperation for Computerization (CICC), administered by MITI, offers assistance to developing countries in terms of computerization and promotion of various systems, as well as knowledge. Since 1987, under commission from the Japanese government, the organization has been jointly developing advanced multi-language machine translation systems with China, Thailand, Malaysia and Indonesia.

For the opening ceremony on December 15, guests, the representatives of organizing corporations, and then persons concerned gathered at the venue. After speeches by the leading officials of the MITI and the Ministry of Labour, Mr. Nagao, the chief organizer of the Fair, Mr. Yuasa, chief executive committee member, and some distinguished guests cut the tape to declare the Fair open.

Symposium 1 consisted of three sessions. First, Prof.Tanaka (Tokyo Institute of Technology) gave a keynote speech on "Research on machine translation and its future tasks to meet globalization." In this speech, Prof. Tanaka introduced past and present trends in research and development of machine translation in various countries. He then revealed that CICC employs the interlingual approach instead of the transfer approach, which is currently on the market. The reason for this choice was: 1) the interlingual approach offers an easier means of developing multi-language translation methods; 2) various AI technique are integrated with this approach; 3) therefore, introducing this approach may have greater impact on technological development in developing countries. In fact, using this approach, these countries were able to construct particular systems for their own languages. Moreover, dictionaries made for machine translation are valuable not just for this original purpose but for understanding natural language processing. Important future tasks for researchers include the construction of large-scale knowledge bases and the integration of such techniques as human interfacing, evaluation method, standardization, networking, and parallel processing. The outline of the CICC projects was then introduced, followed by reports on the status of research and development in joint research countries: Thailand, China, Malaysia and Indonesia. Though these countries face different challenges in diverse environments, researchers in all of these nations have the same position and aspiration to achieve success in this field.

After Prof. Tanaka's speech, a panel discussion was held by the representatives of various institutes from these countries on the theme "International cooperation toward a multi-language translation system."

The theme of the first session was "The gradual introduction of machine translation systems and utilization know-how." In this session, Mr. Makino (associate professor at Toho Univ.) discussed improvement of machine translation process, technique and quality, from the viewpoint of engineering. Mr.Ogawa (Nikkei Printing Inc.) then said: "Those who think that machine translation can be used without editing know nothing about translation. At the same time, those who think machine translation works know nothing about MT." He then introduced the know-how involved in constructing dictionaries. Mr.Sakurai, representing the machine translation industry, said that even professional translators cannot do a perfect job, and explained the way of Japanese sentence edition. Mr. Mori (Babel Inc.) spoke on the important points in utilizing machine translation systems, from the viewpoint of training translators and editors.

The theme for the second session was "This is how I use machine translation." In this session, Mr Harada (IBM) gave a detailed explanation, based on his five-year experience as a user, of the points that need improvement in operability and design, and the improving process. He emphasized that to fully utilize machine translation systems, it is important to know the characteristics of the material to be translated and the particular flow of translation of that material, and to arrange the machine translation program, related programs, and operating method suitable for the material.

Mr.Ando (Toin Corporation) showed some examples of actual usage of machine translation, saying, "You have to use your brain just a little to use machine translation fully." Mr.Ozawa (NCR Japan, Ltd.) said that he uses machine translation to make a draft. In this way, he said, he tries to give an answer to the long-term conflict between competent human translators and machine systems. He also said, "If translators begin to employ machine translation systems, they can release themselves from various restrictions: restrictions of working site and the burden of hard work which demands that translators use their brains merely as part of their bodies." Mr.Furusho (Yumeya), who manages a personal computer network host station for sixty members, introduced a case of textbook preparation for English study by the auditory translation method.

The theme of the third session was "The potential of translation via network." In this session, Mr. Inoue (Iris International, Inc.) introduced the present state of machine translation service between English and several European languages via NIFTY-SERVE. Mr.Harada (Kodensha Co.Ltd.) introduced examples of Japanese-Korean translation system operation via personal computer networks. Mr. Kashiwabara (Subaru International Inc.) explained network services between machine translation systems and commercial databases, and between machine translation systems and contracted translators. Prof. Ishizaki (Keio Univ.) expressed his opinion as to the difficulty of system construction, giving an instance in which his students could not clearly distinguish machine translation from automatic translation.

In the fourth session, Mr. Yoshida, a lawyer, spoke on "Machine translation and copyright and intellectual property."

The theme of the first session was "Future electronic communication." Prof.Nishigaki (Meiji Univ.) gave the keynote speech, saying that toward the 21st century, international communication will be supported by electronic media, and that among those media, machine translation may play a vital role.

In the second session, a panel discussion by senior researchers was held under the theme "Future machine translation technique." Discussion was extremely vivid, presenting various opinions as to challenges in analysis technique and utilization of the pattern recognition technique, both of which machine translation must fully overcome; how to implant to machine systems the know-how that human translators have; how to incorporate tools to bridge the gap between translators and machine translation; opinions on the need to establish automatic pre-editing techniques to enhance translation efficiency; how to input common sense to computers, the range of common sense and so on.

In the third session, a panel discussion was held on "Future translation." Scholars of both science and literature participated, discussing from the cultural viewpoint how machine translation will affect translation work in the future. Some expressed their concerns that technical needs may deteriorate language, and eventually traditional culture, whether machine translation systems can convey human feelings, and whether or not machines force people to write dull sentences without the expression of human emotions. The discussion was developed by the unique and lofty ideas that even though all new techniques evoke varied opposition and serious concern, most of which proves eventually to be beside the point, technological progress has always changed people and their environment, eventually contributing to the creation of a new culture.

On December 15 and 16, an exhibition was held on machine translation systems, related software, and related tools. At this exhibition, three organizations and eleven enterprises displayed their merchandise.

In the booth of the CICC, researchers demonstrated evaluation tests of their systems to reveal the outcome of the past six-year international joint researches. The researchers who participated in this demonstration came from various countries: Thailand, Malaysia, Indonesia and China, as well as Japan. This machine translation system, employing the interlingual approach, can simultaneously translate several languages. Its current performance allows translation in 20 ways among 5 countries. Using international network lines, the system can convey, for instance, the speeches of delegates at the ASEAN meetings, in our own languages, automatically interpreted by the machine. Though this epoch-making system is still in the R & D stage, its early commercialization is expected. Translation demand is rapidly growing, commensurate with the globalization trend seen in the transfer of manufacturing facilities overseas, the establishment of overseas research and development centres, and the growing international information exchange.

To attract visitors to this exhibition, the organizers sent direct mail to the research industry, export/import industry, manufacturers of electric and information appliances, office apparatus, automobiles, and machines, which have relatively high overseas manufacturing ratios. During the two days of the exhibition, the venue was crowded with 3,500 visitors. Since the majority of the visitors had sizeable translation demand in their business, they showed special interest in the displayed products, staying at the venue for a long time and asking questions eagerly. There were numerous people around the booths of several companies which had revealed new products immediately before the exhibition. The staff were exhausted giving explanations to hundreds of visitors, who seemed to have been impressed by the rapid progress of machine translation systems, and expressed their appreciation for the improving operability of products.

In conclusion, the Fair, attracting a large number of visitors, was successful even though there were several problems, including the short period for public information, the date of the fair falling in the busy year-end season, and poor access to the venue.

The Second General Assembly of IAMT was called to order by President Makoto Nagao at 4:40 p.m. and the provisional agenda was adopted.

President Nagao reported that since the creation of IAMT two years earlier at MT Summit III the membership of the three regions had grown to over 700 (615 individual and 86 corporate and institutional members). During this period the Association launched its newsletter, MT News International, under the editorship of John Hutchins, five issues of which had been published to date.

IAMT contributed financially to TMI-92 in Montreal and TMI-93 in Kyoto, and, in addition, it organized a workshop in collaboration with AMTA on the evaluation of MT, held in San Diego on November 1992.

The Secretary (Muriel Vasconcellos) reported that, pursuant to a decision of the 2nd meeting of the Council (Montreal, 25 June 1992), IAMT had been incorporated in Washington, D.C., as a non-profit association on 15 October 1992. The Bylaws, copies of which were circulated to the General Assembly, had been ratified by the Council at its 4th meeting on 21 July 1993. She explained that they provided for two new vice presidential positions in order to guarantee representation of all three regions on the Executive Committee. The two new positions are: Vice President for Programs, which is to be held by a representative of AAMT, and Vice President for Information, to be held by a representative of EAMT. Since the secretariat is in Washington, D.C., the positions of Secretary and Treasurer are filled by representatives of AMTA. The position of President-elect will rotate between the regions in tandem with that of President. Each region shall select its own representatives, including those to fill the respective offices, and the officers are confirmed by the General Assembly.

The financial report, also circulated to the Assembly, had been approved by the Council. As of 10 July there was a balance on hand of $17,627.37. A budget for 1994 is in preparation.

Speaking for AAMT, Makoto Nagao reported that their journal, started three years ago by the Japan Association for Machine Translation, was now being published in both English and Japanese and circulated throughout the Asia-Pacific region. The Association held an "MT Fair" in December 1992 and organized a workshop on MT evaluation in June 1993. Its committees on evaluation and controlled language each conducted a study and prepared a monograph in their area of concern. AAMT had 200 individual members and 54 corporate or institutional members as of July 1993.

Reporting for the American region, Muriel Vasconcellos indicated that AMTA has built up a mailing list of about 2,500 names and has sent out several mailings. This list was enhanced by a large number of inquiries received following publication of a suite of articles on MT by AMTA authors in Byte magazine in January 1993, which included a box on AMTA and IAMT. As of the June there were 330 individual members and 17 corporate or institutional members.

The Regional Editor, Joseph Pentheroudakis, now does the page makeup for MT News International and arranges to print the copies for EAMT as well as AMTA. The Association has just published the MT Yellow Book (advance copies were made available at Summit IV for inspection), which includes a membership list and also serves as a locator of individuals, institutions, and companies involved in MT -- mainly in the Americas. She hoped that it would grow to include the other regions as well.

AMTA organized a two-day workshop on MT evaluation (San Diego, 2-3 November 1992) and co-sponsored a seminar on translators and MT in cooperation with the American Translators Association (San Diego, 4 November). It also contributed US$2,000 to TMI-92 in Montreal (25-27 June 1992). In 1993 AMTA will give a workshop on "Text and MT," dealing mainly with the translatability of MT input, and it will co-sponsor a seminar with ATA on "Matching Up MT Users and MT Environments," to be held in Philadelphia on 5 and 6 October, respectively. AMTA is considering having its own conference in 1994.

The activities of EAMT in the last two years, reported by Margaret King, included organization of a workshop on "Translation and Translation Theory," held immediately after Coling '92 (Nantes, 23-28 July) and workshops on "Computer-integrated Translation and the User" (Saarbrücken, 29-30 October 1992) and "The Lexicon and MT" (Heidelberg, 26-28 April 1993). In addition, EAMT will be lending its patronage to the Bulgarian Summer School on Computational Linguistics (5-11 September 1993), which this year emphasizes machine translation. At each event the opportunity is taken to hold a "rolling meeting" of EAMT members to consider MT-related issues. The EAMT held its first general assembly in Utrecht on 22 April 1993. It is now planning for its own three-day conference in 1994, one day each to be devoted to users, commercial systems, and research. There are 80 individual members and 15 corporate or institutional members.

At this point the following new officers of IAMT were presented for confirmation by the General Assembly: President, Margaret King; President-elect, Muriel Vasconcellos; and Secretary, W. Scott Bennett. Roberta Merchant continues as Treasurer. Outgoing President Nagao then passed the gavel to President-elect Margaret King, who took the chair. She stated that the aims of IAMT would not change and that the Association will continue to support MT users, developers, and researchers, as well as cooperation between them. She expressed admiration for the insightful leadership provided by Makoto Nagao as founder and first president of IAMT and thanked him for his hard work and enthusiasm. Finally, she appealed to members to come forward with new ideas.

John Hutchins presented a proposed resolution urging the regional associations to provide the editor of MTNI with copies of all their publications and requesting the organizers of meetings convened by the regional associations to prepare a report on such events for MTNI and to deposit a copy of the proceedings, if any, to all of the foregoing immediately upon publication. The proposal was seconded and approved unanimously as Resolution GA2/1 (see text below).

A second resolution was then presented by Margaret King, who explained that a member of EAMT had been made aware that a patent exists which might be used to discourage diversity and free competition in research, development, and the sale of machine translation systems. There was concern that the claims of the patent in this case were so general that they could, if they prevailed, affect all research and development in machine translation and natural language processing. In light of this threat to the science of machine translation, the proposed resolution was drafted opposing the inappropriate use of patents and other legal instruments for this purpose. She emphasized that the general principle was of greater concern than the particular case. There being a second and no objections, the proposal was approved as Resolution GA2/2 (see text below).

It was announced that the next General Assembly will take place at the time of MT Summit V, which will be held in Luxembourg on 11-13 July 1995. It was also announced that TMI-95 will be held in Louvain, Belgium.

The General Assembly was adjourned at 5:15 p.m.

Respectfully submitted,

Muriel Vasconcellos, Secretary

PCMT: A New Passion that Changes EverythingTwo years ago, when we met in Washington at MT Summit III, it was obvious that MT was increasingly headed for the personal computer. Today the revolution is upon us. The advent of affordable software that can run on anyone's desktop ("PCMT"2) has totally challenged the received wisdom about MT usage. We must take a new look at the user profile, the purposes of MT, the products and the markets to which they are being directed, and the long-range future of the industry as a whole.

This report addresses the gap in our understanding of current MT usage by attempting an overview of all uses of MT based on the most concrete facts that could be found. It has considered only tried-and-true experiences and cumulative data reported directly by users. Information is particularly nebulous in the area of PCMT. Since there is no major up-front investment that needs to be justified, the user is less motivated to keep statistics. Nevertheless, some impressive facts are already a matter of record.

In the first place, there is now evidence that we are talking in rather large numbers of MT users. The June 1993 issue of WordPerfect Magazine reported the results of a mail-in poll in which readers voted for their favourite PCMT software. A total of 7,865 respondents took the trouble to send in their vote3. Presumably these people have road-tested at least one of the products and may in fact be using MT for practical purposes. The top three choices were Linguistic Products' PC-Translator, MicroTac Software's Translation Assistant, and Globalink's GTS (version unspecified). PC-Translator has doubled in sales each year since it first appeared on the market in 1985. The company periodically introduces improvements in its 12 language combinations and usually has new combinations in the pipeline; the developers have been heartened by the high percentage of registered users who request upgrades and new languages4. Globalink, which offers seven language combinations, went public in June 19935, and their prospectus states that approximately 13,000 units have been sold or placed with dealers since 1990. MicroTac, for its part, leads the market by a wide margin: in May 1993, all-time total sales of its four bidirectional packages reached a staggering 150,000 units6. The Translation Assistants are priced at under US$100 and, in some discount houses, as little as US$60.

In all, there are 10 companies selling PCMT in the United States. Together they translate in a total of 17 different directions, and a number of other systems and language combinations are under development7.

These products are being used in myriad ways. In the long run, translation varies as greatly as the texts that undergo it, the people who perform the process, and the consumers who require it. Each use is somewhat unique.

Even more impressive than the numbers is the fact that many users of the PCMT systems are happy campers. Their ranks include both translators and nontranslators, and it is among the latter that PCMT is cutting its widest swath. From unsolicited testimonials received by the vendors8, we learn that many people are enlisting these packages to prepare letters and memos in languages that are foreign to them. One user of this kind writes: "The PC-Translator is doing wonderfully, we are all satisfied." There seems to be a slight preference for enlisting them to produce translations of texts prepared by the user rather than to comprehend foreign texts, which are typically input by hand or by a pesky process of optical scanning.

Sometimes the users do not know the target language at all. Installed on a laptop, PCMT has served as a practical companion in social situations where language is a barrier, and it has helped travellers to get around in foreign countries. An American in Paris reports that he used French Assistant to explain to the caretaker of his building that the hot water was off. Another MicroTac user, an American priest filling in at the last minute on a cruise ship, relied on this same software to prepare his sermon in French. Most touching, perhaps, is the user of Italian Assistant who wrote: "Through your product I have been able to correspond with my relatives in Italy since my trip in 1990, when I was introduced to them for the first time. My dad passed away two years ago and my mom is too old to write."

Finding out who really uses machine translation is no simple task. A few years ago it was possible, with help from the vendors, to identify at least those customers who were using MT on a significant scale. Today, however, with PCMT selling in large volume and with vendors busy attending to a broader customer base, the picture is far less clear. For the purpose of this report, a strategy was devised for locating a representative sample of MT users, who were to be presented with the following list of questions9.

As the first step, a list was drawn up of known users for whom fax addresses were available10. There were 33 of these (two of whom could not be reached). Next, a list was prepared of individuals who had checked the "User" box on their application form when they joined the Association for Machine Translation in the Americas. This exercise garnered 15 more names. It was known that some of these people were prospective users still investigating the feasibility of MT, so a letter was prepared addressing each one as a "user or potential user of MT" and asking them to report on their plans for using it if they did not already have it installed. The third step was to contact the vendors directly to ask them for the names and fax numbers of "some of [their] principal clients," sharing with them the list of questions that would be asked. Because of multiple sites and contacts, a total of 32 inquiries were sent out to vendors of 23 systems or families of systems. Six additional known vendors could not be reached. Of the 32 who were contacted, 14 replied and provided information about their users. These replies yielded 22 additional users, all of whom were approached. In the end, fax letters went out to 70 users or potential users.

Thus a fairly wide net was cast. Even so, the coverage was far from complete. The information obtained without the assistance of the vendors was not collected in any systematic way. In the vendor cycle, not all of them could be contacted, many who were contacted did not respond, and those who did reply did not necessarily give a full list of their customers. Response from the PCMT vendors, who account for far and away the largest volume of purchased (if not operating) units, was particularly low: only three replied, and only one of these directed us to specific users. Given such large gaps in the coverage, the answers received can only be considered representative of the vendors and users who were reached and had the time and inclination to share their experience. They do not speak for MT as a whole.

Another piece of missing information, which would be difficult for any survey to ferret out, is the user sites that have fallen by the wayside -- and why. This information is important for a full understanding of MT usage. However, it is hard to come by. One usually learns it by chance. Recently, for example, in a translation service that had shown positive results with MT, there was a breakdown in the hardware on which the system depends, and management was unwilling to buy the same equipment again. Elsewhere, an MT operation was eliminated because of a company-wide "reorganization" -- perhaps an indirect victim of the foundering economy. At yet other site the operation was dependent on an individual, and when that person left there is no structure to keep it going. There may also be MT failures in the true sense that the text was not a good match for the system or not enough time and money were being saved to justify the investment. For a variety of reasons, most of this information, which would be very illuminating, is kept dark.

Despite its limitations, however, the material collected for the present report is significant in many ways. Its very abundance gives it a certain authority. A total of 40 responses were received: 33 from actual MT users, one from a user with a commitment to start in July 1993, and six from companies that were in the process of investigating MT -- two were running pilot tests, one had put out an invitation to bid, and three were undertaking feasibility studies. CompuServe was included in this last group, with plans to offer on-line service from English to French starting in the fall of 1993 and other combinations later. In addition, answers to the same questions, gathered within the last nine months, were available from five other users and were included in the study. The analysis that follows covers the 33 responses from actual users and the five additional ones for which information was available, for a total of 38 user sites -- or 54% of those that had been contacted. In all, they represent 15 different systems: Atlas, DP/Translator, Duet Qt, Général TAO, Hicats, Shalt, JICST, Logos, MicroCat, Metal, PC-Translator, Pivot, NHK, Spanam/Engspan, and Systran (including Systran Express, the on-line service that anyone with a PC, a modem, and a checkbook can tap into). There were 16 users in the Americas, 11 from Europe, and 11 from Japan11. This may be the largest body of data ever collected at a single time on the use of MT. While it does not permit hard statistics, some very interesting conclusions can be drawn about how MT stands up to the test of translating texts in the real world.

We can learn a lot about how much MT is being used from the volume of translation being produced and the percentage that this represents of the total workload. The survey yielded some illuminating information in this regard.

Thirty of the 38 users gave information on the volume of translation they produce using MT, the percentage that this represents of their total workload, or both (see table). Many of them had statistics at their fingertips, and it is easy to see that high-volume users, new or pilot users who are keeping a close watch on the effect of MT implementation, and users closely involved with development of the system itself would have reason to keep careful records.

In the category of large-volume users, the figures show that there are some truly industrial-strength MT operations. The European Commission is near the top of the list with 30 million words a year of general translation, for which they use Systran in a total of 13 different language combinations and serve from 400 to 500 end-consumers. These numbers take on special importance because the translations are in a wide range of subject areas and discourse genres. They amount to 15% of the total translation workload of the CEC. Interestingly, only 30% of Systran's output is postedited by professional translators; the rest is delivered "raw" and is used for information purposes only.

Two other very large users are Bull in France, which expects to be using Systran at an annual rate of 45 million words by the end of 1993, and Lexi-tech, which uses Logos for about 25 million words a year. Both these companies are using MT for technical documentation. Météo generates about 17 million words of weather bulletins each year for Environment Canada. The U.S. Air Force/FASTC, in its venerable information-gathering operation, annually translates between 10 and 12.5 million words with Systran. Intergraph relies on their own DP/Translator for about 10 million words. Xerox produces about 9 million words with Systran. Nikkei Printing uses NEC's Pivot and Sharp's Duet Qt for about 4.5 million words. And so on.

Added together, the volume of MT produced by these users -- slightly over half the known users approached in the survey -- comes to about 180 million words a year. MT use in the world undoubtedly exceeds 380 million. These figures translate, respectively, to some 720,000 pages of known use and about 1.2 million pages of estimated use. It is impossible to guess what percentage this represents of total translation in the world, however, since experts recognize that there is really no way to quantify the latter.

It can be seen from the table that the bulk of the work is translations of technical manuals and other material related to localization. The volume produced by the 15 users that provided this information comes to approximately 108 million, or 60% of the total volume reported. Of the entire sample of 38 users, 23, or 61% of them, fall into this category.

Another important parameter to look at is the proportion of the total translation load being handled by MT. The figures on percentage of the overall workload run the gamut. For the 24 who answered this question, the proportions ranged from 5% to 100% and formed an almost perfect bell-shaped curve. The average was 46% and the mean was 50%. Lexi-tech, one of the biggest users, relies on MT for 100% of its workload, and Nikkei Printing, also with a very large volume, uses it for 95%. Environment Canada uses Météo for 85% of all weather bulletins. The U.S. Air Force, which has had an MT installation since 1970, reports 80%. Some respondents seemed unclear on whether they should include languages not offered by their MT system in calculating the percentage, so it should be kept in mind that the figures may not always be referring to the same thing.

The high-percentage users are often high-volume users as well. The 10 respondents in the table that reported at least 50% usage and also reported figures for volume together produce 118.5 million words, or 66% of the total. As might be expected, many of these high-percentage users do technical manuals and other types of localization work: of the 12 users at 50% or higher, seven do this kind of work, and, as noted already, they account for a large share of the total volume. This should be concrete proof of the long-held assumption that there is a comfortable fit between technical manuals/localization and the automation of translation. In other words, MT does seem to work well for these applications.

Another interesting fact that emerges from the table is that most of the respondents have started using MT in the last five years. Of those in the table, 22, or 73%, began to use MT in 1988 or later. For the entire responding population of 38 users, the figure is 82%. In other words, MT use has recently taken quite a spurt. And of course, with the advent of PCMT, this trend can be expected to accelerate sharply.

Closely related to how much of the job MT is doing is the amount of human effort involved in the form of pre- and postediting. (None of the respondents had interactive workstations.)

Pre-editing was cited as a major issue only by the Raytheon user translating software written in the Ada programming language and by two of the three respondents who work with Japanese-English. While one said "pre-editing is basically division of long sentences and we usually don't spend that much time on it," another said that it is contracted out, and the third J-E user reported that pre-editing takes about 40% of total translation time. The other 30 respondents, all working with a Western language as the source, regarded pre-editing to be negligible or at least easily justifiable; 24 said they did little or none -- although interpretations of the term appeared to vary. Five said that they run an automatic spell-checker on the input; five mentioned conversion software or adaptation of the format; one referred to the need to proofread OCR output; and two indicated that pre-editing mainly involved blocking material that does not require translation. One user spends time "cutting overly long sentences into shorter ones, fixing up punctuation, etc." A user in France has tried "end-user sensitization to 'clear writing,' with no evidence of success," while another one gives informal guidance on how to write for MT. Two said that their documents are written originally in a controlled language, and one reported that the input is edited to conform to the company's controlled language at a rate of 3,750 words a day -- which also happens to be their rate of postediting. Estimates of percentage of total translation time were given at 5%-10%, 10% (two respondents), and 20%-30%. One user included terminology research and dictionary maintenance under this heading, for approximately 60% of total MT time.

Postediting, on the other hand, generally accounted for a large share of production time and cost, and it was also the subject of a lot of comments when it came to discussing the disadvantages of MT. A number of respondents said that postediting is done directly on a word processor, one of them preferring commercial off-the-shelf word processing to the product developed by the MT vendor. Many pointed out that the requirement for postediting varies depending on the quality of the output, and that some language combinations give better results than others (e.g., "German-English [is better than] English-German"). The J-E user that did not report very much pre-editing said: "We rewrite the sentences after MT rather than [pre-]editing. Usually it takes a lot of time and manual power." An E-J user, in turn, felt that the main disadvantage of MT was the difficulty of postediting to achieve "acceptable" expressions in Japanese. The system developed by NHK has a user interface that presents several choices of output for the user to pick from, and the user can specify how many choices the system offers. Général TAO, when it gets overly challenged, leaves segments in the source language untranslated, and these passages must then be done by hand.

Several used the word "extensive" in characterizing their postediting. One respondent indicated that 75% or more of the text is touched during the postediting phase, although this proportion might vary depending on the translator, the product, or the language. On the other hand, Météo requires intervention in less than 5% of the output for a translation of good quality.

A number of respondents said that they review the entire text or do a "100% full postedit." This percentage should not be confused with the percentage of text that is actually corrected. A few require very high quality (e.g., for subtitles of television broadcasts, insurance contracts, publications), while some of them settle for an in-between product -- from "clean[ing] up the language, adjust[ing] the format, and review[ing] for technical accuracy," to "editing for accuracy but not for style unless requested," and, finally, to "quick and dirty." The U.S. Air Force has special software developed by Systran, called Editsys, which automatically picks out problem areas and leaves the rest of the text, usually about 80%, to be delivered without review. Some users have two levels of editing -- "information only" (or "for understanding only") versus a full translation. One respondent indicated that they offer both raw and reviewed translation but that only reviewed translation is "marketed" and accounts for 95% of their usage.

In terms of share of the total process, the user who said that terminology and dictionary work accounted for 60% of total MT time went on to attribute 20% of this time to postediting. Another said postediting represented 25% of the time. A third one said the proportion was 30%.

In the discussion of the disadvantages of MT, postediting kept coming up as a sore point. The respondents complained of the high cost, the time it takes, and the lack of user-friendly functions for posteditors.

Underlying the whole question of production is the purpose for which the translation is required. It is important to assess whether or not MT contributes to achieving the user's long-term service objective. As we saw earlier, a large percentage of the respondents are engaged in producing localization materials, often including immense volumes of technical manuals and, in at least three cases, software as well. Their responses definitely show that MT helps to move the process along so that they can get their products to market sooner. Perhaps the contribution of MT is not so much in producing a structurally correct text as it is in keeping terminology consistent and in eliminating the need to reintroduce graphics and format codes in target-language documents. Fisher-Rosemount, a high-volume user and manufacturer of machinery for industrial fluids, said that "translation would be barely feasible for this volume at this speed without it. By retaining formatting attributes, tables, and illustrations, [MT] saves enormous work and money." This user's bottom line: "Cost savings of nearly 50%." The sentiment is echoed by the owner of a commercial translation service that relies heavily on MT, who says: MT is "indispensable for high-volume jobs."

MT is being used for other purposes as well, of course. The sharing of scientific and technical information, especially from on-line databases, is a growing area. The U.S. Air Force (FASTC) has now expanded its MT operation to 17 subject fields and five languages and is starting to translate titles and short abstracts from on-line sources. Since 1990 the Japan Information Center for Science and Technology has been translating the mammoth JICST database into English with its own MT system and reports a 40% reduction in cost. Also in Japan, the Bio Information Center provides up-to-date data in medical and biotechnical fields (medical reports, database abstracts) with the help of MT, while the Pan American Health Organization in Washington, D.C., uses MT for publication-quality texts in similar technical fields as well as others. And Henkel KgaA in Düsseldorf uses MT to translate chemical abstracts, reports, and data sheets.

The Canadian agency DTSB-Statistics recently started using MT to translate technical papers and repetitive texts such as consumer price indexes for dissemination purposes. And of course Météo's weather bulletins for Environment Canada are a well-known example of MT use; translation is now bidirectional, and turnaround time for a given bulletin is less than 6 minutes.

One of the most novel uses of MT was reported at MT Summit III -- namely, NHK's television captioning project. Their MT system is now bundled in a prototype subtitle production system that also includes integrated modules for videotape monitoring on-screen, manual superimpose-timing input, and preview of the completed program. It was unveiled in June 1993.

From the users' responses, it would appear that the issue is not whether MT can meet these needs, but rather how efficiently it can do so. In some cases it has proved to be highly functional, while in others the jury is still out.

The users were forthcoming about both the advantages and disadvantages of MT. Several listed a number of advantages and no disadvantages. The advantages cited most often were consistency of terminology, faster turnaround (to speed up market penetration), and increased productivity. One user commented that the terminology factor directly contributed to increased productivity ("at least 1.8 times better than human-only translation"). It was noted that certain types of errors are avoided -- e.g., skipped passages, numbers incorrectly copied. Filters on publishing systems which eliminate the need to re-enter format codes were very popular. Also cited was MT's ability to quickly process high volumes of material in many languages simultaneously.

Other specific comments were: "When the requester requires FYI translation, we can speed up the edit and still make the translation intelligible." "Less need for top quality translator." "We expect a capacity increase as soon as we have gained more experience with the system" (a user who started at beginning of 1993). "It gets better" (a new user).

And from the operator's perspective: "Lightens the translator's load." "No cumbersome typing." "It also maintains the original format created in WordPerfect." "Beneficial for us because the kind of text we translate is very dry and very repetitive." "I really enjoy working with DP/Translator; it requires a lot of work at the beginning with the creation of custom dictionaries but helps maintain consistency. The machine generates a draft translation, performing the most boring part of the task, so that I can concentrate on perfecting the output."

The respondents were equally expressive about the disadvantages. Many of them complained about the poor quality of the output and the cumbersome process of postediting. They want better interfaces and postediting tools.

From the manager's viewpoint, several respondents cited the high cost of source text preparation and postediting. Two said it was difficult to find texts suitable for MT. One complained that it involves a lot of training, and two of them noted that it's costly for smaller projects. Another remarked that system development is too slow and that there should be more user support. In one case it was noted that inclusion of MT in the production scheme had complicated the workflow. With regard to one particular system, the respondent mentioned that enhancements are very costly because of its size. Two of them regretted that hardcopy input documents were not scannable; "efficiency from the use of MT is largely lost in the time required to manually key in a text." A user of the old Weidner MicroCat workstation reported that the equipment is wearing out and the alternatives seem too expensive. Also cited were the high cost of purchase and maintenance; complicated handling; "an un-ergonomic user interface"; lack of acceptance by internal translators. A new user said: "No improvement in speed so far."

Other comments were: "It somewhat inhibits creativity"; "loss of idiomacy and style"; "resulting text is a little stilted and awkward"; "excessive adherence to MT output changes expression"; "translation system not sufficiently flexible about using one term in one context but another in a different context."

The following response gave real food for thought: "Up to now we have not really been able to make use of the advantages (consistency of terminology, speed, etc.). One of the advantages mentioned by salesmen, etc., [namely] that MT relieves translators of boring, repetitive tasks, is not relevant in my opinion as there are other repetitive tasks instead: text conversion, parameter editing, deformatting, writing Pattern Matcher instructions, reformatting, etc. I enjoy working with MT because it is an interesting tool and you learn a lot, but whether it really beats manual translation remains to be seen."

By and large the users have a positive outlook, a desire to streamline their MT operations, and a keen interest in introducing improvements and trying out new applications. One current user plans to take on a new application, joining the ranks of those who use MT to screen translation requests. Another site is plugging MT into databases on CD-ROM.

They are asking for, and working on, new and better tools. They want to be on high-end workstations instead of mainframes. They want software to test texts ahead of time to see if they lend themselves to MT. Much in demand is a good system for repetitions processing, whereby previously translated texts are matched against the ongoing translation process and displayed for possible pasting in. They need better converters for moving freely between different publishing environments. They are also working on terminology managers. Integration of the workstation seems to be the key. The Canadian Government is putting the finishing touches on a "fully equipped zero-wait-time multimedia workstation on a LAN server" with access to terminology banks, multi-task word-processing packages, automated terminology searching, text analysis, and other specialized software.

They are also asking for, and working on, more language combinations, more domains, and better strategies for controlling the quality of input texts. At least two of them are seriously looking into interlingual MT, and the Unión Fenosa in Spain, working with Carnegie Mellon's Kant system, is dreaming the impossible dream and turning it into reality: MT with no postediting!

1. Updated version of invited lecture presented at MT Summit IV (Kobe, Japan, 19-22 July 1993). Published with permission.

2. "PCMT" is understood here to refer to PC-based MT products that do full-sentence batch translation.

3. From a larger "Reader's Choice" questionnaire, this number of people cast votes specifically for a PCMT package (source: Shannon Harmon, WordPerfect Corporation).

4. Source: Ralph Dessau, Linguistic Products.

5. GLNK U on the National Capitalization Market. Globalink regretted not being able to provide more information for the current report but was under a routine temporary period of silence.

6. Source: Michael Tacelosky, President, MicroTac Software (figure does not include upgrades.)

7. Source: "Report on PC-based MT products", American Translators Association, December 1992, compiled by L.Chris Miller.

8. Copies of the original testimonials provided by Linguistic Products and MicroTac Software.

9. Questions based on a model developed by Joann Ryan for research presented at the seminar "Machine Translation for Translators" (San Diego, 4 November 1992), sponsored jointly by the American Translators Association and the Association for Machine Translation in the Americas.

10. The entry criterion for the study was that the user could be reached by fax.

11. The list of users, together with the type of text they are translating, was published in the proceedings of MT Summit IV. To this list should be added late responses received from the Commission of the European Communities (general and technical translation, including 70% information-only), Inter Group (technical manuals), and JAPO (patent titles and abstracts), which are included in the totals cited in the present version of the paper.

…………………….Estimated……….Percentage

User….Year………no.of words……..of total…………Type of

#…..of startup……per yearb………..volume……………text

……………………(thousands)

1……..1970…………11,250………….80………..Scientific and technical articles

2……..1977…………17,000………….85………..Weather bulletins

3……..1978………….9,000…………..50………..Dissemination

4……..1980………….2,500…………..67………..General and technical

5……..1981…………30,000………….15………..Low-level in-house documents

6……..1982……………………………10………..Technical manuals

7……..1986…………10-100………...100………..Service publications

8……..1987…………………………….20…….....Technical manuals

9……..1988…………25,000…………100………..Technical manuals

10……1988…………10,000………………..…….Software, hardware documentation

11……1988………….4,500………..…95………..Technical manuals

12……1988………….1,600………………..……..Technical manuals

13……1988………………………..….10……..…Customer documentation

14……1989………….2,500-3,000…..40…………Technical manuals

15……1989………….44-60………………………Subtitles for news in English

16……1989…………750-1,000………5…………Internal technical documentation

17……1990…………2,500…………..50…………Insurance and pension contracts

18……1990…………3,445c………….-e………...Titles + abstracts, JICST database

19……1990…………2,000………….. 25………..On-line, hardcopy documentation

20……1990…………..480………………………..Technical manuals

21……1990…………..350……………20………..Technical manuals

22……1991…………1,600…………...67………..Technical manuals

23……1991…………...375…………...30………...Manuals, technical reports

24……1991……………………………80…………Chemical abstracts, data sheets

25……1992…………45,000………….50…………Technical manuals

26……1992…………..1,500……………………….Software, user manuals

27……1992……………345d…………..9…………Titles of unexamined patents

28……1992…………….25…………….5…………Scientific publications

29……1993…………3,300……………30…………Technical manuals, price indexes

30……1993…………………………….90…………Computer manuals

a. Eight of the 38 respondents did not provide the information being compared in this table.

b. Figures for numbers of pages were multiplied by 250 to permit comparison. Those for less than a year were annualized.

c. 85,000 titles plus 15,000 abstracts; average length of title estimated at 10 English words and average length of abstract (200 Japanese characters with upper limit of 300) estimated at 150 English words.

d. About 23,000 titles per year at an average of 15 English words each.

e. 90% of the abstract are written in English by bilingual abstractors; of the remaining 10%, all (100%) are translated by MT.

The analysis of answers to the questionnaire revealed the following facts. The large-scale user who uses the MT system systematically translates documents of a few hundred to a thousand pages per one MT terminal a month. Their target is rather limited, e.g. manuals, technical documents, patents, etc. so translation efficiency gets better.

On the other hand, small-scale users, such as personal users, translate only a few pages a month. Many of them use MT for raw translation of manuals and rough reading of technical documents.

Time spent on pre-editing ranges from a few minutes to fifteen minutes per page (A4 letter size), while postediting takes longer - 10 to 30 minutes. Most pre-editing is done for Japanese input. For example, long sentences are paraphrased as short ones, and omitted elements in sentences are supplied to make translation easier. The main tasks of postediting are the replacement of translated words and the refinement of the output... It seems that many successful users have expertise in editing and dictionary management, although these circumstances depend on the nature of their system. Some users have developed original support tools for MT systems...

Generally, the people working with MT systems are classified as pre-editors, post-editors, rewriters and operators. For small-scale users, one person fills all roles. Even with large-scale users, roles are not clearly separated. It was also found that English abilities are not always equal to those of English specialists (1st class of English examination in Japan). It means that ability to use an MT system is considered more important than personal ability in English. Almost all users train MT workers in their own company or office.

Users of Japanese-English systems are particularly insistent on improvements in translation quality. However, large-scale users with experience of actual use want easier operability and compatibility with other MT systems and software. This is because their target is to raise the efficiency of the MT system or of the total translation operation from an understanding of the translation capability of the present system.

The number of responses to the questionnaire was 31. The major fields are information, communication, electronics, machinery and aviation. The sizes of the Japanese-English and English-Japanese dictionaries are nearly the same. Many dictionaries have about 2,000-6,000 words. In the case of manuals, user dictionaries are often made for specific types of machines or airplanes. Most entries are for nouns. Parts of speech and semantic markers are included, but not the meaning of a word or an example sentence. However, many of the dictionaries stored in spreadsheets, RDB or text files have information about synonyms, explanation of word meanings and example sentences. Many answers indicated that priority of translation equivalents is made according to frequency of use. But human judgement is necessary for final selection...

The following were mentioned as problems in use: entry of verbs and idioms; priority between user dictionaries and basic dictionaries; timing for revision; handling compound names (in chemistry); unifying items and formats of dictionaries between manufacturers. Few users comprehend correctly statistical data about numbers of words, costs of dictionary compilation, and so on...

There is no way to decrease translation costs dramatically other than by using MT. The greatest motivation for introducing MT is to decrease translation costs. The other motivation is pure interest in MT as such. This survey, however, shows that many users do not use MT systems effectively in practice. The effective users of MT systems are companies that have to translate large volumes of documents: document translation companies, translation divisions in large companies, and so on.

The characteristics of effective users are:

1) they manage translation processes and personnel efficiently

2) they have enough experience of MT to stand the learning curve

3) they have examined the capability of MT, and they use MT systems within their capabilities.

4) they have trained their translation engineer within their organization.

In addition, they have reviewed their MT use as follows:

1) they construct a working environment in which resources such as dictionaries are shared among users

2) they have integrated computerized document processing

3) they have examined various methods of pre-editing

We believe it is important to share such experience among MT users in order to increasingly popularize MT systems.

The result of this survey is planned to be published soon. Please contact the editor [Hirosato Nomura, address on front page of MTNI] about its availability in English. The report will be combined with that conducted by another working group "MT System Evaluation Committee".

Developed and enhanced over six years of R&D activity along with actual operation for the "Japan News Retrieval" service [translation service for news and articles of money market and securities over the Nikkei Telecom network], ARGO J/E, a Japanese-English machine translation software package, promises to give a new key to Japan's MT market.

ARGO J/E, which was formally unveiled by CSK (F16 Shinjuku Sumitomo Bldg., Nishi-Shinjuku 2-6-1, Shinjuku-ku, Tokyo, Japan) on December 11, 1992, uses the "modification detection system" that is a CSK-specific translation method. [This is a] syntactic analysis method by which the words in a sentence are analyzed to detect the "modification relationship" between them (or to build a syntactic-semantic structure or tree of them). Parts of speech, semantic primitives, cases, and attributes are used for the establishment of a "modify-modified" structure. It permits analysis of sentences written in the Japanese language which otherwise could not be achieved because of their relatively free word order. The current version of ARGO J/E, features that CSK staff having experience in both development and internal trial operation will, on behalf of each user, take charge of building a special dictionary containing terms and idioms of his own. It will prove a powerful translation support, claims CSK, by clearing most MT users of their protesting complaint, "I have it, but always fail to manage it."

ARGO J/E is open to users. Its basic concept is that, as the user gets accustomed to his system, it should be growing to meet his requirements. The multiwindow-based graphical user interface (GUI) and the user terminology building tool enable the user to add new nouns, verbs and idioms, and special semantics between words to his dictionary database.

ARGO J/E is available with optional dictionaries of technical terminology including geography, biography, and information processing. The user can use more than one of those dictionaries and user dictionaries simultaneously so that he can translate sentences describing information of different fields at a time.

The price of an ARGO J/E package, 5.6 million yen, consists of 2.0 for the software and 3.6 for the organization of a user dictionary by CSK staff.

CSK Marketing is planning to supply both Japanese and overseas markets with 2,000 sets of ARGO J/E in five years over Nihon Sun Microsystems' sales network.

ARGO J/E is designed to work in the following hardware and software environments:

Hardware: SUN SPARCstation2 (main memory, 32 MB; swap area, 100 MB; hard disk, 300 MB minimum)

Software: SUN OS 4.1.1 (OS), OpenWindows 2.0.1-up (window environment), JLE 1.1-up (Japanese language environment)

For further information call +81-3-3342-3047 CSK Machine Translation Development Dept. (c/o Mr. Nagano)

As the internationalization of Japan has progressed, rapid and accurate translation of large volumes of English-language information has become increasingly important for those at the forefront of the business world.

LogoVista E to J is an English-to-Japanese translation support system based on the theories of Professor Susumu Kuno, an internationally recognized linguistics scholar. LogoVista E to J features "high accuracy in syntactic parsing of the input English sentences," "appropriate semantic processing," and "sophisticated synthesis of Japanese sentences." It has been well received as a system that produces "usable translations."

The system's dictionary component, one of the keys to translation accuracy, consists of a Main Dictionary, used in conjunction with high-precision grammar rules for parsing; Technical Dictionaries containing specialized terms used in business and a variety of scientific and technical fields; and a User Dictionary, which individual users can build to fit their needs.

LogoVista E to J has been designed as an "open system" that is not restricted to one platform. It can be used on a variety of workstations and personal computers.

The quality of the translation output is the most important factor in the overall quality of a translation support system. However, a lack of standard criteria impedes the evaluation of the output of translation systems. We will illustrate the linguistic characteristics of LogoVista E to J by using real examples of its translation output. [Omitted - see original article.]

The workstation and personal computer versions of LogoVista E to J have been designed to have identical features. You can use the system on a platform appropriate for the volume of your translation needs.

a) Ease of Use. LogoVista E to J is operated by using a mouse to select commands from pull-down menus. The simplicity of the design allows people with little knowledge of computers to use the system.

b) Overview of the Translation Process Text Input. Users can directly type in text to be translated, or an OCR system can be used when large volumes of text need to be input.

Users can control the following translation options:

* the maximum number of Alternate Translations (from 2 to 20)

* the formality of the Japanese output (da, dearu, or desu)

* output of proper names (Japanese only, English only, or Japanese with English in parentheses)

* translation of definite articles (sono)

* input text type (dialogue or standard)

* text layout (Side-by-side or Formatted)

* use of Technical Dictionaries (several dictionaries can be used at the same time)

The input text is analyzed and translated. LogoVista E to J's progress in translating each sentence is displayed on the screen, as is the system's progress through the entire document being translated.

When the processing of each sentence is completed, its translation is displayed on the screen.

The user can produce a more appropriate translation by using the Alternate Translation function.

The Alternate Translation function allows the user to choose an alternative translation of an entire sentence or of a selected word or phrase.

LogoVista E to J "learns" statistically from the choices made by the user while postediting with the Alternate Translation feature. In this way, the system is customized as it is used.

The user can also edit the translation output directly without using the Alternate Translation function.

The user can further customize LogoVista E to J by building a User Dictionary (including nouns, verbs, adjectives, and adverbs) created through simple, automated procedures.

For inquiries, please contact: LogoVista Corporation, 10-24, Shiomi 2-chome, Koto-ku, Tokyo 135 Japan. Tel: +81-3-5690-8531; Fax: +81-3-5690-1290

PENSÉE-GV is a Japanese-English/English-Japanese translation software developed by Oki Electric Industry Co.Ltd. Main advantage of PENSÉE-GV is to translate documents which may include figures and tables, keeping the original layout. PENSÉE-GV is called from GlobalView, an integrated document processing software developed by Fuji Xerox Corp.

Previous machine translation systems dealt with only texts. So, the user must have extracted texts from the documents beforehand, and returned the texts into the original places after the translation. This sequence of work included time-consuming cut-and-paste and re-formatting. These were routine activities and hoped to be eliminated.

Oki's policy for the machine translation system is to remove as many routine activities as possible, which will lead to inefficiency. Based on the policy, Oki developed a JE/EJ translation support software, PENSÉE-GV, which translates documents into target language keeping the original layout.

*** Easy to use. The user just clicks target source document icon(s) and selects EJ or JE translation operation in the pull-down menu, the PENSÉE automatically translates the source document(s) into the target language and shows the translated document icon(s). Just selecting the target word(s), sentence(s) and folder(s), the operator also easily choose the various modes from word-by-word translation to folder (collection of documents) translation.

*** Preservation of format. The translation retains the layout and font information of the original pages even with figures and lists. Reformatting and re-inputting become unnecessary.

*** High-quality translation. We achieve high-quality translation based on our 'Deep Case Grammar', which is a refinement of Case Grammar by our AI technology.

This enables PENSÉE to capture deep structure of long and complicated sentences.

*** High-speed translation. PENSÉE-GV translates about 15,000 words per hour. It takes only tens of seconds for the 1 page translation.

*** Options of translation. In JE translation, PENSÉE-GV has the alternative translation modes for non-subject sentences, which frequently appear in manuals: (1) to translate them in the passive voice (standard mode); (2) to translate them in the imperative (MEIREI mode). In EJ translation, PENSÉE-GV has the alternative translation modes for the Japanese ending postpositional particle: (1) to use 'DEARU' way (standard mode); to use 'DESU' or 'MASU' way (KEITAI mode).

Translation languages:

Japanese/English, English/Japanese

Desk Top Publishing software:

GlobalView

Translation engine: PENSÉE

Hardware: OKITAC-S

Memory required: 16 MB or more (recommended 32 MB or more)

Hard disk required: 400 MB or more (including OS and GlobalView)

OS etc.: SUN OS 4.1.1 or later

+ JLE 1.1.1 or later

+ OS/FX optional function

+ GlobalView

System dictionary:

JE approximately 90,000 words

EJ approximately 60,000 words

Special dictionary (optional):

JE approximately 50,000 words (16 fields)

EJ approximately 50,000 words (16 fields)

(information processing, automobile, metal, mathematics, etc.)

User dictionary: unlimited within the capacity of the disk

Translation speed: approx. 15,000 words/hour (differ by machine)

Translation option:

JE: standard/MEIREI

EJ: standard/KEITAI

OCR (optional):

Japanese: SPARC Reader

English: ScanWorX

Readable document forms: plain text, JStar, ICHITARO, OASYS, Lotus 123, etc.

(Conversion software is required except for plain text and JStar. Some figures can not be read.)

<Contact> For more detailed information, please inquire at: Oki Electric Industry Co.Ltd. Tel: +81-3-3454-2111

Users of Intergraph Corporation's DP/Translator MT system met for the first time in Huntsville, Alabama, on 4 and 5 May of this year. Four sessions dealing with MT were held during the annual meeting of the Intergraph Graphics Users Group, which is the worldwide organization for users of Intergraph products.

On the first day, the DP/Translator product development group gave presentations on the new features of version 1.4 just released, and on future product direction. The role of user feedback in development was emphasized, along with the means by which it can be provided through the customer hotline, a database tracking system for problem reports, and the Special Interest Group organization.

A panel of users convened on the second day to discuss their implementations of a number of DP/Translator language pairs in different translation environments. Users from corporate translation departments as well as a translation service bureau participated. The type of translation being done varied from information-only translation for internal use to the translation of user documentation for publication.

One area of much discussion concerned typical problems in MT output. Guidelines for writers and translators in avoiding these problems were a topic of particular interest. A report on a joint experiment by technical writers and translators indicated that adhering to writing guidelines that allowed for better machine translation output also improved the quality of the original document. The results were seen as significant for organizations that can exercise control over the writing process and important in achieving cooperation between writers and translators. Other users had also observed that the quality of the computer translation varied with documents written by different writers or in different styles.

Productivity was another important topic of discussion. One user provided some recent statistics on translator productivity when employing DP/Translator to translate scientific articles on request for information only. After 3 months of concentrated work in the lexicon, productivity gains were beginning to be realized, and client acceptance of lightly edited computer output was favorable. With no control over the input for translation, building large dictionaries was seen as the most important factor in improving translation quality and increasing productivity in the post-editing phase. Comparisons of translator productivity in terms of words per day of finished translation seemed difficult, however, since in each organization translators had different duties.

Also on the second day was a presentation on using DP/Translator in a translation service bureau. Mentioned as key requirements in using MT successfully were the development large lexicons, having the ability to accept documents from clients using many different publishing applications, and having translators experienced in MT post-editing. Business growth and increased hiring of translators underscore the success of MT in the service bureau environment when properly managed.

For further information about DP/Translator please contact Gary Thornton, Electronic Publishing Division, Intergraph Corporation, Mail Stop LR23A2, Huntsville, Alabama 35894-0001, USA.

Linguistic Products has announced that it will host a seminar in Florida on the correct use of PC-TRANSLATOR. The seminar will be held in Boca Raton from 11-13 October, immediately following the annual meeting of the American Translators Association (ATA), to allow overseas ATA attendees to participate. The seminar will be conducted with big-screen demonstrations and will include the following topics:

- PC-Translator and the Competition

- PC-Translator Architecture

- Language Availability

- Dictionary Composition

- Dictionary Modification and Enhancement

- Database Adaptation to PC-Translator Format

- Word Processor and DTP Format Preservation

- Conventional and Special Applications

- PCT and Large Scale Translation Projects

- Conventional Text Acquisition and Preparation

- Source Text Acquisition by OCR

The cost is US$1,500, payable in advance. In return participants will receive a free copy of PC-TRANSLATOR (a US$895 value) in the language pair and direction of their choice. The language pairs and directions available are English into Spanish, French, Italian, Swedish, Danish, Norwegian, Dutch and Portuguese; and Spanish, French, Italian, Swedish, Danish, Norwegian, and German into English.

Hotel rooms will not exceed $100 per night. Lunch will be provided, but evenings will be free for sight-seeing.

The company points out that the seminar will provide an ideal opportunity for actual or potential users to become familiar with PC-TRANSLATOR, learn how to use it correctly and take home a free copy for further testing and use. For reservations, please contact Evelyn Smith, telephone: (713) 298-2565, fax: (713) 298-1911.

All reservations must be accompanied by payment of the $1,500 fee, and must specify which version of PC-TRANSLATOR is selected. The fee may be charged to a MASTERCARD or VISA account.

An internati