MT News

This article orginally appeared in the July-Aug 1992 issue of Language Industry Monitor

MT System for Caterpillar now under development

Carnegie Group, the Pittsburg software house which developed Reuter’s renown online story classification system, tis, and Caterpillar, the world’s largest manufacturer of earth-moving equipment, announced in May a multimillion dollar, five-year project to develop a fully automatic translation system. The system, called Automated Translation System (ats), will be used to translate the approximately two million pages of manuals which Caterpillar distributes with its heavy machinery into eleven, as yet unspecified target languages. In the announcement, Caterpillar explained that it initiated the project to facilitate its global operations and to comply with the cec’s recent directive stipulating that all operations manuals for equipment and hardware shipped into the ec countries after 1992 must be written in the respective languages of those countries.
The Carnegie Group says that ats will not require postediting because it is based on a form of Knowledge-based Machine Translation (KBMT), a method developed at Carnegie- Mellon’s Center for Machine Translation (also participating in the project). KBMT relies on a restricted domain of subject matter together with a limited vocabulary and syntax. The ats system will be implemented using Caterpillar Technical English (cte), a restricted subset of English covering the domain of Caterpillar service information.
“Although these restrictions do not hamper the writing of documentation, they allow us to remove the ambiguity normally inherent in English through a process of machine-encoding the knowledge domain,” said Phil Hayes, director of Natural Language Systems at Carnegie Group. “By eliminating ambiguity, we eliminate the need for post-editing.” As yet unspecified software will be provided to assist technical writers in producing cte texts.

Carnegie Group, Five PPG Place, Pittsburg, PA 15222, USA Tel +1 412 642 6900, Fax +1 412 642 6906

Japanese to demo Asian interlingua MT system this fall

The Machine Language Laboratory Centre of the International Cooperation of Computing (CICC) in Tokyo, Japan, has announced that it intends to begin demonstrating its MT system in November. In a brief news item posted in the Newsbytes section of Compuserve, it was indicated that the system would be experi mentally linked via a packet-switching network among a number of Asian countries.
The CICC has partners in China, Thailand, Malaysia, and Indonesia; it is an initiative of the Japanese Ministry of International Trade and Industry (MITI) and the Overseas Development Agency. Backers of the Machine Language Laboratory include Fujitsu, Hitachi, Matsushita, Mitsubishi, nec, ntt, Oki, Ricoh, Sharp, and Toshiba. The MT project has cost three billion yen (US$23 million) so far and is scheduled to be completed in two years. John Hutchins, editor of Machine Translation International News, the publication of the International Translation Association, comments, “All previous interlingua systems have been loosely Indo-European based. This is an interesting project because it is an attempt to develop an intermediary language/ representation for non-European languages. Quite a lot has been written about the project in the research literature,” he adds, “but there is little factual information about performance. It will be interesting to hear what comes out of these demonstrations.”

Pc-based translators: they keep on coming

Faithful readers of Language Industry Monitor will know that we take great interest in low-end, PC-based translation systems, not to deliver definitive verdicts on their linguistic capabilities but to learn more about the aims of the developers and to hear the claims they make about their packages.

Sun Data Systems, a small software house near Delft, recently announced Vertaler (Translator), a translation package for pcs which it deems suitable for the automatic translation of invoices, price lists, bids, and other “standard” texts. Vertaler is a Clipper application, meaning it is written in dBase, the popular dos database programming language, and compiled as a standalone program.
Rather than offering Vertaler as a ready-to-run package, Sun Data Systems positions it as a framework for a specific application, thus skirting such irksome details as dictionary size, performance, and language pairs. Although Sun Data Systems says that the program can be used independently, the company itself incorporates Vertaler into the financial adminis tration packages which it develops for customers. Vertaler evolved out of an application developed for a flower exporter in Aalsmeer which wanted to be able to print invoices, labels, and instructions in Japanese.
Peter Verweij, director of Sun Data Systems, recently explained in an interview with the Dutch Software Magazine that he was interested in practical solutions. “People,” he declared, “are waiting for working products.” Asked how his approach differed from that of bso/Language Technology (formerly dlt), the natural language group of bso, a major Dutch systems house, he replied: “They seem primarily interested in what is not possible, while I concentrate on what is.” Correspondingly, he dispenses with such matters as syntax, morphology, and semantics and simply builds up bilingual dictionaries out of words and phrases. He declared, not surprisingly, that “the quality of the translation is determined by the quality of a given dictionary. The program works with brute force.”

Sun Data Systems, Vlielandsweg 24A, 2624 KC Pijnacker, The Netherlands; Tel +31 1736 94346

[ return | top | home ]