Machine Translation Today and Tomorrow

John Hutchins

The field of machine translation (MT) was the pioneer research area in computational linguistics during the 1950s and 1960s. When it began, the assumed goal was the automatic translation of all kinds of documents at a quality equalling that of the best human translators. It became apparent very soon that this goal was impossible in the foreseeable future. Human revision of MT output was essential if the results were to be published in any form. At the same time, however, it was found that for many purposes the crude (unedited) MT output could be useful to those who wanted to get a general idea of the content of a text in an unknown language as quickly as possible. <…>

Many large organizations have large volumes of technical and administrative

documentation that have to be translated into many languages. For many years, MT with human assistance has been a cost-effective option for multinational corporations and other multilingual bodies (e.g. the European Union). MT systems produce rough translations which are then revised (post-edited) by translators. But post-editing to an acceptable quality can be expensive, and many organizations reduce costs and improve MT output by the use of ‘controlled’ languages, i.e. by reducing (or even eliminating) lexical ambiguity and simplifying complex sentence

structures – which may itself enhance the comprehensibility of the original texts.

At the same time as organizations have made effective use of MT systems, human translators have been greatly assisted by computer-based translation support tools, e.g. for terminology management, for creating in-house dictionaries and glossaries, for indexing and concordances, for post-editing facilities, and above all (since the end of the 1980s) for storing and searching databases of previously translated texts (‘translation memories’). Most commonly these tools are combined in translator workstations – which often incorporate full MT systems as well. <…>

However, translation does not have to be always of publishable quality. Speed and accessibility may be more important. From the beginnings of MT, unrevised translations from MT systems have been found useful for low-circulation technical reports, administrative memoranda, intelligence activities, personal correspondence, indeed whenever a document is to be read by just one or two people interested only in the essential message and unconcerned about stylistic quality or even exact terminology. The range of options has expanded significantly since the early 1990s, with the increasing use and rapid development of personal computers and the Internet.

More powerful PCs have encouraged the marketing of translation software for the general public. As general-purpose systems, the quality is inevitably poor. Input texts often contain high proportions of non-technical, colloquial language of the kind which MT systems have always found most problematic. Quality is usually not good enough for professional translators (although some use the output for drafts), but it is found adequate for individual ‘occasional’ users, e.g. for gists of foreign texts in their own language, for communicating with others in unknown languages, and for translating Web pages and electronic mail

It is the coming of online translation on the Internet, however, that has brought the most significant changes, with potentially far-reaching implications for the future. Exposure to information in many languages has created a rapidly growing demand, and this may well be MT’s niche market: the real-time online supply of rough translations to support personal communication and information needs. <…>

From: Gerd Willée, Bernhard Schröder, Hans-Christian Schmitz (eds.) Computerlinguistik: was geht,

was kommt? Computational linguistics: achievements and perspectives. Festschrift für Winfried Lenders

(Sankt Augustin: Gardez! Verlag, 2002), p. 159-162

Интервью для абзацно-фразового перевода

Интервью №1

Не нашли, что искали? Воспользуйтесь поиском: