A time-honoured pursuit
Machine translation (MT) is the application of computers to the task of translating texts from one natural language to another. One of the very earliest pursuits in computer science, MT has proved to be an elusive goal, but today a reasonable number of systems are available which produce output which, if not perfect, is of sufficient quality to be useful in a number of specific domains.
The archaic-sounding term “machine translation” is — for historical reasons — nowadays primarily associated with standalone translation programs, whereas the translation software now available runs the gamut from simple dictionary lookup programs used as word-processor add-ons to sophisticated batch-translation systems and fast translation on the web. The EAMT maintains an inclusive point of view.
In recent years, translation software packages which are designed primarily as an aid for the human translator in the production of translations, rather than a standalone “black box”, have become popular within professional translation organizations. These programs, referred to as computer-aided translation (CAT), use a variety of linguistic tools to improve the productivity of translators, particularly when translating highly repetitive texts, such as technical documentation.
Another viable application for MT is content scanning, that is, using a translation system simply to obtain a rough draft so as to be able to get the general gist of a text. MT is widely used in the European Commission for this purpose, for example, and it is widely used on the Internet.
Software manuals, not literature
Human language is enormously complex, and translation between languages is certainly not just a matter of replacing the words. Texts where the style and nuance of the language is especially important are a challenge even for professional translators. So we should not be surprised that the inherent limitations of the current generation of translation programs mean that they are less able to translate some kinds of texts. They do much better on straightforward text like technical manuals or informative web pages, and much worse on literary text. They also only work on well-written texts with correct spelling and conventional grammar.
Is MT any good?
Users of MT programs should be aware of these limitations when they judge the translation. In particular, round-trip translation (where a text is translated into a foreign language and back into your own language) is a very bad way to judge whether the translation is any good. It may be good fun, but it won’t tell you much about the translation quality: a bad round-trip does not mean the outward journey was bad; and a good round-trip might mask a meaningless word-for-word translation.For a more in-depth discussion on some of these topics, we recommend you visit the following links:
- “Machine Translation” page on Wikipedia: here
- Statistical Machine Translation: here
- Neural Machine Translation: here
The Internet has proven to be a huge stimulus for MT, with hundreds of millions of pages of text and an increasingly global — and linguistically diverse — public. What role will MT play in bridging languages barriers in cyberspace? Stay tuned…