Рубрика:
57
medical translation and for patent documents, but the Internet would be the obvious
home for such specialized MT systems. They will probably not be free (as many
online translation services are now), but users will surely accept charges for better
quality.
2. On the other hand, the ready availability of low-quality MT from Internet services
and from commercial stand-alone software could well increase the demand for
higher-quality human translation, particularly from those with no previous experience
of translation. Some suppliers of online translation are already providing add-on
human translation services (e.g. post-editing or full translation). Currently they are
used mainly by organizations without their own translation services, but wider use
may be expected in the future.
3. For Internet users, a desirable development would be integration with other
language applications. What users are seeking is information, in whatever language it
may have been written or stored – translation is just a means to that end. Many would
welcome the seamless integration of translation with summarization, database
mining, document retrieval, information extraction, etc. There is already research on
cross-lingual information retrieval, multilingual summarization, multilingual text
generation from databases, and so forth, and before many years there may well be
systems available on the market and on the Internet.
Perhaps most desired of all are systems capable of translating spoken language – not
just for trained speakers in restricted domains (e.g. hotel booking and business
negotiations, as in current research projects in Japan, USA and Germany), but for all
speakers in all situations.
4. Users will want reliable and accurate results – poor quality text can re-read and
puzzled over, spoken output must be understood immediately. Automatic speech
translation of open-ended communication will not come in the near future, and may
never be possible, but in the medium term we may expect to have systems capable of
translating the utterances of most speakers in well defined situations (banks, theatres,
airports, rail stations, etc.)
5. At a more mundane level, the language coverage of all MT systems needs to be
wider. Currently, most concentrate on the major commercial languages (English,
French, German, Spanish, Japanese, Chinese, Korean); and many languages spoken
by large populations in developing countries have been ignored by software
companies and even by research groups.
Equally, there is a real need for systems to deal with the kind of colloquial (often ill
formed and
badly spelled) language found in emails and chatrooms.
6. The traditional rule-based approaches found in current systems are probably not
equal to these tasks on their own. In MT research, there is much interest in exploring
new techniques in neural networks, parallel processing, and particularly in corpus-
based approaches: statistical text analysis (alignment, etc.), example-based machine
translation, hybrid systems combining traditional linguistic rules and statistical
methods, and so forth. Above all, the crucial problem of lexicon acquisition (always a
bottleneck for MT) is receiving major attention by many research groups, in
particular by exploiting the large lexical and text resources now available (e.g. from
Страницы
- « первая
- ‹ предыдущая
- …
- 55
- 56
- 57
- 58
- 59
- …
- следующая ›
- последняя »
