11.05.2013 Views

últimas corrientes teóricas en los estudios de traducción - Gredos ...

últimas corrientes teóricas en los estudios de traducción - Gredos ...

últimas corrientes teóricas en los estudios de traducción - Gredos ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

PABLO CANCELO–EVALUATION OF MACHINE TRANSLATION SYSTEMS<br />

EVALUATION OF MACHINE TRANSLATION SYSTEMS<br />

1. GENERAL CONSIDERATIONS<br />

92<br />

PABLO CANCELO<br />

Universida<strong>de</strong> da Coruña<br />

Machine translation products are curr<strong>en</strong>tly receiving a consi<strong>de</strong>rable amount of<br />

hype. At one <strong>en</strong>d of the scale are mass media reports on one product after another that use<br />

the latest magical technique to produce nearly perfect translations. Unfortunately, these<br />

reports are usually based on the manufacturers’ promotional press releases, and make it<br />

into print without any attempt at verification or review. At the other <strong>en</strong>d of the spectrum<br />

are the <strong>de</strong>tractors of machine translation, those who assert that all translation programs are<br />

useless, and the whole effort is a meaningless waste of time. In the middle, however, is<br />

another group of people – of which this researcher is one – who hold that machine<br />

translation technology, while not perfect, has progressed in rec<strong>en</strong>t years and some of the<br />

systems can r<strong>en</strong><strong>de</strong>r a source language docum<strong>en</strong>t into an un<strong>de</strong>rstandable, though rough,<br />

target language translation.<br />

In or<strong>de</strong>r to a<strong>de</strong>quately answer the question of whether machine translations of<br />

docum<strong>en</strong>ts is practical, it is necessary not only to rigorously test the curr<strong>en</strong>t commercial<br />

systems but also to look a little bit more c<strong>los</strong>ely at the translation process in g<strong>en</strong>eral.<br />

Translation is a highly creative process. For each semantic i<strong>de</strong>a in one language,<br />

there are several possibilities for expressing the same i<strong>de</strong>a in another language. It is the job<br />

of the translator to find the best match possible, without sacrificing the overall flow of the<br />

finished docum<strong>en</strong>t. Giv<strong>en</strong> the high <strong>de</strong>gree of creativity and complexity involved, it is naï ve<br />

to view the translation process as a “black box” mechanism whereby a docum<strong>en</strong>t in the<br />

source language goes in one <strong>en</strong>d and the translation in the target language comes out of the<br />

other, no matter how poor the input might be. Unfortunately there is no “black box”<br />

system that can remove all the difficulties inher<strong>en</strong>t in language translation.<br />

Ev<strong>en</strong> if machine translation systems can never duplicate human translations, we<br />

may won<strong>de</strong>r whether they can at least g<strong>en</strong>erate output that is un<strong>de</strong>rstandable and useful for<br />

in-house memos and the like. The answer is a guar<strong>de</strong>d “maybe” assuming that users<br />

proceed with caution. The danger with machine translation is not that the output may be<br />

incompreh<strong>en</strong>sible, but rather that it may be ambiguous or ev<strong>en</strong> erroneous, leading the<br />

rea<strong>de</strong>r to misun<strong>de</strong>rstand the int<strong>en</strong>t of the original writer. That said, though, wh<strong>en</strong> used<br />

correctly, and for appropriate purposes, MT systems can be extremely useful.<br />

Any translation tool might be evaluated according to the economic b<strong>en</strong>efits or<br />

fitness of a particular system as part of a process in an organization (Sparck Jones, K. &<br />

Gailliers: 1996). This evaluation is typically performed by pot<strong>en</strong>tial users and/or purchasers<br />

of a system (individuals, companies or ag<strong>en</strong>cies)<br />

Researchers and <strong>de</strong>velopers might also focus on criteria such as accuracy,<br />

intelligibility and the linguistic constructions the system can successfully handle. This

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!