Machine Translation Capabilities And Limitations

As of today, machine translation (MT) can still be defined as the holy grail of computational linguistics. While undeniable that this discipline has seen a significant progress in the last decade (especially with the introduction of statistical methods), translation between morphologically-rich languages remains an extremely challenging task. Statistical MT tries to generate translations through statistical procedures based on huge bilingual text corpora.

Since automatic translation translates texts without the aid of professional translators or native speakers who can discern the nuances and influence of context, it is important to remember that the accuracy and appropriateness of the results is not guaranteed. Human language is full of ambiguities, exceptions, plays on words, subtle expressions, mistakes, and logical associations that computers cannot handle (at least in 2011).

Machine translation always returns a translation result for any given source text, sometimes one that is useful, very often one that does not fit the context, and in some cases even a direct copy of the source text (whenever the translation cannot be resolved).

All this said, there are good things to say about machine translation. In certain scenarios, it can be sufficient to get an imprecise translation that reveals what the text is about without everything being translated correctly. And, at the same time, there are circumstances where it can be more important to get the result without delay than to get a good translation. One easy way to understand what these systems can offer is by having them translate into your native language.

You can see some machine translation examples from Spanish and Japanese into English at:

Alternatively, you can also visit the following popular sites to run your own tests.

The advantages of machine translation compared to the work of human translators could be summarized as: cost, speed and mathematical consistency.