Limits of automatic translation

machine translationEven the „smartest” software can’t replace the work of a professional translator. The reasons for machine translations often offering unsatisfactory results are quite diverse. Ambiguity in translation appears because a single word can have several meanings, depending on the context.

The process of eradicating ambiguity can be based on statistical approach or on a deeper understanding of the meaning of the word. The first option leaves a lot of room for translation errors, and the second is quite hard to carry out in practice. Therefore, machine translation would never replace the work of a written or oral translator. Language can be used for expressing emotion, demonstrating superiority, jokes, expression of welcome in relation to a stranger, etc. Besides, conversations don’t necessarily have to be sincere. It is impossible to take all that into account when doing a machine translation.

Culture of people from different countries and regions varies considerably, and therefore, there is no universal structure which could be used on every occasion for translation. Mathematical theory of groups states that a computer would never be able to know the language to the fullest extent. This is related to the fact that no single subgroup can be bigger than the whole group of which it’s a part. Therefore, language and translation aren’t included in mathematically solved issues.

It’s irrelevant how fast the computer performs, because the real translation problems are related to the linguistic aspect and not to the technical one. It’s hard to describe a language adequately with the help of modules, perfectly executing an enormous amount of connections between the words of the original and target languages, reflecting them as a larger scheme. Linguists have been brainstorming these issues for many years, but still haven’t found a satisfactory way to solve this problem.

There is no way to qualitatively assess the accuracy of a translation, but accuracy is what is required from a machine translation. That’s why the evaluations are often riddled with errors. You could see a message about a translation being written in perfect English, but that’s subjective and the assessment would never be accurate. Quite often, a machine translation contains a single error per line, yet assessed as though it has a 90 per cent accuracy.

Any text that has any kind of significance requires the work of a professional translator. It’s wrong to think that a machine translation is cheaper than ordering from a translator, because in reality, the support of productivity of automatic translation service might require high costs, and translation would require additional editing. Machine translation can merely serve as additional assistance, if you need to translate a large amount of texts in a short amount of time. However, you would not be able to use such systems as a single translation instrument.