From peripherals that can translate instantly from one language to another to able to keep a conversation in a language we do not handle, to applications for cameras which translate text from signs, posters, etc.; and Google with its neural translator which constantly feeds itself, Automatic or (MT) has advanced by leaps and bounds in recent years. It has become a translator’s tool and one more factor in the client-company relationship.

However, despite all the breakthroughs, machine translation is still lacking.

The most obvious scenario is its complete inefficiency in translating literary texts. Google may understand the syntagmatic relationships between the sentences, but (for now) it cannot understand the subtle nuances between synonyms, how a metaphor or metonymy, humor, etc., are structured. Surely we can translate a 100-page employee handbook but we will not be able to translate the first line of Leaves of Grass.

Furthermore, because it is not able to interpret text, machine translation engines simply do not work with texts that badly written. If the writing of a text is poor, ambiguous or has spelling mistakes, the MT cannot do too much about it. This is why an editing step must be taken into account prior to using the engine, to ensure that the result is not unreadable. One example (from Spanish into English) is that MT engines cannot infer when the deictic elements are not in place: “si” or “sí” (if or yes), “que” or “qué” (that or what), “como” or “cómo” (like or how), etc.

It’s not advisable either to use machine translation when the text to be translated is too technical. Surely, we can use GTM to translate cell phone’s user manual, but it would be impossible to translate the manual for a hydraulic crane, since for the latter we must have extensive knowledge on the subject to translate it. The same can be said about many of the legal texts we come across: the legal systems are different in each country, so the translation of the same word will also be different depending on the country. “Court,” for example, is not “Corte” but “Tribunal” in Argentina (except with the Supreme Court), while in Spain “Corte” is used as a standard translation.

This blog has no intention of criticizing the use of MT, but simply to warn about some of its limitations. At Trusted Translations, we encourage its use whenever possible, since every day we witness a constant improvement.

Tagged with: