Machine translation is the act of translating a whole text with one or several softwares, without the help of any human being.
An automatic translation software analyzes the text in the original language (the one that has to be translated) and creates automatically a text in an other language (this is the translated text). In order to do so, it uses precise rules to transfer the grammatical structure (words from the dictionnary, linguistic rules) or is based on translation memories.
Here are the links of two famous automatic translation tools :
- Google Translate : http://translate.google.fr/. When in English we have men should clean the house, it says in French what can be translated by « women should ». This sexist proposition can be explained by the frequency of the word « women » used with the verb « to clean » in the memories of Google : http://translate.google.com/#auto/fr/Men%20are%20men%20and%20men%20should%20clean%20the%20house.
- Systran : http://www.systranet.com/fr/fr/translate. Its main rival is Reverso http://www.reverso.net/text_translation.aspx?lang=FR, which is as bad, if not worse. Here are some comparisons of translations more or less good by Google and Systran : http://www.cahiersdugepe.fr/index1367.php#tocto2n14.
- The rapidity of the translation process (thanks to big volumes of memory).
- The possibility to integrate this system to a Computer Assisted Translation softare (CAT) for repetitive texts with short sentences.
- The machine doesn’t « get tired », it uses mechanically the parameters which were set.
- The low cost of the translation (since the work is « easy » for the machine and you only have to buy the software once).
- No more need to find translators (for a basic translation ) who will propose different prices, services and array of expertise.
- The lack of reliability (eternal debate between the translation of the word or of its meaning, the machine cannot yet translate a meaning or determine the right meaning among several ones).
- The machine is limited to a unique equivalent for each word or cannot choose between several equivalents for the same original word, it doesn’t think of choosing the most used word in such context or such region.
- The machine translates too literally.
- It doesn’t know about the style, syntax, contact, register, punctuation, etc.
- Most of spelling softwares are not completely efficient (conjugation, prepositions with or without accents, grammar, etc.).
- The machine cannot distinguish what has to be translated or not.
- The machine is unable to spot mistakes in the original document before translating.
Automatic translation still has to improve, and the translations have not yet reached the quality one can expect from a person. The first step is to determine to which kind of text should the system be adapted, and then check the pre-translation. « There are today a certain number of softwares that provide a translation, if not perfect, of a quality good enough to be used in some specific areas, generally for technical information. » says the website of the EAMT (European Association for Machine Translation). In any case, no matter the future improvements in this area, there is no doubt that better the translation memory of the software (the largest and most accurate possible), better the translation.