The thesis entitled translation and programs used was divided into two main chapters: the theoretical part and the practical one. Into the first one, composed of five chapters, the concept of machine translation was defined and was integrated as being part of the big domain of Natural Language Processing. Further was presented the evolution of machine translation and types of MT. More precise how evolved the MT, from a program of translation (Systran) that translated 60 sentences from Russian to English to the multitude of programs that translate millions of documents daily. After these two steps was needed to elaborate the architecture of each system: rule-based, statistical, data-driven and eventually the hybrid. I have presented the specific characteristics for the architectures and illustrated them with at least one program, who they work. The fourth chapter was current state of machine translation. The last one, Machine Translation programs, treats at least one example for each architecture with their caractheristics, languages and technology used.
The practical section was also divided into two main chapters: a part with the work methodology and examples from the legal-administrative texts where were identified the main mistakes of translation and the second one with analysis and interpretation. This
In all translated documents the main problem was the one of equivalence, the second one was that some sentences didn’t had sense. The words were decontextualised and the programs translated word by word, from there the non-senses. The worse programs of translation were Babylon and SDL, the best with the fewest mistakes was Systran and for the wham that had Romanian into the combination of languages was Google Translate, currently the most widely used translation software. None of the programs of translation not recommended at translating large amounts of text that’s because in the case of the smaller sentences the quality of the translation can be controlled and checked. Google Translate is able to give synonyms for the term but only for small phrases, from there it can be selected the correct equivalent. In my work I put an emphasis on the layout that is because it is very important to insert a text into a program of traslation and at the end to not having to correct the stucture of a text, especially in the case of lega-administrative texts that have a fixed structure.
These programs would help a human translator to make an opinion over the text, to understand it, but even so there are situation in which the translation is incomprehensible. What I have noticed over my word is that sentences must be short and complete, not very complex, ambiguous words must be avoided, the register must be as simple as possible, the sentences from the source text must be logical and very simple otherwise the translation will not be that good.
The aim of this thesis was to show where the machine fails to render a good translation and to identify which are the main mistakes that this makes, more precise the behavior of the programs regarding some type of texts.
Indeed the machine makes plenty of mistakes, but if you know on which type of text could work, would save a lot of time, money and energy. Over time with the actual technology and information, the quality of the translation could be improved.
The domain of the machine translation is very vast and there are still a lot of things that need to be clarified and descovered. I believe that this work could be useful to the translators, if they know how to use the machine translation, because it would make their work easier, save time and energy.’
...(download the rest of the essay above)