Neural Machine Translation

Abstract

Imagine that you are a translator. You are asked to translate from German to English and you come across the word Sitzpinkler. Its literal meaning is someone who pees sitting down, but its intended meaning is wimp. The implication is that a man who sits down to pee is not a real man.But there is more going on here. This word was popularized on a comedy show that coined several other terms in this fashion. One is Warmduscher, someone who takes a warm shower, or even Frauenversteher, someone who understands women. In fact, a whole fad emerged to come up with new terms like this. All these terms are used as insults, but not as real serious insults. They are used very much in jest, a slight mocking.These terms are also firmly a reflection of the current zeitgeist, when the expectations of what it means to be a man are changing. Using such terms is a light-hearted commentary on this change. It is not really unmanly to sit down to pee, although it is something that women do and hence a man who wants to be a traditional "real" man loses some of his identity this way. As you can see, there is a lot going on here.So, what is a translator going to do? Probably use wimp and move on. This example demonstrates that translation is basically impossible. The meaning of words in a language are tied to their prior use in a specific culture. Four score and seven years is not just any way to say 87 years. And I have a dream implies much more than just announcing a vision of the future. Words carry not only an explicit meaning but also an undercurrent of implications that often does not have any equivalent in another language and another culture.

Pregled metoda obrade prirodnih jezika i strojnog prevođenja

“…Kako problem prevođenja zahtijeva više razina apstrakcije (riječi, klauze, složene klauze, POS, morfologija riječi, sintaksa, semantika, diskurs, veza između klauza) nad prirodnim jezikom, ali i puno stručnog znanja i zaključivanja, to MT čini jednim od težih problema kod umjetne inteligencije (Koehn, 2020). Također, budući da se jedna rečenica može prevesti na više način i procjena kvalitete prevođenja predstavlja dodatni izazov (Majcunić, Matetić, Brkić Bakarić,2019).…”

Traducción humana vs. traducción automática: análisis contrastivo e implicaciones para la aplicación de la traducción automática en traducción jurídica

“…Hasta hace poco, la tae era el sistema más utilizado en la industria de la traducción, pero las mejoras en procesamiento computacional y la consiguiente adopción de redes neuronales han permitido adoptar con éxito la tan (Bentivogli et al, 2016). Según Koehn (2017), los modelos de redes neuronales ya se habían propuesto de manera muy similar anteriormente (Castaño et al, 1997;Forcada y Ñeco, 1997), aunque ninguno se había podido entrenar con cantidades de texto como las que están disponibles actualmente y, además, la potencia de procesamiento de las máquinas y los ordenadores de entonces no era la de ahora. Por tanto, solo los avances en procesamiento actuales han permitido "resucitar" los antiguos modelos neuronales, que se han abierto hueco en el mundo de la ta.…”

Recent Progress, Emerging Techniques, and Future Research Prospects of Bangla Machine Translation: A Systematic Review

“…In the output layer, the decoder outcome is interpreted as a probability distribution. A softmax activation function is used to ensure proper probability distribution [20]. NMT is a data-driven approach where a NN model is trained with a parallel corpus of SL and TL.…”