current tech can translate decently a sentence, but a sentence has no context so if there's a double meaning, it becomes a matter of probability to obtain a cor

current tech can translate decently a sentence, but a sentence has no context so if there's a double meaning, it becomes a matter of probability to obtain a correct translation. i.e. "I never said she stole my money" there are 7 possible meaning conveyed by stress alone, which will need to be translated differently. also current tech has limits when translating whole pages. paragraphs introduce context and meaning that's later referenced. sentence by sentence translation is bound to lose all of that info along the way. I don't think current neural network used in mapping translations have enough complexity to keep context as they go. they might get there eventually alone as the computer scale or with the help of preprocessors.