language detection and translation

[lchunleo]

1. am referring to language detection and language translation for short text such as tweets. does opennmt work well?

2. How much words do i need to achieve a 90% and above accuracy and low perplexity? i have been doing for 20K words as input but manage to get only 50+% accuracy using the default settings of learning rate, epoch and layers? What is the generally expected accuracy for a default settings should i be expecting?
   The result from my work is that the translated words are more readable but the meaning is not correct which is better than my initial input of 4K words which is totally gibberish and no structure at all.