Foreign language characters should be in training data or not

[nijin...@gmail.com]

Hello.. Currently I have a lot of news domain data to train in tesseract for non-english language. But what I'd like to know is that in my news data, there are many english words and should I remove or add these english words  to get the better accuracy.  ( What I learned is that in tesseract, there are already english trained model and to predict the english words, we can use -l eng+(my_language)).