LSTM encoder-decorder model for Sinhala spell checking.

Hello!

I really need your help on this one. I am a student and i am doing my final year research and I'm doing an automated spell checker for the sinhala language. currently i have trained a LSTM encoder-decorder model with 100,000 words and have gotten 94% accuracy. but when it comes to the predictions it's always wrong.

It would be great if i can get an insight on this.

thank you in advance.

deeplearning encoder lstm spell