This repo contains the source code and other details for a neural machine translation based on attention using pytorch. This model translates Korean into English.

• Weekly Report : check here :)
  From February 2020, the weekly report can be found there.

• BLEU(Bilingual Evaluation Understudy) score

• Translation Sentence

차를 마시러 공원에 가던 차 안에서 나는 그녀에게 차였다.

> I was dumped by her in a car on the way to the park to drink tea .  

사과의 의미로 사과를 먹으며 사과했다.

> I apologize while eating an apple for the meaning of an apology .

내가 그린 기린 그림은 긴 기린 그림이냐, 그냥 그린 기린 그림이냐?

> Is the giraffe I drew a long giraffe picture or just a giraffe picture ?

• Preprocess

  • Delete the sentence with the length of 149(Korean) or more and 387(English) or more based on space.
  • Delete the sentence containing some special characters.

• Configuration

!python preprocess.py

The source text file(src) and target text file(tgt) are tokenized through Mecab+SentencePiece.

!python train.py

If you want to continue training the model, add --train_from (model path)/model.pt later.

!python translate.py -model data/model/model.pt -src data/src-test.txt -tgt data/tgt-test.txt -replace_unk -verbose -gpu 0

!perl tools/multi-bleu.perl data/tgt-test.txt < data/pred.txt

!pyhton gui.py

You have to change from "data/src-test.txt" to "data/demo/KoreanTokenInput.txt" of translate_opts > --src in opts.py and "data/pred.txt" to "data/demo/EnglishTokenOutput.txt" of translate_opts > --output.

https://github.com/OpenNMT/OpenNMT-py