The en-ko model cannot obtain accurate results
The en-ko model cannot obtain accurate results。
I used ct2-transformers-converter to convert en-ko to en-ko-ct, but sometimes the translation results were quite confusing. However, I was able to obtain the correct results using other models of Helsinki NLP. Look forward to your reply.
Some test results:
Fail
Tokenized input: ▁hello .
Translated output: hello 안녕하세요.
Tokenized input: ▁High - performance ▁Deep ▁Learning ▁models ▁for ▁Text 2 S pe ech ▁tasks .
Translated output: 논문 제목은 ‘High-Performance Deep Learning Models for Text2Speech Tasks’이다.
success
Tokenized input: ▁Please ▁use ▁our ▁dedicated ▁channels ▁for ▁questions ▁and ▁discussion .
Translated output: 질문 및 토론을 위해 전용 채널을 사용하십시오.
Tokenized input: ▁Help ▁is ▁much ▁more ▁valuable ▁if ▁it ' s ▁shared ▁publicly ▁so ▁that ▁more ▁people ▁can ▁benefit ▁from ▁it .
Translated output: 더 많은 사람들이 혜택을 누릴 수 있도록 공개적으로 공유하면 도움이 훨씬 더 가치가 있습니다.
Please continue conversation in original issue report;
https://github.com/FricoRico/Versta.Models/issues/2