This is where I made my talk. |
It's pretty big! |
About my presentation, afterwards, I was told "Bold move!" by one of the participants. I'll take it as a good thing...
There was catering.
I could have had my breakfast here! |
I learned many things at this summit, but below are the main things I learned.
In neural machine translation of De/Ja and Ja/De, the translation might be done in the order of: De>En>Ja or Ja>En>De. Therefore, for Ja/De translation, some English is mixed in the translated German. This is something I heard from one person, there is no verification. He mentioned that it is possible that the language is translated to English first.
I also learned that domain-specific machine translation (domain-specific MT) works well. For example, in the MT developed by WIPO, sentences like "I am a boy" is not included in the corpus of the machine translation. This is because no such sentence will appear in patents. Therefore, when doing patent translation, MT designated for patent might be better. It seems that there will be a notification from WIPO around next week (I didn't hear this from any representative, so I will not publicize any contents)
0 件のコメント:
コメントを投稿