dosubot[bot]
commented
4 months ago 🤖
Thank you for providing the detailed error log. It seems like the issue is related to the text processing part of the ingestion process. The yarn run ingest command uses a script that includes a text splitting step, which might be where the punctuation conversion is happening.
The script uses RecursiveCharacterTextSplitter to split the text into chunks. If this splitter is not handling Japanese punctuation correctly, it could be converting it to Korean punctuation during the splitting process.
However, without access to the specific implementation of RecursiveCharacterTextSplitter and the rest of the codebase, it's hard to pinpoint the exact cause of the issue. It would be helpful if you could provide more information about the RecursiveCharacterTextSplitter and how it's implemented, or any other part of the code that might be handling text processing.
In the meantime, you might want to check the configuration of the text splitter and see if there's a setting that's causing this issue. If the splitter is using a library or service for text processing, you might also want to check the documentation for that to see if there's any known issues or settings that could cause this.
If you're unable to find a solution, it might be worth considering using a different method for text splitting that can handle Japanese punctuation correctly. There are many libraries and services available that can handle text processing for different languages, so you might be able to find one that works better for your needs.
I hope this helps! If you have any more questions or need further clarification, please don't hesitate to ask.
yarn run ingest result:
Original document: 日本人は，世界でいちばん寝不足 睡眠不足が慢性化すると，睡眠負債に 近年，「睡眠負債」という言葉が注目を集めるようになっています。睡眠負債とは，どのような状態のことをいうのでしょうか。 毎日8時間の睡眠を必要とする人が，ある日，何らかの理由で6時間しか眠らなかったとします。この2時間の睡眠不足は，それだけでは睡眠負債とはいいません。しかし，睡眠不足が何日も重なり，数日から数週間の単位で睡眠不足が慢性化した状態になると，睡眠負債とよばれるようになります。
error details: 。， was converted to Korean, like 셆